Split (3)

train · 4.35k rows

article stringlengths 1.98k 169k	summary stringlengths 1.01k 4.15k	section_headings listlengths 2 38	keywords listlengths 0 12	year stringclasses 11 values	title stringlengths 30 189
During translation elongation , the ribosome ratchets along its mRNA template , incorporating each new amino acid and translocating from one codon to the next . The elongation cycle requires dramatic structural rearrangements of the ribosome . We show here that deep sequencing of ribosome-protected mRNA fragments reveals not only the position of each ribosome but also , unexpectedly , its particular stage of the elongation cycle . Sequencing reveals two distinct populations of ribosome footprints , 28–30 nucleotides and 20–22 nucleotides long , representing translating ribosomes in distinct states , differentially stabilized by specific elongation inhibitors . We find that the balance of small and large footprints varies by codon and is correlated with translation speed . The ability to visualize conformational changes in the ribosome during elongation , at single-codon resolution , provides a new way to study the detailed kinetics of translation and a new probe with which to identify the factors that affect each step in the elongation cycle . To accomplish the huge task of translation elongation—in each cycle , accurately incorporating a new amino acid into a nascent peptide every 1/6th of a second , then moving precisely three nucleotides along the mRNA template—the ribosome undergoes a series of major structural rearrangements ( Figure 1 ) ( reviewed in Chen et al . , 2012 and Noeske and Cate , 2012 ) . During the initial decoding step of elongation , aminoacylated tRNAs are delivered to the decoding site ( A site ) as part of a ternary complex with EF-Tu ( in prokaryotes ) or the orthologous eEF1A ( in eukaryotes ) . When the anticodon of one of these aminoacylated tRNAs is able to base-pair stably with the specific mRNA codon in the decoding site ( A site ) , a new peptide bond is formed between the nascent polypeptide and the specified amino acid . The ribosome then undergoes a massive rearrangement in which the ribosomal subunits rotate relative to each other ( Frank and Agrawal , 2000; Zhang et al . , 2009 ) . Along with this rotation , the A and P site tRNAs move from ‘classic’ to ‘hybrid’ states: the anticodon ends stay in their original A and P sites and the acceptor ends move to the P and E sites ( Moazed and Noller , 1989; Munro et al . , 2007 ) . This rotated state of the ribosome undergoes additional conformational changes in preparation for translocation ( Zhang et al . , 2009; Fu et al . , 2011 ) . The ribosome can fluctuate between rotated and non-rotated states until EF-G ( eEF2 in eukaryotes ) binds and stabilizes the rotated ribosome ( Agirrezabala et al . , 2008 ) . GTP hydrolysis by EF-G then promotes translocation of the mRNA along the ribosome , coupled to a large intra-subunit rotation of the 30S head ( Ratje et al . , 2010 ) , after which the ribosome subunits rotate back to a closed formation for the next cycle ( Gao et al . , 2009 ) . Structural and biochemical studies have revealed many of the atomic-level changes that allow this complicated process to occur ( Pulk and Cate , 2013; Tourigny et al . , 2013; Zhou et al . , 2013 ) , and new details continue to emerge , reshaping models , raising new questions , and leaving other questions still unanswered . 10 . 7554/eLife . 01257 . 003Figure 1 . Schematic representation of the eukaryotic elongation cycle . Blue overlay denotes stages at which the ribosome has undergone a large inter-subunit rotation . Ribosome shapes are for illustration only , not a literal representation of the structure or degree of rotation . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 003 Recently , ‘ribosome profiling’ by high-throughput sequencing of ribosome-protected fragments has provided a powerful tool for identifying the position of ribosomes on mRNAs across the entire transcriptome ( Ingolia et al . , 2009 ) . Cell lysates are treated with nuclease to degrade all mRNA not physically protected by ribosomes , and the ribosome-protected fragments are extracted , sequenced , and mapped back to the genome to show ribosome positions , revealing the overall translation level of each gene as well as the distribution of ribosomes along the mRNA . Nucleotide-level precision of ribosome positions is possible because of the very consistent size of ribosome footprints in the conditions assayed . The authors of the method used a nuclease protection assay to establish that , in yeast treated with the elongation inhibitor cycloheximide , each ribosome protects a footprint of 28 nucleotides ( nt ) , confirming earlier reports ( Steitz , 1969; Wolin and Walter , 1988 ) . While performing ribosome-profiling experiments in Saccharomyces cerevisiae , we serendipitously noticed a population of smaller ribosome-protected fragments . To better capture these fragments and to investigate their origins , we revised the ribosome-profiling protocol originally established by Ingolia et al . Our experiments revealed that , in the absence of cycloheximide , the small ribosome-protected fragments were abundant , consistent with an early observation of short ribosome footprints in the absence of cycloheximide ( Wolin and Walter , 1988 ) . We show here that the small fragments originate from ribosomes in a conformation distinct from that previously observed in the presence of cycloheximide . The ability to discern distinct ribosomal structural states by ribosome profiling has given us insight into how codon , tRNA , and amino acid identity and translational speed relate to ribosome structure . This additional dimension of ribosome-profiling data will provide a valuable new layer of molecular and mechanistic information , at codon resolution , for future studies of translation . We began our investigation of ribosome footprint size by isolating ribosome-protected mRNA fragments from yeast using a modified ribosome-profiling procedure . The standard ribosome-profiling protocol includes a size selection for RNA fragments of around 28 nt . To eliminate the bias against smaller fragments , we broadened the initial size range and selected RNA fragments between 18 and 32 nt after RNase I digestion . By selecting fragments in this broader size range , and , importantly , by carrying out the entire procedure in the absence of cycloheximide or other inhibitors , we observed two clearly distinct , abundant populations of ribosome-protected mRNA fragments ( ‘footprints’ ) , 28–30 nt and 20–22 nt long . We visualized fragment lengths and positions with a three-dimensional ‘metagene’ representation: sequence reads representing the ribosome-protected fragments from all expressed genes were aligned relative to the start codon of the corresponding gene and tallied by fragment length and position to show the average pattern of translation along all annotated coding regions ( Figure 2A–C , Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 01257 . 004Figure 2 . Ribosome-protected fragment positions and size distributions from yeast not treated with elongation inhibitors . ( A ) The position of each fragment was calculated relative to the start codon of its gene . The 5′ end positions ( x axis ) and lengths of all fragments ( y axis ) were tallied across all genes with a coding region of at least 300 nt . Higher color intensity reflects more fragments . RNA fragments between 18 and 32 nucleotides were selected after gel electrophoresis; shorter and longer fragments are not entirely excluded but their read counts are presumed to be unrepresentative of their true abundance . ( B ) Profiles of the 5′ end positions of all 20 nt and 28 nt fragments relative to the start codon of their genes , as in ( A ) . ( C ) Total counts of mapped fragment lengths . ( D ) Distribution of 21 nt and 28 nt fragments in coding regions and untranslated regions of mRNAs . ( E ) Positions of 21 nt and 28 nt fragments relative to the reading frame . ( F ) Interpretation of fragment positions on an arbitrary gene fragment . Arrowheads show hypothetical nuclease cleavage sites relative to a ribosome in a non-rotated or rotated conformation ( shape is for illustration only ) . The resulting fragments are shown with the inferred decoding site ( A site ) , and their positions in a grid as in Figure 2A are shown with corresponding colors . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 00410 . 7554/eLife . 01257 . 005Figure 2—figure supplement 1 . Ribosome-protected fragment positions and size distributions from yeast not treated with elongation inhibitors . Two biological replicates of ribosome-protected fragment distribution , as in Figure 2A , C . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 005 We found overwhelming evidence that both populations of fragments came from translating ribosomes . The 21 and 28 nt fragments were both found almost entirely within annotated coding regions ( CDS ) and not in 5′ or 3′ UTRs; 98 . 3–99 . 7% of mappable 21 nt fragments , and 96 . 5–99 . 6% of mappable 28 nt fragments , mapped within the annotated CDS in three replicates ( Figure 2D ) . Both populations also showed the 3-nucleotide periodicity expected of fragments originating from elongating ribosomes ( Figure 2E ) . We conclude that fragments of both sizes are footprints of translating ribosomes . The 5′-most peaks in the metagene represent ribosomes with the start codon in the P site and the second codon in the A site ( Kapp and Lorsch , 2004; Ingolia et al . , 2009 ) . Using this as a reference for phasing all the footprints , we inferred that for ribosomes with a given codon in the A site , small and large footprints generally had the same 5′ ends positioned 15–16 nt upstream of the A-site codon , and differed at their 3′ ends: extending 2–3 nt beyond the A-site codon in the small footprints and 10 nt beyond the A-site codon in the large footprints , respectively ( Figure 2F ) . During elongation , at each codon , the ribosome cycles through a stereotyped sequence of steps as it incorporates the specified amino acid and translocates to the next codon . These steps are accompanied by major rearrangements of the ribosome structure , including a rotation of the large subunit relative to the small subunit upon peptide bond formation . We hypothesized that the non-rotated , pre-peptide-bond ribosomes and rotated , post-peptide-bond ribosomes might protect different lengths of mRNA , and that the two resulting footprint sizes might , therefore , represent these two conformations . To determine what footprint sizes were protected by ribosomes in distinct stages of elongation , we performed ribosome profiling on yeast treated with inhibitors that block different steps of the cycle . Cycloheximide is an elongation inhibitor that binds to the E site of ribosomes , preventing the E site tRNA from leaving the ribosome . When cycloheximide was added to the yeast immediately before harvest and was present throughout lysis and RNase I treatment , the most prevalent footprints were 28–30 nt long and were distributed along the coding sequence with a 3-nt periodicity ( Figure 3A–C , Figure 3—figure supplement 1 ) . Apart from a distinct peak at the start codon , there were very few 20–22 nt footprints . 10 . 7554/eLife . 01257 . 006Figure 3 . Ribosome-protected fragment positions and size distributions from yeast treated with elongation inhibitors . ( A and B ) As in Figure 2A , B , fragment position and size distribution for yeast treated with cycloheximide . ( C ) Distribution of mapped fragment lengths for yeast treated with cycloheximide . ( D and E ) Fragment position and size distribution for yeast treated with anisomycin . ( F ) Distribution of mapped fragment lengths for yeast treated with anisomycin . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 00610 . 7554/eLife . 01257 . 007Figure 3—figure supplement 1 . Ribosome-protected fragment positions and size distributions from yeast treated with elongation inhibitors . ( A ) Biological replicate of ribosome-protected fragment distribution after cycloheximide treatment . ( B ) Biological replicate ( top ) and technical replicate ( bottom; independent fractionation and library preparation from the same lysate as Figure 3 ) of fragment distribution after anisomycin treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 007 Our data confirmed previous evidence that the ribosome predominantly protects a 28 nt footprint in the presence of cycloheximide , and suggest that cycloheximide stabilizes one stage of the elongation cycle . Previous work shows that cycloheximide bound alongside a tRNA in the E site prevents either the incorporation of the next aminoacylated tRNA in the A site or peptide bond formation ( Schneider-Poetsch et al . , 2010 ) . In either case , it is expected to trap the ribosome in a non-rotated conformation , suggesting that the non-rotated conformation protects 28–30 nt of mRNA . We next conducted ribosome-profiling experiments using yeast treated with anisomycin , an elongation inhibitor that binds to the peptidyl transferase center ( Grollman , 1967; Hansen et al . , 2003 ) . We observed almost exclusively small footprints in yeast treated with anisomycin ( Figure 3D–F , Figure 3—figure supplement 1 ) . By comparison to the effects of cycloheximide treatment , we inferred that anisomycin stabilizes a distinct conformation of the ribosome that protects 20–22 nt of mRNA . Although anisomycin's precise mechanism is not characterized , it has higher affinity for post-translocation ribosomes than for pre-translocation , cycloheximide-treated ribosomes , suggesting that it preferentially binds a ribosome conformation distinct from that stabilized by cycloheximide ( Barbacid and Vazquez , 1974 , 1975 ) . Lincomycin and other antibiotics that bind the peptidyl transferase center induce translocation , and lincomycin-treated ribosomes prefer a rotated conformation in in vitro FRET experiments ( Fredrick and Noller , 2003; Ermolenko et al . , 2013 ) . It is possible that anisomycin acts similarly to stabilize a rotated conformation . We have thus demonstrated that two distinct ribosome conformations can be stabilized using elongation inhibitors . Stabilization of distinct conformations by two drugs resulted in a nearly complete reciprocal bias in the size of ribosome footprints , providing evidence that large and small footprints originate from distinct ribosomal conformations . We hypothesize that each ribosome cycles through both conformations , protecting first a large footprint and then a small footprint at each codon . The footprints identified by high-throughput sequencing in a ribosome-profiling experiment represent a deep sampling of ribosomes in different states , and thus the ratio of large to small footprints in untreated cells could show , at single-codon resolution , how many ribosomes are in each stage of elongation . To enrich for ribosomes in a single , defined stage of the elongation cycle , we induced conditions expected to result in the depletion of a specific aminoacyl-tRNA and thus to increase the decoding time when the cognate codon is in the A site . We treated yeast with 3-amino-1 , 2 , 4-triazole ( 3-AT ) , an inhibitor of histidine biosynthesis , to create a specific shortage of His-acylated tRNA and cause ribosomes to pause on histidine codons ( Figure 4A ) . We would therefore expect ribosomes to accumulate at histidine codons in a pre-peptide-bond conformation . Estimating codon-specific occupancy as described in more detail below , we found that the shortage of His-tRNA dramatically increased the relative abundance of large footprints from ribosomes with His codons in the A site , with minimal effect on the abundance of small footprints ( Figure 4B , C , Figure 4—figure supplement 1 ) . During the decoding phase of elongation , before peptide bond formation , the ribosome is in a non-rotated conformation ( Frank and Agrawal , 2000; Gao et al . , 2009 ) ; these results therefore strongly suggest that the decoding phase of elongation ( the non-rotated conformation ) is represented by large footprints . 10 . 7554/eLife . 01257 . 008Figure 4 . Effect of 3-amino 1 , 4 triazole on translation of histidine codons . ( A ) Schematic representation of the hypothesized effect of 3-AT . 3-AT reduces intracellular concentrations of histidyl-tRNA and thus is expected to increase time spent decoding histidine codons ( i . e . , in the decoding phase of the cycle , with a His codon in the A-site ) . ( B ) All 61 sense codons are plotted by the log2 of the relative abundance of large footprints with the specified codon in the A-site for untreated cells ( x axis ) against the log2 relative abundance of large footprints for yeast treated with 3-AT ( y axis ) . Values shown are the average of three untreated replicates and two 3-AT treatments ( 10 min and 60 min ) . Histidine codons are denoted in red ( CAT ) and cyan ( CAC ) . ( C ) As in ( B ) , showing the relative abundance of small footprints . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 00810 . 7554/eLife . 01257 . 009Figure 4—figure supplement 1 . Effect of 3-amino 1 , 4 triazole on translation of histidine codons . As in Figure 4 , log2 relative occupancy , log2 large footprint abundance , and log2 small footprint abundance in comparisons of three untreated replicates and two 3-AT treated samples . Histidine codons are denoted in red ( CAT ) and cyan ( CAC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 009 Recently , ribosome profiling has revealed that translation speed varies systematically by codon ( Tuller et al . , 2010; Stadler and Fire , 2011; Dana and Tuller , 2012 ) ; we hypothesized that there might be distinct codon-specific effects on the rate of the two distinct phases of elongation represented by small and large footprints . Using data from untreated cells , we calculated the number of large and small footprints corresponding to ribosomes with a given codon in the A site , for each codon position in the yeast transcriptome . Large footprints were defined as 28 or 29 nt and small footprints were defined as 20 , 21 , or 22 nt with 5′ ends positioned relative to the inferred A site as depicted in Figure 2F . We found substantial variation in the characteristic length distribution between codons: small footprints ranged from 38 ± 12% ( UUU ) to 87 ± 9% ( CGG ) of the total footprints for a given codon identity , averaged across three replicates . To explore this codon effect , we computed the relative occupancy of each of the 61 sense codons in the A site . We started by considering an individual gene and calculated the over- or underrepresentation of footprints at each codon position compared to the average for all codon positions in that gene , including both small and large footprints ( an example from a highly expressed gene is shown in Figure 5A ) . After performing this computation for every gene , we averaged these multipliers across all occurrences of a given codon in the genome to provide the ‘relative occupancy’ for that codon , representing , on a relative scale , how frequently we observed ribosomes with that codon positioned at the A site . The relative occupancies varied over a fivefold range , from 0 . 48 ± 0 . 04 ( GGU ) to 2 . 6 ± 0 . 67 ( CCG ) ( unitless , average of three replicates ) and were highly correlated between independent replicates ( Figure 5B ) . As a control , we also analyzed the occupancy based on the codon one position 3′ of the A site , which has not yet entered the decoding site . We found that the range of occupancies relative to the codon in the A site was much broader than the range of occupancies relative to the next codon , suggesting that the A-site occupancies reflect an aspect of translation , not merely confounding factors such as biases in fragment capture ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 01257 . 010Figure 5 . Codon-specific variation in large and small footprint abundance . ( A ) Distribution of ribosome footprint counts on the highly expressed gene FBA1 , highlighting an arbitrary window , codons 250–279 . Ribosome footprint counts per position were consistent between replicates and varied between instances of the same codon in this window . Relative occupancy was estimated based on the codon in the inferred A site . Total ( large + small ) footprint coverage at each codon of a gene was computed relative to the average coverage for that gene , then averaged by codon across all genes to provide per-codon relative occupancies . Relative abundance of small or large footprints was computed similarly , comparing the count of small or large footprints at each codon of a gene against the average coverage ( large + small ) for that gene , then averaged by codon across all genes . Examples of small and large footprint abundance values at two specific TTC codons in FBA1 are shown . ( B ) Relative occupancies of all 61 codons compared between two replicates , with Spearman correlation of 0 . 81 . Stop codons and the first 50 codons of each gene were excluded from analysis . Similarly , small footprint abundance ( C ) and large footprint abundance ( D ) compared between replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 01010 . 7554/eLife . 01257 . 011Figure 5—figure supplement 1 . Codon-specific variation in large and small footprint abundance . ( A ) Relative occupancies based on the codon downstream of the inferred A site , compared to the A-site occupancies as in Figure 5 . Similarly , small footprint abundance ( B ) and large footprint abundance ( C ) for the inferred A site and downstream codon . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 011 Codon-specific differences in ribosome occupancy could have been driven by variation in small footprint counts , variation in large footprint counts , or both , potentially revealing the variability of each stage of elongation . We inferred the relative abundance of ribosomes in each state at each codon using a model similar to the one we used to estimate overall relative occupancy , but considering counts of either small or large footprints separately ( Figure 5A ) . As with overall occupancy , the relative abundances of small footprints and the relative abundance of large footprints were both highly correlated between replicates ( Figure 5C , D ) . This suggests that codon identity affected both the pre-peptide-bond and post-peptide-bond stages of elongation . However , the effect of codon identity on the inferred duration of these two phases of the elongation cycle was distinct: the codon-specific relative abundances of small and large footprints were almost uncorrelated ( Spearman's r = 0 . 11 , average of three replicates ) . This led us to search for physical correlates of the codon-specific differences . We found that a major and unexpected determinant of the abundance of footprints from each conformation was the identity of the amino acid encoded by the A-site codon . We found a much greater density of small footprints at codons encoding smaller , polar amino acids than at codons encoding large , aromatic amino acids . The relative abundance of small footprints at codons encoding a given amino acid was correlated with measures of polarity of the cognate amino acid , such as the Kd of transfer of side chains from vapor to water ( Spearman's r = −0 . 75 when grouped by amino acid , r = −0 . 58 by codon , Figure 6A ) , while the relative abundance of large footprints showed no correlation to amino acid polarity ( Spearman's r = 0 . 11 by amino acid , r = 0 . 02 by codon ) ( Wolfenden , 2007 ) . These data strongly suggest that the chemical properties of the amino acid specified by the codon in the A site affect the stability of the rotated , post-peptide-bond conformation of the ribosome . We hypothesize that interactions between the ribosome and polar amino acids acylated to the A-site tRNA can slow translocation substantially . 10 . 7554/eLife . 01257 . 012Figure 6 . Correlates of footprint abundance . ( A ) Small footprint abundance , averaged for all codons encoding the same amino acid plotted against Kd of transfer of side chain from vapor to water as a measure of polarity ( Wolfenden , 2007 ) , with Spearman correlation from the average of three samples . ( B ) Relative occupancy of directly paired codons vs relative occupancy of codons that recognize the same tRNA with wobble pairing . Values are the average of three replicates . Dashed line shows y = x , the expected relationship if occupancy were determined solely by tRNA identity . ( C and D ) As in ( B ) , showing small and large footprint abundance . DOI: http://dx . doi . org/10 . 7554/eLife . 01257 . 012 Many factors have been proposed to affect translation speed at a given codon , particularly tRNA abundance . In yeast , the number of genes encoding a specific tRNA has been shown to be highly correlated with both codon usage and cellular tRNA concentrations ( Percudani et al . , 1997 ) . A related measure of codon optimality is the tRNA adaptation index ( tAI ) , which attempts to rank codons in translational efficiency by accounting for tRNA copy number , wobble pairing constraints , and codon usage ( dos Reis et al . , 2004 ) . We found that the relative occupancy per codon was only weakly correlated with tAI and with tRNA genomic copy number ( Spearman's r = −0 . 39 and −0 . 28 , respectively; average of three replicates ) and that the tAI was not particularly correlated with the relative abundance of either small footprints or large footprints ( r = −0 . 34 and r = −0 . 20 , respectively; average of three replicates ) . Thus , unexpectedly , codon ‘optimality’ , as represented by the tAI , does not appear to be a major determinant of relative ribosome occupancy under the conditions tested here . The 3-AT data show that in an extreme case , a limited supply of the tRNA cognate to the A-site codon slows translation during the large-footprint stage . In contrast , our overall results in untreated yeast suggest that the differences in abundance among tRNAs in wild-type cells have only a minor effect on relative ribosome occupancy of the cognate codons under optimum growth conditions . We also investigated the relationship between wobble base pairing , relative occupancy , and the density of large and small footprints . Wobble base pairing at the A site has recently been linked with slowed elongation in humans and worms ( Stadler and Fire , 2011 ) . We compared codons with perfect Watson-Crick complementarity vs the synonymous codons that pair imperfectly with the same tRNA ( Johansson et al . , 2008 ) . While we found no consistent trend toward increased occupancy at wobble-paired codons , we observed notably higher occupancy on a subset of wobble-paired codons comprising proline CCG ( G-U base pairing ) , leucine CUG ( G-U ) , and arginine CGA ( A-I ) ( Figure 6B ) . For these three wobble codon outliers , we see a dramatic increase in short footprints , representing post-decoding stages of translation ( Figure 6C , D ) . The arginine CGA codon is known to be a strong inhibitor of translation in yeast , and its inhibitory effect is due more to wobble decoding than tRNA abundance and may include interactions after the initial decoding ( Letzring et al . , 2010 ) . Our data confirm that CGA is indeed one of the most slowly translated codons , and its high relative occupancy is due to increased abundance of small footprints , suggesting that its slow elongation is primarily due to a prolonged post-decoding stage . Overall , the abundance of footprints from each step of elongation was clearly affected by several distinct codon-specific features with sometimes synergistic and sometimes opposing effects . A ribosome must cycle through a series of consecutive associations with mRNA to decode the message one codon at a time . The stability of the ribosome-mRNA association allows one to observe precisely where ribosomes reside on transcripts—down to the codon being decoded—by isolating and sequencing ribosome-protected mRNA fragments . We were quite surprised to discover that the ribosome protects two different footprint sizes ( 28–30 nt and 20–22 nt ) , as the original ribosome-profiling experiments and nuclease protection assays only captured the longer footprints ( Ingolia et al . , 2009 ) . The difference is explained by the experimental conditions: the small footprints were revealed only after we left out cycloheximide , a translation inhibitor commonly used to stabilize ribosomes on mRNA for ribosome profiling . Indeed , early study of ribosome pausing found that when cycloheximide was omitted , 20–24 nt footprints accumulated in addition to the larger footprints they saw from cycloheximide-treated ribosomes ( Wolin and Walter , 1988 ) . As in our own experiments , the small and large footprints they observed had the same 5′ terminus and differed at the 3′ end . We propose that the two footprints sizes originate from two ribosome conformations corresponding to different stages of elongation: large footprints from non-rotated ribosomes during the decoding stage before peptide bond formation , and small footprints from rotated ribosomes during the translocation stage after peptide bond formation . Additional biochemical and structural studies will be required to pinpoint the exact stages of elongation and ribosome conformations responsible for the two footprint sizes . It is not clear which of the known conformational changes during the elongation cycle are most relevant: the inter-subunit rotation after peptide bond formation , the intra-subunit swivel of the 30S head during translocation , or smaller rearrangements such as movement of the L1 stalk . As for the physical origin of the small and large mRNA fragments , crystal structures of rotated and non-rotated ribosomes show that mRNA accessibility is not likely to be dramatically different between the two conformations ( Ben-Shem et al . , 2010 , 2011 ) . RNAse I may be small enough to penetrate into the mRNA entrance channel and cleave the mRNA just two nucleotides from the A site . Alternately , the ribosome itself may be more susceptible to RNAse degradation in the rotated conformation , allowing ribosomal RNA cleavage that in turn enables RNAse I to access the mRNA entrance channel , yielding a smaller mRNA footprint . Importantly , however , both small and large footprints have also been observed in wheat germ extract treated with micrococcal nuclease , indicating that the two footprint sizes are neither species- nor nuclease-specific ( Wolin and Walter , 1988 ) . We hypothesize that the relative abundance of large and small footprints reflects the relative duration of different stages of elongation at each codon . ( We use the A site codon by default in this discussion , though in principle we could compile results based on the codon in the P site or any other frame of reference . ) Comparing our relative occupancy values to an estimated bulk elongation rate of 5 . 6 amino acids per second ( Ingolia et al . , 2011 ) , our model would predict variation in average codon elongation time from as little as 0 . 08 s ( GGU ) to as much as 0 . 5 s ( CCG ) . A number of caveats apply to this interpretation , and any hypotheses must be pursued with complementary approaches . Ribosome footprint data have inherent biases from ligation and other steps of the library preparation . Further , the overall balance of small and large footprints varied between replicates , leaving open the question of which conformation is more populated in vivo . Some variability arises from the mRNA fragment isolation . In this work , we chose size markers of 18 and 32 nt , but size selection from polyacrylamide gel is imprecise . ( This choice also limits what we can observe: recent work found distinct 16 nt fragments from ribosomes stalled on truncated mRNAs , Guydosh and Green , 2014 . ) The size distribution may also reflect differential efficiency of library preparation from smaller or larger fragments . Nonetheless , although the overall ratio of small to large footprints varied , the codon-specific variation in this ratio was robust . Our results also highlight the effects of harvest methods and inhibitors such as cycloheximide on footprint distribution . Ribosomes are depleted from the first 50 codons when yeast are harvested by the procedure we used without inhibitors . We interpret this as evidence that elongation continues for around 10 s after initiation ceases during the harvest process . Because the selective depletion of ribosomes from this part of the mRNA could enrich for special cases , we excluded the first 50 codons from our analysis of per-codon footprint distributions . Different harvest methods had large effects on the precise footprint locations even when the overall translation per gene was highly reproducible ( data not shown ) . Similarly , the average occupancies per codon with and without cycloheximide were surprisingly uncorrelated ( Spearman's r = 0 . 02 , comparing the average of three untreated samples and the average of two cycloheximide-treated samples ) , though the total footprints per gene correlated quite well ( Spearman's r = 0 . 97 between the average fpkm in three untreated samples and the average fpkm in two cycloheximide-treated samples ) . Ribosomes in different positions may be differentially affected either by the drug treatment or by runoff elongation during harvest without inhibitors . In either case , some ribosomes may halt while others undergo several more rounds of elongation . There are many potentially rate-controlling steps of elongation and many factors necessary for each cycle , including aminoacylated tRNA and elongation factors eEF1 , eEF2 , and the yeast-specific eEF3 ( Kapp and Lorsch , 2004 ) . For example , interactions between the tRNA anticodon and the mRNA codon , the tRNAs and the ribosome , the amino acids and the peptidyl transferase center , and the nascent peptide and the tunnel , as the tRNAs move through the A , P , and E sites , can all presumably affect the speed of each step . Thus , the speed of each elongation cycle is expected to be influenced by codon , tRNA , and amino acid identity . One of the surprising aspects of this study is that tRNA abundance or codon optimality failed to predict variation in observed ribosome occupancy and , further , that much of the variation in codon-specific occupancy was in the steps following decoding and peptide bond formation . Biochemical evidence suggests that evolution has tuned tRNA sequence and modifications to balance the contributions of amino acid identity , codon pairing strength , and tRNA structure to binding affinity of a given tRNA , such that most aminoacylated tRNAs have similar affinity to ribosomal A sites ( Olejniczak et al . , 2005; Dale et al . , 2009; Shepotinovskaya and Uhlenbeck , 2013 ) . While this affinity tuning is a plausible result of selection for fidelity in decoding , ribosome profiling has revealed a lack of uniformity both in decoding and post-decoding steps . Once the interactions that determine the codon-specific rate of decoding are decoupled and replaced by a new set of codon-specific interactions in the subsequent steps of elongation , the great diversity in physical properties of amino acids and in the intrinsic stability of the codon–anticodon interaction may lead to wide variation in the kinetics of post-decoding steps . New methods for high-throughput measurement of translation have led to renewed interest in modeling the constraints on coding sequence and the effects of codon choice on translation efficiency ( Tuller et al . , 2010; Plotkin and Kudla , 2011; Dana and Tuller , 2012; Charneski and Hurst , 2013; Shah et al . , 2013 ) . The ability to distinguish ribosome conformations at codon resolution now allows us to map these effects to specific phases of the elongation cycle , initiation , or termination . Future in vivo and in vitro experiments using this approach to monitor the decoding and translocation steps at each codon should provide new precision in dissecting the mechanisms by which mRNA sequence , core translation factors and regulatory factors control initiation , elongation , and termination of translation . For all experiments , excluding 3-AT drug treatment experiments , BY4741 was grown overnight in YPD at 30°C; two 500 ml cultures of YPD were inoculated from the overnight culture to an OD600 of ∼0 . 2 . For experiments involving 3-AT , S288C was grown as above in SC-His media at 30°C . Cells were then grown to mid-log phase , OD600 ∼0 . 6 , prior to harvest . ( Strain information: BY4741 derived from S288C: MATa his3Δ1/his3Δ1 leu2Δ0/leu2Δ0 lys2Δ0/LYS2 MET15/met15Δ0 ura3Δ0/ura3Δ0 . S288C: MATa SUC2 gal2 mal mel flo1 flo8-1 hap1 . ) Cells were harvested by filtration at room temperature and then quickly frozen in liquid N2 . Resulting cell pellets were then pulverized using a MM301 Retsch mixer mill at 30 Hz for 3 min . All chambers and tubes were pre-frozen in liquid N2 or dry ice . Approximately 400–500 µl of cold lysis buffer ( 20 mM Tris pH 8 . 0 , 140 mM KCl , 1 . 5 mM MgCl2 , 1% Triton ) was added to cell powder . Resulting lysates were pre-cleared by centrifugation at 2000 rpm for 5–10 min at 4°C . Lysate was transferred to a clean pre-chilled tube and further clarified by centrifugation at 20 , 000×g for 10 min at 4°C . Lysate was then stored at −80°C until RNase digestion . For the cycloheximide experiments , cycloheximide was added to cells prior to harvest at 100 µg/ml and was also present at 100 µg/ml in the lysis buffer . For the anisomycin experiment , anisomycin was added to mid-log cells at 100 µg/ml and cells were allowed to grow for an additional 30 min prior to harvest . Anisomycin was also present at 100 µg/ml in the lysis buffer . For the 3-AT experiments , 3-amino-1 , 2 , 4-triazole was added to mid-log cells to reach a final concentration of 100 mM , then cells were grown with shaking for 10 and 60 min prior to harvest . RNase digestion and monosome isolation were performed similar to Ingolia et al . ( 2009 , 2012 ) . Cell lysate ( ∼800 µg total RNA measured by Nanodrop ) was allowed to thaw on ice . 600 U of RNase I ( AM2294; Life Technologies , Carlsbad , CA ) was added to cell lysate and placed on a nutator at room temperature for 1 hr . A second cell lysate served as an undigested control; 120 U of SUPERase-In was added and placed on a nutator as above . Linear 10–50% sucrose gradients were prepared using a BioComp Gradient Master ( Biocomp Instruments , Fredericton , Canada ) according to manufacturer's instructions . Sucrose was dissolved in 20 mM Tris pH 8 . 0 , 140 mM KCl , 5 mM MgCl2 , 0 . 5 mM DTT , 20 U/ml SUPERase-In; 100 µg/ml cycloheximide or 100 μg/ml anisomycin were added to buffer for corresponding experiments . After RNase digestion , lysate was added to the top of gradients and sedimented at 35 , 000 rpm in a SW41 rotor for 3 hr . Gradients were fractionated at 0 . 17 mm per second using the BioComp Gradient Master while the A260 was continuously monitored . Fractions corresponding to the monosome peak were collected and pooled . RNA was then purified using a miRNeasy Mini kit from Qiagen ( cat# 217004; Qiagen , Venlo , Netherlands ) as per manufacturer's instructions . Ribosome footprint libraries were prepared similar to Ingolia et al . ( 2012 ) . Purified RNA was separated on a 15% TBE-Urea gel . RNA oligonucleotides of 18 and 34 nucleotides were run side by side with isolated RNA and used as size markers to cut RNA of desired size for gel extraction . Size-selected RNA fragments were then treated with polynucleotide kinase to remove the 3′ phosphate . After isopropanol precipitation , dephosphorylated fragments were ligated to Universal miRNA cloning linker from New England Biolabs , Ipswich , MA ( cat# S1315S ) . Ligated fragments were separated from excess linker by gel electrophoresis on a 15% TBE-Urea gel . After gel extraction , ligated fragments were then reverse transcribed using SuperScript III from Life Technologies ( cat# 18080-085 ) according to manufacturer's instructions . Reverse transcriptase reactions were primed with 1 µl of 1 . 25 µM NI-NI-9 primer ( Supplementary file 1 ) . Additionally 20 U of SUPERase-In was added to each RT reaction . Reactions were incubated at 48°C for 30 min . After reverse transcription , RNA template was removed by the addition of 2 . 2 µl of 1 N NaOH and incubation at 98°C for 20 min . After precipitation , cDNA was separated from excess primer by gel electrophoresis on a 5% TBE-Urea gel . cDNA was then circularized using CircLigase ssDNA ligase from Epicentre , Madison , WI ( cat# CL4115K ) according to manufacturer's instructions . After circularization , 5 µl of the circularization reaction was added to 1 µl of pooled ribosomal subtraction oligos ( Supplementary file 1 ) , 1 µl of 20x SSC , and 3 µl of water . Each sample was then denatured for 90 s at 100°C and then annealed to 37°C . MyOne Streptavidin C1 DynaBeads ( 25 µl per reaction ) were washed three times in 1x Bind/Wash buffer ( 1 M NaCl , 1 mM EDTA , 10 mM Tris , pH 8 . 0 ) . Beads were then resuspended in 2x Bind/Wash buffer ( 10 µl per reaction ) . Beads were added to each cDNA/oligo mixture and incubated for 15 min at 37°C in an Eppendorf ThermoMixer at 1000 rpm . Beads were collected on a magnetic stand and ∼17 . 5 µl of eluate was recovered for each reaction . Resulting eluate was then used as a template for PCR amplification . Pilot PCR reactions were prepared in order to determine the number of cycles necessary for adequate amplification . PCR reactions consisted of 20 µl of 5x HF buffer , 2 µl of 10 mM dNTPs , 0 . 5 µl of 100 µM NI-NI-2 primer , 0 . 5 µl of 100 µM indexing primer ( Supplementary file 1 ) , 5 µl of eluate template , 71 µl of water and 1 µl of Phusion polymerase ( cat# M0530L; NEB ) . Each 100 µl reaction was separated into five 16 . 7 µl aliquots . PCR conditions were as follows: initial denaturation for 30 s at 98°C , followed by cycles of 10 s at 98°C , 10 s of annealing at 65°C , and 5 s of extension at 72°C . One aliquot was removed after 8 , 10 , 12 , and 14 cycles . Amplification was examined by gel electrophoresis on an 8% TBE polyacrylamide gel . Once optimal cycle was determined , an additional 100 µl PCR was performed and run on an 8% TBE polyacrylamide gel . The product band was then cut out and DNA extracted from the gel slice . Libraries were quantified by Bioanalyzer using a DNA High Sensitivity kit ( cat# 5067-4626; Agilent , Santa Clara , CA ) . Libraries were then sequenced on an Illumina Genome Analyzer 2 according to manufacturer's instructions by the Stanford Functional Genomics Facility . Raw sequence data are available in the Gene Expression Omnibus under accession GSE58321 . Cloning linker sequences were trimmed from Illumina reads and the trimmed fasta sequences were aligned to S . cerevisiae ribosomal and noncoding RNA sequences using bowtie v . 0 . 12 . 7 or v . 1 . 0 . 0 to remove rRNA reads ( Langmead et al . , 2009 ) . The non-rRNA reads were aligned to the S . cerevisiae genome as a first pass to remove any reads that mapped to multiple locations . Reads that passed this filter ( those that mapped uniquely to the genome , or those that did not map at all , such as splice junction reads ) were then aligned to the S . cerevisiae transcriptome with bowtie , allowing two mismatches and only reporting alignments of reads that mapped uniquely in the transcriptome ( bowtie -v 2 -m 1 -a --norc --best –strata ) . The S . cerevisae transcriptome sequences were based on CDS sequences downloaded from the UCSC genome browser , sacCer2 assembly , in August 2011 . Untranslated region coordinates were taken from supplemental table S4 of Nagalakshmi et al . ( 2008 ) . When no UTR was annotated , 50 nt upstream and/or downstream of the CDS was included by default . A list of read counts and read lengths per nucleotide position in the transcriptome , based on the 5′ end of the mapped read , was generated . From that list , metagene grids as in Figure 2A were made by tabulating all footprints 11–36 nt long within the following regions: last 25 nt of 5' UTR , first 200 nt of CDS , last 100 nt of CDS , and first 50 nt of 3' UTR , for all genes with a CDS of at least 300 nt . Non-unique positions in the transcriptome were filtered by splitting the yeast transcriptome into all overlapping 20mers , mapping this set of all 20mers back to the transcriptome with bowtie , and collecting the mapped locations of any 20mers with more than one perfect match in the transcriptome . The counts of small and large footprints from ribosomes with each codon positioned in the inferred A site were generated from the list of reads at each nucleotide position as depicted in Figure 2F . The large footprints were defined as 28 nt reads with the 5′ end 15 nt upstream of the codon at position i , and 29 nt reads with the 5′ end 16 nt upstream of i . Small footprints included 20 nt and 21 nt reads with the 5′ end 15 nt upstream of i and 21 nt and 22 nt reads with the 5′ end 16 nt upstream of i . For each gene , the analysis included codons 51 through the second codon before the stop codon , to avoid the region at the beginning of genes from which ribosomes have been depleted by runoff elongation during harvest . Genes with fewer than 10 footprints in total were excluded , as were any non-unique positions within genes . Processed data are available in the Gene Expression Omnibus under accession GSE58321 . The ‘relative occupancy’ per codon was generated by first computing the average number of footprints ( large + small ) across the gene . Then , at each position i in gene g , compute ( large + small at position i ) / ( average large + small in gene g ) . These ratios were then averaged across all instances of a given codon ( e . g . , CGA ) in the transcriptome to give the relative occupancy . The densities of small and large footprints were computed as above: ( small at i ) / ( average large + small in gene g ) and similarly ( large at i ) / ( average large + small in gene g ) .	To make a protein from a gene , the gene is first transcribed to produce a molecule of messenger RNA ( mRNA ) , which then passes through a molecular machine called a ribosome . The ribosome reads the genetic code in the mRNA in groups of three letters at a time , and each triplet of letters ( or codon ) represents an amino acid . The ribosome then joins the relevant amino acids together to build a protein . The ribosome processes about six amino acids per second , on average , but the mRNA is not fed through at a constant rate . Instead , the ribosome changes its shape to ratchet along the mRNA from one codon to the next: it then reads the new codon and adds another amino acid to the protein . However , many of the details of this ratcheting process are not fully understood . In this study , Lareau , Hite et al . have used a technique called ‘ribosome profiling’ to explore the movement of ribosomes along mRNA molecules . First , all of the pieces of mRNA molecules that are not protected inside a ribosome were chemically destroyed . The sequences of the protected fragments were then read and matched to the full-length gene sequences . The protected fragments came in two different sizes: some were about 28–30 letters long , and others were about 20–22 letters long . Lareau , Hite et al . suggest that these different fragment sizes occur because the ribosome switches between two shapes at each codon as it ratchets along the mRNA , and so it protects different lengths of mRNA . In previous ribosome-profiling experiments , the fragments had all been about 28 letters long; but these experiments had used a chemical to halt the progress of the ribosomes along the mRNAs before measuring the length of the fragments . Lareau , Hite et al . argue that this chemical locks the ribosome in the same shape when it brings the ribosome to a halt , and so the protected fragments always have the same length . Further , other chemicals that halt ribosomes appear to lock this molecular machine in the other shape , and so it can only protect the shorter fragments . The findings of Lareau , Hite et al . show that ribosomal profiling experiments can reveal much more than simply where a ribosome is on an mRNA molecule . Further study into the different stages of the ribosome ratcheting process will help uncover how the speed that a ribosome translates an mRNA into a protein can be encoded in the mRNA sequence itself .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2014	Distinct stages of the translation elongation cycle revealed by sequencing ribosome-protected mRNA fragments
It has long been thought that the life cycle of Streptomyces bacteria encompasses three developmental stages: vegetative hyphae , aerial hyphae and spores . Here , we show interactions between Streptomyces and fungi trigger a previously unobserved mode of Streptomyces development . We term these Streptomyces cells ‘explorers’ , for their ability to adopt a non-branching vegetative hyphal conformation and rapidly transverse solid surfaces . Fungi trigger Streptomyces exploratory growth in part by altering the composition of the growth medium , and Streptomyces explorer cells can communicate this exploratory behaviour to other physically separated streptomycetes using an airborne volatile organic compound ( VOC ) . These results reveal that interkingdom interactions can trigger novel developmental behaviours in bacteria , here , causing Streptomyces to deviate from its classically-defined life cycle . Furthermore , this work provides evidence that VOCs can act as long-range communication signals capable of propagating microbial morphological switches . Our current understanding of microbial growth and development stems largely from investigations conducted using single-species cultures . It is becoming clear , however , that most bacteria and fungi exist as part of larger polymicrobial communities in their natural settings ( Scherlach et al . , 2013; Traxler and Kolter , 2015 ) . Microbial behavior is now known to be modulated by neighbouring organisms , where interspecies interactions can have profound and diverse consequences , including modifying virulence of human pathogens ( Peleg et al . , 2010 ) , altering antibiotic resistance profiles of mixed-species biofilms ( Oliveira et al . , 2015 ) , enhancing bacterial growth ( Romano and Kolter , 2005 ) , and increasing production of specialized metabolites by fungi and bacteria ( Schroeckh et al . , 2009; Stubbendieck and Straight , 2016 ) . Consequently , an important next step in advancing our developmental understanding of microbes will be to expand our investigations to include multi-species cultures , and in doing so , unveil new and unexpected microbial growth strategies . The soil is a heterogeneous environment that is densely populated with bacteria and fungi , and as such , represents an outstanding system in which to study the effects of bacterial-fungal interactions . Within the polymicrobial communities occupying the soil , Streptomyces represent the largest genus of the ubiquitous actinomycetes group . These Gram-positive bacteria are renowned for both their complex developmental life cycle ( Elliot et al . , 2008 ) and their ability to produce an extraordinary range of specialized metabolites having antibiotic , antifungal , antiparasitic , and anticancer properties ( Hopwood , 2007 ) . The Streptomyces life cycle encompasses three developmental stages ( Figure 1A ) . First , a spore germinates to generate one or two germ tubes . These grow by apical tip extension and hyphal branching , ultimately forming a dense vegetative mycelial network that scavenges for nutrients . Second , in response to signals that may be linked to nutrient depletion , non-branching aerial hyphae extend into the air away from the vegetative cells . These aerial hyphae are coated in a hydrophobic sheath that enables escape from the aqueous environment of the vegetative mycelium ( Claessen et al . , 2003; Elliot et al . , 2003 ) , and their emergence coincides with the onset of specialized metabolism within the vegetative cells ( Kelemen and Buttner , 1998 ) . Aerial development requires the activity of the ‘bld’ gene products , where mutations in these genes result in colonies lacking the fuzzy/hydrophobic characteristics of wild type . The final developmental stage involves the differentiation of aerial hyphae into spores through a synchronous cell division and cell maturation event . This process is governed by the whi ( for ‘white’ ) gene products , whose mutants fail to form mature , pigmented spores ( McCormick and Flärdh , 2012 ) . In addition to being highly stress-resistant , spores also provide a means of dispersing Streptomyces to new environments , as all characterized Streptomyces cell types are non-motile . 10 . 7554/eLife . 21738 . 003Figure 1 . Physical association with yeast triggers Streptomyces exploratory behaviour . ( A ) Developmental life cycle of Streptomyces . Germ tubes emerge from a single spore , and grow by apical tip extension and hyphal branching , forming a dense network of branching vegetative hyphae . In response to unknown signals , non-branching aerial hyphae coated in a hydrophobic sheath , escape into the air . Aerial hyphae differentiate into chains of dormant , stress-resistant non-motile spores . The bld gene products are required for the transition from vegetative growth to aerial hyphae formation , while the whi gene products are required for the differentiation of aerial hyphae into spore chains . ( B ) S . venezuelae grown alone ( top row ) and beside S . cerevisiae ( middle row ) on YPD ( yeast extract-peptone-dextrose ) medium over 14 days . Bottom panels: scanning electron micrographs of S . venezuelae grown alone ( left ) , S . venezuelae on S . cerevisiae ( middle ) , and S . venezuelae beside S . cerevisiae ( right ) for 14 days on YPD agar medium . White bars: 5 µm . ( C ) S . venezuelae explorer cells growing up a rock embedded in agar ( left ) , and over a polystyrene barrier within a divided petri dish ( right , and schematic below ) . ( D ) S . venezuelae wild type and developmental mutants grown beside S . cerevisiae on YPD agar medium for 14 days . Top: S . cerevisiae , together with wild type and ∆bld mutant strains ( bld mutants cannot raise aerial hyphae and sporulate ) . Bottom: S . cerevisiae grown next to ∆whi mutant strains ( whi mutants can raise aerial hyphae but fail to sporulate ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 00310 . 7554/eLife . 21738 . 004Figure 1—figure supplement 1 . Explorer cells are hydrophilic . S . venezuelae growing on top of S . cerevisiae cells raise hydrophobic aerial hyphae and spores . These structures effectively repel aqueous droplets ( bromophenol blue dye dissolved in water ) . In contrast , explorer cells are hydrophilic , and application of aqueous droplets ( as above ) results in liquid dispersion . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 00410 . 7554/eLife . 21738 . 005Figure 1—figure supplement 2 . Phylogeny of exploratory streptomycetes . The phylogeny was created using aligned rpoB sequences from wild Streptomyces isolates ( WAC strains ) that exhibited exploratory growth . For comparison , we included the non-spreading S . coelicolor , S . lividans , S . avermitilis , S . griseus and S . clavuligerus . A maximum likelihood tree was built using RaxML with a GTRGAMMA model of nucleotide substitution , with 500 bootstrap replicates to infer support values of nodes . Output was created using FigTree . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 00510 . 7554/eLife . 21738 . 006Figure 1—figure supplement 3 . S . venezuelae grown beside diverse yeast strains . S . venezuelae was grown beside the indicated yeast strains on YPD agar medium for 14 days . Z . florentinus , S . castellii , S . cerevisiae , D . honsenii , and P . fermentas are soil isolates , while C . parapsilosis , C . albicans and C . neoformans are laboratory strains . Yeast strains able to induce S . venezuelae exploratory growth are labelled in blue text , and yeast strains unable to induce S . venezuelae exploratory growth are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 006 In this work , we identify a novel interaction between Streptomyces venezuelae and fungal microbes that induces a previously unknown mode of bacterial growth . We refer to this as ‘exploratory growth’ , whereby cells adopt a non-branching vegetative hyphal conformation that can rapidly traverse both biotic and abiotic surfaces . We show that part of the mechanism by which fungi induce exploratory growth involves glucose depletion of the growth medium . Remarkably , this novel mode of growth can be communicated to other – physically separated – streptomycetes through a volatile compound . Volatile signalling further alters cell propagation and survival of other bacteria . To explore interactions between Streptomyces and fungi , we cultured Streptomyces venezuelae alone or beside the yeast Saccharomyces cerevisiae on solid agar ( Figure 1B ) , and incubated these cultures for 14 days . As expected , during this time S . venezuelae on its own formed a colony of normal size . In contrast , when S . venezuelae was grown beside S . cerevisiae , its growth was radically different . During the first five days , the cells appeared to consume S . cerevisiae , before initiating a rapid outgrowth that led to S . venezuelae colonizing the entire surface of a 10 cm agar plate after 14 days . Remarkably , growth did not cease when physical obstructions were encountered: S . venezuelae cells were able to spread over rocks and polystyrene barriers ( Figure 1C ) . To gain insight into this phenomenon , we visualized the leading edge of the rapidly migrating S . venezuelae cells ( Video 1 ) . We found it initially progressed at a rate of ~1 . 5 µm/min . This is an order of magnitude faster than would be explained by growth alone , given that hyphal tip extension has been calculated to occur at a rate of 0 . 13 µm/min ( Richards et al . , 2012 ) . We refer to this rapid movement as 'exploratory growth' , and these spreading cells as 'explorers' , based on their ability to effectively transverse both biotic and abiotic surfaces . To further investigate the morphology of these explorer cells , we used scanning electron microscopy ( SEM ) to visualize S . venezuelae grown alone , S . venezuelae at the yeast interface , and S . venezuelae explorer cells , after 14 days of growth ( Figure 1B ) . We found S . venezuelae alone grew vegetatively , albeit without any obvious branches ( branching vegetative cells were observed during growth on other media types , as expected ) , whereas S . venezuelae growing on S . cerevisiae raised aerial hyphae and sporulated . Microscopic analysis of explorer cells revealed that they failed to branch and were reminiscent of aerial hyphae . Unlike aerial hyphae , however , these filaments were hydrophilic , based on their inability to repel aqueous solutions ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 21738 . 007Video 1 . Leading edge of S . venezuelae explorer cells over a 17 hr time course . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 007 To determine whether exploratory growth required classic developmental regulators ( the bld and whi gene products ) , we grew a suite of S . venezuelae developmental mutants beside S . cerevisiae to evaluate whether these mutations impacted colony spreading ( Figure 1D ) . Four S . venezuelae bld mutants ( bldC , D , M , N ) and five S . venezuelae whi mutants ( whiB , D , G , H , I ) were inoculated beside S . cerevisiae . Unexpectedly , all developmental mutant strains displayed a similar exploratory behaviour as wild type after 14 days , although the bldN mutant exhibited slower exploration than the other strains . The mutant strains did , however , differ in their growth on yeast , with the bld mutants failing to raise aerial hyphae , and the whi mutants failing to sporulate . This demonstrated that exploratory growth was distinct from the canonical Streptomyces life cycle , and represented a new form of growth for these bacteria . To determine whether this exploratory behaviour was unique to S . venezuelae , we inoculated other commonly studied streptomycetes beside S . cerevisiae . We found that well-studied Streptomyces species , including S . coelicolor , S . avermitilis , S . griseus , and S . lividans , failed to exhibit an analogous spreading behaviour when plated next to S . cerevisiae . We next tested 200 wild Streptomyces isolates , growing each beside S . cerevisiae . Of these , 19 strains ( ~10% ) exhibited exploratory growth similar to S . venezuelae . To determine whether this behaviour was confined to a particular Streptomyces lineage , we performed a phylogenetic analysis of these explorer-competent strains using rpoB sequences , and included non-exploratory model Streptomyces species as outgroups ( Figure 1—figure supplement 2 ) . We found S . venezuelae and these wild Streptomyces did not form a monophyletic group , suggesting that exploratory growth is wide-spread in the streptomycetes . We next sought to determine whether Streptomyces exploratory behaviour could be triggered by other fungi . S . venezuelae was inoculated beside laboratory strains of Candida albicans , Candida parapsilosis , and Crypotococcus neoformans , and beside wild soil isolates of S . cerevisiae , Zygosaccharomyces florentinus , Saccharomyces castellii , Pichia fermentans and Debaryomyces hansenii ( Figure 1—figure supplement 3 ) . We observed that all species , apart from C . neoformans and P . fermentans , induced S . venezuelae exploratory behaviour . This indicated that a broad range of microbial fungi could trigger exploratory growth . To understand how fungi could stimulate exploration , we took advantage of an S . cerevisiae haploid knockout collection containing 4309 individual knockout strains . Each S . cerevisiae mutant was pinned adjacent to S . venezuelae and after 10 days , S . venezuelae exploratory growth was assessed . We identified 16 mutants that were unable to promote S . venezuelae exploration ( Figure 2A ) . Of these , 13 had mutations affecting mitochondrial function , including eight in genes coding for enzymes in the tricarboxylic acid ( TCA ) cycle ( Figure 2A ) , three in genes whose products contribute to the mitochondrial retrograde signalling pathway , as well as two whose products are involved in mitochondrial metabolism . 10 . 7554/eLife . 21738 . 008Figure 2 . Yeast stimulates S . venezuelae exploratory growth by consuming glucose and inhibits it by acidifying the medium . ( A ) S . cerevisiae mutants that fail to stimulate S . venezuelae exploratory growth . Left: functional grouping of the exploration-deficient S . cerevisiae mutations . Asterisks indicate genes also identified in C . albicans as affecting S . venezuelae exploratory growth . Right: Mutations in S . cerevisiae TCA cycle-associated genes affect exploration after citrate production . For each interaction , the indicated S . cerevisiae mutant was grown beside wild type S . venezuelae for seven days on YPD agar medium . ( B ) Glucose concentration and pH associated with wild type and mutant S . cerevisiae strains grown on YPD agar medium . Glucose concentrations ( grey bars ) and pH ( blue squares ) were measured from medium alone , and beneath wild type , ∆LPD1 or ∆KGD2 S . cerevisiae strains grown on YPD medium for seven days . All values represent the mean ± standard error for four replicates . ( C ) Top: schematic of the experimental set up , with S . cerevisiae grown to the left of S . venezuelae on YPD medium . Two replicates are grown on each agar plate . Bottom: wild type , ∆LPD1 , and ∆KGD2 S . cerevisiae strains grown for 14 days beside wild type S . venezuelae on unbuffered YPD agar and YPD agar buffered to pH 7 . 0 with MOPS . ( D ) Wild type S . cerevisiae spotted beside wild type S . venezuelae and grown for 14 days on YPD agar medium plates supplemented with acetate or citrate , each buffered to pH 5 . 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 00810 . 7554/eLife . 21738 . 009Figure 2—figure supplement 1 . C . albicans gene mutations that affect S . venezuelae exploratory growth . S . venezuelae ( S . ven . ) was inoculated alone or beside wild type ( WT ) , ΔLPD1 , or ΔKGD2 C . albicans strains on YPD agar for 10 days . C . albicans mutant strains were from the GRACE collection of tetracycline repressible deletion mutants . The indicated amount of tetracycline was added to YPD medium to induce the C . albicans gene deletion phenotype . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 00910 . 7554/eLife . 21738 . 010Figure 2—figure supplement 2 . S . venezuelae grown alone on glucose-deficient medium exhibits similar exploratory growth to S . venezuelae growing next to yeast on glucose medium . S . venezuelae was grown either alone or beside S . cerevisiae on G+ ( glucose-containing ) agar medium , and alone on G- ( no glucose ) . Two replicates were spotted per plate , and plates were incubated for 10 days . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 010 We confirmed these mutant effects using Candida albicans strains carrying tetracycline-repressible haploid mutations ( Figure 2—figure supplement 1 ) . We grew four mutant strains adjacent to S . venezuelae , and found that two of them , ∆LPD1 and ∆KGD2 , also failed to stimulate S . venezuelae exploratory behaviour . As the products of these two genes act in the TCA cycle ( Figure 2A ) , these data collectively suggest that fungal respiration , and in particular TCA cycle function , influences exploratory growth in S . venezuelae . In considering how TCA cycle defects could influence S . venezuelae behaviour , we hypothesized that glucose uptake and/or consumption might play a role . We measured glucose levels of a YPD agar control , and compared this with YPD agar underneath S . cerevisiae . Uninoculated medium had 3 . 8 times as much glucose as S . cerevisiae-associated agar ( Figure 2B ) , confirming that S . cerevisiae consumed glucose during growth on YPD agar . This suggested that either glucose depletion by yeast , or some product of glucose metabolism , may trigger S . venezuelae exploratory growth . To test these possibilities , we first asked whether exploratory growth could be triggered by lowering glucose concentrations . We plated S . venezuelae on YP ( yeast extract-peptone ) in the presence ( G+ ) and absence ( G− ) of glucose ( Figure 2—figure supplement 2 ) . After 10 days , we found growth on G− medium permitted S . venezuelae exploration , irrespective of whether yeast was present . This implied that glucose repressed exploratory growth . We also tested glucose consumption by the S . cerevisiae LPD1 and KGD2 mutants . The products of these genes , along with that of KGD1 , comprise the 2-oxoglutarate dehydrogenase complex responsible for converting 2-oxoglutarate into succinyl-CoA in the TCA cycle ( Przybyla-Zawislak et al . , 1999 ) ( Figure 2A ) . We found wild type , ∆LPD1 and ∆KGD2 S . cerevisiae strains consumed similar levels of glucose ( Figure 2B ) , suggesting that other factors must be inhibiting S . venezuelae exploration when grown adjacent to these TCA cycle mutants . All TCA cycle-associated S . cerevisiae mutants that failed to stimulate S . venezuelae exploratory behaviour were blocked after the production of citrate in the TCA cycle ( Figure 2A ) . We hypothesized that this disruption might result in an accumulation of organic acids , and that S . cerevisiae mutants secreted these acids to maintain a neutral intracellular pH . We measured the pH of wild type , ∆LPD1 , and ∆KGD2 strains when grown on YPD ( G+ ) agar , and found wild type S . cerevisiae raised the agar pH from 7 . 0 to 7 . 5 , whereas both TCA cycle mutants lowered the agar pH to 5 . 5 ( Figure 2B ) . To test whether acid secretion by the S . cerevisiae LPD1 and KGD2 mutants prevented S . venezuelae exploratory growth , the two mutants were grown beside S . venezuelae on non-buffered YPD agar , and equivalent medium buffered to pH 7 . 0 ( Figure 2C ) . After 14 days growth on non-buffered plates , the S . cerevisiae mutants failed to stimulate S . venezuelae exploratory behaviour , whereas the same strains on buffered agar – which would counter the pH-lowering effects of the secreted acids – could now promote S . venezuelae exploration . To further verify this pH-dependent effect , we grew wild type S . cerevisiae beside S . venezuelae on YPD agar supplemented with citrate or acetate ( Figure 2D ) . We found that after 14 days , S . venezuelae spreading was inhibited , confirming that secreted acids inhibited S . venezuelae exploration . Collectively , these results suggested that S . venezuelae exploratory growth is a glucose- and acid-repressible phenomenon . Consistent with these observations , we also determined that S . venezuelae exploration was associated with a significant rise in pH: as S . venezuelae consumed the yeast , the medium pH rose from 7 . 0 to 8 . 0 , and once S . venezuelae exploratory growth initiated ( day 5 ) , the pH rose further to 9 . 5 ( Figure 3A ) . This increase in pH was also observed for S . venezuelae grown on G- medium ( in the absence of yeast ) ( Figure 3—figure supplement 1 ) , suggesting that the rise in pH was mediated by the Streptomyces cells . To determine whether high pH was sufficient to promote exploration , we inoculated S . venezuelae cells on YPD agar medium buffered to pH 9 . 0 . Exploration was not induced under these growth conditions ( Figure 3—figure supplement 2 ) . These data indicated that alkaline conditions were important but not sufficient for exploration , and further suggested that an adaptation phase was required during the transition to exploratory growth . 10 . 7554/eLife . 21738 . 011Figure 3 . The alkaline stress response is associated with S . venezuelae exploratory behaviour . ( A ) The surface area and medium pH associated with S . venezuelae explorer cells beside S . cerevisiae on YPD agar were measured and plotted every day for 14 days . ( B ) Schematic of the method used to identify genes required for S . venezuelae exploratory growth . S . venezuelae spores were subject to chemical mutagenesis , then screened on G- agar ( no glucose , exploration-permissive without S . cerevisiae ) for a lack of exploratory growth . Static colonies ( beige ) were grown beside S . cerevisiae ( pink ) on YPD medium to confirm a lack of exploratory growth . Genomic DNA was isolated from strains unable to initiate exploratory growth on G- agar , and when inoculated beside S . cerevisiae on YPD medium . Whole genome sequencing was performed to identify mutations responsible for the lack of exploratory growth . ( C ) Morphology of a mutant cytochrome bd oxidase S . venezuelae strain ( ∆cydCD ) and the corresponding complemented strain grown on YPD agar for 14 days . ( D ) Transcript levels for alkaline stress-responsive genes in S . venezuelae explorer cells ( grown beside S . cerevisiae on YPD medium ) , divided by levels for non-exploratory S . venezuelae cells ( grown alone on YPD medium ) . Transcript levels were normalized and differential expression was log2-transformed . The associated sven gene numbers are shown above the bar graphs . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 01110 . 7554/eLife . 21738 . 012Figure 3—figure supplement 1 . S . venezuelae grown alone raises the pH of glucose-deficient medium . S . venezuelae was grown alone on G- ( no glucose ) or G+ ( glucose-containing ) agar medium containing the pH indicator bromothymol blue . Two replicates were inoculated on each plate , and these were grown for 14 days . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 01210 . 7554/eLife . 21738 . 013Figure 3—figure supplement 2 . High pH alone does not stimulate S . venezuelae exploration . S . venezuelae was grown alone on YPD agar medium buffered to pH 9 . 0 using 50 , 100 or 200 mM borate . Two replicates were inoculated on each plate , and these were grown for 14 days . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 01310 . 7554/eLife . 21738 . 014Figure 3—figure supplement 3 . Complementation of explorer mutant phenotypes . Top row: EMS ( ethyl methanesulfonate ) mutagenesis-derived S . venezuelae explorer mutants containing point mutations in the cydABCD operon , grown beside S . cerevisiae . Bottom row: Explorer mutants complemented with a cosmid carrying the wild type cydABCD operon . Two replicates were inoculated on each plate . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 014 To investigate the genetic basis for this phenomenon we employed chemical mutagenesis , and screened for S . venezuelae mutants that failed to display exploratory behaviour when grown on G- medium ( where yeast is not required ) ( Figure 3B ) . Candidate non-spreading mutant colonies were identified , and were tested in association with S . cerevisiae on YPD ( G+ ) medium to confirm their inability to spread . Of the 48 exploration-defective mutants identified on G− medium , only three were also unable to spread when grown on YPD medium beside S . cerevisiae . This indicated that exploratory growth on G− agar may have distinct genetic requirements from exploratory growth on YPD ( G+ ) medium . We sequenced the genomes of wild type S . venezuelae and the three non-spreading mutants of interest ( those unable to spread on both G- medium alone and YPD ( G+ ) medium beside S . cerevisiae ) . Each mutant harbored point mutations in the sven_3713-3716 operon . This operon is predicted to encode subunits of the cytochrome bd oxidase complex ( cydA/sven_3713 and cydB/sven_3714 ) , along with an ABC transporter required for cytochrome assembly ( cydCD/sven_3715 ) ( Brekasis and Paget , 2003 ) . One strain had a mutation in sven_3715 ( H673Y ) , while the other two strains had mutations in sven_3713 ( Q186stop ) and were likely clonal . To ensure that these mutations were responsible for the exploration-defective phenotype , we complemented the exploratory growth defect in each mutant with a cosmid carrying an intact cydABCD operon , and confirmed that exploration was restored ( Figure 3—figure supplement 3 ) . We also deleted cydCD in a wild type S . venezuelae background , and confirmed that this strain was unable to initiate exploration when grown beside S . cerevisiae . As before , spreading could be restored to the mutant after introducing cydABCD on an integrating plasmid vector ( Figure 3C ) . These data indicated that the cytochrome bd oxidase complex was essential for S . venezuelae exploration . S . venezuelae , like many other bacteria , encodes two cytochrome oxidase complexes . The cytochrome bd oxidase catalyzes terminal electron transfer without a concomitant pumping of protons across the membrane , while the cytochrome bc1-aa3 complex requires proton transfer from the cytoplasm . The cytochrome bd oxidase functions as part of the alkaline stress response in other bacteria ( Krulwich et al . , 2011 ) . As we had established that alkaline conditions were a prerequisite for S . venezuelae exploration , we questioned whether other alkaline stress-responsive genes might be associated with exploratory growth . Using RNA-sequencing ( RNA-seq ) , we examined the transcription profiles of S . venezuelae alone , compared with S . venezuelae exploratory cells grown beside S . cerevisiae on YPD medium ( Figure 3D ) . The five gene clusters mostly highly upregulated in S . venezuelae explorer cells encoded the ATP synthase complex ( sven_5018-26; 7 . 6-fold increase relative to non-spreading ) , two predicted cation/proton antiporter complexes ( 95 . 9- and 85 . 3-fold increase relative to non-spreading for sven_5668-72 and sven_5764-68 , respectively ) , and two peptide transporters ( 17 . 4- and 38 . 3-fold increase relative to non-spreading for sven_4759-63 and sven_5150-54 , respectively ) ( Figure 3D ) . Higher expression of the cation/proton antiporters , alongside increased ATP synthesis , would be expected to enhance proton uptake into the cell; equivalent genes are upregulated as part of the alkaline stress response in other bacteria ( Krulwich et al . , 2011 ) . Amino acid catabolism is also upregulated under alkaline growth conditions in other bacteria ( Padan et al . , 2005 ) . Given the dramatically increased expression of the peptide transporters , we confirmed that exploratory growth required an amino acid source ( Supplementary file 1 ) . Collectively , these results suggest that exploration is coupled with a metabolic reprogramming that permits robust growth under highly alkaline conditions . S . venezuelae exploration is associated with high pH conditions , and our data suggested this rise in pH was promoted by S . venezuelae itself . We hypothesized that this pH effect could be mediated either through the secretion of diffusible basic compounds , or through the release of volatile organic compounds ( VOCs ) . To differentiate between these possibilities , we set up a two-compartment petri plate assay , where S . venezuelae was grown beside S . cerevisiae on YPD agar in one compartment , while the adjacent compartment contained uninoculated YPD agar ( Figure 4A ) . As a negative control , we set up an equivalent set of plates , only with S . venezuelae alone ( no yeast ) on YPD agar in the first compartment . In each case , the two compartments were separated by a polystyrene barrier . After 10 days , we measured the pH of the uninoculated YPD compartment , and found the compartment adjacent to S . venezuelae alone remained at pH 7 . 0 , whereas the one adjacent to S . venezuelae explorer cells had risen from pH 7 . 0 to 9 . 5 , indicating the explorer cells produced a basic VOC ( Figure 4A ) . 10 . 7554/eLife . 21738 . 015Figure 4 . Volatile organic compounds released by S . venezuelae raise the medium pH and induce exploratory growth in physically separated Streptomyces . ( A ) Effect of S . venezuelae explorer cells on pH of physically separated medium . Each compartment is separated by a polystyrene barrier . S . venezuelae and S . cerevisiae were grown in the left compartment of one plate ( left ) , while S . venezuelae alone was grown in the left compartment of the other plate ( right ) . After 10 days , bromothymol blue pH indicator dye was spread on the agar in the right compartment of each plate . Blue indicates VOC-induced alkalinity . ( B ) S . venezuelae was grown alone on YP ( G- agar ) in the left compartment , while the right compartment contained uninoculated YP ( G- ) agar . After seven days , the same pH indicator dye as in Figure 4A was spread over the agar in the right compartment . Blue represents a rise in pH above 7 . 6 . ( C ) Left: S . venezuelae alone was inoculated in each compartment . Right: S . venezuelae was grown beside S . cerevisiae in the left compartment , and S . venezuelae alone was grown in the right compartment . All strains were grown on YPD ( G+ ) agar medium for 10 days . ( D ) Top left: Wild Streptomyces isolate WAC0566 was grown alone in each compartment . Top right: WAC0566 was grown beside S . cerevisiae in the left compartment , and grown alone in the right compartment . Bottom left: S . venezuelae was grown beside S . cerevisiae in the left compartment , and WAC0566 was grown alone in the right compartment . Bottom right: WAC0566 was grown beside S . cerevisiae in the left compartment , while S . venezuelae was grown alone in the right compartment . All strains were cultured on YPD ( G+ ) agar medium for 10 days . ( E ) Schematic of the plate-based assay used to assess the effects of volatile-emitting solutions ( and controls ) on nearby Streptomyces colonies . H2O , TMA , or ammonia solutions were placed in a blue plastic dish , and S . venezuelae was spotted around each dish on YPD medium . Plates were incubated at room temperature for seven days . ( F ) Surface area and pH of S . venezuelae colonies grown on YPD medium around small dishes containing H2O or TMA solutions , as shown in Figure 4E . S . venezuelae was grown at room temperature for seven days on either unbuffered YPD medium or YPD medium buffered to pH 7 . 0 using MOPS . All values represent the mean ± standard error for four replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 01510 . 7554/eLife . 21738 . 016Figure 4—figure supplement 1 . The S . venezuelae cydCD mutant strain can explore in response to volatile signals produced by neighbouring explorer cells . Left: Wild type ( WT ) S . venezuelae was grown beside S . cerevisiae in the left compartment , and the S . venezuelae cydCD mutant strain was grown alone in the right compartment . Middle: Wild type S . venezuelae was grown beside S . cerevisia in the left compartment , and wild type S . venezuelae was grown alone in the right compartment . Right: The cydCD mutant was grown adjacent to S . cerevisiae in the left compartment ( where it sporulated but did not spread ) , while wild type S . venezuelae was grown alone in the right compartment . All strains were grown on YPD ( G+ ) agar medium for 10 days . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 01610 . 7554/eLife . 21738 . 017Figure 4—figure supplement 2 . Wild explorer Streptomyces species promote exploration in S . venezuelae using volatile signals . Using our two-quadrant assay , 13 independent wild Streptomyces isolates ( WAC strains ) were inoculated on G- agar , adjacent to S . venezuelae inoculated on G+ agar medium ( where no exploration was observed on its own ) . Each WAC strain was able to promote S . venezuelae exploration through the release of a volatile compound . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 01710 . 7554/eLife . 21738 . 018Figure 4—figure supplement 3 . The VOC produced by S . venezuelae explorer cells can be produced by liquid-grown ( G- ) S . venezuelae and WAC0566 cultures . All strains were grown in 48-well plates , with ( l ) indicating liquid cultures ( top rows ) , and ( s ) indicating solid YPD agar ( bottom rows ) . Liquid cultures were either G+ ( glucose-containing ) or G- ( no glucose ) , while all solid medium was G+ ( exploration repressive condition ) . Plates were grown shaking for three days . For all plates , we monitored exploration by strains growing on G+ agar ( bottom rows ) , in response to VOCs produced by the liquid-grown cultures . The top panel shows S . venezuelae ( left ) and the wild Streptomyces strain WAC0566 ( right ) grown in G+ liquid ( a condition where the VOC of interest is not expected to be produced ) . The middle panel shows the same strains , only grown in G- liquid ( where the VOC was predicted to be produced ) . The bottom panel shows the test for interspecies VOC production/response , with S . venezuelae and WAC0566 grown in G- liquid , opposite WAC0566 and S . venezuelae , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 018 To verify that the VOC was produced by S . venezuelae explorers and not by S . cerevisiae , we repeated the two-compartment assay with S . venezuelae grown alone on G- agar , a condition that also induced exploratory behaviour . We found that S . venezuelae growing alone on G- agar could alkalinize the adjacent YPD compartment . This confirmed that a basic VOC was produced by S . venezuelae explorer cells ( Figure 4B ) . Bacterial VOCs can influence a wide range of cellular behaviours . To determine whether the VOC produced by explorer cells represented an exploration-promoting signal for physically separated Streptomyces colonies , we leveraged our two-compartment assay , inoculating one with S . venezuelae beside S . cerevisiae on YPD agar , and the adjacent compartment with S . venezuelae on the same medium ( a condition where exploration by S . venezuelae otherwise requires yeast association ) . As expected , after 10 days , the S . cerevisiae-associated cells were actively spreading . Remarkably , the adjacent S . venezuelae cells ( in the absence of yeast ) had also initiated exploratory growth ( Figure 4C ) . As a negative control , S . venezuelae alone was grown in both compartments on YPD agar; spreading was not observed for cells grown in either compartment after 10 days ( Figure 4C ) . These data implied that S . venezuelae explorer cells released a VOC that effectively promoted exploratory growth in distantly located S . venezuelae cells . We tested whether our exploration-deficient cydCD mutant was able to respond to this VOC , and observed that despite its inability to explore when grown next to yeast , this mutant was capable of exploration when stimulated by neighbouring explorer cells ( Figure 4—figure supplement 1 ) . To determine whether S . venezuelae explorers used VOCs to potentiate exploration in other species , we again used our two-compartment assay . We cultured S . venezuelae with S . cerevisiae in one compartment , and tested whether these cells could stimulate exploratory growth of the wild Streptomyces isolate WAC0566 in the adjacent compartment ( Figure 4D ) ( WAC0566 initiates exploratory growth when cultured next to yeast , but fails to spread on its own; Figure 4D ) . Negative control plates were set up in the same way as before , with WAC0566 alone in both compartments . After 10 days , WAC0566 grown adjacent to S . venezuelae explorers initiated exploratory growth , and this was not seen for the negative control ( Figure 4D ) . This indicated that exploratory growth could be communicated to unrelated streptomycetes . We tested the volatile-mediated communication between these strains in a reciprocal experiment , and found that S . venezuelae exploration could also be stimulated by a VOC produced by yeast-associated WAC0566 ( Figure 4D ) . This inter-species promotion of S . venezuelae exploration was observed for at least 13 other wild Streptomyces strains ( Figure 4—figure supplement 2 ) . Importantly , VOC communication of exploratory growth was confined to those species with exploratory capabilities ( S . coelicolor failed to respond to the VOC elicitor ) . We determined that the exploration-promoting VOC could be produced by liquid-grown ( G- ) cultures , and that it stimulated exploratory growth by both S . venezuelae and WAC0566 ( Figure 4—figure supplement 3 ) . To rule out the possibility that any liquid-grown culture could promote exploration , we also grew S . venezuelae and WAC0566 in YPD ( G+ ) liquid medium , and found these cultures were unable to stimulate exploration . This suggested that VOC production was glucose-repressible , and its production correlated with growth conditions that promoted exploration . To determine the identity of the VOC , we grew S . venezuelae and WAC0566 in G+ and G- liquid culture for three days . We collected the supernatants of each culture , and assayed them using two-dimensional gas chromatography time-of-flight mass spectrometry ( GC×GC-TOFMS ) . From this , 1400 unique compounds were identified . To determine which compound ( s ) were responsible for promoting exploration , we applied a stringent filter , requiring the compound ( s ) to be: ( i ) present in at least 50% of S . venezuelae and WAC0566 exploration-inducing ( G- ) cultures; ( ii ) present in at least 10-fold greater abundance in exploration-inducing ( G- ) cultures versus static ( G+ ) cultures; and ( iii ) have at least a 60% similarity score to known compounds in the 2011 National Institute of Standards and Technology ( NIST ) Mass Spectral Library . We arrived at a list of 21 candidate compounds ( Supplementary file 1 ) . Of these , 12 were not detected in the negative controls ( G+ cultures ) . Within this group of 12 , only four were detected in 100% of S . venezuelae and WAC0566 exploration-promoting cultures: trimethylamine ( TMA ) , thiocyanic acid , 6-methyl-5-hepten-2-one , and 2-acetylthiazole . Notably , TMA was >10 fold more abundant than the other three compounds , and thus we focussed our initial investigations on this molecule . TMA is a volatile nitrogen-containing metabolite with a high pKa ( 9 . 81 ) . As we knew S . venezuelae produced a basic VOC , we hypothesized that TMA was responsible for promoting exploration . To test this possibility , we placed commercially-available TMA in a small plastic container at the centre of a YPD ( G+ ) agar plate , and then inoculated S . venezuelae at defined positions around this container ( Figure 4E ) . After seven days , S . venezuelae cultured adjacent to the TMA-emitting solutions had initiated exploratory growth , while those grown next to a water-containing control failed to spread . This implied that TMA was the VOC used by S . venezuelae and WAC0566 to elicit exploratory growth . TMA production is not well understood , although recent work has revealed two mechanisms by which it can be generated from quaternary amines . Acinetobacter sp . employ a carnitine oxygenase ( product of the cntAB gene cluster ) in converting L-carnitine into TMA ( Zhu et al . , 2014 ) , while Desulfovibrio desulfuricans converts choline into TMA using a choline-trimethylamine lyase ( encoded by the cutCD genes ) ( Craciun and Balskus , 2012 ) . S . venezuelae lacks any gene with similarity to cntA , and thus does not use an equivalent pathway to generate TMA . It does possess homologues of cutCD; however , these genes were more highly expressed ( ~5 fold ) in static S . venezuelae cultures ( where no TMA was ever detected ) , than in spreading cultures . This suggested that these gene products may not direct TMA production in S . venezuelae . TMA can also be produced upon biogenic reduction of trimethylamine N-oxide ( TMAO ) by TMAO reductases . Bacteria known to carry out this reaction typically encode one or more TMAO reducase operons , including some combination of torSTRCAD ( or torSTRCADE ) , torYZ , dmsABC , and ynfEFGH ( Dunn and Stabb , 2008; McCrindle et al . , 2005 ) . S . venezuelae encodes homologs to some of these genes [specifically torA ( top hit: SVEN_1326 ) , dmsAB ( top hit: SVEN_3040-3039 ) , and ynfEFG ( top hit: SVEN_3040 , 3040 and 3039 ) ] . In our RNA seq data , however , all of these genes ( along with more divergent homologs ) were expressed at extremely low levels , with equivalent levels for each gene being observed in both static and exploratory cultures . This suggested these gene products were unlikely to be involved in converting TMAO to TMA in S . venezuelae . To confirm that TMA could raise the pH of the growth medium in the same way as explorer cells , we measured the pH of non-inoculated YPD agar around dishes containing TMA , and found the pH rose from 7 . 0 to 9 . 5 . To test whether TMA induced exploratory growth by raising the pH , we repeated our plate assays described in Figure 4E , and buffered the agar to 7 . 0 using 50 or 200 mM MOPS ( Figure 4F ) . The pH of these plates rose to 8 . 0 ( as opposed to 9 . 5 on non-buffered plates ) , and TMA failed to induce S . venezuelae exploration to the same extent as on non-buffered plates . To further validate the pH-mediated effect of TMA , we tested whether ammonia ( another basic VOC ) had the same effect ( Figure 4E ) . After seven days , ammonia induced S . venezuelae exploratory growth , suggesting that VOC-mediated alkalinity stimulated Streptomyces exploration . TMA can alter the developmental program of streptomycetes , and is known to modify the antibiotic resistance profiles of bacteria ( Letoffe et al . , 2014 ) . Given the antibiotic production capabilities of Streptomyces bacteria , we wondered whether the release of TMA might also inhibit the growth of other bacteria . To explore this possibility , we set up a small petri dish of YPD agar inside a larger dish of YPD agar ( Figure 5 ) . S venezuelae and S . cerevisiae ( exploratory cultures ) or S . venezuelae alone ( static cultures ) were inoculated on the smaller dish , and plates were incubated for 10 days . The soil-dwelling bacteria Bacillus subtilis or Micrococcus luteus were then spread on the larger petri dish . Growth of B . subtilis and M . luteus in association with exploratory or static S . venezuelae cultures , was then assessed after overnight incubation . B . subtilis and M . luteus colony numbers were reduced by an average of 17 . 4% and 25 . 1% , respectively , on plates exposed to VOCs produced by exploratory S . venezuelae , relative to those grown adjacent to static cultures . We determined that the pH of medium adjacent to exploratory S . venezuelae had risen to 9 . 5 , suggesting that TMA and its pH-modulatory effects could be responsible for the growth-inhibition of these bacteria . To directly test the inhibitory potential of TMA , we set up an equivalent assay , where the TMA-producing S . venezuelae-S . cerevisiae combination was substituted with aqueous TMA solutions of varying concentrations . We spread B . subtilis , and M . luteus around the TMA-containing receptacles , and after seven days , quantified growth ( Figure 5C ) . We observed an approximately 50% drop in viable cells when exposed to 0 . 9% TMA , and in the case of B . subtilis , a further drop in viability was observed as TMA concentrations increased . This confirmed that TMA adversely affected the growth and survival of other soil bacteria . 10 . 7554/eLife . 21738 . 019Figure 5 . S . venezuelae VOCs inhibit the growth of other bacteria . ( A ) S . venezuelae was grown beside S . cerevisiae ( left ) or alone ( right ) on YPD agar in a small dish placed within a larger dish containing YPD medium . After 10 days , an indicator strain ( B . subtilis or M . luteus ) was spread around the dish . ( B ) Quantification of B . subtilis and M . luteus colonies following growth adjacent to static or explorer S . venezuelae cultures . Values represent the mean ± standard error for three replicates . The asterisk ( * ) indicates p<0 . 05 , as determined by a Student’s t-test . ( C ) Quantification of B . subtilis and M . luteus survival following incubation around small dishes containing TMA solutions at concentrations ranging from 0–22 . 5% . Plates were incubated at room temperature for two days . Percent survival indicates the OD600 of strains around wells containing 0 . 9% , 5 . 6% , or 22 . 5% TMA solutions compared to the OD600 of strains around wells containing H2O ( 100% survival ) . Values represent the mean ± standard error for three biological replicates , and each biological replicate is the average of four technical replicates . The asterisks ( *** ) indicate p<0 . 005 , as determined by a Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 019 S . venezuelae exploration is triggered by two key metabolic cues: glucose depletion and a rise in pH . We observed exploratory growth under low glucose conditions . In low-glucose areas of the soil , Streptomyces may initiate exploratory growth in an attempt to colonize environments with more readily available nutrients , whereas in high-glucose areas ( e . g . near plant roots , or in association with fruit ) ( Kliewer , 1965; Lugtenberg et al . , 1999; Romano and Kolter , 2005 ) , exploration may be less advantageous , initiating only after nearby fungi – or other microbes – consume the existing glucose supply . Microbial alteration of nutrient profiles is likely to be common in the soil environment ( e . g . Romano and Kolter , 2005 ) , and we expect that the exploratory growth away from glucose-depleted areas would provide a benefit analogous to that of motility systems in other bacteria . Although the mechanism underlying exploration remains to be elucidated , it may be linked to sliding motility given its apparently passive nature ( no appendages involved ) , and the fact that Streptomyces are known surfactant producers . S . venezuelae exploration is also promoted by a self-induced rise in extracellular pH . Alkaline growth conditions trigger morphological switches in a range of fungi , including the human pathogens C . albicans , C . neoformans , and Aspergillus fumigatus ( Bertuzzi et al . , 2014; Davis et al . , 2000; O'Meara et al . , 2014 ) . This is this first time this phenomenon has been observed in bacteria . Exploratory growth by Streptomyces cells is coordinated by the airborne compound TMA . TMA can further induce exploration in physically distant streptomycetes . Importantly , this volatile signal is not limited to S . venezuelae , and can be both transmitted and sensed by other Streptomyces species . Consequently , it is possible for Streptomyces to respond to TMA produced by other bacteria and initiate exploratory growth under conditions where glucose concentrations are high and/or glucose-titrating organisms are absent . Developmental switching in response to VOC eavesdropping has not been previously reported , but exploiting community goods in this way is not unprecedented . For example , quorum signals and siderophores produced by one organism can be taken up or used by others ( Lyons and Kolter , 2015; Traxler et al . , 2012 ) . The VOC repertoire of microorganisms appears to be vast ( Chuankun et al . , 2004; Insam and Seewald , 2010; Kai et al . , 2009; Schöller et al . , 2002; Schulz and Dickschat , 2007; Wilkins and Schöller , 2009 ) . Volatile compounds have historically been implicated in the ‘avoidance responses’ of fungi , promoting their growth away from inanimate objects ( Cohen et al . , 1975; Gamow and Böttger , 1982 ) . Increasingly , these compounds are now being found to have important roles in communication between physically separated microbes ( Audrain et al . , 2015; Bernier et al . , 2011; Briard et al . , 2016; Kim et al . , 2013; Letoffe et al . , 2014; Schmidt et al . , 2015 , 2016; Tyc et al . , 2015; Wang et al . , 2013; Wheatley , 2002 ) . A range of fungi use the volatile alkaline compound ammonia to induce morphological switches in other fungi , and to mediate inhibition of neighbouring colonies ( Palková et al . , 1997 ) . Our observations suggest that VOCs may also be key bacterial morphological determinants , communicating developmental switches both within and between different microbial species . In addition to serving as communication signals , VOCs may also provide their producing organisms with a competitive advantage in the soil . Volatile molecules can modulate the antibiotic resistance profiles of bacteria ( Letoffe et al . , 2014 ) , and can themselves have antifungal or antibacterial activity ( Schmidt et al . , 2015 ) . TMA is a particularly potent example . Here , we show that exposure of other bacteria to TMA inhibits their growth , while previous work has revealed that TMA exposure increases bacterial sensitivity to aminoglycoside antibiotics . Notably , Streptomyces synthesize an extraordinary range of antibiotics , including many aminoglycosides . Thus in the soil , Streptomyces-produced TMA may have direct antibacterial activity , in addition to sensitizing bacteria to the effect of Streptomyces-produced antibiotics . The ability of Streptomyces to modulate the growth of other soil-dwelling bacteria during exploratory growth would maximize their ability to colonize new environments , and exploit whatever nutrients are present . Exploratory growth represents a powerful new addition to the Streptomyces developmental repertoire , and one that appears to be well-integrated into the existing life cycle . When grown next to yeast , explorer cells emerge from a mass of sporulating cells ( Figure 6 ) . This functional differentiation represents an effective bet-hedging strategy , whereby spreading explorer cells scavenge nutrients for the group , while the sporulating cells provide a highly resistant genetic repository , ensuring colony survival in the event of failed exploration . Explorer cells resemble vegetative hyphae , in that their surface is hydrophilic; however , unlike traditional vegetative hyphae , explorer cells do not appear to branch . We presume that explorer cells dispense with frequent branching as a trade-off for the ability to rapidly spread to new environments . Exploratory growth also occurs independently of the typical bld- and whi-developmental determinants , supporting the notion that this is a unique growth strategy . It is possible , however , given the slower exploration observed for bldN mutants ( where bldN encodes a sigma factor ) , that BldN regulon members help to facilitate the exploration process . 10 . 7554/eLife . 21738 . 020Figure 6 . New model for Streptomyces development . When S . venezuelae is grown alone on glucose-rich medium S . venezuelae exploratory growth is repressed ( left ) . When S . venezuelae is grown beside S . cerevisiae or other yeast on glucose-rich medium ( right ) , the yeast metabolizes glucose , relieving the repression of S . venezuelae exploration . S . venezuelae explorer cells produce the volatile pheromone TMA , which raises the pH of the medium from 7 . 0 to 9 . 5 . Explorer cells activate alkaline stress genes to withstand the alkaline pH . TMA , and its associated medium alkalinisation , can induce exploratory growth in physically separated Streptomyces . DOI: http://dx . doi . org/10 . 7554/eLife . 21738 . 020 While we observed exploratory growth in a subset of Streptomyces species , it is possible that this capability is more broadly conserved and is stimulated by different conditions than those investigated here . Indeed , microbes are abundant in the soil , and interactions between different organisms within these communities are likely to be more the norm than the exception . Our work illustrates the importance of inter-species interactions in bacterial development , as a key to revealing novel growth strategies . It also emphasises the need to consider long-range communication strategies , in the form of volatile compounds , which may play widespread roles in regulating development and metabolic activities in microbial communities . Strains , plasmids and primers used in this study are listed in Supplementary file 1 . S . venezuelae ATCC 10712 was grown on MYM ( maltose-yeast extract-malt extract ) agar medium for spore stock generation . Spreading was investigated during growth on the surface of YPD ( yeast extract-peptone-dextrose/glucose ) agar , glucose-deficient YP ( G- ) agar , yeast extract agar supplemented with different amino acid sources ( tryptone or 2% casamino acids ) or YPD/G- agar medium supplemented with citrate , acetate , borate or MOPS buffer . All strains were grown at 30°C , apart from the TMA experiments which were conducted at room temperature in a fume hood . S . cerevisiae strain BY4741 ( MATa; his3∆1; leu2∆0 ura3∆0 met15∆0 ) was grown on the same spreading-investigative media at 30°C or room temperature . Prior to plating S . venezuelae and S . cerevisiae together , S . venezuelae was cultured in liquid MYM at 30°C , while S . cerevisiae was grown in liquid YPD at 30°C overnight . Three microliters of S . venezuelae cultures were applied to the right of 3 µL S . cerevisiae on the surface of YPD agar medium , and plates were then incubated at 30°C or room temperature for up to 14 days SEM was used to examine strains grown on YPD or MYM agar for up to 14 days . Samples were prepared and visualized using a TEMSCAN LSU scanning electron microscopy as described previously ( Haiser et al . , 2009 ) . To monitor the rate of exploratory growth ( Video 1 ) , an Olympus SZX12 Sterioscope and CoolSNAP HQ photometric camera were used to capture 70 frames of growth over the course of 17 hr . rpoB ( Guo et al . , 2008 ) was amplified from each of the 19 exploration-competent wild isolates using primers RpoBPF and RpoBPR ( Supplementary file 1 ) , before being sequenced using RpoBF1 and RpoBR1 ( Supplementary file 1 ) . Trimmed rpoB sequences were aligned using Mafft version 7 . 2 . 6 . 6 . A maximum likelihood tree was built using RAxML version 8 . 2 . 4 ( Stamatakis , 2006 ) , using a GTRGAMMA model of nucleotide substitution , with 500 bootstrap replicates to infer support values of nodes . Outputs were visualized using FigTree . Overnight cultures of S . venezuelae were spotted onto rectangular plates containing YPD agar ( OmniTray: Nunc International ) using a 384-pin replicator . Each strain of a S . cerevisiae BY4741 haploid deletion library was inoculated beside an individual S . venezuelae colony using a 384-pin replicator . Plates were grown for five days at 30°C and screened for an absence of S . venezuelae exploratory growth . Yeast mutants unable to stimulate S . venezuelae exploratory growth were re-tested on individual YPD agar plates . For C . albicans deletion screens , C . albicans GRACE collection tetracycline repressible deletion mutants ( Roemer et al . , 2003 ) were inoculated beside S . venezuelae on YPD agar plates . Mutants were induced using 1 or 5 µg/mL tetracycline , which is below the minimum inhibitory concentration of tetracycline for S . venezuelae . Measurements of glucose levels beneath S . cerevisiae colonies and in YPD alone were performed using a Glucose ( GO ) Assay Kit ( Sigma ) . For all experiments , pH levels of solid agar were measured using one or a combination of pH sticks and the pH indicator dye bromothymol blue ( Sigma , St Louis , MO ) . Approximately 108 S . venezuelae spores were added to 1 . 5 mL 0 . 01 M KPO4 at pH 7 . 0 . Spores were centrifuged and resuspended in 1 . 5 mL 0 . 01 M KPO4 at pH 7 . 0 . The spores were then divided into two 750 µL aliquots in screw-cap tubes . As a control , 25 µL H2O was added to one aliquot , while 25 µL ethyl methanesulfonate ( EMS , Sigma , M0880 ) was added to the other aliquot . Tubes were vortexed for 30 s , and incubated shaking at 30°C for 1 hr , with an additional inversion being performed every 10 min . Spores were centrifuged at 3381 ×g for 3 min at room temperature , prior to being resuspended in 1 mL freshly made and filter-sterilized 5% w/v sodium thiosulfate solution . Spores were washed twice in 1 mL H2O , after which they were resuspended in 1 mL H2O . For each tube , a dilution series ranging from 10−4 to 10−8 was made using H2O , and 100 µL of each dilution was then spread onto MYM agar plates and incubated for three days at 30°C . Individual colonies were counted to ensure that survival of the EMS-treated spores was , at most , 50% that of the untreated ( H2O ) control . Colonies were collected from plates inoculated with EMS-treated spores , and were screened for loss of spreading capabilities on G- agar plates . Select mutants were then tested for their inability to spread when plated next to yeast; those mutants that also failed to initiate spreading in the presence of S . cerevisiae were grown in liquid MYM , and chromosomal DNA was extracted using the Norgen Biotek Bacterial Genomic DNA Isolation kit for downstream sequencing . Using the Illumina Nextera XT DNA sample preparation kit , DNA libraries were prepared for three non-exploratory S . venezuelae mutants , alongside their wild type S . venezuelae parent . Whole genome-sequencing was performed on an Illumina MiSeq instrument ( Illumina , San Diego , CA , USA ) using 150 bp paired-ends reads . Reads were aligned to the S . venezuelae reference genome using Bowtie 2 ( Langmead and Salzberg , 2012 ) and were converted to BAM files using SAMtools ( Li et al . , 2009 ) . Single nucleotide polymorphisms ( SNPs ) were called using SAMtools mpileup and bcftools , and SNP locations , read depth , and identities were generated using VCFtools ( Danecek et al . , 2011 ) . An in-frame deletion of sven_3715-3716 was generated using ReDirect technology ( Gust et al . , 2003 ) . The coding sequence was replaced by an oriT-containing apramycin resistance cassette . The gene deletion was verified by PCR , using combinations of primers located upstream , downstream and internal to the deleted genes ( see Supplementary file 1 ) . The cydCD mutant phenotype was complemented using a DNA fragment encompassing the WT genes , sven_3713-3714 , and associated upstream and downstream sequences ( see Supplementary file 1 ) , cloned into the integrating plasmid vector pSET152 . To control for any phenotypic effects caused by plasmid integration , pSET152 alone was introduced into wild type and the cydCD mutant strains , and these strains were used for phenotypic comparison with the complemented mutant strain . RNA was isolated as described previously from two replicates of S . venezuelae explorer cells growing beside S . cerevisiae for 14 days , and two replicates of S . venezuelae alone grown for 24 hr on YPD agar plates ( we were unable to isolate high quality RNA from S . venezuelae alone at later time points ) . For all four replicates , ribosomal RNA ( rRNA ) was depleted using a Ribo-zero rRNA depletion kit . cDNA and Illumina library preparation were performed using a NEBnext Ultra Directional Library Kit , followed by sequencing using unpaired-end 80 base-pair reads using the HiSeq platform . Reads were aligned to the S . venezuelae genome using Bowtie 2 ( Langmead and Salzberg , 2012 ) , then sorted , indexed , and converted to BAM format using SAMtools ( Li et al . , 2009 ) . BAM files were visualized using Integrated Genomics Viewer ( Robinson , 2011 ) , and normalization of transcript levels and analyses of differential transcript levels were conducted using Rockhopper ( McClure et al . , 2013 ) . RNA-seq data has been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE86378 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? token=idmrgcmexranpun&acc=GSE86378 ) . S . venezuelae and WAC0566 were grown in liquid YPD ( G+ ) or YP ( G- ) for three days . For each strain and condition , six biological replicates were grown , and for each , three technical replicates were analyzed . Four milliliters of each culture supernatant were transferred to 20 mL air-tight headspace vials , which were stored at −20°C prior to volatile analysis . Headspace volatiles were concentrated on a 2 cm triphasic Divinylbenzene/Carboxen/Polydimethylsiloxane ( DVB/CAR/PDMS ) solid-phase microextraction ( SPME ) fiber ( Supelco , Bellenfonte , PA ) ( 30 min , 50°C , 250 rpm shaking ) . Volatile molecules were separated , identified , and relatively quantified using two-dimensional gas chromatography time-of-flight mass spectrometry ( GC×GC-TOFMS ) , as described previously ( Bean et al . , 2012; Rees et al . , 2016 ) . The GC×GC-TOFMS ( Pegasus 4D , LECO Corporation , St . Joseph , MI ) was equipped with a rail autosampler ( MPS , Gerstel , Linthicum Heights , MD ) and fitted with a two-dimensional column set consisting of an Rxi−624Sil ( 60 m × 250 μm×1 . 4 μm ( length × internal diameter × film thickness ) ; Restek , Bellefonte , PA ) first column followed by a Stabilwax ( Crossbond Carbowax polyethylene glycol; 1 m × 250 μm×0 . 5 μm; Restek , Bellefonte , PA ) second column . The main oven containing column one was held at 35°C for 0 . 5 min , and then ramped at 3 . 5 °C/min from 35°C to 230°C . The secondary oven containing column 2 , and the quad-jet modulator ( 2 s modulation period , 0 . 5 s alternating hot and cold pulses ) , were heated in step with the primary oven with +5°C and +25°C offset relative to the primary oven , respectively . The helium carrier gas flow rate was 2 mL/min . Mass spectra were acquired over the range of 30 to 500 a . m . u . , with an acquisition rate of 200 spectra/s . Data acquisition and analysis was performed using ChromaTOF software , version 4 . 50 ( LECO Corp . ) . Chromatographic data was processed and aligned using ChromaTOF . For peak identification , a signal-to-noise ( S/N ) cutoff was set at 100 , and resulting peaks were identified by a forward search of the NIST 2011 Mass Spectral Library . For the alignment of peaks across chromatograms , maximum first and second-dimension retention time deviations were set at 6 s and 0 . 15 s , respectively , and the inter-chromatogram spectral match threshold was set at 600 . Analytes that were detected in greater than half of exploration-promoting Streptomyces cultures ( grown in G- medium ) and not detected in media controls or S . venezuelae grown in G+ medium ( failed to promote exploration ) , were considered candidate molecules associated with the phenotype of interest . Aqueous solutions ( 1 . 5 mL ) of commercially available TMA solutions ( Sigma ) , ammonia solutions ( Sigma ) or water ( negative control ) were added to small , sterile plastic containers and placed in a petri dish containing 50 mL YPD agar . TMA solutions were typically diluted to 11 . 5% w/v , although concentrations as low as 0 . 9% were able to promote spreading and inhibit the growth of other bacteria . Ammonia solutions of 0 . 1–1 M were used , and all were able to induce spreading . S . venezuelae was inoculated around the small vessels , after which the large petri dish was closed and incubated in the fume hood at room temperature for up to 10 days . For buffering experiments , YPD plates were supplemented with 50 or 200 mM MOPS buffer ( pH 7 . 0 ) . Medium pH was measured as above , while colony surface areas were measured using ImageJ ( Abràmoff et al . , 2004 ) . For bacterial survival assays around TMA-containing vessels , B . subtilis and M . luteus strains were grown overnight in LB medium , before being subcultured to an OD600 of 0 . 8 . One hundred microliters of each culture were then spread on YPD agar plates , adjacent to water or TMA-containing vessels . For assays to measure how S . venezuelae explorer VOCs affect the survival of other bacteria , S . venezuelae was grown alone or beside S . cerevisiae in a small petri dish containing YPD agar . This small dish was placed inside a larger dish containing YPD agar . Plates were grown for 10 days , before B . subtilis and M . luteus were subcultured to an OD600 of 0 . 8 , and diluted 1/10 000 . Fifty microliters of each culture were then spread on the larger plate containing YPD agar , and colonies were quantified after overnight growth . To test the effect of TMA on B . subtilis and M . luteus growth , these indicator strains were grown overnight in LB medium , before being subcultured to an OD600 of 0 . 8 . One hundred microlitres were spread around wells containing 1 . 5 mL solutions of TMA at different concentrations on YPD ( water control , 0 . 9% , 5 . 6% and 22 . 5% ) . Plates were incubated for two days at room temperature in the fume hood , before cells were scraped into 2 mL YPD and vigorously mixed . Dilution series were used to measure the OD600 of the resulting cell suspensions . Error bars indicate standard error of three biological replicates , and four technical replicates of each .	Soil is home to many bacteria . In fact , soil gets it characteristic ‘earthy’ smell from a common type of soil bacteria known as Streptomyces . Remarkably , Streptomyces are also the original sources of most of the antibiotics that are prescribed by doctors to treat bacterial infections . Scientists have been studying Streptomyces for over 70 years , and in all this time , there has been unanimous agreement on how these bacteria grow . That is to say that , unlike most other bacteria , Streptomyces grow like plants: they don’t move , and instead produce spores that are dispersed like seeds . This stationary lifestyle makes these bacteria somewhat vulnerable to predators , and so it is thought that Streptomyces make antibiotics to help protect themselves from other bacteria that are able to move around in the soil . However , this established view of Streptomyces growth has now been turned on its head because Jones et al . have discovered that Streptomyces bacteria can indeed move when grown in the presence of fungi . Specifically , when a species of Streptomyces is grown with yeast , some of the bacteria start to explore their environment , move over top of other bacteria and up hard surfaces to heights that would be the equivalent of humans scaling Mount Everest . Unexpectedly , Jones et al . also found that these “explorer” Steptomyces can communicate with nearby Streptomyces bacteria with a perfume-like airborne signal and convince their relatives to begin exploring too . Furthermore , while this volatile signal promotes the growth of Streptomyces , it adversely affects other bacteria and makes them sicker such that they are less able to grow and survive . Together these findings reveal new ways that bacteria and other microbes can interact and communicate with each other . They also emphasise that researchers will need to consider such long-range communication strategies if they hope to better understand microbial communities .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2017	Streptomyces exploration is triggered by fungal interactions and volatile signals
Cytoplasmic dynein , a microtubule-based motor protein , transports many intracellular cargos by means of its light intermediate chain ( LIC ) . In this study , we have determined the crystal structure of the conserved LIC domain , which binds the motor heavy chain , from a thermophilic fungus . We show that the LIC has a Ras-like fold with insertions that distinguish it from Ras and other previously described G proteins . Despite having a G protein fold , the fungal LIC has lost its ability to bind nucleotide , while the human LIC1 binds GDP preferentially over GTP . We show that the LIC G domain binds the dynein heavy chain using a conserved patch of aromatic residues , whereas the less conserved C-terminal domain binds several Rab effectors involved in membrane transport . These studies provide the first structural information and insight into the evolutionary origin of the LIC as well as revealing how this critical subunit connects the dynein motor to cargo . Molecular motors transport a variety of cargos , including membranous organelles , proteins , mRNA , and chromosomes , along cytoskeletal tracks throughout the cell . Long distance transport in animal cells occurs primarily along microtubule tracks using kinesins ( primarily , plus-end-directed ) and cytoplasmic dyneins ( minus-end-directed ) as motors ( Allan , 2011 ) . The cytoplasmic dyneins are divided into two distinct subclasses . Cytoplasmic dynein 1 is employed broadly for many different types of retrograde microtubule transport within animal cells ( Allan , 2011 ) . Cytoplasmic dynein 2 , in contrast , specifically carries out retrograde intraflagellar transport in cilia and flagella ( Ishikawa and Marshall , 2011 ) . Cytoplasmic dynein is a large homodimer that consists of a heavy chain ( >500 kDa ) and several smaller associated subunits , each present in two copies . The dynein heavy chain includes an N-terminal elongated ‘tail’ domain followed by ∼350 kDa motor domain that also contains the microtubule-binding domain ( Carter et al . , 2011; Kon et al . , 2012; Schmidt et al . , 2012 ) . The heavy chain ‘tail’ domain binds the associated subunits , which include the intermediate chain ( IC ) , the light intermediate chain ( LIC ) , and the light chains ( LC ) : Tctex1 , LC8 , and LC7/roadblock ( Allan , 2011 ) . A variety of studies have implicated these associated subunits in cargo binding , either directly , such as with rhodopsin ( Tai et al . , 1999 ) , or indirectly by adaptors , such as dynactin ( Karki and Holzbaur , 1999; Schroer , 2004 ) . The LIC subunits , which are present in all cytoplasmic dyneins described thus far but absent from axonemal dyneins ( Inaba , 2007 ) , are thought to play important roles in cargo transport . Invertebrates contain a single cytoplasmic dynein 1 LIC isoform while mammals have two LIC genes ( LIC1 and LIC2 ) ( Hughes et al . , 1995 ) , which may define two distinct cytoplasmic dynein 1 populations ( Tynan et al . , 2000a; Tan et al . , 2011 ) . Cytoplasmic dynein 2 is associated with a third LIC isoform ( LIC3 ) ( Grissom et al . , 2002 ) that is required for retrograde intraflagellar transport ( Hou et al . , 2004 ) . Knockdown studies have implicated LIC1 and 2 in membrane trafficking toward the endosomal-recycling compartment ( ERC ) in the cell center ( Horgan et al . , 2010b ) , ER export ( Kong et al . , 2013 ) , lysosomal localization and morphology ( Tan et al . , 2011 ) , and axonal retrograde transport ( Koushika et al . , 2004 ) . During mitosis , LIC1/2 are required for many cytoplasmic dynein 1 functions including centrosome anchoring , dynein localization to the kinetochore , progression through the spindle assembly checkpoint , and chromosome alignment ( Mische et al . , 2008; Sivaram et al . , 2009; Raaijmakers et al . , 2013 ) . Several proteins implicated in membrane trafficking have been suggested to interact with LICs , including Rab4a ( Bielli et al . , 2001 ) and FIP3 , a Rab11-family interacting protein ( Horgan et al . , 2010a , 2010b ) . LIC1 and LIC2 appear to have largely redundant roles ( Kong et al . , 2013; Raaijmakers et al . , 2013 ) , although the centrosome protein pericentrin was reported to only bind to LIC1 ( Tynan et al . , 2000a ) and Par3 binds specifically to LIC2 ( Schmoranzer et al . , 2009 ) . In addition to mediating cargo binding , the LIC may be important for the stability of the dynein heavy chain; the LIC is the most stably bound subunit to the heavy chain ( King et al . , 2002 ) and is necessary for the stable expression and solubility of the recombinant human dynein heavy chain ( Trokter et al . , 2012 ) . Sequence comparisons show that the LICs are most conserved in the N-terminal half of the protein and that this conserved region contains a P-loop , a canonical nucleotide-binding sequence ( Perrone et al . , 2003 ) . Based upon the high similarity to the P-loops of the ATP-hydrolyzing ABC transporters , it was suggested that the LIC may be an ATPase ( Hughes et al . , 1995 ) . However , bioinformatic databases , such as Pfam ( Finn et al . , 2014 ) , place the LIC in the same family as Ras-like , GTP-binding proteins . Several studies investigated the role of potential nucleotide hydrolysis by mutating a critical lysine residue in the P-loop but did not find a phenotype on cytoplasmic dynein function ( Tynan et al . , 2000a , 2000b; Yoder and Han , 2001; Hou et al . , 2004 ) . However , beyond sequence analysis , little biochemical or structural information exists for the LIC subunits . In this study , we report the crystal structure of the conserved N-terminal domain of the LIC from a thermophilic hyphal fungus , Chaetomium thermophilum , and show that it is composed of a canonical G protein fold . However , unlike most small GTP-binding proteins , the nucleotide pocket is empty , the P-loop exhibits a closed conformation and our biochemical experiments confirm that the fungal LIC G domain does not bind nucleotide . In contrast , we find that the human LIC1 G domain is capable of binding guanine nucleotides , particularly GDP . Our results reveal that the cytoplasmic dynein LIC evolved from the small G protein superfamily and show biochemical differences between fungal and metazoan LICs that may play a role in dynein cargo transport . We further show how the LIC links the dynein motor domain to its multiple cargos . We originally attempted to crystallize full-length human LIC1 but were unable to obtain diffraction-quality crystals . Given prior success in crystallizing proteins from the thermophilic fungus Chaetomium thermophilum ( Amlacher et al . , 2011 ) , we were inspired to crystallize the LIC from this organism . C . thermophilum possesses a gene that encodes a 4413 a . a . protein with ∼50% and 30% sequence identity to the heavy chain of cytoplasmic dynein 1 from human and Saccharomyces cerevisiae , respectively . It also contains a LIC gene ( EGS22626 . 1 ) that encodes for a 547-residue protein with 23% sequence identity and of similar size to human LIC1 . For comparison , the S . cerevisiae LIC gene , which was originally identified based on a dynein-like nuclear migration phenotype ( Lee et al . , 2005 ) , has 18% sequence identity with human LIC1 . The sequence identity was higher with LICs from other hyphal fungi such as Neurospora crassa ( 75% ) and Aspergillus nidulans ( 55% ) ( Figure 1—figure supplement 1 ) . Sequence alignments show that the conservation is greatest in an N-terminal region of ∼300 residues ( predicted molecular weight of ∼33 kDa ) ( Figure 1A; Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 03351 . 003Figure 1 . The dynein light intermediate chain has a Ras-like fold . ( A ) Diagram depicting the approximate range of conservation among all LICs ( residue numbers with respect to the C . thermophilum LIC sequence ) . ( B ) The purified full-length C . thermophilum LIC ( FL LIC ) and the crystallized protein ( xtals ) were resolved on an SDS-PAGE gel and silver-stained , revealing proteolysis during crystallization . Proteolysis with chymotrypsin ( +Chy ) ( overnight at 1:250 moles protease: LIC ) produced similar sized fragments to those seen in the crystal . The asterisk marks a contaminating 75 kDa protein . ( C ) The 2 . 1 Å structure of the C . thermophilum LIC is shown with the N-terminus oriented to the front and the C-terminus towards the back . β-strands and α-helices are labeled with respect to comparable elements in Ras . Elements that align with Ras are teal , and elements not found in Ras are yellow . ( D ) A topology map of LIC secondary structure is shown , and the color scheme corresponds to ( C ) . Numbers with ‘a’ are additional inserts not seen in Ras . The P-loop , switch 1 , switch 2 , G4 , and G5 motifs are labeled based on where they are found structurally ( not based on sequence ) . Regions absent from the electron density are labeled with a dashed line . ( E ) Structural alignment of LIC with Ras-GMPPNP ( PDB 52P1 ) ( Pai et al . , 1990 ) . Alignment was performed using chimera after removing the C-terminal helices and loops in the LIC structure . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 00310 . 7554/eLife . 03351 . 004Figure 1—figure supplement 1 . Sequence alignment of full-length LICs . The full-length sequences of C . thermophilum LIC , Neurospora crassa LIC , Aspergillus nidulans LIC , H . sapiens LIC1 , and H . sapiens LIC2 were aligned using MafftWS ( algorithm E-INS-I , accuracy oriented ) ( Katoh and Standley , 2013 ) . Percentage identity is depicted with a gradation of blue shading ( dark blue is 100% identical ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 00410 . 7554/eLife . 03351 . 005Figure 1—figure supplement 2 . Structural and sequence similarity with the Rabs . A ) The C . thermophilum LIC G domain was aligned with Mus musculus Rab33 ( PDB: 2G77 ) , H . sapiens Rab28 ( PDB: 3E5H ) , and H . sapiens Rab32 ( PDB: 4CYM ) using the Dali server ( Holm and Rosenstrom , 2010 ) . Only the core of the LIC G domain is shown as in Figure 1E . ( B ) The sequences of C . thermophilum LIC and the Rabs in ( A ) were structurally aligned by the Dali server ( Holm and Rosenstrom , 2010 ) . If at least two of the three Rabs have amino acids that are similar to the aligned C . thermophilum LIC residue , the column is shaded light purple; if all proteins have an identical residue , the column is shaded blue . Common secondary structure and the G motifs are denoted . The switch 2 loop ( G3 motif ) is much longer in the C . thermophilum LIC structure and is underlined in red . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 00510 . 7554/eLife . 03351 . 006Figure 1—figure supplement 3 . Phylogenetic analysis of the LIC in the Ras superfamily . An unrooted maximum likelihood phylogenetic tree of 198 sequences was generated by PhyML to reveal the placement of dynein LIC in the Ras superfamily . The red circles denote branches with greater than 80% bootstrap support ( 300 bootstraps total ) . The Dali server's top three hits for structure similarity to the C . thermophilum LIC structure are red . Ras subfamilies are labeled , and the LICs are denoted in blue . The abbreviations are as follows: ARATH , Arabidopsis thaliana; PLAF7 , Plasmodium falciparum; SCHPO , Schizosaccharomyces pombe; CAEEL , Caenorhabditis elegant; DROME , Drosophila melanogaster; Nve , Nematostella vectensis; Bfl , Branchiostoma floridae; Cin , Ciona intestinalis; Xtr , Xenopus tropicalis . The numbers adjacent to the three-lettered codes Nve , Bfl , Xtr , and Cin are accession numbers found in the DOE Joint Genome Institute database . The uppercase abbreviations are Uniprot codes . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 006 The C . thermophilum LIC expressed well in Escherichia coli and could be purified to near homogeneity ( see ‘Materials and methods’ ) . Crystals of the C . thermophilum LIC , which appeared after approximately 1 month , diffracted to 2 . 1 Å and a complete X-ray diffraction dataset was obtained ( Table 1 ) . However , these crystals were difficult to reproduce . When the crystals were analyzed by SDS-PAGE , three polypeptides corresponding to molecular weights of approximately 33 , 27 , and <15 kDa were observed but not the full-length 60 kDa LIC protein ( Figure 1B ) . Mass spectrometry showed that the 33 kDa and 27 kDa polypeptides contained sequences that resided within the conserved N-terminal domain . These results suggested that the full-length LIC was being digested by a minor contaminating protease over the course of a month and that the protease-resistant region that crystallized corresponded to the most conserved region of the LIC ( Figure 1A ) . 10 . 7554/eLife . 03351 . 007Table 1 . Crystallographic data and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 007NativeSeMetData collection Space groupC 2 2 21P3221 Cell dimensions a , b , c ( Å ) 59 . 37 , 138 . 67 , 112 . 5458 . 81 , 58 . 81 , 198 . 23 α , β , γ ( ° ) 90 , 90 , 9090 . 00 , 90 . 00 , 120 . 00PeakRemote Wavelength1 . 1158690 . 979730 . 95696 Resolution ( Å ) 50–2 . 10 ( 2 . 15–2 . 10 ) 50–3 . 50 ( 3 . 56–3 . 50 ) 50–3 . 50 ( 3 . 56–3 . 50 ) * I/σI12 . 3 ( 1 . 7 ) 12 . 1 ( 1 . 6 ) 11 . 9 ( 1 . 5 ) * Completeness ( % ) 99 . 9 ( 99 . 7 ) 99 . 8 ( 96 . 8 ) 99 . 8 ( 98 . 0 ) * Redundancy7 . 3 ( 7 . 4 ) 21 . 1 ( 11 . 9 ) 21 . 0 ( 12 . 0 ) * †Rsym0 . 18 ( 1 . 35 ) 0 . 23 ( 0 . 68 ) 0 . 24 ( 0 . 73 ) * ‡Rpim0 . 07 ( 0 . 48 ) 0 . 11 ( 0 . 22 ) 0 . 11 ( 0 . 23 ) * CC1/299 . 6 ( 53 . 3 ) * Phasing Resolution50–4 . 2 No . of SeMet sites4 Initial figure of merit0 . 32Refinement Resolution ( Å ) 50–2 . 10 No . reflections27 , 513 §Rwork/Rfree17 . 4/22 . 0 No . non-hydrogen atoms Protein2428 Water122 B-factors Protein35 . 6 Water33 . 1 R . m . s deviations Bond lengths ( Å ) 0 . 012 Bond angles ( ° ) 1 . 23 Ramachandran favored ( % ) 98 . 0 Ramachandran outliers ( % ) 0 . 0 PDB code4W7G*Numbers in parentheses refer to the highest resolution shell . †Rsym = ∑hkl∑i\|Ii ( hkl ) − 〈Ihkl〉\|/∑hkl∑i Ii ( hkl ) , where Ii ( hkl ) is the scaled intensity of the ith measurement of a reflection and 〈Ihkl〉 is the average intensity for that reflection . ‡Rpim = ∑hkl [1/ ( n−1 ) ]1/2 ∑i∣Ii ( hkl ) − 〈Ihkl〉∣/∑hkl∑i Ii ( hkl ) , where n is the number of times a single reflection has been observed . §R = ∑hkl∣Fobs , hkl − Fcalc , hkl∣/∑hkl∣Fobs , hkl∣x 100 , where Rfree was calculated on a test set comprising approximately 6% of the data excluded from refinement . Due to the difficulty in reproducing the crystals , we were unable to obtain crystals with selenomethionine-labeled protein or heavy metal derivatives in order to obtain experimental phases . To produce the crystals with selenomethionine-labeled protein , we attempted in situ proteolysis during crystallization , as reported in prior studies ( Bai et al . , 2007 ) . After testing a number of proteases , we found that chymotrypsin produced similar proteolysis products to those observed with our original crystals ( Figure 1B ) . Next , we performed in situ proteolysis of selenomethionine-labeled C . thermophilum LIC in crystallization drops ( molar ratio of 1:1000 , chymotrypsin to LIC ) . Selenomethionine crystals appeared in 3 days and diffracted to 3 . 6 Å , and the subsequent dataset was phased by Multi-wavelength Anomalous Dispersion ( MAD ) ( Table 1 ) and used to build an initial low-resolution model for the structure . This model was then successfully used in a molecular replacement search to phase the native 2 . 1 Å data set , previously obtained from our initial unlabeled protein crystals . After multiple rounds of refinement , we obtained a final 2 . 1 Å structure with an R-work of 17 . 4 and R-free of 22 . 0 ( Table 1 ) . Arg44 and Thr394 were the N-terminal and C-terminal residues that were visible in the electron density map; the molecular weight of the intervening polypeptide chain corresponds approximately to the highest molecular weight band ( ∼33 kDa ) seen from the crystals by SDS-PAGE ( Figure 1B ) . However , the additional two polypeptide fragments seen by SDS-PAGE of the crystals ( Figure 1B ) suggest that internal cleavage also occurred , most likely in regions where electron density is missing from our structure ( 74–88 , 202–210 , and 346–374 ) . While the only sequence analysis in the literature suggested that the LIC might be an ATPase ( Hughes et al . , 1995 ) , the Pfam sequence database classifies the LICs as belonging to the Ras-like , GTP-binding protein superfamily . The crystal structure of the conserved 33-kDa fragment of the LIC indeed supports this similarity with Ras . The LIC structure revealed a central β-sheet flanked by α-helices ( Figure 1C ) , which in comparison with other protein structures in the Protein Data Bank ( PDB ) using the Dali server ( Holm and Rosenstrom , 2010 ) , is most similar to the G domain of small GTP-binding proteins ( Wittinghofer and Vetter , 2011 ) . Like Ras ( Pai et al . , 1990 ) , the prototype of small G proteins , LIC has six β-strands in the core of the structure ( five parallel and one anti-parallel ) as well as five similarly placed α-helices ( Figure 1C , E ) . As a result of this structural similarity , we refer to this region as the LIC G domain . However , the LIC G domain also has conspicuous differences from Ras ( Figure 1C , E ) . In addition to the five-α-helices found in Ras , the LIC G domain has two-α-helical insertions ( α-helix 3a and 4a ) that extend from the core of the structure ( Figure 1C–E ) . Rab28 and Rab33 also have a similar α-helix 4a insertion , and the Dali server identifies these Rabs , followed by Rab32 , as the most similar to the LIC structure ( Figure 1—figure supplement 2A ) . The LIC G domain also has three additional α-helices at the C-terminus ( α-helix 6 , 7 , and 8; Figure 1C , D ) that interact with the core G domain structure ( Figure 1E ) and has a short seventh β-strand that extends the β-sheet ( Figure 1C , D ) . These results indicate that the LIC is related to the Ras GTPase superfamily . Despite structural similarity with G proteins , the LIC sequence has highly diverged from G proteins . Using a structure-based sequence alignment , the sequence similarity is only 12% , 8% , 8% , and 10% between the C . thermophilum LIC and Rab33 , Rab28 , Rab32 , and H-Ras , respectively; the similar residues are mostly hydrophobic amino acids populating the central β-sheet ( Figure 1—figure supplement 2B ) . We also performed a phylogenetic analysis of fungal and metazoan LICs within the Ras superfamily ( Rojas et al . , 2012 ) . The phylogenetic tree , which reveals similar subfamilies as seen in past work ( Rojas et al . , 2012; Basilico et al . , 2014 ) , suggests that the LIC shares a common ancestor with the Ras/Rab subfamilies , whereas the SRPRBs and ARF families , basal subfamilies of the Ras superfamily , exclude the LIC clade ( Figure 1—figure supplement 3 ) . G proteins have five unstructured loops with highly conserved residues that contribute to nucleotide-binding and hydrolysis ( Wittinghofer and Vetter , 2011 ) . These five motifs are G1 ( or the P-loop , GxxxxGKS/T ) , G2 ( or switch 1 , a loop that includes a conserved threonine ) , G3 ( or switch 2 , DxxG ) , G4 ( N/TKxD ) , and G5 ( SAK ) . These canonical motifs are present in metazoan LIC1/2 , but diverged considerably in fungal LICs , including C . thermophilum , suggesting that the pocket of fungal LICs might not bind the nucleotide ( Figure 2 ) . Consistent with this sequence information , our structure of the C . thermophilum LIC G domains shows a lack of electron density corresponding to a nucleotide in the binding pocket ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 03351 . 008Figure 2 . Conservation of LIC sequences and alignment with the LIC structure . 20 LICs were aligned via Promals ( Pei and Grishin , 2007 ) ( the C . thermophilum LIC pdb aided the alignment ) , and only the conserved sequence of the G domain is shown with the numbering based on C . thermophilum LIC . The P-loop , G3 , G4 , and G5 motifs were identified by the LIC structure and are labeled . The secondary structure of the C . thermophilum LIC is depicted above the sequences , and the residues missing in the structure are underlined in red . The red asterisks denote where sequence was taken out for space . Residues that were 80% conserved among only LIC1 and 2 sequences ( 12 out of 15 ) , only LIC3s ( 4 out of 5 ) , or universally conserved among all LICs ( 16 out of 20 ) are highlighted light blue , purple , and pink , respectively . Only the C . thermophilum LIC sequence extends to 394; all the other sequences were truncated with respect to C . thermophilum LIC , a . a . 343 , because their predicted α-helix 8 extends beyond the alignment shown here . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 00810 . 7554/eLife . 03351 . 009Figure 2—figure supplement 1 . Electron density map of the P-loop and switch 2 . The 2Fo − Fc electron density map ( blue ) of the C . thermophilum LIC G domain is shown ( contoured at 1 . 50 σ ) with all protein atoms visible ( the color scheme: oxygen in red , nitrogen in blue , carbon in yellow , and hydrogen in gray ) . Gly54 and Gln60 of the P-loop and Thr116 of switch 2 are labeled . Molecules of water are depicted as pink crosses . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 009 Remarkably , analysis of the C . thermophilum LIC binding pocket reveals significant structural deviations from Ras , which likely explain the absence of bound nucleotide . A similar glycine demarks the start of the P-loop at the end of β1 . However , three residues ( V57 , D58 , and S59 ) form an extra turn of α1 that shortens the ‘P-loop’ of C . thermophilum LIC compared to the canonical P-loop of Ras ( Figure 3A ) . Unusual for G proteins , the ‘P-loop’ and ‘switch 2’ of the C . thermophilum LIC G domain interact with one another; the side chain of Q60 in the P-loop hydrogen bonds with the main chain carbonyl of G54 and the side chain of T116 in switch 2 ( Figure 3B ) . This glutamine is found only in the P-loops of fungal LIC but is not present in metazoan LIC P-loops ( Figure 2 ) . As a result of this architecture , the C . thermophilum LIC P-loop occupies the space of the bound GTP analogue GMPPNP in Ras ( PDB: 5P21 ) ( Pai et al . , 1990 ) and sterically clashes with the α- and β-phosphates ( Figure 3A , inset ) . These observations suggest that the P-loop and switch 2 loops are stabilized in a closed conformation that prevents nucleotide from binding to C . thermophilum LIC . 10 . 7554/eLife . 03351 . 010Figure 3 . The LIC G domain binding pocket exhibits a closed conformation that is not compatible with nucleotide binding . ( A ) The C . thermophilum LIC was aligned with Ras-GMPPNP ( PDB 5P21 ) and a view of the GTP-binding pocket is shown with corresponding G motifs labeled . GMPPNP is not shown . The inset shows the aligned P-loops with Ras in complex with GMPPNP . ( B ) Interactions between the P-loop and switch 2 of C . thermophilum LIC are shown with a dashed yellow line . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 010 With regard to the other G motifs , density for the homologous G2 loop ( switch 1 ) is absent in the LIC G domain structure , although this is also the case for many G proteins in a GDP-bound form ( Wittinghofer and Vetter , 2011 ) . The G3 motif in G proteins follows after β3 and contains a conserved Asp that coordinates the magnesium ion and a glycine that hydrogen bonds with the γ-phosphate ( DxxG ) ( Wittinghofer and Vetter , 2011 ) . However , the same loop in the C . thermophilum LIC G domain lacks an Asp and a Gly in a comparable position . The loops corresponding to the G4 motif and G5 motif are in similar positions in the C . thermophilum LIC G domain as in Ras ( Figure 3A ) . Collectively , the structural data suggest that the G domain of the fungal LICs have lost their ability to bind the nucleotides . The sequence and structural data described above suggest that the C . thermophilum LIC G domain may not be able to bind nucleotide while metazoan LICs , which contain a more canonical P-loop motif , may have nucleotide-binding capability . To test these ideas , we directly assayed for nucleotide-binding by C . thermophilum LIC and human LIC1 G domains purified after E . coli expression ( see ‘Materials and methods’ ) . When the C . thermophilum LIC G domain was injected onto a C18 column for reverse-phase liquid chromatography ( RPLC ) with acetonitrile , no 260 nm adsorbing molecule eluted from the column ( Figure 4A ) , which is consistent with the lack of nucleotide in the crystal structure . In contrast , when the same experiment was performed with the human LIC1 G domain , a small molecule eluted with a retention time that did not match either GDP or GTP but which had an adsorption spectrum characteristic of a guanine nucleotide ( Figure 4B , C; Figure 4—figure supplement 1A ) . The buffer used in the RPLC included tetrabutylammonium hydroxide , which binds to negatively charged entities , making them hydrophobic . Since the column retains hydrophobic molecules longer , more negatively charged molecules exhibit longer retention times . The molecule extracted from the LIC1 G domain has a longer retention time than even guanosine tetraphosphate , suggesting that it is slightly more negatively charged ( Figure 4B ) . When the small molecule was separated from the LIC1 G domain before injecting it on the C18 column , it had the same retention time ( Figure 4—figure supplement 1B ) , suggesting that the long retention time was not an artifact of simultaneously injecting it along with protein on the column . 10 . 7554/eLife . 03351 . 011Figure 4 . The human LIC1 G domain binds guanine nucleotide . ( A ) The C . thermophilum LIC G domain ( a . a . 45–394 ) was injected on the C18 column with an increasing gradient of acetonitrile and detection with a wavelength of 260 nm . Standards are at 1 mM , and kinetin triphosphate ( KTP ) , a non-biological nucleotide ( Hertz et al . , 2013 ) , was added for an internal control . KTP was used as a positive control for sample injection because it elutes after all other nucleotides due to its high negative charge . ( B ) The human LIC1 G domain ( amino acids 65–354 ) was analyzed by RPLC as done in ( A ) . Nucleotide standards are at 0 . 5 mM , and the LIC was simultaneously injected with an equal concentration of guanosine tetraphosphate ( GTP4 ) for an internal reference . An arrow indicates LIC nucleotide . ( C ) The wavelength spectrum of the LIC nucleotide in ( B , arrow ) is shown from 210 nm to 350 nm . ( D ) The human LIC1 G domain at 150 µM and ppGpp at 150 µM were analyzed by RPLC separately . LIC1 and ppGpp , each at 150 µM were then injected simultaneously . The structure of guanosine-3ʹ , 5ʹ-bisdiphosphate ( ppGpp ) is shown . ( E ) The human LIC1 G domain and C . thermophilum LIC G domain were incubated with 5 mM EDTA and 1 mM GDP for 1 hr at room temperature . An excess of MgCl2 was then added at a final concentration of 10 mM and the resulting protein , with KTP as an internal control , was analyzed by RPLC . The inset shows the GDP standard alone superimposed with the two chromatograms . ( F ) The human LIC1 G domain ( 150 µM ) was incubated with 5 mM EDTA , 0 . 5 mM GDP , and 5 mM GMPPNP for 1 hr at room temperature and analyzed by RPLC as done in ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 01110 . 7554/eLife . 03351 . 012Figure 4—figure supplement 1 . Guanine nucleotide extraction from human LIC1 . ( A ) The absorbance of ADP and GDP was measured with a wavelength spectrum ranging from 215 nm to 350 nm . ( B ) The human LIC1 G domain was boiled for 10 min or methanol-extracted ( by adding an equal volume of 100% methanol ) . The protein was filtered from the supernatant with a 10 kDa molecular-weight cutoff concentrator , and the flow-through was analyzed by RPLC . The extracted nucleotide is denoted with an arrow . ( C ) The average retention time and standard deviation of each nucleotide from all RPLC runs are displayed ( n = 5 for LIC nucleotide , n = 4 for commercial ppGpp , n = 3 for GMP , n = 5 for GDP , and n = 5 for GTP ) . A t test was done with the human LIC1 nucleotide and the commercial ppGpp , resulting in p = 0 . 21 . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 01210 . 7554/eLife . 03351 . 013Figure 4—figure supplement 2 . Nucleotide exchanges with GTP and ADP . ( A ) The human LIC1 G domain ( 150 µM ) was incubated with 5 mM EDTA , 0 . 5 mM GDP , and 5 mM ADP for 1 hr at room temperature . MgCl2 was then added at a final concentration of 10 mM , followed by buffer exchange and RPLC analysis . ( B ) The human LIC1 G domain was incubated with 0 . 5 mM GDP and 5 mM GTP and analyzed as done in ( A ) . ( C ) The dead volume of the RPLC equipment was increased , leading to increased retention times of all nucleotides in comparison to data collected for Figure 4 . The average and standard deviation of the retention time were calculated for each standard and the LIC1 nucleotide ( n = 5 for LIC nucleotide , n = 9 for GDP , n = 5 for GTP , n = 3 for ADP ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 01310 . 7554/eLife . 03351 . 014Figure 4—figure supplement 3 . Human LIC1 co-purifies with GDP from human cells . ( A ) Strep-tagged human LIC1 ( a . a . 1–354 ) was purified from HEK-293T cells using Strep-Tactin beads . A sample ( 1 μL ) of the final concentrated LIC1 ( 50 μL total ) is displayed beside varying amounts of BSA . The 1/50th sample of purified human LIC1 is approximately 150 ng , which allows for an estimation of approximately 7 . 5 μg total . ( B ) Purified porcine brain tubulin ( 175 μg ) was buffer exchanged into 50 mM NH4OAc , concentrated to approximately 3 . 5 μg/μl . A 10 μl sample of a 4-fold dilution was analyzed by LC-MS for GDP and GTP based on mass , fragmentation and retention time . In the final scan , GDP and GTP ( with approximate retention times of 2 . 3 min and 2 . 5 min , respectively ) were both detected . ( C ) The 7 . 5 μg of human LIC1 shown in ( A ) was buffer exchanged into 50 mM NH4OAc and concentrated to approximately 188 ng/μl . LC-MS analysis was conducted as done with tubulin in ( B ) . A distinct peak was seen for GDP , yet only background noise was present in the scan for GTP . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 01410 . 7554/eLife . 03351 . 015Figure 4—figure supplement 4 . Instability of human LIC1 G domain without nucleotide . Point mutations were made in the human LIC1 G domain ( a . a . 65–354 ) , specifically K80A of the P-loop and D248N of the G4 motif . The mutants and wild-type protein ( WT ) were expressed in BL21 DE3 RIPL cells for 4 hr following induction with 0 . 5 mM IPTG . The test expressions of the mutants were analyzed following lysis and centrifugation . Samples of the pellet ( P ) and supernatant ( S ) were resolved by SDS-PAGE and Coomassie stained . The red boxes indicate LIC1 . The results show that the K80A and D248N mutants are mostly insoluble when expressed in bacteria . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 015 A guanine nucleotide that is more charged than guanosine tetraphosphate is guanosine-3ʹ , 5ʹ-bisdiphosphate ( termed ppGpp ) , which has two diphosphates connected to the 5ʹ and 3ʹ hydroxyls of ribose ( Figure 4D ) . ppGpp , which is produced as part of the stringent response in E . coli ( Chatterji and Ojha , 2001 ) , has been co-crystallized with the bacterially expressed GTPases Arl5 ( bound to a likely ppGpp remnant GDP3’P; PDB: 1ZJ6 ) ( Wang et al . , 2005 ) and Obg ( PDB: 1LNZ ) ( Buglino et al . , 2002 ) . Within the margin of variation of different chromatographic runs ( Figure 4—figure supplement 1C ) , the retention time of 150 µM ppGpp was similar to that of the guanine nucleotide co-purifying with 150 µM human LIC1 ( Figure 4D; Figure 4—figure supplement 1C ) . To determine if the nucleotides were truly identical , we combined and injected 150 µM ppGpp and 150 µM human LIC1 G domain . Only one peak was detected , which corresponded to the sum of the absorbances of the commercial ppGpp and the nucleotide co-purifying with the human LIC1 G domain injected separately ( Figure 4D ) . Thus , the nucleotide in the human LIC1 G domain co-purifies precisely with ppGpp in the same chromatographic run . The eluted human LIC1 nucleotide was not successfully identified by mass spectrometry , most likely due to its high negative charge and low quantity following extraction , and thus additional information on its identity could not be obtained by this method . However , with indistinguishable RPLC elution times and a guanine-like absorption spectrum , we believe that ppGpp is the most likely candidate for the LIC-bound nucleotide . The nucleotide ppGpp can reach millimolar concentrations under bacterial stress ( Srivatsan and Wang , 2008 ) , as can occur during the growth conditions for protein overexpression . However , ppGpp is thought to only exist in bacteria and plants and not human cells ( Takahashi et al . , 2004; Srivatsan and Wang , 2008 ) . Thus , we do not believe that ppGpp is a natural ligand for the human LIC , but that binding of this ligand reflects an unnatural situation of bacterial expression . Nevertheless , ppGpp binding indicates that the metazoan LIC is indeed competent to bind nucleotide and also raised questions of its nucleotide specificity , especially since we did not detect bound GTP , which is present in millimolar concentrations in bacteria ( Buckstein et al . , 2008 ) . Thus , we hypothesized that human LIC1 might have a higher affinity for guanosine diphosphates than triphosphates . As was observed in the crystal structure of Arl5 ( Wang et al . , 2005 ) , LIC1 might bind only one of the two diphosphates of ppGpp , which can be more abundant than GDP during bacterial stress ( Srivatsan and Wang , 2008 ) . To test whether the human LIC1 G domain can indeed also bind GDP , the protein was incubated with 1 mM GDP and 5 mM EDTA ( to promote the release of bound ppGpp ) . After quenching the exchange reaction with 10 mM MgCl2 and removing the unbound nucleotide , the protein was subjected to RPLC to analyze the composition of the bound nucleotide . Two peaks were now detected corresponding to the retention times of GDP and ppGpp , indicating that GDP in solution had partially exchanged for the ppGpp bound to the protein after bacterial expression ( Figure 4E ) . In a similar nucleotide exchange experiment , we found that ADP did not bind to human LIC1 indicating that this G domain is indeed guanine-specific ( Figure 4—figure supplement 2A ) . Furthermore , when the identical GDP exchange was performed with the C . thermophilum LIC G domain , the GDP in solution did not bind to the fungal protein ( Figure 4E ) . Therefore , the human LIC1 G domain , but not C . thermophilum LIC , is capable of binding GDP and is most likely binding one diphosphate of ppGpp , as found previously with Arl5 ( Wang et al . , 2005 ) . Note , approximately 2 . 5 mg/ml of LIC1 was required to detect nucleotide above the noise in the described RPLC method; therefore , it is possible that GDP also may co-purify with bacterially expressed human LIC1 , but its abundance is below our detection limit . We next examined relative affinity of the LIC G domain for guanosine diphosphates versus triphosphates . We incubated 150 µM human LIC1 G domain with 5 mM EDTA and a 10-fold excess of GMPPNP ( 5 mM ) to GDP ( 0 . 5 mM GDP ) ; the non-hydrolyzable GMPPNP was used to prevent any possibility of binding and conversion to GDP by enzymatic hydrolysis . GMPPNP binds to many G proteins and has been widely used as a non-hydrolyzable analogue ( Wittinghofer and Vetter , 2011 ) . After quenching the exchange reaction with 10 mM MgCl2 , the protein was analyzed by RPLC . The eluted material contained a mixture of GDP and non-exchanged ppGpp but no GMPPNP was detected despite its 10-fold excess to GDP in the starting reaction ( Figure 4F ) . We also repeated the same exchange reaction with hydrolyzable GTP ( again in 10-fold excess to GDP ) but did not detect any GTP bound to human LIC ( Figure 4—figure supplement 2B ) . Thus , these experiments reveal that the human LIC1 G domain has a higher affinity for guanosine diphosphates than guanosine triphosphates . The nucleotide exchange experiments suggested that human LIC1 most likely is bound to GDP and not GTP in its native environment . To test this further , we expressed the human LIC1 G domain in HEK-293T cells , a human cell line . After purifying the protein , any potentially bound nucleotide was released by boiling . LC-MS analysis of the non-proteinaceous supernatant clearly revealed the presence of GDP , but no detectable GTP ( Figure 4—figure supplement 3 ) . Porcine brain tubulin was used as a positive control to demonstrate that the LC-MS could detect both GDP and GTP . Thus , human LIC1 expressed in human cells co-purifies with GDP alone . Thus , our results collectively suggest that metazoan LIC preferentially binds guanosine nucleotides containing a diphosphate ( either GDP or ppGpp ) that extends into the binding pocket . We next sought to examine the positions of conserved residues on the surface of the LIC G domain , which might be candidate residues that participate in binding to other proteins ( Figure 5A ) . We classified residues as ( 1 ) conserved throughout all LICs , ( 2 ) LIC1/2-specific , or ( 3 ) LIC3-specific ( considered conserved if 80% similar in each category ) ( Figure 2 ) . Relatively few universally conserved residues are on the surface of the LIC G domain , with the exception of a hydrophobic groove of universally conserved residues composed of three aromatic residues ( Y93 , Y95 [β2] , and Y113 [β3] ) and Leu125 ( α2 ) ( residue numbers refer to the position in C . thermophilum LIC ) ( Figure 5A ) . A group of highly conserved , class-specific LIC3 residues surround this universally conserved , hydrophobic patch ( Figure 5A ) . In LIC1/2 , the corresponding residues are often reversed in charge or neutralized . A patch of LIC1/2-specific acidic residues ( including E325 , E304 , and D306 ) lie on the opposite side of the molecule that includes the C-terminal loops and α-helix 4 , 4a , and 6 ( Figure 5A ) . Three LIC3-specific patches also lie in this same region ( Figure 5A ) . 10 . 7554/eLife . 03351 . 016Figure 5 . The G domain contains the binding interface of the dynein heavy chain . ( A ) The conservation of residues shown in Figure 2 is mapped onto the surface of the C . thermophilum LIC G domain . The surface is shown in two different orientations with each orientation showing universally conserved residues ( pink ) and LIC1/2-specific residues ( blue ) vs LIC3-specific residues ( purple ) . The LIC is oriented either toward the N-terminus or toward the C-terminal loops ( as in Figure 1C ) . Conserved amino acids and the corresponding residue numbers are labeled according to the C . thermophilum LIC sequence . ( B ) HA-tagged fragments of human LIC1 ( a . a . 1–389 , 389–523 ) were expressed in HEK-293T cells , immunoprecipitated with an anti-HA antibody , and immunoblotted for the dynein heavy chain or HA tag . The asterisk denotes a non-specific band that reacts with the anti-HA antibody . ( C ) HA-tagged double and triple mutants of human LIC1 were expressed , immunoprecipitated , and analyzed as in ( B ) with additional immunoblotting for the dynein intermediate chain . The residue numbers shown correspond to human LIC1 . Homo sapiens ( H . s . ) LIC1 residues correspond to C . thermophilum ( C . t . ) LIC as follows: H . s . D110 , D112 , D113 = C . t . D102 , E104 , D105; H . s . Y101 , Y103 , W120 = C . t . Y93 , Y95 , Y113; H . s . V316A , E317A , D319A = C . t . I303 , E304 , D306; H . s . E317A , K332A = C . t . E304 , K319; H . s . R260A , E262A , E338A = C . t . E325 , E248 , K246 . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 016 The sequence alignment in Figure 2 also revealed unique insertions or deletions in different LIC species or isoforms . Hyphal fungi have a longer loop between α1 and β2 . The LIC3s have a longer loop between α6 and α7 and a shorter loop between α3A and β5 than other LICs . We next sought to examine which domain of the LIC interacts with the dynein heavy chain by expressing full-length or fragments of HA-tagged human LIC1 in HEK-293T cells , immunoprecipitating the expressed LIC fragment , and probing for the dynein heavy chain by immunoblot analysis . The fragments tested were an N-terminal fragment of the LIC G domain ( a . a . 1–389 , which terminates after a predicted helix that may correspond to α8 ) and a C-terminal fragment ( a . a . 389–523 ) . We found that the N-terminal fragment , a . a . 1–389 , bound the heavy chain with a comparable efficiency to the full-length human LIC but the C-terminal domain did not bind ( Figure 5B ) . We also tested a longer C-terminal fragment starting after α7 ( a . a . 355–523 ) and also saw no binding ( data not shown ) . These results suggest that the heavy chain-binding interface lies within the highly conserved G domain and not in the C-terminal domain . Our sequence analysis identified conserved solvent-exposed residues on the surface of the G domain ( Figure 5A ) , and we wished to determine if any of these regions were involved in heavy chain binding . Double or triple alanine mutations were made in full-length human LIC1 targeting LIC1/2-specific regions ( 1: D110A/D112A/D113A; 2: V316A/E317A/D319A; 3: E317A/K332A; 4: R260A/E262A/E338A ) and the one patch of universal conservation ( 5: Y101A/Y103A; 6: Y101A/Y103A/W120A ) . The mutant LIC1s were then expressed in HEK-293T cells and tested for heavy chain binding by immunoprecipitation and immunoblot analysis . Strikingly , the mutations of the universally conserved aromatic residues exhibited dramatically decreased binding to the heavy chain; the double mutant Y101A/Y103A exhibited decreased binding to the dynein heavy chain , and the binding appeared to be almost entirely disrupted in the triple mutant Y101A/Y103A/W120A ( Figure 5C ) . The inability of these mutants to incorporate into the dynein holoenzyme complex was also verified by probing for the dynein intermediate chain ( Figure 5C ) . In contrast , the other five double or triple mutations tested bound normally to the dynein heavy chain . These results indicate that the universally conserved hydrophobic groove on the LIC is involved in binding to the dynein heavy chain . We next investigated how the LIC binds to proteins that link dynein to Rabs involved in membrane transport . FIP3 and RILP are both adapter proteins that have been shown previously to bind both to mammalian LIC ( Horgan et al . , 2010a; Horgan et al . , 2010b; Scherer et al . , 2014 ) and Rab GTPases , Rab11 in the case of FIP3 ( Hales et al . , 2001 ) and Rab7 in the case of RILP ( Cantalupo et al . , 2001 ) . LIC1 has also been implicated in dynein's interaction with another Rab effector—BicD2 ( Splinter et al . , 2012 ) , an effector of Rab6 ( Matanis et al . , 2002 ) , although whether this occurs through a direct interaction or not is unknown . We purified the LIC1 G domain ( a . a . 1–389 ) , the C-terminal domain ( a . a . 389–523 ) , and the full-length human LIC1 from bacteria to determine the binding site for cargo in vitro . The GST-tagged domains and full-length LIC1 were bound to glutathione beads and incubated with recombinant GFP-tagged FIP3 , RILP , or BicD2 . Binding was determined by centrifuging the beads and immunoblotting with an anti-GFP antibody . All three Rab effectors bound to the full-length LIC1 but not to untreated beads . This result also clearly establishes that BicD2 can bind directly to the LIC . Interestingly , FIP3 , RILP , and BicD2 each bound the C-terminal domain almost as equally well as to full-length LIC1 but showed no interaction with the G domain ( Figure 6A–C ) . The assay was also conducted with GFP alone , which showed no binding to the full-length LIC1 or its truncations ( Figure 6—figure supplement 1 ) . Thus , while the G domain contains the primary binding site for the dynein heavy chain , the LIC C-terminal domain appears to serve as a docking site for cargo adaptors . 10 . 7554/eLife . 03351 . 017Figure 6 . The LIC C-terminus alone binds Rab effectors . ( A ) Human GST-tagged full-length LIC and LIC truncations , including the G domain ( a . a . 1–389 ) and the C-terminus ( a . a . 389–523 ) , were purified from E . coli bound to glutathione beads and incubated with recombinant GFP-FIP3 . The beads were centrifuged , washed , and probed with an anti-GFP antibody to assess binding of the GFP-tagged protein . ( B ) The same experiment in ( A ) was done with GFP-RILP . ( C ) The same experiment in ( A ) was done with GFP-BicD2 . ( D ) A model depicts the LIC G domain bound to the dynein heavy chain ( HC ) with the C-terminus bound to a Rab effector , allowing for membrane transport by dynein . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 01710 . 7554/eLife . 03351 . 018Figure 6—figure supplement 1 . Controls for GST pulldowns . ( A ) The supernatants following the three GST pulldowns shown in Figure 6 were probed with anti-GFP to show the unbound prey ( FIP3 , RILP , and BicD2 ) . ( B ) A GST pulldown was done as in Figure 6A–C except with superfolder GFP alone . This negative control was done in parallel with the pulldown in Figure 6C . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 01810 . 7554/eLife . 03351 . 019Figure 6—figure supplement 2 . Disorder probability of human LIC1 . The Protein Disorder Prediction System ( PrDOS ) ( Ishida and Kinoshita , 2007 ) was used to assess the probability of disorder of full-length human LIC1 . The red lines , at residues 65 and 389 , indicate the approximate range of where the G domain exists . DOI: http://dx . doi . org/10 . 7554/eLife . 03351 . 019 The evidence that human LIC1 can bind nucleotide , unlike its fungal homologues , raises the question as to whether or not metazoan LIC acts as a GTPase switch . Most small GTP-binding proteins change their binding affinity for protein partners between GTP and GDP states , and one could envision a similar role for a nucleotide switch in cargo selection and binding by metazoan LICs . However , our data suggest that LICs do not function as GTPase switches . One reason is that the canonical GTPase switch 2 motif ( DxxG ) , which plays a critical role in GTP hydrolysis ( Wittinghofer and Vetter , 2011 ) , appears to be absent from the equivalent loop in metazoan LICs ( Figure 2 ) . Furthermore , human LIC1 binds guanosine diphosphate preferentially over guanosine triphosphate; even when GMPPNP or GTP is in 10-fold excess over GDP , little or no GTP binding to the LIC is detected ( Figure 4F ) . This preference for guanosine diphosphates explains why human LIC1 co-purifies with ppGpp from bacteria , even though the GTP concentration is approximately 1 . 6 mM in E . coli during exponential growth ( Buckstein et al . , 2008 ) . In contrast , the well-characterized GTPase Ras has a 10-fold higher affinity for GTP than GDP ( John et al . , 1990 ) . There are examples of G proteins that prefer guanosine diphosphates , such as ADP-ribosylation factor ( ARF ) , which has a >25-fold higher affinity for GDP than GTPγS ( Weiss et al . , 1989 ) . ARF can exchange GTP for GDP under particular conditions ( Weiss et al . , 1989 ) , and it is possible that metazoan LICs can do so as well . However , given the combination of preferential GDP binding and the lack of a canonical switch 2 sequence , we find it unlikely that the LIC is a true GTPase . While metazoan LIC may not switch between a GDP-bound and a GTP-bound state , it is possible that it reversibly exchanges GDP and adopts an apo form . Small GTPases are stable in the apo form only when bound to a guanine-nucleotide-exchange factor , or GEF ( Cherfils and Chardin , 1999 ) ; by analogy , LIC1 may release GDP upon binding the dynein heavy chain . This idea could be tested in the future by examining the nucleotide state of the LIC when in complex with the dynein heavy chain . Another hypothesis for the role of nucleotide binding by metazoan LIC is simply to provide structural stability . In support of a role of nucleotide in structural stability , we have found that human LIC1 is unstable with point mutations that interfere with nucleotide binding ( K80A in the P-loop or D248A in the G4 motif; Figure 4—figure supplement 4 ) . Surprisingly , overexpression of the P-loop K80A mutation in Caenorhabditis elegans rescued the null LIC1 phenotype ( Yoder and Han , 2001 ) , but endogenous expression levels may reveal a phenotype associated with instability of the mutant protein . The fungal LICs appear to have evolved another strategy for achieving protein stability in the absence of bound nucleotide . Two residues , Gln60 and Thr116 , in the C . thermophilum LIC appear to play an important role in stabilizing a nucleotide-free state and preventing nucleotide binding ( Figure 3B ) . Gln60 replaces the P-loop lysine , which is critical for nucleotide binding and hydrolysis , and Thr116 replaces the aspartate found in switch 2 ( DxxG ) , which contacts the nucleotide-bound Mg2+ ion via a water molecule ( Wittinghofer and Vetter , 2011 ) . ‘Pseudo-ATPase’ is a recently coined term for proteins with a structure typical of ATPases but which does not catalyze ATP hydrolysis and uses nucleotide for protein stability ( Adrain and Freeman , 2012 ) . For example , the kinetochore nucleotide-binding protein BUBR1 does not hydrolyze ATP in carrying out its function in the mitotic checkpoint complex but rather requires ATP for its stability ( Suijkerbuijk et al . , 2012 ) . We propose that metazoan LIC may be a ‘pseudo-GTPase’ , where nucleotide is used for stability rather than in a cycle of conformational change . Another protein that also has a common G-protein fold with no nucleotide-binding or hydrolysis capability is CheY , a bacterial chemotactic protein ( Artymiuk et al . , 1990; Chen et al . , 1990 ) . A recent study also identified the kinetochore protein , CENP-M , as a ‘pseudo-GTPase’ , since it contains a G protein fold but does not contain bound nucleotide ( Basilico et al . , 2014 ) . Phylogenetic analysis suggests that the three identified ‘pseudo-GTPases’ , LIC , CENP-M , and CheY , are all most similar to the Rab and Ras subfamilies of the G protein superfamily ( Figure 1—figure supplement 3 ) ( Basilico et al . , 2014 ) . In this study , we have shown that the G domain alone binds the heavy chain and identified a groove of highly conserved , aromatic residues that are involved in this interaction . However , since this hydrophobic patch is conserved among all LICs , other residues must determine the targeting of LIC1/2 to cytoplasmic dynein 1 and LIC3 to cytoplasmic dynein 2 . Our structure , combined with sequence analysis , provides some clues on residues that might be involved in the selective recognition of a particular dynein heavy chain . For example , the LIC3s have several class-conserved and mostly hydrophobic residues ( conserved among LIC3s but not LIC1/2 ) that surround the hydrophobic groove ( Figure 5A ) . LIC1/2 exhibits fewer class-conserved residues on the face of the G domain with the hydrophobic groove , although a few such residues exist ( Figure 5A ) . Future structure–function binding studies with cytoplasmic dynein 1 and 2 and LIC1 and LIC3 will be needed to fully understand the nature of this selective interaction . Our studies show that the LIC's C-terminal domain is involved in binding three different cargo adapter proteins , FIP3 , RILP , and BicD2 , and thus appears to contain multiple recognition sites for distinct adapter proteins . Interestingly , the C-terminal domain is predicted to be disordered ( Figure 6—figure supplement 2 ) . The LIC domain structure has interesting parallels with the dynein intermediate chain , which has a well-ordered WD repeat domain that interacts with the heavy chain and a less conserved disordered region that binds the dynein light chains ( Mok et al . , 2001; Nyarko et al . , 2004 ) and cargo-adaptors like dynactin ( Vaughan and Vallee , 1995 ) . A number of disordered proteins become ordered upon binding other proteins . It will be interesting to determine what structure the C-terminal tail of the LIC adopts before and after binding to Rab effectors . While this study points to the C-terminal tail of LIC as a cargo binding site , it is possible that adapter proteins may be discovered in the future that bind to the G domain , perhaps via LIC1/2 specific residues . Overall , our data suggest a working model in which the LIC domains are specialized for distinct functions . The G domain is docked onto the dynein heavy chain , while the C-terminal domain binds many of the LIC's protein partners , specifically Rab effectors involved in membrane transport ( Figure 6D ) . Recent studies have shown that these Rab effectors , in addition to linking dynein to membrane cargos , link dynein to dynactin to create an ultraprocessive motor ( McKenney et al . , 2014; Schlager et al . , 2014 ) . Thus , the LIC may have a critical role in the mechanism of generating an ultraprocessive dynein–dynactin complex . While the LIC may simply serve as a docking site for Rab effectors , it is also possible that the LIC N- and C-terminal domains communicate allosterically in some way with other dynein chains and dynactin to affect their activities . The homologous dynein LIC in C . thermophilum ( EGS22626 . 1 ) was PCR-amplified from C . thermophilum cDNA , generously given by Peter Walter's lab at UCSF . The full-length protein was cloned into the vector pGEX-6P-1 , which encodes an N-terminal GST tag . The protein was expressed in BL21 DE3 RIPL cells and induced with 0 . 5 mM IPTG for 3 hr at 37°C . The protein was purified using glutathione agarose 4B ( USB ) and then cleaved with GST-tagged human rhinovirus 3C protease overnight at 4°C . The cleaved product was further purified by gel filtration into 10 mM Tris–HCl ( pH 7 ) and 25 mM NaCl using a HiPrep 16/60 Superdex S-200 HR column ( GE Healthcare , Piscataway , NJ ) with an AKTA FPLC system ( GE Healthcare ) . The full-length protein eluted as a monomer , which was verified by static light scattering . Crystal trials were setup with 16 . 6 mg/ml protein by hanging drop vapor diffusion . Native crystals were obtained at 20°C in a condition including 0 . 1 M MES ( pH 6 . 5 ) and 20% ( wt/vol ) PEG 10000 in a Nextal PEGs screen ( Qiagen Inc . , Valencia , CA ) . The crystals were cryoprotected by the addition of 18% glycerol to the well solution and flash-cooled by plunging in liquid nitrogen . To express selenomethione-labeled C . thermophilum LIC , protein was expressed in M9 minimal media . Before induction , the culture was incubated for 30 min with a cocktail including lysine , phenylalanine , threonine , isoleucine , leucine , valine , and selenomethione to inhibit methionine biosynthesis . The culture was then induced with 0 . 5 mM IPTG for 3 hr at 37°C . The labeled protein was purified in the same way as native protein except with the addition of 5 mM TCEP in the gel filtration buffer . Chymotrypsin ( Sigma Aldrich , St . Louis , MO ) was added to the protein immediately before setting up crystal trays at a ratio of 1:1000 moles ( chymotrypsin to LIC ) with the LIC at 8 . 6 mg/ml . Crystals were obtained at 20°C from a condition including 0 . 2 M calcium acetate , 0 . 1 M Tris ( pH 7 ) , and 20% ( wt/vol ) PEG 3000 ( Qiagen ) . The crystals were cryoprotected by the addition of 18% glycerol to the well solution and flash-cooled by plunging in liquid nitrogen . For bacterial expression , strep-tagged human LIC1 ( NM_016141 . 3 ) and its truncations were cloned into pGEX-6P-1 , which encodes an N-terminal GST tag . For mammalian cell expression ( HEK-293T cells ) , strep-tagged human LIC1 ( a . a . 1–354 ) was cloned into a pHR vector for lentiviral expression . Strep-tagged LIC1 was purified with Strep-Tactin resin ( IBA , Germany ) and eluted via a commercial Strep-Tactin elution buffer containing 2 . 5 mM desthiobiotin ( Novagen , Germany ) . Human LIC1 was purified for RPLC with glutathione agarose 4B ( USB ) and then cleaved with GST-tagged human rhinovirus 3C protease overnight at 4°C . The cleaved product was further purified by gel filtration into 10 mM Tris ( pH 7 ) , 50 mM NaCl , 2 mM MgCl2 , and 2 mM TCEP . FIP3 ( human; AB383948 . 1 ) , RILP ( human; NM_031430 . 2 ) , and BicD2 ( mouse; NM_029791 . 4 ) were cloned into the vector pET28a , which encodes an N-terminal His tag with an additional N-terminal StrepII tag and superfolder-GFP . The proteins were expressed in BL21 DE3 RIPL cells overnight at 18°C and purified via Strep-Tactin resin ( IBA ) . RILP and BicD2 were eluted via a commercial Strep-Tactin elution buffer ( Novagen ) , and FIP3 was eluted using 2 . 5 mM d-Desthiobiotin ( Sigma Aldrich ) in 100 mM Hepes ( pH 7 . 4 ) , 10% glycerol , 0 . 5 mM EGTA , 5 mM MgCl2 , and 300 mM NaCl . RILP was purified further by gel filtration into 50 mM Tris–HCl ( pH 7 ) , 150 mM NaCl , 2 mM MgCl2 , 1 mM EGTA , and 2 mM TCEP . FIP3 was further purified by gel filtration into 30 mM Hepes ( pH 7 . 4 ) , 50 mM K-Acetate , 2 mM Mg-Acetate , 1 mM EGTA , and 10% glycerol . GST-tagged superfolder GFP was purified via glutathione beads , followed by cleavage of the GST tag with human rhinovirus 3C protease . GFP was then further purified by gel filtration into 25 mM Hepes ( pH 7 . 5 ) , 150 mM NaCl , 10% glycerol , and 2 mM TCEP . Diffraction data were collected at the Advanced Light Source ( ALS ) ( Lawrence Berkeley National Laboratory ) , beamline 8 . 3 . 1 . Multi-wavelength anomalous dispersion ( MAD ) datasets were collected from selenomethionine ( SeMet ) -derivatized LIC crystals at two wavelengths , a high-energy remote wavelength and at a wavelength halfway between the peak and inflection point . A native data set was also collected from underivatized protein . The SeMet data sets were indexed to P3221 and merged using HKL2000 ( Otwinowski and Minor , 1997 ) , and substructure determination ( four selenium sites ) and initial phases were obtained using AutoSol in Phenix ( Adams et al . , 2010 ) . The structure was built using Coot ( Emsley and Cowtan , 2004 ) , and AutoBuild in Phenix ( Adams et al . , 2010 ) improved the initial model . After several rounds of refinement ( via phenix . refine ) , the initial 3 . 6 Å structure was used as a search model for molecular replacement of the 2 . 1 Å native dataset using Phaser ( Adams et al . , 2010 ) . The 2 . 1 Å native dataset was integrated and indexed to C 2 2 21 using XDS ( Kabsch , 2010 ) and scaled and merged using XSCALE ( Kabsch , 2010 ) . After successful molecular replacement , several rounds of model building and refinement were carried out using Coot ( Emsley and Cowtan , 2004 ) and phenix . refine ( Adams et al . , 2010 ) . Final data collection and refinement statistics can be found in Table 1 . The HPLC system ( Waters ) was used with a C18 column ( Phenomenex , Torrance , CA ) . The solutions used were as follows: buffer A consisted of 5% acetonitrile , 5 mM tetrabutylammonium hydroxide , 25 mM KH2PO4 ( pH 6 ) ; buffer B consisted of 60% acetonitrile , 5 mM tetrabutylammonium hydroxide , 25 mM KH2PO4 ( pH 6 . 0 ) . The gradient for an RPLC run was 0% to 65% buffer B over 44 min with an 18 μl injection of sample . Nucleotides GMP , GDP , and GTP were purchased from Sigma Aldrich , and ppGpp was purchased from TriLink Biotechnologies ( San Diego , CA ) . The synthetic nucleotide KTP was purchased from BIOLOG Life Science Institute ( Germany ) and used as a positive control for sample injection . HEK-293T cells were cultured in DMEM media containing 10% FBS and 5% penicillin/streptomycin/glutamine . Mammalian expression LIC constructs were encoded with an mCherry reporter in the following construct: pHR-mCherry-p2A-LIC . The pHR construct was co-transfected with the plasmids pMD2 . G and pCMVΔ8 . 91 for lentivirus production . To transfect cells , cells in one well of a six-well dish were transfected with 10 µl of 2 mg/ml polyethylenimine ( Polysciences , Inc . Warrington , PA ) and 2 µg of total DNA . The virus-containing media were collected after 3 days of incubation following transfection , and cell particulates were spun out . The virus was then concentrated 10-fold in PBS ( Gibco , Grand Island , NY ) using Lenti-X concentrator ( Clontech , Mountain View , CA ) . To amplify protein expression , the lentivirus was used to infect HEK-293T cells for immunoprecipitations . Infections were conducted by adding 10–20 µl of the concentrated virus to a six-well dish of HEK-293T cells at ∼50% confluency . The cells were then passaged to expand the cell quantity for immunoprecipitations . Approximately , 182 sequences of members of the Ras superfamily were used to compare the LIC's placement within this protein family . Most sequences included those used in a phylogenetic study of the Ras superfamily ( Rojas et al . , 2012 ) . Other sequences included the sequences of purported ‘pseudo-GTPases’ , CheY , and CENP-M . In our analysis , a structure-based sequence alignment was made with C . thermophilum LIC and the Dali server's top hits for structural similarity to the C . thermophilum LIC G domain ( Rab33 , Rab28 , and Rab32 ) . This alignment , made by the Dali server ( Holm and Rosenstrom , 2010 ) , was used as a profile , and the collection of Ras superfamily sequences was aligned to this profile using MafftWS ( Katoh and Standley , 2013 ) . The resulting alignment was modified by deleting segments in which less than 80% of the sequences did not align . The modified alignment was then used to generate a phylogenetic tree using Maximum Likelihood ( Guindon et al . , 2010 ) . The ATGC PhyML was run using the LG substitution model , SPR and NNI tree improvement , 5 random starting trees , and 300 bootstraps . The resulting tree ( Figure 1—figure supplement 3 ) was displayed using iTOL ( Letunic and Bork , 2007 ) . Strep-tagged human LIC1 and porcine brain tubulin , which was purified as described in Castoldi and Popov ( 2003 ) , was buffer exchanged into 50 mM NH4OAc and concentrated to approximately 0 . 2–4 μg/μl . The protein was then boiled for 8 min . The denatured protein was pelleted , and the supernatant was isolated for LC-MS analysis by the Vincent Coates Foundation Mass Spectrometry Laboratory ( Stanford University ) . All LC-MS analyses were carried out by negative electrospray ionization using Agilent 1100 HPLC and linear ion trap mass spectrometer LTQ XL ( Thermo Fisher Scientific , Waltham , MA ) . HPLC conditions: Merck ZIC-HILIC , PEEK Column , 100 mm × 2 . 1 mm ID , 3 . 5 µm , 200 Å; flow rate 0 . 3 ml/min . The gradient elution was from 70% to 10% ( B ) in 5 min . The mobile phase consisted of A: 90 mM ammonium acetate in water and B: acetonitrile/10 mM ammonium acetate buffer ( 9:1 vol/vol ) ; the total run time was 9 min . Samples were diluted 4-fold with acetonitrile , and 10 μl of the sample was injected . The retention times were 2 . 32 min and 2 . 57 min for GDP and GTP , respectively . Ionization efficiency and the fragmentation pattern were evaluated by infusion of standard solutions of 10 μM GDP and GTP . Mass spectrometry method: three scan events were monitored . Scan 1: full scan with the mass range of 160–800 Da . Scan 2: CID ( collision induced dissociation ) of m/z = 442 Da ( GDP ) . Scan 3: CID of m/z = 522 Da ( GTP ) . LCquan software was used for data analysis . Fragment ions m/z = 344 Da ( GDP ) and m/z = 424 Da ( GTP ) were utilized for GDP and GTP detection . The analyte identification was performed based on the retention time of analytes and their fragmentation pattern comparing unknown and standard samples . After each sample or standard , a blank reagent was injected to minimize possible carryover . Protein Data Bank Accession Code is 4W7G .	Living cells are constantly bustling with activity . They take in nutrients , carefully split their genetic information between new cells when they divide , and move their internal components into the right positions . To move these cargos around , the cell uses proteins—such as dynein—that essentially walks along the cell's internal scaffolding by making step-like movements . However , how a dynein motor protein is tethered to its cargo is not known in detail . One part of the dynein structure thought to play an important role in binding the motor to its cargo is called the light intermediate chain ( LIC ) . Schroeder et al . used X-ray crystallography to solve the structure of the light intermediate chain of dynein motors from a fungus . This information with other experimental techniques reveals that the LIC subunit has two distinct regions: one that binds to three different proteins that serve as adapters for cargo attachment , and one that binds to the rest of the dynein motor . The structure of the LIC includes a fold that is also found in many proteins belonging to a family of enzymes called GTPases , suggesting that the LIC evolved from this family . GTPases use a molecule called GTP to release energy and often act as on–off switches for various processes inside cells . However , the fungal LIC subunit cannot bind to molecules called nucleotides—which can act as energy sources—the way GTPases do . This prevents the LIC subunit from acting as a molecular switch . In contrast , the human version of the LIC is able to bind to some nucleotides , in particular one called GDP . However , since the LIC cannot bind to the high-energy nucleotide GTP , the human LICs most likely also do not act as on–off switches: Schroeder et al . instead propose that the LIC may use GDP only to stabilize the protein . It remains to be seen how cargo attachment to the LIC is regulated . Further structural work and biochemistry with the LIC bound to the dynein motor and cargo will provide more insight into the mechanism of intracellular cargo transport .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	A Ras-like domain in the light intermediate chain bridges the dynein motor to a cargo-binding region
Theory predicts that sexual reproduction can either facilitate or restrain transposable element ( TE ) accumulation by providing TEs with a means of spreading to all individuals in a population , versus facilitating TE load reduction via purifying selection . By quantifying genomic TE loads over time in experimental sexual and asexual Saccharomyces cerevisiae populations , we provide direct evidence that TE loads decrease rapidly under asexual reproduction . We show , using simulations , that this reduction may occur via evolution of TE activity , most likely via increased excision rates . Thus , sex is a major driver of genomic TE loads and at the root of the success of TEs . Self-replicating transposable elements ( TEs ) can occupy large fractions of genomes in organisms throughout the tree of life ( reviewed in Hua-Van et al . , 2011 ) . Their overwhelming success is driven by their ability to proliferate independently of the host cell cycle via different self-copying mechanisms involving ‘cut-and-paste’ or ‘copy-and-paste’ systems . These mechanisms allow TEs to invade genomes in a similar way to parasites , despite generally not providing any advantage to the individual carrying them ( Doolittle and Sapienza , 1980; Orgel and Crick , 1980 ) . To the contrary , TEs generate deleterious effects in their hosts by promoting ectopic recombination and because most new TE insertions in coding or regulatory sequences disrupt gene functions ( Finnegan , 1992; Montgomery et al . , 1991 ) . Theory predicts that sexual reproduction can either facilitate or restrain the genomic accumulation of TEs , and it is currently unclear whether the expected net effect of sex on TE loads is positive or negative . Sexual reproduction can facilitate the accumulation of TEs because it allows TEs to colonise new genomes and spread throughout populations ( Hickey , 1982; Zeyl et al . , 1996 ) . Because the colonisation of new genomes is more likely for active TEs , sexual reproduction should favour the evolution of highly active TEs ( Charlesworth and Langley , 1986; Hickey , 1982 ) , even though increased activity generates higher TE loads in the host genome . At the same time , sexual reproduction facilitates the evolution of host defences and increases the efficacy of purifying selection against deleterious TE copies by reducing selective interference among loci ( Ågren and Wright , 2011; Arkhipova and Meselson , 2005; Crespi and Schwander , 2012; Wright and Finnegan , 2001 ) . In the absence of sex , reduced purifying selection can thus result in the accumulation of TEs , unless TE copies get eliminated via excision at sufficiently high rates ( Burt and Trivers , 2006; Dolgin and Charlesworth , 2006 ) . Genomic TE loads have been empirically estimated for natural populations of asexual and related sexual organisms , but no consistent difference emerges ( Ågren et al . , 2015; Bast et al . , 2016; Jiang et al . , 2017; Szitenberg et al . , 2016 ) , probably because many confounding factors not related to reproductive mode such as hybridisation and polyploidisation can affect TE loads and mask the effect of sex ( Arkhipova and Rodriguez , 2013; Hua-Van et al . , 2011 ) . Here , to quantify whether the net effect of sexual reproduction on genomic TE loads is positive or negative , we study the evolution of genomic TE loads in experimental yeast ( Saccharomyces cerevisiae ) populations generated in a previous study ( McDonald et al . , 2016 ) . McDonald et al . maintained four sexual and four asexual strains originating from the same haploid ancestral strain ( W303 ) under constant conditions over 1000 generations . For sexual strains , a mating event ( meiosis ) was induced every 90 generations . Sequencing of each strain was conducted at generation 0 and every 90 generations prior to mating ( for details see Materials and methods , and McDonald et al . , 2016 ) . In the present study , we use the published Illumina data to quantify TE loads in each strain for each sequenced generation . TEs in S . cerevisiae are well characterised ( Carr et al . , 2012; Castanera et al . , 2016; Voytas and Boeke , 1992 ) . S . cerevisiae TEs consist solely of ‘copy-and-paste’ elements that are flanked by long terminal repeats ( LTRs ) and are grouped into the families Ty1-Ty5 ( Voytas and Boeke , 1992 ) . The 12 . 2 Mb genome of the studied yeast strain comprises approximately 50 full-length , active Ty element copies , and 430 inactive ones ( Carr et al . , 2012 ) . Inactive copies include truncated elements as well as remnants from TE excisions ( i . e . , solo-LTR formation; Carr et al . , 2012 ) . Excisions occur by intra-chromosomal recombination between the two flanking LTRs of a TE , and result in the removal of protein-coding genes that allow for transposition . Using different computational approaches to quantify genomic TE loads in experimental yeast strains , we show that sex is required for the success of TEs , as TE loads decrease over time under asexual reproduction . For the first approach , we quantified total TE loads without distinguishing between active and inactive TEs . This was done by computing the fraction of reads that mapped to a curated S . cerevisiae TE library ( see Materials and methods ) for each yeast strain and sequenced generation . This analysis revealed that the total TE load in sexual strains remained constant over 1000 generations , but decreased in asexual strains over time ( resulting in a total reduction of 23 . 5% after 1000 generations; generation effect p<0 . 001 , reproductive mode effect p=0 . 081 , and interaction between generation and mode p<0 . 001; permutation ANOVA , Figure 1—figure supplement 1 ) . For the second approach , we focused on full-length TE copy insertions , because only those are active and can lead to increased genomic TE loads over time . Detecting specific TE insertions by aligning short-read data to a reference genome is difficult and associated with a detection bias towards TEs present in the reference genome . Moreover , because sequencing was done with population pools and not individual clones within populations , it is not possible to analyse turnover or activity of TEs within specific genomic backgrounds . Instead , we analysed the presence versus absence of specific TE insertions in each population over time . With a pipeline that combines different complementary approaches ( Nelson et al . , 2017 , see Materials and methods ) , the available sequencing data allowed us to detect 24 out of the 50 full-length insertions that are present in the reference genome of the ancestral strain at the start of the experiment ( generation 0 ) . As with the first approach , we found that the number of ( detectable ) full-length TE copies remained constant in sexual yeast strains , but decreased in asexual strains over time ( generation effect p=0 . 006 , reproductive mode effect p=0 . 033 , and interaction between generation and mode p<0 . 001; permutation ANOVA ) . In asexual strains , the estimated average number of full-length TEs decreased from approximately 50 to 41 over 1000 generations ( Figure 1 ) . This decrease could be generated by either increased TE excision rates in asexual as compared to sexual yeast , reduced transposition rates , or a combination of both mechanisms . To evaluate the relative importance of the two mechanisms , we estimated the number of losses of TEs present in the ancestral yeast strain , as well as the number of novel insertions , at each assayed generation ( Figure 2 ) . These analyses revealed that ‘ancestral’ TE insertions are lost at a higher rate in asexual than sexual strains ( generation effect p=0 . 002 , reproductive mode effect p=0 . 027 , and interaction between generation and mode p<0 . 001; permutation ANOVA ) , while we detected similar numbers of novel TE insertions ( indicating similar transposition rates ) under both reproductive modes ( generation effect p=0 . 338 , reproductive mode effect p=0 . 271 , and interaction between generation and mode p=0 . 599; permutation ANOVA ) . Taken together , our empirical observations indicate that even very rare events of sex ( here just 10 out of 990 reproduction events ) are sufficient to maintain genomic TE loads , while asexuality results in the reduction of TE loads . The parallel reduction of TE loads in different asexual strains suggests that the evolution of reduced TE activity ( the ratio of transposition to excision ) in asexual strains influences genomic TE loads more strongly than purifying selection , which should act to reduce TE loads most effectively in sexual strains . To evaluate whether these findings are plausible , we tested whether the net loss of TEs under asexualitly is predicted by a simple model of TE dynamics . As explained above , different theoretical approaches have shown that both purifying selection and activity rate evolution can affect TE loads under sexual or asexual reproduction ( Charlesworth and Langley , 1986; Dolgin and Charlesworth , 2006; Hickey , 1982 ) . However , no theoretical study has considered TE load evolution under the joint effects of the different processes . To fill this gap , we extended the individual-level simulation program of Dolgin and Charlesworth ( 2006 ) . This program allows to study the evolution of TE copy numbers in an asexual lineage as a function of TE activity ( the joint effects of transposition and excision rates ) , as well as of the strength of selection against TE insertions , which depends on the fitness cost per TE insertion . To compare TE loads in sexual and asexual lineages , we first extended the program to include events of sexual reproduction and parameterised the simulations with empirically determined values from yeast ( Blanc and Adams , 2004; Carr et al . , 2012; Garfinkel et al . , 2005 ) . We ran individual-based simulations with a range of transposition rates , excision rates and selection coefficients with and without epistasis between TE copies as pertinent for yeast ( see Supplementary file 2A ) . For all simulations , TE loads in populations undergoing sex every 90 generations decreased faster than in asexual populations , contrary to our empirical observations . This occurs because sexual events generate variation among individuals in TE loads ( and thus variation in fitness ) , which facilitates selection against deleterious TEs ( see also Dolgin and Charlesworth , 2006 ) . Different transposition rates under meiosis ( sex ) or mitosis ( asex ) did not affect this finding . Indeed , increased TE activity during meiosis only transiently increases TE loads in sexual strains . Because such activity also generates increased variation in TE loads ( and therefore in fitness ) among strains , the additional TE copies generated during meiosis are rapidly removed by purifying selection ( Figure 1—figure supplement 3 ) . In short , none of the simulations generated the empirically observed pattern of lower TE loads in asexual than sexual strains . In a second step , we therefore allowed TE activity rates to evolve over time , by introducing a modifier allele that increases excision rates . The allele has no direct fitness effect , so it can only be fixed in a population via genetic hitchhiking . In simulations that included the modifier allele , the modifier spreads rapidly to fixation in asexual strains , because it is associated with genomes that have fewer TE copies , and therefore have a higher relative fitness . As a consequence , TE activity rates decrease in asexual populations ( Figure 1—figure supplement 4 ) . By contrast , the modifier cannot spread as rapidly in sexual populations because recombination constantly breaks up the association between the modifier and less TE loaded backgrounds . By allowing for the evolution of TE activity rates in our simulations , we were able to identify parameter values representative for yeast that result in simulations with a very close fit to our empirical results ( Figure 1B , Supplementary file 2B ) . These analyses thus corroborate our empirical findings that a likely mechanism driving genomic TE load reduction in asexual yeast strains is the rapid evolution of increased TE excision rates . A similar effect would be expected if our modifier acted on transposition rather than excision rates , since the net TE activity depends on the relative rates of transposition vs excision . However , our empirical results do not suggest major differences in transposition rates between sexual and asexual yeast strains . In combination with our findings that , in the absence of TE activity evolution , sexual strains always lose TEs faster than asexual ones , the empirical results are best explained by an increase in TE excision rates under asexuality ( Figures 1 and 2 ) . Our study shows that sexual reproduction permits the maintenance of TEs in S . cerevisiae , while in its absence , TE loads decrease , likely via the evolution of TE activity rates . The findings are consistent with empirical findings of low TE activity in old asexuals ( Bast et al . , 2016 ) and the idea that TEs should evolve to be benign in asexual species , because the evolutionary interests of TEs and their host genome are aligned ( Charlesworth and Langley , 1986 ) . While the exact mechanisms causing TE activity change in the asexual yeast populations cannot be assessed in the empirical data , our simulations suggest that there is some form of TE defense mechanism ( a ‘modifier locus’ ) that either segregates in the ancestral yeast strain used in the experiments or repeatedly appeared de novo during experimental evolution . Independently of the exact mechanism , we confirm that TE loads do not increase , but decrease , in asexual populations . This contrasts with the hypothesis that most asexual species are evolutionarily short lived because they are driven to extinction via negative consequences of accumulating TE copies ( Arkhipova and Meselson , 2005 ) . Instead , sex is at the root of the evolutionary success of parasitic TEs . We used data generated in a previous study based on experimental evolution of the yeast S . cerevisiae ( for in-depth details see McDonald et al . , 2016 ) . In short , 12 different strains were initiated from the same pool of ancestral strains ( derived from haploid W303 strains ) and kept under constant conditions . Sexual reproduction in yeast depends on the presence of two separate mating types . Only individuals with different mating types can fuse and go through meiosis . Asexual reproduction occurs through budding . For the experiment , six haploid strains consisting of mating type a ( MATa ) and six haploid strains of mating type α ( MATα ) , were grown over 990 generations . Of these , four strains were grown exclusively asexually ( two of MATa , two of MATα ) , while the eight others ( four of MATa , four of MATα ) were mixed for mating events every 90 generations , resulting in four sexual strains . Paired-end Illumina reads were generated for each of the 12 different strains every 90 generations during 990 generations ( for a total of 11 sequencing events per strain ) . Read numbers per sample ranged from 12 , 775 to 10 , 270 , 312 , averaging 2 , 964 , 869 reads per sample , with a total of 818 , 303 , 966 reads . Details of the read data can be found at BioProject PRJNA308843 and in the original study ( McDonald et al . , 2016 ) . The genome of the haploid W303 S . cerevisiae strain was retrieved from Lang et al . ( 2013 ) . All Illumina paired-end raw reads of the 12 replicate strains generated in McDonald et al . ( 2016 ) were downloaded from the SRA ( BioProject identifier PRJNA308843 ) . Raw reads were quality filtered by first removing adapter sequences ( with the script used in the original study; McDonald et al . , 2016; provided by Daniel P Rice , Harvard University ) , followed by removing the first 10 bases and quality trimming using trimmomatic v0 . 33 ( Bolger et al . , 2014 ) with parameters set to LEADING:3 TRAILING:3 HEADCROP:10 SLIDINGWINDOW:4:15 MINLEN:36 . Additionally , non-overlapping paired-reads were constructed in silico from the subset of the original paired-reads that were overlapping , as a prerequisite to run the insertion detection pipeline . For this , overlapping reads ( on average overlapping by 16 bp ) were merged using PEAR v0 . 9 . 6 with standard parameters ( Zhang et al . , 2014 ) . Merged reads were split in half and 20 bp deleted from each read at the overlapping ends using the fastx_toolkit v 0 . 0 . 13 . 2 ( Hannon Laboratory , 2010 ) . This resulted in mean read lengths of 72 bp . These ‘artificial’ non-overlapping read pairs were afterwards merged with the read set fraction that was non-overlapping . A S . cerevisiae specific , curated and updated TE library that contained all consensus sequences of all TE families found in this species is available from Carr et al . ( 2012 ) . With this library , we identified TE content and specific copy insertions in the W303 genome using RepeatMasker v4 . 02 ( Smit et al . , 2015 ) with parameters set to -nolow -gccalc -s -cutoff 200 -no_is -nolow -norna -gff -u -engine rmblast . For overall TE load estimates , the fraction of reads mapped to TEs out of total mappable reads was calculated . For this , the TE library was appended to the masked W303 genome and all reads for all strains and generations were mapped using BWA v0 . 7 . 13 with standard parameters ( Li , 2013 ) . For all strains , mean per-base coverage was checked with bedtools genomecov v2 . 26 ( Quinlan and Hall , 2010 ) , upon which the asexual strain sample 3D-90 was excluded from all further analyses , as coverage was lower than one-fold for this sample . Following this analysis , stat-reads from the PopoolationTE2 v1 . 10 . 04 program ( Kofler et al . , 2016 ) was utilised to extract the number of total mapped reads and reads mapped to TEs . For statistics , a permutation ANOVA with the formula lm ( coverage ~generationmode ) was utilised; for details see github repository ( Bast and Jaron , 2019 , copy archived at https://github . com/elifesciences-publications/reproductive_mode_TE_dynamics ) . To detect specific reference ( present in the reference genome ) and non-reference TE insertions in all samples , the McClintock pipeline was utilised ( Nelson et al . , 2017 ) . This pipeline combines six different , benchmarked programs in a standardised fashion . McClintock was run with the non-overlapping read set , the curated TE library , and the W303 assembly using default parameters . The nonredundant insertions output file per sub-program was collected . Next , we utilised a custom python script to collect all information on insertions detected by all different programs and counted insertions with evidence from different programs only once . To identify full-length TEs and solo LTR insertions from the McClintock custom filtered output , we tagged insertions by length according to the typical TY TE properties found in S . cerevisiae ( i . e . a full TE is a combination of internal sequence and two LTRs within a 500 bp range; solo LTRs are between 220 and 420 bp; see Supplementary file 1 ) . Because TE insertion detection was influenced by the coverage , coverage was taken into account when calculating the number of insertions , by adding it as random factor ( coverage effect p<0 . 001 , generation effect p=0 . 006 , reproductive mode effect p=0 . 033 , and interaction between generation and mode p<0 . 001; permutation ANOVA with the formula lm ( counts ~ coverage+generationmode ) ; for details see github repository , Bast and Jaron , 2019 ) . We then calculated the number of lost TEs in asexual strains from the regression slope in asexuals after correcting for coverage ( i . e . computing residuals ) over 1000 generations , with 50 full-length TEs in the ancestor . To additionally check for a bias due to coverage differences between sexual and asexual strains , we randomly subsampled read data for each sample corresponding to the mean coverage of the asexual strains for each generation ( Figure 1—figure supplement 2 ) . To model TE dynamics in yeast we adjusted an individual based , forward in time simulator by Dolgin and Charlesworth ( 2006 ) . We extended the model to include sexual cycles via fusion of two haploid individuals and recombination , with on average one crossover on each of the 16 modelled chromosomes ( yeast has 16 chromosomes; Goffeau et al . , 1996; McDonald et al . , 2016 ) . Each chromosome carries 200 loci that are potential targets for a TE insertion . A simulation is initiated with a single individual with 50 TEs randomly placed in the 3200 loci of the genome . The founder individual then populates clonally the whole simulated deme of explicitly simulated 100 , 000 individuals . With currently available computational resources , there was no need to scale deterministic parameters of the model as was done in the original study by Dolgin and Charlesworth ( 2006 ) . To account for mutations during this phase , we ran 20 burn-in generations of transposition and excision cycles on every individual separately without applying selection . One generation in the simulation consists of a round of selection and reproduction with transposition occurring during reproduction , followed by excision . The relative fitness wn of an individual carrying n TEs was modelled as wn= exp ( -an-12bn2 ) , where a and b are parameters representing the strength of selection and the strength of epistatic interactions between TEs respectively ( Dolgin and Charlesworth , 2006 ) . The simulation was then continued for 990 generations . We performed 10 replicates of each simulation . Using the average TE load in the population measured every 10 generations , we fitted a linear model to estimate average TE loss across the ten replicates of each simulation . Parameters were derived from yeast experimental measurements and simulations were run with perturbation in the surrounding parametric space ( see Supplementary file 2A ) . We further explored the effects of different transposition rates during meiosis vs asexual reproduction , but this did not change the dynamics even for meiotic transposition rates that were not biologically plausible ( up to 10% of TEs transposing during meiosis ) . The last extension included the introduction of an unlinked , general modifier allele increasing the excision rates of all elements by the same amount . The parameters related to this extension are the initial frequency of the modifier allele and the excision rate increases when the modifier allele is present ( see Supplementary file 2B ) . See the code documentation for details . Raw read data of the experiment are available at SRA ( BioProject identifier PRJNA308843 ) . The code used for both the analyses of empirical data and for the theoretical prediction of TE dynamics together with explanations are available online at https://github . com/KamilSJaron/reproductive_mode_TE_dynamics ( copy archived at https://github . com/elifesciences-publications/reproductive_mode_TE_dynamics ) .	The genetic information of most living organisms contains parasitic invaders known as transposable elements . These genetic sequences multiply by copying and pasting themselves through the genome , but this process can disrupt the activity of important genes and put the organism at risk . How transposable elements proliferate in a population depends on the way organisms reproduce . If they simply clone themselves asexually , the selfish elements cannot spread between the different clones . If the organisms mate together their respective transposable elements get mixed , which helps the sequences to spread more easily and to potentially become more virulent . However , sexual reproduction also comes with mechanisms that keep transposable elements in check . Bast , Jaron et al . took advantage of the fact that yeasts can reproduce with or without mating to explore whether sexual or asexual organisms are better at controlling the spread of transposable elements . The number of copies of transposable elements in the genomes of yeast grown sexually or asexually was assessed . The results showed that sexual populations kept constant numbers of selfish elements , while asexual organisms lost these genomic parasites over time . Simulations then revealed that this difference emerged because a defense gene that helps to delete transposable elements was spreading more quickly in the asexual group . The work by Bast , Jaron et al . therefore suggests that sex is responsible for the evolutionary success of transposable elements , while asexual populations can discard these sequences over time . Sex therefore helps genetic parasites , somewhat similar to sexually transmitted diseases , to spread between individuals and remain virulent .	[ "Abstract", "Introduction", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "short", "report" ]	2019	Asexual reproduction reduces transposable element load in experimental yeast populations
Taste circuits are genetically determined to elicit an innate appetitive or aversive response , ensuring that animals consume nutritious foods and avoid the ingestion of toxins . We have examined the response of Drosophila melanogaster to acetic acid , a tastant that can be a metabolic resource but can also be toxic to the fly . Our data reveal that flies accommodate these conflicting attributes of acetic acid by virtue of a hunger-dependent switch in their behavioral response to this stimulus . Fed flies show taste aversion to acetic acid , whereas starved flies show a robust appetitive response . These opposing responses are mediated by two different classes of taste neurons , the sugar- and bitter-sensing neurons . Hunger shifts the behavioral response from aversion to attraction by enhancing the appetitive sugar pathway as well as suppressing the aversive bitter pathway . Thus a single tastant can drive opposing behaviors by activating distinct taste pathways modulated by internal state . Gustatory systems have evolved to identify appetitive substances of nutritional value and to elicit avoidance of toxic compounds . Organisms may encounter food sources that also contain harmful substances , and this poses an interesting perceptual problem . Drosophila melanogaster , for example , dines on fermenting fruit that contains both appetitive and aversive compounds . During fermentation , yeast and bacteria enzymatically convert six-carbon sugars into ethanol and acetic acid . These fermentation products can be toxic to the fly , yet the scent of decaying fruit is attractive and moreover the fly feeds despite the presence of these toxic compounds ( McKenzie and Parsons , 1972; McKenzie and McKechnie , 1979; Zhu et al . , 2003 ) . These observations suggest that adaptive mechanisms may have evolved to ensure that toxic products do not impair the hungry fly from approaching and feeding on decaying fruits . Cider vinegar , for example , contains the potentially toxic metabolite acetic acid but elicits strong odor-evoked attraction ( Semmelhack and Wang , 2009 ) . Drosophila melanogaster , termed the vinegar fly , is more resistant to the toxic effects of acetic acid than other Drosophila species that do not depend upon fermenting food sources and shows greater tolerance to acetic acid than structurally similar carboxylic acids ( McKenzie and McKechnie , 1979; Parsons , 1980; Chakir et al . , 1993 ) . Moreover , flies can utilize acetic acid as a caloric source when deprived of other food sources ( Parsons , 1980; Hoffmann and Parsons , 1984 ) ; acetic acid is converted into acetyl-CoA and metabolized by the tricarboxylic acid cycle . Thus D . melanogaster may have evolved specific adaptations that allow the fly to recognize acetic acid as an appetitive tastant despite its potential toxicity . We have explored the seemingly paradoxical effects of acetic acid on feeding behavior in the vinegar fly . Feeding is initiated by extension of the proboscis , a behavior that allows the fly to taste a potential food source ( Dethier , 1976 ) . Flies recognize a relatively small number of basic taste categories , including sweet , salty , bitter , sour , fat , and carbonation ( Liman et al . , 2014; Zhang et al . , 2013; Jaeger et al . , 2018; Chen and Amrein , 2017; Masek and Keene , 2013; Fischler et al . , 2007 ) . As in mammals , most tastants excite only one class of sensory neurons and each class is thought to activate determined neural pathways to elicit innate behavioral responses ( Liman et al . , 2014 ) . For example , activation of sugar-responsive neurons drives appetitive feeding responses , whereas bitter-responsive neurons elicit aversion and suppress feeding ( Liman et al . , 2014; Marella et al . , 2006 ) . Sour taste , evoked by acids , is less well-understood than other taste modalities . Acids are potentially toxic to animals and may also indicate that food is unripe or spoiled . Both flies and mammals generally exhibit taste aversion to strongly acidic food ( Charlu et al . , 2013; DeSimone et al . , 2001 ) . However , flies do not uniformly avoid acidic stimuli: they show greater sugar consumption at an acidic pH than at a neutral or basic pH ( Deshpande et al . , 2015 ) , and at low concentration acids may counteract the repulsive effect of bitter compounds on feeding ( Chen and Amrein , 2014 ) . Flies also prefer to lay eggs on carboxylic acids such as acetic and citric acid ( Joseph et al . , 2009; Chen and Amrein , 2017 ) . This ovipositional preference is mediated by taste sensory neurons in the legs that respond specifically to acids ( Chen and Amrein , 2017 ) . Dedicated acid-sensing neurons in the proboscis have not been identified , although acid responses in bitter-sensing neurons have been observed ( Charlu et al . , 2013; Rimal et al . , 2019 ) . The activation of gustatory neurons in flies and mammals elicits innate behavioral responses , but these responses can be modulated by internal states such as satiety or hunger . Hunger elicits several adaptive changes in behavior: increased food-seeking and food consumption , enhanced locomotor activity , decreased sleep , and altered olfactory and taste sensitivity ( Sternson et al . , 2013; Itskov and Ribeiro , 2013; Pool and Scott , 2014; Yang et al . , 2015 ) . In flies , starvation increases sugar sensitivity , which promotes feeding , and decreases bitter sensitivity , which enhances acceptance of food sources that contain aversive tastants ( Inagaki et al . , 2012; Inagaki et al . , 2014 ) . Starved flies also show enhanced olfactory attraction to cider vinegar , which facilitates food search behavior ( Root et al . , 2011 ) . These hunger-dependent changes in both olfactory and gustatory sensitivity result , at least in part , from alterations in sensory neuron activity ( Inagaki et al . , 2012; Inagaki et al . , 2014; Root et al . , 2011 ) . Acetic acid , a product of fruit fermentation , signals the presence of food preferred by the fly and may also serve as a caloric source ( Parsons , 1980; Hoffmann and Parsons , 1984 ) . Acetic acid , however , can be toxic and flies avoid residing on food containing acetic acid ( Parsons , 1980; Joseph et al . , 2009 ) . We have examined the behavioral and neural responses to the taste of acetic acid and observe that hunger induces a dramatic switch in the behavioral response to this metabolite . Fed flies show taste aversion to acetic acid whereas starved flies exhibit a strong appetitive response . Genetic silencing demonstrates that the bitter-sensing neurons mediate acetic acid aversion whereas the sugar-sensing neurons mediate the appetitive response to acetic acid . Hunger shifts the response from aversion to attraction by enhancing the sugar pathway as well as suppressing the bitter pathway . Calcium imaging reveals that acetic acid activates both sugar- and bitter-sensing neurons in both the fed and starved state . Thus , a single tastant activates two distinct neural pathways that elicit opposing behaviors dependent upon internal state . This hunger-dependent switch may reflect an adaptive response to acetic acid , a potential toxin that can also afford nutritional value under extreme conditions . The taste response to acetic acid was analyzed by examining the proboscis extension response ( PER ) , an appetitive response that initiates feeding ( Dethier , 1976 ) . Appetitive tastants elicit PER when applied to the legs or labellum , the distal segment of the proboscis . Aversive tastants do not elicit PER and diminish the PER elicited by an attractive tastant ( Dethier , 1976 ) . In fed flies , exposure of acetic acid ( 1–10% ) to the labellum elicited very weak levels of PER similar to those induced by water ( Figure 1A ) , even though these flies showed strong PER to sucrose ( Figure 1B ) . Moreover , when acetic acid was mixed with either 300 mM or 50 mM sucrose it strongly reduced sucrose-evoked PER in fed flies ( Figure 1C–D ) . 79% of fed flies exhibited PER to 300 mM sucrose alone , and this response was reduced to 44% when 10% acetic acid was added ( Figure 1C ) . These data demonstrate that acetic acid elicits taste aversion in fed flies . A dramatic switch was observed in the behavioral response of starved flies . When acetic acid was applied to the labellum of one- or two-day starved flies , strong , dose-dependent PER was observed , with 86% of flies exhibiting PER to 10% acetic acid after two days of starvation ( Figure 1A ) . Moreover , when acetic acid was added to concentrations of sucrose ranging from 5 mM to 300 mM , no suppression of sugar-evoked PER was observed in two-day starved flies; in fact , the addition of acetic acid enhanced PER to 5 or 10 mM sucrose ( Figure 1C–D and Figure 1—figure supplement 1 ) . In one-day starved flies , acetic acid weakly suppressed PER at 50 mM but not 300 mM sucrose ( Figure 1C–D ) . Thus the behavior of starved flies , especially after two days of food deprivation , contrasts sharply with the strong PER suppression induced by acetic acid in fed flies , indicating that acetic acid is aversive to fed flies but becomes appetitive after starvation . This switch from an aversive response in fed flies to an appetitive response in starved flies is specific for acetic acid and was not observed for other aversive tastants , such as the bitter compounds quinine and lobeline ( Figure 1—figure supplement 2 ) . Appetitive compounds such as sugar also do not elicit a qualitative switch in behavior , since sugar elicited PER in both fed and starved flies ( Figure 1B ) . Thus , acetic acid appears unique in its ability to elicit opposing behavioral responses dependent upon internal state . We next performed experiments to demonstrate that PER elicited by acetic acid is a component of an appetitive feeding response . When a fly is stimulated asymmetrically with an appetitive tastant on only one leg , extension of the proboscis is observed in the direction of the stimulus ( Saraswati , 1998; Schwarz et al . , 2017 ) . We first confirmed that acetic acid elicits strong PER in starved flies when applied to the legs instead of the labellum ( Figure 1E ) . We then stimulated the legs asymmetrically and observed that in 72% of trials , starved flies that showed PER extended the proboscis in the direction of the stimulus ( Figure 1F , Video 1 ) . Proboscis extension was never observed in the direction opposing the stimulus , and in 28% of trials flies exhibited PER neither toward nor away from the stimulus ( Figure 1F ) . When afforded the option to consume 5% acetic acid following PER , 7 of the 10 flies tested consumed it ( Video 2 ) . Thus , the majority of flies extend their proboscis in the direction of an acetic acid stimulus and voluntarily consume it , suggesting that this response is an appetitive component of feeding behavior . Acetic acid exists in solution as three chemical species: undissociated acetic acid , which partially dissociates to produce acetate and protons . We asked whether PER to acetic acid reflects a more general taste response to low pH . Starved flies failed to show PER to hydrochloric acid at pH values equivalent to those of 5% or 10% acetic acid , which elicit strong PER , indicating that low pH is not sufficient to induce an appetitive response ( Figure 1—figure supplement 3A ) . We also tested the response of starved flies to potassium acetate at molarities equivalent to those of 5% or 10% acetic acid and failed to observe a response ( Figure 1—figure supplement 3B ) . These experiments suggest that neither protons nor acetate ions are capable of eliciting PER , suggesting that undissociated acetic acid is recognized by taste cells . In accord with this suggestion , propionic acid , a simple carboxylic acid structurally similar to acetic acid , elicited strong PER in starved flies whereas the more distantly related citric acid elicited a weaker response ( Figure 1—figure supplement 3C ) . Thus , undissociated small aliphatic acids may be recognized by gustatory neurons to elicit PER in starved flies . Proboscis extension can be elicited by the taste organs , but it remains possible that other sensory modalities such as olfaction contribute to this behavioral response . Acetic acid activates olfactory sensory neurons ( Ai et al . , 2010 ) . We therefore removed the olfactory organs , the third antennal segment and maxillary palp , from two-day starved flies and observed that PER to acetic acid was unperturbed by this manipulation ( Figure 2A ) . 76% of flies lacking olfactory organs showed PER to 10% acetic acid , a value close to that observed with control flies ( 72%; Figure 2A ) . Fed flies lacking olfactory organs failed to show PER to acetic acid , mirroring the behavior of control flies ( Figure 2B ) . Fed flies lacking olfactory organs also continued to show aversion to acetic acid , as we observed significant suppression of PER when acetic acid was added to sucrose ( Figure 2C ) . These experiments demonstrate that both the appetitive and aversive proboscis extension responses to acetic acid are observed in the absence of olfactory organs , and are likely to be mediated by the gustatory system . We demonstrated the requirement for taste neurons by examining the response to acetic acid in pox-neuro ( poxn ) mutants , in which taste bristles are transformed into mechanosensory bristles lacking gustatory receptors ( Boll and Noll , 2002 ) . Starved poxn∆M22-B5 homozygous mutants failed to display PER to any concentration of acetic acid tested ( Figure 2D ) . In contrast , wild-type flies , poxn∆M22-B5/+heterozygotes ( which have normal bristles ) , and rescue flies ( poxn∆M22-B5 mutants carrying the SuperA rescue transgene; Boll and Noll , 2002 ) showed strong PER , with up to ~50–70% of flies responding ( Figure 2D ) . Interpretation of these experiments must be tempered by the observation that the poxn∆M22-B5 mutants often appeared physically smaller than control flies and are known to have central nervous system abnormalities in addition to their lack of taste bristles ( Boll and Noll , 2002 ) . Nonetheless , these experiments suggest that the appetitive and aversive responses to acetic acid require the taste organs and are largely independent of olfaction . Neurons in the chemosensory bristles of the labellum detect distinct taste modalities , including sugar , bitter , water , and low and high concentrations of salt ( Liman et al . , 2014; Marella et al . , 2006; Cameron et al . , 2010; Zhang et al . , 2013; Jaeger et al . , 2018 ) . We employed genetic silencing to identify the neuronal classes responsible for the appetitive and aversive responses to acetic acid . Sugar-sensing neurons express multiple chemoreceptors and elicit PER in response to sugars ( Dahanukar et al . , 2007; Slone et al . , 2007; Fujii et al . , 2015 ) . The receptor Gr64f is expressed in all sugar-responsive taste neurons ( Fujii et al . , 2015 ) . We therefore silenced the sugar-sensing neurons by expressing UAS-Kir2 . 1 , encoding an inwardly rectifying potassium channel ( Baines et al . , 2001 ) , under the control of the transcriptional activator Gr64f-Gal4 . Starved flies harboring both the Gr64f-Gal4 and UAS-Kir2 . 1 transgenes showed very low frequencies of PER ( 12–28% ) in response to increasing concentrations of either sucrose or acetic acid ( Figure 3A–B ) . Control flies containing either the Gr64f-Gal4 or UAS-Kir2 . 1 transgenes alone resembled wild-type flies and exhibited strong PER to both sucrose and acetic acid , with 100% of flies responding to sucrose and ~70–80% responding to acetic acid ( Figure 3A–B ) . These experiments demonstrate that the appetitive response to acetic acid observed in starved flies is mediated by the sugar-sensing neurons . We asked whether the acetic acid response is mediated by the gustatory receptors ( Grs ) that detect sugars . Eight Grs are expressed in sugar sensory neurons of the labellum ( Fujii et al . , 2015; Yang et al . , 2015 ) . As expected , homozygous flies carrying deletions in all eight sugar-sensing Gr genes ( ∆8Grs/∆8Grs; Yavuz et al . , 2014 ) showed a strong reduction in PER to sucrose as compared with control heterozygous flies ( Figure 3C ) . Homozygous mutant flies did exhibit some residual PER to sucrose , which may reflect the presence of additional uncharacterized sugar receptors . In contrast to their strongly reduced response to sugar , homozygous mutant flies continued to show PER to acetic acid ( Figure 3D ) . Interestingly , acetic acid-evoked PER in homozygous mutant flies was significantly greater than in control flies ( Figure 3D ) . This increase in PER may reflect the possibility that the sugar-sensing circuit is upregulated in the mutants due to diminished sensory activity or intensified hunger . Alternatively , the absence of sugar receptors at the dendritic membrane may allow for increased accumulation of acetic acid receptors , or acetic acid and sucrose transduction pathways may employ a common limiting component that is no longer limiting in mutant sugar-sensing neurons . Overall , these experiments demonstrate that the response to acetic acid in starved flies is mediated by the sugar-sensing neurons but does not employ the sugar receptors . The sugar-sensing neurons also elicit PER in response to fatty acids , such as hexanoic and octanoic acid , through a molecular mechanism distinct from sugar detection ( Masek and Keene , 2013; Tauber et al . , 2017; Ahn et al . , 2017 ) . We therefore asked whether sugar neurons recognize acetic acid and fatty acids through the same mechanism , since hexanoic and octanoic acids also are aliphatic carboxylic acids . A previous study showed that PER induced by fatty acids requires phospholipase C ( PLC ) signaling in sugar-sensing neurons whereas PLC is dispensable for PER to sucrose ( Masek and Keene , 2013 ) . We therefore tested whether PLC signaling in sugar neurons is required for PER to acetic acid . An RNAi transgene targeting the gene norpA , a fly ortholog of PLC , was expressed in sugar neurons under the control of Gr64f-Gal4 . RNAi inhibition of norpA expression in the sugar neurons severely reduced PER to fatty acids whereas PER to either sucrose or acetic acid was unaffected ( Figure 3E–G ) . These data suggest that the appetitive response to acetic acid is mediated by sugar-sensing neurons , but engages molecular pathways distinct from those employed in the detection of either sugars or fatty acids . We next identified the neurons that mediate the aversive response to acetic acid in fed flies . Multiple classes of bitter sensory neurons reside in the labellum , and each of these neurons expresses the receptor Gr66a ( Weiss et al . , 2011 ) . We therefore employed the regulatory sequences of Gr66a to drive the expression of Kir2 . 1 to silence the bitter neurons . In initial experiments we demonstrated the efficacy of Kir2 . 1 silencing . Starved control flies exhibit PER to sucrose , and this response is strongly diminished by the addition of the bitter compounds quinine or lobeline ( Figure 4—figure supplement 1A–B ) . This suppression of PER by bitter compounds is no longer observed when Kir2 . 1 is expressed in bitter neurons , demonstrating the efficacy of Kir2 . 1 silencing ( Figure 4—figure supplement 1A–B ) . We therefore employed Kir2 . 1 to examine the effect of silencing bitter neurons on the aversive responses to acetic acid in fed flies . In control fed flies acetic acid suppressed PER to sucrose , but this suppression was largely eliminated when the bitter neurons were silenced ( Figure 4A ) . In controls , ~80% of flies exhibited PER to sucrose alone , and this was reduced to ~20–30% upon addition of 10% acetic acid . In contrast , when bitter neurons were silenced over 80% of flies continued to exhibit PER upon exposure to a mixture of sugar and 10% acetic acid ( Figure 4A ) . These results indicate that bitter-sensing neurons mediate the aversive response to acetic acid in fed flies . We also examined the consequences of bitter neuron silencing on the responses to acetic acid alone . Starved flies exhibit PER to acetic acid alone , but this response is not observed in fed flies ( Figure 1A ) . Acetic acid elicited PER in ~20% of control fed flies , a value near baseline ( Figure 4B ) . Silencing of the bitter neurons resulted in a striking increase in the percentage of fed flies that exhibited PER to acetic acid ( ~60–80%; Figure 4B ) . Silencing bitter neurons in fed flies did not affect PER to sucrose alone , indicating that the bitter neurons do not exert nonspecific suppression of PER ( Figure 4—figure supplement 1C ) . These results afford an explanation for the observation that fed flies normally fail to exhibit PER to acetic acid . Our data suggest that acetic acid activates sugar-sensing neurons , which promote an appetitive response , but in the fed state simultaneous activation of bitter-sensing neurons completely suppresses this response . Silencing the bitter neurons eliminates this suppression , unmasking the appetitive response . Silencing the bitter neurons resulted not only in the emergence of PER to acetic acid in fed flies but also enhanced PER to acetic acid in starved flies ( Figure 4C ) . PER to 10% acetic acid was observed in 70–80% of control flies and this value increased to 96% upon bitter neuron silencing ( Figure 4C ) . Silencing bitter neurons in starved flies did not affect PER to sucrose ( Figure 4—figure supplement 1D ) . These results demonstrate that even in the starved state , bitter neurons suppress PER to acetic acid . The observation that bitter neuron silencing has a stronger effect on acetic acid-induced PER in fed flies ( Figure 4B ) than in starved flies ( Figure 4C ) suggests that activity within the bitter-sensing circuit is suppressed in the starved state . This inhibition by hunger could occur either within or downstream of bitter sensory neurons . The striking increase in PER to acetic acid after starvation results from the hunger-dependent suppression of the bitter-sensing circuit but may also reflect enhancement of the appetitive sugar-sensing pathway . We and others observe that PER to sucrose is increased by starvation , indicating that the neural pathway for sugar-sensing is upregulated by hunger ( Figure 1B; Inagaki et al . , 2012 ) . We therefore examined the relative contributions of the sugar- and bitter-sensing pathways to acetic acid-induced PER . We compared responses to acetic acid in both fed and starved flies with and without bitter neuron silencing . In control flies , starvation strongly increased PER to acetic acid:~20% of fed flies and 70–80% of starved flies responded at the highest concentration ( Figure 4D–E ) . Upon bitter neuron silencing the difference between PER in fed and starved flies was still observed , but was much smaller in magnitude: 68% of fed flies and 96% of starved flies responded at the highest concentration ( Figure 4F ) . This starvation-dependent enhancement of PER in bitter-silenced flies is likely to reflect the enhancement of the sugar-sensing circuit . These results reveal a state-dependent interaction between the bitter- and sugar-sensing circuits that affords a logic for the behavioral switch . In fed flies , bitter neurons strongly suppress the appetitive response to acetic acid mediated by sugar neurons . Hunger results in a behavioral switch that increases PER both by suppressing the bitter-sensing circuit and enhancing the sugar-sensing circuit . Two members of the ionotropic receptor ( IR ) family of chemoreceptors , IR25a and IR76b , function in tarsal sugar-sensing neurons to detect fatty acids and in a separate population of tarsal sour-sensing neurons to detect organic and inorganic acids ( Ahn et al . , 2017; Chen and Amrein , 2017 ) . These IRs also mediate salt detection in multiple classes of labellar taste neurons , including sugar-sensing neurons ( Zhang et al . , 2013; Jaeger et al . , 2018 ) . We therefore tested whether IR25a and IR76b are required for appetitive or aversive taste responses to acetic acid . We tested the Ir25a and IR76b mutant lines that exhibit impairments in fatty acid and sour taste detection as well as the control strain ( w1118 ) used in these studies ( Ahn et al . , 2017; Chen and Amrein , 2017 ) . Starved flies carrying mutations in Ir25a or Ir76b showed robust PER to 1–10% acetic acid , indicating that these receptors are not required for the appetitive response to acetic acid ( Figure 4—figure supplement 2A and D ) . Fed flies carrying IR25a or IR76b mutations showed very low levels of PER to acetic acid , indicating that the aversive pathway that suppresses acetic acid-evoked PER in the fed state remains intact ( Figure 4—figure supplement 2B and E ) . We note that both fed and starved IR25a mutant flies showed slightly lower PER to acetic acid than controls , suggesting that IR25a may contribute to this response even though it is not strictly required for acetic acid detection . We also tested the ability of acetic acid to suppress sucrose-evoked PER in fed flies . In IR25a mutants , the suppression of sucrose-evoked PER was similar or stronger than we observed in control w1118 flies ( Figure 4—figure supplement 2C ) . IR76b mutants also showed a decrease in sucrose-evoked PER as the acetic acid concentration was increased from 1% to 10% , but low concentrations of acetic acid unexpectedly enhanced their response to sucrose ( Figure 4—figure supplement 2F ) . This enhancement is likely due to the genetic background of the flies and not to the loss of IR76b because IR76b/+ heterozygotes , which carry one functional copy of IR76b , showed the same effect , and their responses to sucrose containing acetic acid did not significantly differ from homozygotes ( Figure 4—figure supplement 2G–H ) . Taken together , these experiments indicate that IR25a and IR76b are not required for either appetitive or aversive responses to acetic acid . The observation that the appetitive response to acetic acid is mediated by sugar-sensing neurons whereas the aversive response is mediated by bitter neurons suggests that acetic acid is an unusual tastant capable of activating two opposing classes of sensory cells . We therefore performed two-photon imaging of taste sensory neurons to confirm whether acetic acid activates both sugar- and bitter-sensing cells . The genetically encoded calcium indicator GCaMP6f ( Chen et al . , 2013 ) was expressed in sugar- ( Gr64f-Gal4 ) or bitter- ( Gr66a-Gal4 ) sensing neurons . Imaging was performed on sensory axon termini in the subesophageal zone ( SEZ ) of the fly brain ( Figure 5—figure supplement 1 ) . Strong GCaMP responses to acetic acid were observed in sugar neurons in both fed and starved flies , with the peak response to acetic acid about half of that observed with 500 mM sucrose ( Figure 5A–D ) . We noted that the responses to acetic acid stimuli often appeared more variable across trials and flies than responses to sucrose ( Figure 5D; Figure 5—figure supplement 2 ) . Despite this variability , sugar neurons in 27 of 28 fed or starved flies responded to acetic acid at levels greater than the water response . We also examined the acetic acid response in sugar neurons of fed homozygous mutant flies carrying deletions in all eight sugar receptors ( ∆8Grs/∆8Grs ) . The response to sucrose in these mutants was reduced to the level of the response to water , whereas the response to acetic acid was not affected ( Figure 5—figure supplement 3 ) . Acetic acid activation of sugar neurons was variable across flies , but the proportion of flies responding to acetic acid did not differ by genotype: sugar neurons in 6 of 9 control flies and 6 of 9 mutant flies showed responses to at least one concentration of acetic acid . Thus the response to acetic acid in sugar neurons does not require sugar receptors , a result consistent with the observation that sugar receptor mutants still show PER to acetic acid ( Figure 3D ) . Acetic acid elicits stronger PER in these mutants than in controls but activates the sugar neurons to similar levels in fed flies . These results suggest that the enhanced PER of the mutants is not likely due to an accumulation of acetic acid receptors or increased acetic acid transduction in sugar-sensing neurons . Instead , the diminished response of sensory neurons to sugar as well as intensified hunger may lead to upregulation of the downstream sugar circuit and result in enhanced PER without changes in sensory neuron activation . We next imaged the bitter-sensing neurons and observed significant GCaMP responses to both 1% and 5% acetic acid ( Figure 5E–H ) . As in the sugar neurons , acetic acid responses in bitter neurons often appeared more variable across trials and flies than responses to bitter ( Figure 5H; Figure 5—figure supplement 2 ) , but neurons in the majority of both fed and starved flies responded to acetic acid ( 15/18 fed flies and 16/18 starved flies ) . Acetic acid activated the bitter neurons with peak responses about 30% of those obtained with 1 mM lobeline ( Figure 5E–H ) . The difference between the levels of activity elicited by lobeline and acetic acid may reflect different sensitivities of bitter neurons to the two compounds or the activation of a smaller subset of neurons by acetic acid . We imaged the acetic acid response of four different subclasses of bitter neurons ( Weiss et al . , 2011 ) and observed that only the S-b class exhibited a significant peak response when compared to the response to water ( Figure 5—figure supplement 4 ) . This finding is consistent with studies showing the strongest acetic acid responses in bitter cells of S-type sensilla ( Rimal et al . , 2019 ) and specifically the S-b class ( Charlu et al . , 2013 ) . We do not rule out the possibility that other classes of bitter neurons also respond to acetic acid ( see Figure 5—figure supplement 4 ) . Overall , these results show that acetic acid activates both sugar- and bitter-sensing taste neurons . In both sugar- and bitter-sensing neurons , the average response magnitudes to 5% acetic acid were similar or slightly lower than the response to 1% ( Figure 5 ) . We therefore tested a broader range of acetic acid concentrations . Responses to higher concentrations such as 10% could not be accurately quantified because they frequently activated sensory neurons even before making contact with the labellum , likely due to volatile acetic acid molecules , so we focused on lower concentrations . In sugar neurons , we observed dose-dependent responses as the concentration was increased from 0 . 01% to 1% , but the response diminished slightly at 5% ( Figure 5—figure supplement 5A ) . Bitter neurons failed to show dose dependence in this concentration range ( Figure 5—figure supplement 5B ) . A diminished response over trials may contribute to the apparent lack of dose dependence , since we always tested acetic acid concentrations in ascending order . Indeed , testing 5% acetic acid prior to any other concentrations induced a much higher bitter neuron response ( 113 ± 15% ∆F/F0 , n = 18 trials , six flies ) than in other experiments where 5% acetic acid was tested last ( 54–56% ∆F/F0; Figure 5H and Figure 5—figure supplement 5B ) . Overall , we find that bitter neurons fail to show clear dose-dependent responses to acetic acid and sugar neurons show dose dependence only at low concentrations . These results suggest that acetic acid may exert a more complex effect on sensory neurons than other tastants . For example , secondary effects on neuronal activity could be induced by low pH or by undissociated acetic acid molecules , which may cross the cell membrane and directly affect intracellular pathways ( DeSimone et al . , 2001; Liman et al . , 2014 ) . Because acetic acid activates both sugar- and bitter-sensing neurons , we confirmed that acetic acid does not promiscuously activate all classes of taste neurons by imaging responses of water-sensing neurons , sensory cells in the labellum that respond to low osmolarity tastants ( Cameron et al . , 2010 ) . Water-sensing neurons responded broadly to several taste stimuli at levels that were inversely related to their osmolarity ( Figure 5—figure supplement 6 ) . When the low-osmolarity response of water-sensing neurons was blocked by adding the high molecular mass polymer polyethylene glycol ( PEG ) to each taste solution , acetic acid did not activate the water-sensing neurons beyond the level elicited by PEG alone ( Figure 5—figure supplement 6C–D ) . Acetic acid therefore activates water-sensing neurons solely by the osmolarity-sensing mechanism . Thus , the responses to acetic acid in sugar- and bitter-sensing neurons are specific and are likely to be mediated by receptors that recognize acetic acid . We next compared the responses of sugar- and bitter-sensing neurons in fed and two-day starved flies . We confirmed that the GCaMP6f-expressing flies used for imaging show behavioral changes in response to hunger: like wild-type flies , they show strong hunger-dependent increases in PER to both acetic acid and sucrose ( Figure 5—figure supplement 7 ) . Previous studies suggest that hunger increases sugar neuron responses to sucrose and suppresses bitter neuron responses to lobeline ( Inagaki et al . , 2012; Inagaki et al . , 2014; LeDue et al . , 2016 ) . We observed a trend toward increased sucrose responses in sugar neurons after starvation ( p=0 . 063 for the effect of starvation , two-way ANOVA ) , but lobeline responses of bitter neurons did not differ between fed and starved flies ( Figure 5C–D and G–H ) . Because Inagaki et al . ( 2012 ) detected differences between sucrose responses of fed and starved flies by quantifying the integrated ∆F/F0 response rather than the peak response , we also quantified integrated ∆F/F0 responses to sucrose , which showed a significant difference between fed and starved flies at 500 mM sucrose ( p<0 . 05 , two-way ANOVA followed by Bonferroni post-tests ) . Analyzing the integrated ∆F/F0 response did not reveal any significant differences between lobeline responses of bitter neurons in fed and starved flies . We then examined sensory neuron responses to acetic acid in fed and starved flies . In both sugar and bitter neurons , starved flies showed significantly higher responses than fed flies to 1% but not 5% acetic acid ( Figure 5C–D and G–H ) . Hunger increased the average peak response to 1% acetic acid from 117% to 166% ∆F/F0 in sugar neurons and from 46% to 87% ∆F/F0 in bitter neurons . In sugar neurons , but not bitter neurons , starved flies also showed a trend toward higher responses to 5% acetic acid than fed flies . Our genetic silencing data suggest that hunger elicits a behavioral switch in the acetic acid response by upregulating the sugar-sensing circuit and downregulating the bitter-sensing circuit . Thus the enhancement of the sugar neuron response to acetic acid in starved flies may contribute to the increase in PER , but the enhancement of the bitter neuron response is not consistent with the behavioral change . This effect on bitter neurons along with the lack of significant hunger-dependent changes at 5% acetic acid in either sugar or bitter neurons , despite the fact that PER to 5% acetic acid shows even greater hunger modulation than at 1% ( Figure 1A; Figure 5—figure supplement 7 ) , suggests that sensory neuron modulation is not likely to account for the behavioral switch in the acetic acid response . The striking effects of internal state on this behavior may therefore reflect state-dependent modulation of both the sugar and bitter circuits downstream of the sensory neurons . The adaptive response to acetic acid is dependent on two biological features , the ability of acetic acid to activate two classes of sensory neurons that elicit opposing behaviors and the state-dependent modulation of these two taste pathways . We observe that acetic acid activates both the bitter- and sugar-sensing neurons . The activation of bitter neurons by organic acids has been observed previously by electrophysiologic recording of labellar sensilla , and these acids result in taste aversion ( Charlu et al . , 2013; Rimal et al . , 2019 ) . The observation that bitter neurons respond not only to organic acids but also to hydrochloric acid suggested that a subset of bitter neurons serve as a pH sensor eliciting taste aversion ( Charlu et al . , 2013 ) . However , Rimal et al . ( 2019 ) recently identified IR7a as a narrowly tuned acetic acid receptor acting in bitter neurons . We have not tested whether IR7a mediates the aversive responses to acetic acid that we observe . We also observe that acetic acid activates sugar neurons , eliciting an appetitive taste response . This behavioral response is not observed upon exposure to low pH or acetate , suggesting the presence of a receptor on sugar neurons recognizing short chain aliphatic acids . Two studies have reported that whereas acetic acid activates bitter neurons , it fails to activate sugar neurons ( Charlu et al . , 2013; Rimal et al . , 2019 ) . In these studies acetic acid activated S-type sensilla , which contain both bitter and sugar neurons , but failed to activate L-type sensilla , which contain sugar neurons but not bitter neurons . These results obtained by sensillar recordings contrast with our observations that acetic acid activates the axon termini of sugar neurons . One possible explanation for this discrepancy is that sugar neurons in S-type sensilla respond to acetic acid . Rimal et al . ( 2019 ) observe that in mutants lacking IR7a , S-type sensilla continue to exhibit weak responses to acetic acid , which may represent responses of sugar neurons . We are imaging axonal activity across the entire population of ~62 labellar sugar neurons , whereas electrophysiologic studies record single cells . Weaker dendritic responses not detected by extracellular sensilla recordings may be summated to produce observable axonal responses . In addition , nonlinear amplification of spike rates into calcium responses or presynaptic facilitation by modulatory inputs could also transform weak dendritic responses into stronger axonal signals . We are confident in our result that sugar neurons show axonal GCaMP responses to acetic acid . Acetic acid activated sugar neurons in 27 of 28 flies initially tested ( Figure 5A–D ) , and overall we have observed acetic acid activation of sugar neurons with three different Gal4 drivers and two different GCaMP variants ( Figure 5A–D , Figure 5—figure supplement 3 , and data not shown ) . Moreover , the observation that acetic acid elicits PER in starved flies and this response is eliminated upon silencing activity of the sugar neurons is most consistent with the fact that sugar neurons are activated by acetic acid . It is puzzling , however , that the dose-dependence of sugar neuron responses does not parallel the dose-dependence of acetic acid-evoked PER , which increases from 1% to 10% acetic acid ( Figure 1A; Figure 5—figure supplement 7 ) . One possibility is that a subset of sugar neurons are activated dose-dependently and contributes most strongly to PER , and we may not observe this dose-dependence when imaging activity across all sugar neurons . Alternatively , despite our finding that sugar neuron activity is required for PER to acetic acid , additional sensory neurons may contribute to this behavior and detect acetic acid in a dose-dependent manner . Silencing sugar or bitter neurons largely abolishes the appetitive or aversive response to acetic acid , respectively , suggesting that we have identified the primary neurons that mediate these behaviors . However , our experiments do not rule out the possibility that other labellar taste neurons contribute to these behaviors . The fact that the leg contains dedicated acid-sensing neurons raises the possibility of whether such neurons also exist in the labellum ( Chen and Amrein , 2017 ) . The tarsal acid-sensing neurons utilize IR76b and IR25a , which are dispensable for the acetic acid-induced behaviors we have studied ( Figure 4—figure supplement 2 ) , suggesting that putative labellar acid-sensing neurons would utilize a different molecular mechanism . The activation of different classes of sensory neurons by a single tastant has been observed for salt ( Zhang et al . , 2013; Jaeger et al . , 2018 ) , the long chain fatty acid , hexanoic acid ( Ahn et al . , 2017 ) , as well as for acetic acid in this study . In both flies and mammals , low salt concentrations elicit attraction whereas high salt results in aversion and these opposing behaviors are mediated by distinct classes of sensory neurons ( Zhang et al . , 2013; Jaeger et al . , 2018; Chandrashekar et al . , 2010; Oka et al . , 2013 ) . Hexanoic acid , a caloric source , activates fatty acid receptors in sugar neurons and at high concentrations activates a different receptor in bitter neurons ( Ahn et al . , 2017 ) . A logical pattern emerges in which tastants of potential value to the fly activate attractive taste pathways . These compounds may also be toxic and also activate aversive pathways either at higher concentrations or in different internal states . This affords the fly protection from the potential toxicity of excess , a protection that can be ignored under extreme conditions to assure survival . Internal state can elicit profound behavioral changes that allow the organism to adapt to a changing internal world . Hunger , for example , results in enhanced food search and consumption , increased locomotion , changes in food preference , and altered olfactory and taste sensitivity ( Sternson et al . , 2013; Itskov and Ribeiro , 2013; Pool and Scott , 2014; Yang et al . , 2015 ) . Previous studies have observed that hunger enhances olfactory attraction to cider vinegar , increases sugar sensitivity , and decreases bitter sensitivity ( Root et al . , 2011; Inagaki et al . , 2012; Inagaki et al . , 2014 ) . These effects of hunger represent gain changes; stimuli become more attractive or aversive . By contrast , our experiments reveal a qualitative change in the valence of acetic acid: hunger induces a switch from taste aversion to attraction . Genetic silencing experiments indicate that this switch in starved flies results from enhancement of the appetitive pathway , mediated by sugar neurons , and inhibition of the aversive pathway mediated by bitter neurons . Conversely , in fed flies the appetitive pathway is inhibited whereas the aversive pathway is enhanced . Previous studies suggest that hunger modulates behavioral responses to sugar and bitter at least in part by modulating the activity of the initial neurons in these pathways , the sensory neurons ( Inagaki et al . , 2012; Inagaki et al . , 2014; LeDue et al . , 2016 ) . Our experiments imaging sensory neuron projections in the SEZ revealed a trend toward increased sugar neuron responses to sucrose in starved flies but did not show hunger modulation of the bitter neuron response to lobeline . It is possible that hunger modulation would have been apparent if we tested a greater range of concentrations or that fly to fly variability precluded us from detecting subtle differences . We did observe that hunger significantly increased the responses of both sugar and bitter neurons to 1% acetic acid , and sugar neurons also showed a trend toward enhanced responses at 5% . Thus the enhancement of sugar neuron responses may contribute to increased acetic acid-evoked PER in starved flies . However , modulation of sensory neurons is unlikely to entirely account for the behavioral switch in the acetic acid response for multiple reasons . First , behavioral data suggest that hunger suppresses the bitter circuit , but bitter-sensing neurons show an enhanced response to acetic acid . Second , flies show a greater hunger-dependent change in PER to 5% than 1% acetic acid , but sensory neuron responses only show significant modulation at 1% . Our data therefore suggest that the striking state-dependent switch in the behavioral response to acetic acid may reflect modulation of taste pathways downstream of the sensory neurons . The circuit from sensory neurons leading to proboscis extension remains largely uncharacterized but multiple nodes subject to modulation can be anticipated . The response to sugar extends to multiple behaviors beyond PER including ingestion , swallowing , and suppression of locomotion , each of which is likely to be modulated by hunger ( Pool and Scott , 2014 ) . Modulation at the level of sensory neurons affords gain control that will result in changes in all behaviors elicited by gustatory neurons . Modulation of downstream taste neurons facilitating sensorimotor transformations could afford a flexibility enabling independent control of different behavioral programs driven by the same taste stimulus . This affords genetically determined neural circuits mediating innate behaviors the opportunity for more complex modulation dependent on perception , motivation , and internal state . Flies were reared at 25°C and 70% relative humidity on standard cornmeal food . The wild-type control strain was 2U ( isoCJ1; Dubnau et al . , 2001 ) . All lines used for behavior were outcrossed into this background for at least five generations , with the exception of the ∆8Grs line which contained too many mutations to outcross and the IR25a and IR76b mutants which were tested with the w1118 controls that other studies have used ( Chen and Amrein , 2017; Ahn et al . , 2017 ) . PER assays were generally performed on 3–6 day-old mated females . Calcium imaging was performed on >1 week-old flies to ensure robust GCaMP6f expression , and PER assays for GCaMP6f-expressing flies were performed using flies of the same age . All fly strains have been described previously: Gr64f-Gal4 ( Dahanukar et al . , 2007 ) ; Gr66a-Gal4 ( Scott et al . , 2001 ) ; ppk28-Gal4 ( Cameron et al . , 2010 ) ; Gr98d-Gal4 , Gr22f-Gal4 , Gr59c-Gal4 , and Gr47a-Gal4 ( Weiss et al . , 2011 ) ; UAS-Kir2 . 1 ( Baines et al . , 2001 ) ; UAS-GCaMP6f ( Chen et al . , 2013 ) ; UAS-norpARNAi ( Masek and Keene , 2013 ) ; poxn∆M22-B5 and poxn∆M22-B5 + SuperA rescue ( Boll and Noll , 2002 ) ; ∆8Grs ( R1 , ∆Gr5a;; ∆Gr61a , ∆Gr64a-f ) and ∆8Grs with transgenes for GCaMP imaging ( R1 , ∆Gr5a; Gr61a-Gal4 , UAS-GCaMP6m; ∆Gr61a , ∆Gr64a-f ) ( Yavuz et al . , 2014 ) ; IR25a1 and IR25a2 ( Benton et al . , 2009 ) ; IR76b1 and IR76b2 ( Zhang et al . , 2013 ) . Fed flies were taken directly from food vials for testing . Starved flies were food-deprived with water ( using a wet piece of Kimwipe ) for the specified amount of time before testing . Flies were anesthetized on ice and immobilized on their backs with myristic acid . Unless otherwise specified , PER experiments were conducted by taste stimulation of the labellum . To ensure that we could deliver tastants to the labellum without contacting the legs , we immobilized the two anterior pairs of legs with myristic acid . For leg stimulation experiments ( Figure 1E–F , Video 1 , and Video 2 ) , all legs remained free . Flies recovered from gluing for 30–60 min in a humidified chamber before testing . Before testing PER , flies were water-satiated so that thirst would not affect their responses . PER to water ( the negative control ) was tested after water-satiation , followed by taste stimuli in ascending order of concentration . Flies were water-satiated again before each test . Each test consisted of two trials in which the solution was briefly applied to the labellum or legs using a small piece of Kimwipe . PER on at least one of the two trials was considered a positive response . Only full proboscis extensions , not partial extensions , were counted as PER . Flies were tested in groups of 15–20 , and the percent of flies showing PER to each tastant was manually observed and recorded . Flies that did not respond to any taste stimuli were tested with 500 mM sucrose at the end of the assay . For experiments using only wild-type starved flies , which should always respond to high concentrations of sugar unless they are extremely unhealthy , flies that failed to respond to 500 mM sucrose were excluded from analysis . For experiments comparing fed and starved flies or starved controls and mutants , flies were only excluded from analysis if they appeared very sick . For statistical analyses of PER , each group of 15–20 flies was considered to be a single data point ( ‘n’ ) . A minimum of three groups per genotype or condition were tested for each PER experiment . Because PER can vary substantially from day to day ( possibly due to changes in ambient temperature or humidity ) , control and experimental flies for a given experiment were always tested on the same days , and all experiments were repeated over multiple days . To test directional PER , we contacted the left or right forelegs with acetic acid , alternating between sides every 1–2 trials . Flies were filmed and the videos were analyzed later . We only analyzed trials in which flies showed full PER to the stimulus . Flies often showed repeated extension to a single stimulation; at least one proboscis extension toward the left or right side was considered to be a lateralized response . To test the role of olfaction , the third antennal segments and maxillary palps were removed with forceps while flies were anesthetized on ice . Surgery was performed prior to starvation , and after surgery flies were given ~30 min to recover in food vials before starvation . Control flies were anesthetized for the same duration as antennectomized flies . Flies for calcium imaging were taped on their backs to a piece of clear tape in an imaging chamber ( see Figure 5—figure supplement 1 ) . Fine strands of tape were used to restrain the legs , secure the head , and immobilize the proboscis in an extended position for tastant stimulation . A small hole was cut into the tape to expose the anterior surface of the fly’s head . A square hole along the anterior surface of the head was then cut through the cuticle , including removal of the antennae , to expose the anterior ventral aspect of the brain that encompasses the SEZ . The esophagus was cut in order to visualize the SEZ clearly . The dissection and imaging were performed in modified artificial hemolymph in which 15 mM ribose is substituted for sucrose and trehalose ( Wang et al . , 2003; Marella et al . , 2006 ) . Calcium imaging experiments were performed using a two-photon laser scanning microscope ( Ultima , Bruker ) equipped with an ultra-fast Ti:S laser ( Chameleon Vision , Coherent ) that is modulated by pockel cells ( Conoptics ) . Emitted photons were collected with a GaAsP photodiode detector ( Hamamatsu ) through a 60X water-immersion objective ( Olympus ) . A single plane through the brightest area of axonal projections was chosen for imaging . Images were acquired at 925 nm at a resolution of 256 by 256 pixels and a scanning rate of 3–4 Hz . Tastants were delivered to the labellum via a custom-built solenoid pinch valve system controlled by MATLAB software . Pinch valves were opened briefly ( ~10 ms ) to create a small liquid drop at the end of a 5 µL glass capillary , positioned such that the drop would make contact with the labellum . Tastants were removed after a fixed duration by a vacuum line controlled by a solenoid pinch valve . Proper taste delivery was monitored using a side-mounted camera ( Veho VMS-004 ) , which allowed for visualization of the fly and tastant capillary using the light from the imaging laser . At least three trials of each stimulus were given , with at least one minute rest between trials to avoid habituation . Calcium imaging data were analyzed using custom MATLAB code based largely on the code used in Hattori et al . ( 2017 ) . Images were registered within and across trials to correct for movement in the x-y plane using a sub-pixel registration algorithm ( Guizar-Sicairos et al . , 2008 ) . Regions of interest ( ROIs ) were drawn manually around the area of axonal projections . Average pixel intensity within the ROI was calculated for each frame . The average signal for 20 frames preceding stimulus delivery was used as the baseline signal ( F0 ) , and the ∆F/F0 values for each frame were then calculated . The peak stimulus response was quantified as the average of the ∆F/F0 values for the two highest consecutive frames during tastant presentation . No trials were excluded from analysis unless the tastant drop failed to make proper contact with the labellum . For fly by fly analyses , we defined a fly as responding to a tastant if the average peak response across at least three trials was higher than the average peak response to water by a magnitude of at least 15% . We also considered thresholds of 10% or 20% but found that 15% appeared to be a reasonable ( and likely conservative ) threshold for defining a fly’s response . Statistical analyses were performed using GraphPad Prism , Version 4 . The most relevant statistical results are reported in the figures and legends , and all statistical results are reported in Supplementary file 1 . All graphs represent mean ± SEM . For Gal4/UAS experiments , statistical significance was attributed only to data points for which experimental flies that differed from both the Gal4/+ and UAS/+ controls in the same direction . Sample sizes are listed in the figure legends . No explicit power analyses were used to determine sample sizes prior to experimentation . Minimum sample sizes were decided prior to experimentation based on previous experience knowing how many samples are usually sufficient to detect reasonable effect sizes . Additional samples were added if the initial results were inconclusive or more variable than expected , but never with the intent to make a non-significant p-value significant or vice versa . For experiments in which the same genotype was tested under different conditions ( e . g . fed vs . starved ) , flies from the same vials were randomly allocated into each experimental group . In general , the experimenter was not explicitly blinded to the group or genotype .	Our sense of taste is critical to our survival . Taste helps us to consume nutritious foods and avoid toxins . There are five basic taste categories: sweet , salty , bitter , sour , and umami or savory , a taste typical of protein-rich foods . Each taste category activates a distinct pathway in the brain , triggering specific feelings and behaviors . We normally find sugar , salt , and components of protein pleasant , and seek out foods with these tastes . By contrast , we often find overly bitter or sour tastes unpleasant and try to avoid them . As sour and bitter-tasting substances often contain toxins , this response helps to protect us from poisoning . Across the animal kingdom , these preferences are largely hardwired from birth . But the relationship between taste and nutrients is not always straightforward . Some substances can be toxic despite also containing useful nutrients . Overripe fruit , for example , is broken down by yeast and bacteria to produce acetic acid , or vinegar . Like other acids , acetic acid can be toxic . But for the fruit fly Drosophila melanogaster , also known as the vinegar fly , acetic acid from rotten fruit can be a valuable source of calories . So how do flies react to the taste of acetic acid ? Devineni et al . show that , unlike other chemicals , acetic acid triggers different taste responses in flies depending on whether the insects are hungry . Well-fed flies find the taste repulsive , probably because it signals toxicity . But hungry flies find it attractive , presumably because of their overriding need for calories . Devineni et al . show that acetic acid activates both sugar-sensing and bitter-sensing pathways in the fly brain . Hunger increases activity in the sugar pathway and reduces it in the bitter pathway . As a result , hungry flies are attracted to acetic acid , whereas fully fed flies are repulsed . Flexibility in the taste system enables animals to react to the same substance in different ways depending on their current needs . Related to this , evidence suggests that obesity may be associated with altered sensitivity to certain tastes , such as sweet , as well as a blunted response to satiety signals . Understanding how the brain combines information about taste and hunger to control food consumption may ultimately help us to understand and treat obesity .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	Acetic acid activates distinct taste pathways in Drosophila to elicit opposing, state-dependent feeding responses
How the brain controls the need and acquisition of recovery sleep after prolonged wakefulness is an important issue in sleep research . The monoamines serotonin and dopamine are key regulators of sleep in mammals and in Drosophila . We found that the enzyme arylalkylamine N-acetyltransferase 1 ( AANAT1 ) is expressed by Drosophila astrocytes and specific subsets of neurons in the adult brain . AANAT1 acetylates monoamines and inactivates them , and we found that AANAT1 limited the accumulation of serotonin and dopamine in the brain upon sleep deprivation ( SD ) . Loss of AANAT1 from astrocytes , but not from neurons , caused flies to increase their daytime recovery sleep following overnight SD . Together , these findings demonstrate a crucial role for AANAT1 and astrocytes in the regulation of monoamine bioavailability and homeostatic sleep . We generated antiserum to AANAT1 ( known previously as Dopamine acetyltransferase ( Dat ) ) and confirmed its specificity with immunohistochemistry ( IHC ) in the embryonic central nervous system ( CNS ) . AANAT1 immunoreactivity was observed in the cytoplasm of many cells ( Figure 1A ) but was absent in age-matched ventral nerve cords of embryos that were homozygous for a deletion of the entire AANAT1 gene ( Figure 1B ) . In adult brains co-immunostained with anti-Bruchpilot ( nc82 , Figure 1C ) , a presynaptic marker that labels neuropil regions , AANAT1 was present in distinct populations of cells throughout the brain . We found that AANAT1 was expressed in sub-populations of neurons ( anti-Elav positive ( + ) , Figure 1D , D' ) and glia ( anti-Repo+ , Figure 1E , E' ) throughout the brain . In glial cells AANAT1 was primarily cytoplasmic , but in neurons AANAT1 often appeared to localize to the nucleus . With the astrocyte-specific Alrm-Gal4 driving expression of a Red Fluorescent Protein ( RFP ) reporter ( Alrm >nuRFP ) , we confirmed that all AANAT1-positive glial cells in the central brain are astrocytes ( Figure 1F–F'' ) . Only a subset of astrocytes in the optic lobes that reside between the medulla and lobula did not express AANAT1 ( Figure 1—figure supplement 1A ) . With RNA interference ( RNAi ) -mediated knockdown of AANAT1 from all neurons using the driver nSyb-Gal4 , the AANAT1-positive glial cells in the central brain could be identified more clearly as astrocytes by their ramified morphology , where AANAT1 could be observed in their thin , neuropil-infiltrating processes ( Figure 1—figure supplement 1D , G , H ) . Their identity as astrocytes was further confirmed by the morphologies of single cells labeled with the Multi-Color Flp-OUT ( MCFO ) system ( Figure 1G ) . In contrast to astrocytes , AANAT1 expression was absent from ensheathing glia marked by R56F03-Gal4 ( Figure 1—figure supplement 1B ) . Labeling of astrocytes was confirmed to be specific for AANAT1 because it was lost upon knockdown of AANAT1 from astrocytes with Alrm-Gal4 or Repo-Gal4 ( Figure 1—figure supplement 1C , E , I , J ) . This also revealed more clearly the several clusters of AANAT1-positive neurons and their axon tracts in the central complex of the brain , which we examined in brain regions associated with sleep regulation ( Figure 1—figure supplement 1C , E ) . AANAT1 expression was largely absent from the neuropils of the mushroom body ( MB ) and fan-shaped body ( FSB ) , though there were scattered AANAT1-positive astrocytes nearby . AANAT1 expression in the neuropil of the ellipsoid body ( EB ) came almost exclusively from neurons , as revealed by neuron-selective RNAi knockdown ( Figure 1—figure supplement 1K–Q ) . Elsewhere , it appeared that astrocytes contributed far more to AANAT1 labeling of brain neuropil regions than did neurons; for example , in the antennal lobe ( Figure 1E , F ) and subesophageal ganglion ( Figure 1—figure supplement 1D ) , AANAT1 expression within neuropil regions came primarily from the infiltrative processes of astrocytes . The monoamines serotonin , dopamine , and octopamine ( the insect equivalent of norepinephrine ) are known to act in the fly brain to regulate the quantity and timing of sleep ( Nall and Sehgal , 2014 ) . Pharmacological , genetic , and thermogenetic approaches have converged to demonstrate that serotonin signaling in the fly brain increases sleep , whereas dopamine or octopamine signaling promote waking ( Andretic et al . , 2005; Artiushin and Sehgal , 2017; Kume et al . , 2005; Nall and Sehgal , 2014; Qian et al . , 2017; Wu et al . , 2008; Yuan et al . , 2005 ) . Previous studies have suggested AANAT1 to be expressed in dopaminergic neurons ( Ganguly-Fitzgerald et al . , 2006; Shao et al . , 2011 ) , but this has not been tested directly . With IHC , we examined AANAT1 co-labeling of serotonergic , dopaminergic and octopaminergic neurons using a mCD8-GFP reporter driven by either Trh-Gal4 ( Alekseyenko et al . , 2010 ) , TH-Gal4 ( Friggi-Grelin et al . , 2003 ) , or Tdc2-Gal4 ( Monastirioti et al . , 1995 ) , respectively . AANAT1 was expressed in an average of 14 . 5 ± 4 . 8 of 60 ± 3 . 4 ( 25% ) of serotonergic cells labeled with Trh >mCD8 GFP ( Figure 1H , I–I'' , N ) , which were found largely in a cluster within the medial subeosophageal ganglion ( Figure 1H ) . However , AANAT1 did not co-label cells expressing TH >mCD8 GFP or Tdc2 >mCD8 GFP ( Figure 1J , K–K'' , L , M–M'' , N ) , indicating AANAT1 is not expressed in dopaminergic or octopaminergic neurons . These results are corroborated by single-cell RNA sequencing data showing AANAT1 transcripts in astrocytes and serotonergic neurons ( Croset et al . , 2018 ) . To identify the other types of neurons expressing AANAT1 , we used a mCD8-GFP reporter driven by either MiMIC-vGlut , Gad1-Gal4 , or Cha-Gal4 and found AANAT1 in sub-populations of neurons that release glutamate , gamma-aminobutyric acid ( GABA ) , or acetylcholine , respectively ( Figure 1—figure supplement 1R-W'' ) . Monoamines are mainly synthesized in the neurons that release them , and it is generally understood that their re-uptake into these same neurons occurs via specific transport proteins to prevent their accumulation at synapses ( Martin and Krantz , 2014 ) . Absence of AANAT1 from dopaminergic or octopaminergic neurons showed that cells that produce and release monoamines do not necessarily contribute to their catabolism via AANAT1 . However , the presence of AANAT1 in subsets of glutamatergic , GABAergic and cholinergic neurons suggests that , along with astrocytes , these non-monoaminergic neurons could contribute to regulation of monoamine bioavailability in the brain . The AANAT1 gene produces two protein isoforms , the shorter of which ( FlyBase AANAT1-PA , 240aa in length ) , previously known as aaNAT1b , is more predominant ( Brodbeck et al . , 1998 ) . This shorter isoform was observed to be lost in AANAT1lo mutants ( Hintermann et al . , 1996 ) . AANAT1lo is a spontaneous mutation that arose from insertion of a transposable element into the AANAT1 gene , and tissue extracts from these flies have reduced AANAT1 activity ( Hintermann et al . , 1996; Maranda and Hodgetts , 1977 ) . Using our new AANAT1 antiserum to perform western blotting of brain extracts , we observed only the shorter isoform in controls ( Figure 2A ) . In AANAT1lo homozygotes and hemizygotes ( AANAT1lo/In ( 2LR ) Px[4] ) , AANAT1 protein levels were reduced to 13 and 8% of iso31 controls , respectively ( Figure 2A , B ) . This was confirmed with IHC in the brains of AANAT1lo flies ( Figure 2C–E ) , where we noted residual AANAT1 expression in some Elav+ neurons , but complete loss of AANAT1 from astrocytes ( Figure 2F–F' , G–G' ) . In vitro studies have shown that serotonin and dopamine are substrates for AANAT1 with similar affinities ( Hintermann et al . , 1995 ) . Whether the levels of these and/or other monoamines are regulated by AANAT1 in vivo remains to be determined . We used High Performance Liquid Chromatography - Mass Spectrometry ( HPLC-MS ) to measure levels of serotonin , dopamine , and octopamine in the brains of AANAT1lo flies and controls ( iso31 ) ( Figure 2H ) . Under baseline sleep-cycle conditions , where brain tissues were collected in a 3 hr window after lights-ON ( ZT0 ) , serotonin and dopamine levels in AANAT1lo flies were similar to controls ( Figure 2I ) . Octopamine was undetectable in controls and was found at low levels in brains of AANAT1lo flies ( Figure 2—figure supplement 1B ) . However , if this window was preceded by 12 hr ( ZT12-ZT24 ) of SD overnight , AANAT1lo brains had a robust increase in the levels of serotonin and dopamine compared to controls ( Figure 2I ) , but this had no effect on octopamine levels ( Figure 2—figure supplement 1B ) . Importantly , in control animals the SD protocol itself did not appear to affect the levels of measured monoamines , or of AANAT1 itself ( Figure 2J , K ) . Further , we did not observe changes in the levels of another monoamine catabolic enzyme known to be expressed in astrocytes ( Ebony ) in either AANAT1lo flies , or in flies subjected to SD ( Figure 2—figure supplement 1A , C ) . We conclude that catabolism of serotonin and dopamine in the brains of flies lacking AANAT1 is severely compromised upon SD , leading to inappropriate accumulation of these monoamines . The AANAT1lo mutation increases homeostatic sleep following deprivation ( Shaw et al . , 2000 ) , suggesting AANAT1 could be key to how the brain limits the homeostatic response to sleep need . AANAT1lo is also interesting because these flies were reported to have normal motor activity , and intact daily patterns of sleep ( Shaw et al . , 2000 ) , allowing genetic dissection of homeostatic sleep- control independent of the regulation of baseline sleep . We wondered whether the increased recovery sleep seen in AANAT1lo animals could be explained by loss of AANAT1 function from neurons or astrocytes . To test this , we selectively knocked down AANAT1 in distinct cell types with RNAi and measured both baseline and homeostatic sleep with the Drosophila Activity Monitoring System ( DAMS ) . To evaluate the contribution of neuronal AANAT1 to sleep , we tested nSyb-Gal4 >UAS-AANAT1-RNAi flies , using two independent RNAi lines that target the AANAT1 transcript at distinct sites . Knockdown of AANAT1 in neurons with either line resulted in normal patterns of baseline sleep ( Figure 3A–H ) , as reported for the AANAT1lo allele ( Shaw et al . , 2000 ) . While awake during daytime , these animals had lower levels of activity than controls carrying either the GAL4 or UAS transgene alone ( Figure 3—figure supplement 1A , B ) , but their total daytime sleep amount was similar to at least one of the controls ( Figure 3B , F ) , as were the length of daytime sleep bouts ( Figure 3C , G ) and their number ( Figure 3D , H ) . When we examined the amount of nighttime sleep compared to controls , we found that one RNAi line ( AANAT-RNAi 2 , JF02142 ) , but not the other ( AANAT-RNAi 1 , HMS01617 ) , led to increased amount of nighttime sleep ( Figure 3B , F ) . For knockdown with this RNAi line only , sleep bouts during the night were increased in duration and decreased in number , suggesting improved sleep consolidation at night ( Figure 3C–D;G-H ) . We then assessed whether AANAT1 knockdown in neurons ( nSyb >AANAT1 RNAi ) would impact sleep recovery after SD , as was observed in AANAT1lo flies . For this , flies were subjected to overnight mechanical SD and we found , somewhat surprisingly , that these flies did not display enhanced recovery sleep the next day ( Figure 3I–L ) . Next , we used Alrm-Gal4 to selectively deplete AANAT1 expression from astrocytes with RNAi ( Alrm >AANAT1 RNAi ) . This had no effect on the numbers of astrocytes present in the brain ( Figure 4A ) . These flies showed normal baseline patterns and amounts of daytime and nighttime sleep compared to controls ( Figure 4B–E ) , but while awake they were less active than controls ( Figure 4—figure supplement 1A , B ) . However , upon overnight mechanical SD , these flies had increased recovery sleep the next day ( Figure 4F–I ) , mimicking AANAT1lo flies ( Figure 4—figure supplement 1C–E ) . Like AANAT1 loss-of-function , AANAT1 overexpression in astrocytes also increased recovery sleep following deprivation ( Figure 4—figure supplement 1F ) , underscoring the importance of regulated astrocytic AANAT1 levels in sleep homeostasis . We characterized AANAT1 expression in astrocytes during pupal development with immunochemistry , and found AANAT1 to be expressed weakly in only a few astrocytes at 48 hr after puparium formation ( APF ) , then gradually more strongly in most but not all astrocytes at 72 hr and 96 hr APF ( Figure 4—figure supplement 1G-I'' ) . To investigate when AANAT1 functions in astrocytes for sleep recovery , we used the Temporal And Regional Gene Targeting ( TARGET ) system ( McGuire et al . , 2004 ) to knock down AANAT1 in adult astrocytes with Eaat1-Gal4 . In the brain , Eaat1-Gal4 is a driver line for astrocytes ( which express AANAT1 ) and cortex glia ( which do not ) . When adult flies were raised at 32°C to deplete AANAT1 from glia using RNAi , these animals showed increased recovery sleep compared to the UAS control but not the Gal4 control ( Figure 4—figure supplement 1J-K ) . Our results demonstrate that AANAT1 acts in astrocytes , but not in neurons , to restrict daytime recovery sleep in response to overnight SD . With HPLC-MS , we noted that astrocyte-selective AANAT1 knockdown led to increased levels of brain serotonin and dopamine after SD in most samples compared to controls ( Figure 4J ) . This did not reach statistical significance however , perhaps because of an outlier in each of the control groups , or perhaps because AANAT1 knockdown in astrocytes affects the levels of only a portion of serotonin and dopamine in the brain , albeit an important portion with respect to sleep homeostasis . Together with the clear increases seen in AANAT1lo flies ( Figure 2I ) , these data suggest that Drosophila astrocytes employ AANAT1 to limit accumulation of serotonin and dopamine upon SD . Since loss of AANAT1 does not affect astrocyte numbers or their terminal differentiation , we think it unlikely to play a developmental role and favor the idea that AANAT1 functions in mature astrocytes to limit recovery sleep . To shed light on when AANAT1 might be active with respect to sleep homeostasis , with IHC we examined AANAT1 under baseline sleep-cycle conditions at 3 hr intervals during the light and dark period . In the dark period ZT12-ZT24 , the patterns of AANAT1 expression in sleep regulatory regions the MB , FSB and EB were unchanged compared to the light period patterns described above ( data not shown ) . In addition , we never observed obvious changes of AANAT1 levels in astrocytes over the course of the light period ZT0-ZT12 . Interestingly , during the dark phase most astrocytes throughout the brain showed obvious changes of AANAT1 levels . Dark period AANAT1 expression in astrocyte cell bodies peaked at ZT15 , declined markedly to undetectable levels by ZT21 ( Figure 4K ) , and was re-established at lights-ON ( ZT24 ) . From this we speculate that the loss of AANAT1 might have profound influence on sleep homeostasis near ZT15 , when AANAT1 levels in astrocytes are usually highest , or during daytime when the increased recovery sleep occurs . AANAT1 is expressed in astrocytes that reside throughout the brain , and so it remains unclear whether it modulates sleep homeostasis by acting within a particular region of the brain , or more broadly . Interestingly , the neuropils of key sleep centers ( MB , EB and FSB; Figure 1—figure supplement 1K–Q ) had no AANAT1 staining from infiltrative astrocytes , raising the likelihood it acts elsewhere . Finally , it remains to be established whether the effect of AANAT1 on sleep homeostasis is due to serotonin , dopamine , or both . In Drosophila , serotonergic signaling in the brain promotes sleep , while dopaminergic signaling promotes waking . Levels of both serotonin and dopamine are upregulated in AANAT1lo mutants upon SD , where increased sleep prevails . It stands to reason that AANAT1 could act in astrocytes to limit the deprivation-dependent accumulation of sleep-promoting serotonin . It is also possible that dopamine accumulation plays a role , since thermogenetic activation of wake-promoting dopaminergic neurons at night promotes compensatory sleep the next day . This suggests these particular neurons are upstream of circuits that produce homeostatic responses to extended wakefulness ( Dubowy and Sehgal , 2017; Seidner et al . , 2015 ) , and astrocytic AANAT1 could somehow restrict dopaminergic signaling from these neurons overnight . Our findings illustrate a newly discovered role for astrocytes in the control of monoamine bioavailability and homeostatic sleep drive , where they are specifically engaged to catabolize monoamines whose levels are elevated by overnight SD . Drosophila astrocytes also express the enzyme Ebony , which couples dopamine to N-β-alanine ( Suh and Jackson , 2007 ) , and a receptor for octopamine and tyramine ( Ma et al . , 2016 ) , reinforcing how they are well-equipped to metabolize monoamines , and to monitor and respond to monoaminergic neuronal activity . Neither gene expression studies nor RNA sequencing databases provide evidence for monoamine-synthesizing enzymes in Drosophila astrocytes , so it appears likely that monoamines inactivated by AANAT1 in astrocytes are brought into these cells by an unidentified transporter . Astrocytes are particularly well-suited for regulating sleep in this way because they have ramified processes that infiltrate neuropil regions to lie in close proximity to synapses . SD can seemingly reduce the degree of contact between astrocytes and neurons in the fly brain ( Vanderheyden et al . , 2019 ) , and so it is possible that these structural changes could influence monoamine uptake and inactivation by astrocytes . In neurons , AANAT1 may function to limit sleep consolidation at night , but evidence for this came from only one of the two RNAi lines used in this study and was not observed in AANAT1lo mutants ( Figure 4—figure supplement 1A ) . Further studies are needed to characterize sleep-control functions of AANAT1 in neurons , if any , to understand better how cellular context can impact AANAT1 function in sleep regulation . In light of this , we note that loss of the related enzyme aanat2 in zebrafish larvae decreases baseline sleep ( Gandhi et al . , 2015 ) , which could be attributed to a loss of melatonin since the AANAT1 product N-acetylserotonin is an intermediate in the synthesis of melatonin in vertebrates ( Hintermann et al . , 1996 ) . Clearly , the appropriate balance and cellular context of AANAT activity is critical for the regulation of sleep , and we show here in Drosophila that astrocytes are an important contributor to this balance . Interestingly , astrocytes in rodents express the monoamine transporters and receptors for dopamine and serotonin ( Bacq et al . , 2012; Baganz et al . , 2008; Huang et al . , 2012; Petrelli et al . , 2020; Sandén et al . , 2000; Vaarmann et al . , 2010 ) , raising the possibility that astrocytes in mammals might also participate in mechanisms of sleep regulation involving monoaminergic neural signaling . Drosophila melanogaster stocks were obtained from the Bloomington Drosophila Stock Center ( BSC ) : Trh-Gal4 ( BSC-52249 ) , TH-Gal4 ( BSC-8848 ) , Tdc2-Gal4 ( BSC-9313 ) , Ddc1-Gal4 ( BSC-7010 ) , UAS-mCD8-GFP ( BSC-32186 ) , UAS-RFP . nls ( BSC-30558 ) , Mi{MIC} VGlutMI04979 ( BSC-38078 ) , Gad1-Gal4 ( BSC-51630 ) , Cha-Gal4 ( BSC-6793 ) , R56F03-Gal4 ( BSC-39157 ) , AANAT1lo ( BSC-3193 ) , Df ( 2R ) BSC356 ( BSC-24380 ) , deficiency In ( 2LR ) Px4 ( BSC-1473 ) , tubGal80ts ( BSC-7018 ) , AANAT1 RNAi lines UAS-HMS01617 ( BSC-36726 ) , UAS-JF02142 ( BSC-26243 ) and MCFO stock hs-FlpG5 . Pest; 10xUAS ( FRT-stop ) myr::smGdP-HA , 10xUAS ( FRT-stop ) myr::smGdP-V5-THS-10xUAS ( FRT-stop ) myr::smGdP-FLAG ( BSC-64085 ) . Alrm-Gal4 and Eaat1-Gal4 was provided by Dr . Marc Freeman , and nSyb-Gal4 by Dr . Stefan Thor . For RNAi-mediated knockdown of gene expression , control animals carried only a Gal4 driver , while experimental groups also carried a single copy of the transgene to elicit RNAi . The chromosome carrying Alrm-Gal4 also bore a transgene encoding the nuclear reporter UAS-nuRFP . To mitigate the effects of genetic background for sleep experiments , control Gal4 and UAS flies were crossed to the iso31 stock . In using the TARGET system , we combined GAL80ts , a temperature-sensitive inhibitor of GAL4 , with EAAT1-GAL4 to selectively knock down AANAT1 ( UAS-AANAT1-RNAi 1 ( HMS01617 ) ) during adulthood . Animals were raised at the permissive temperature ( 18°C ) to repress Gal4 , then 4-day-old adult animals were shifted to 32°C for another 5 days to induce RNAi for AANAT1 , before exposing them to SD experiments and sleep monitoring as outlined below . The morphology of single astrocytes was determined by the MCFO technique ( Nern et al . , 2015 ) , where three differently tagged reporters under UAS control ( HA , FLAG and V5 ) were silenced by FRT-flanked transcriptional terminators . Heat shock-induced FLPase expression removed terminators randomly in individual cells , driven by astrocyte-specific Alrm-GAL4 . This created a mosaic of astrocytes of distinct colors . For this experiment , 3–5 days old flies raised at 18°C were heat-shocked at 37°C for 5–8 min and dissected 2–3 days later . To create UAS-AANAT1 , the AANAT1 coding sequence from the cDNA clone GH12636 ( Drosophila Genomic Research Centre ) was PCR-amplified and cloned in- frame into a modified pJFRC-MUH vector ( Pfeiffer et al . , 2010 ) . Transgenic flies with site-specific insertions at VK0005 site on chromosome three were generated using standard microinjection ( BestGene , Inc ) . A KLH-coupled peptide RRPSPDDVPEKAADSC ( amino acids ( aa ) 94–109 of isoform AANAT1-PA ( FlyBase ) , or 129–144 of isoform AANAT1-PB ) was synthesized and injected into rabbits according to guidelines of the Canadian Council for Animal Care ( MEDIMABS , Montreal , QC ) . Adult fly brains were dissected between ZT3-9 ( unless specified otherwise ) in cold phosphate-buffered saline ( pH 7 . 4 ) and fixed in 4% paraformaldehyde for 30 min ( min ) . After three washes of 15 min each with PBS containing 0 . 3% Triton-X-100 ( PBTx-0 . 3% ) , the tissues were blocked in 5% normal goat serum ( Jackson Laboratories ) in PBTx-0 . 5% for 45 min . Tissues were incubated in primary antibodies: rabbit anti-AANAT1 ( 1:2000; this study ) , rat anti-Elav ( 1:100 ) ; Developmental Studies Hybridoma Bank ( DSHB ) , mouse anti-Repo ( 1:50; DSHB ) , mouse anti-nc82 ( 1:50; DSHB ) , mouse anti-GFP ( 1:200; Clontech #632381 ) overnight at 4°C . After three washes ( 15 min each , PBTx-0 . 3% ) , tissues were incubated with secondary antibodies overnight at 4°C: goat anti-mouse ( Rhodamine Red-X , Jackson ImmunoResearch #115-295-146 ) , goat anti-rabbit ( Alexa Fluor 488 , Thermo Fisher Scientific , #A11008 ) , goat anti-mouse ( Alexa Fluor 488 , Thermo Fisher Scientific ) , goat anti-rat ( Alexa Fluor 568 , Thermo Fisher Scientific , #A11077 ) , goat anti-rabbit ( Alexa Fluor 647 , Thermo Fisher Scientific , #A21245 ) . Tissues were again washed ( 3 × 15 min , PBTx-0 . 3% ) , followed by a final wash in PBS . Tissues were mounted in SlowFade Diamond Antifade Mountant ( Thermo Fisher Scientific , #S36964 ) . Fluorescence images were acquired with an Olympus BX-63 Fluoview FV1000 confocal laser-scanning microscope and processed using Fiji . For MCFO labeling , brains were dissected in ice-cold PBS , fixed with 4% paraformaldehyde/PBS for 1 hr at room temperature followed by three successive washes in 0 . 5% PBTx for 20 min each . Simultaneous incubation ( 48 hr at 4°C ) with rat anti-FLAG ( 1:100; Novus Biologicals NBP1-06712 , A-4 ) and rabbit anti-AANAT1 was followed by another 48 hr at 4°C with goat anti-rabbit ( 1:1000; Alexa Fluor 488 , Thermo Fisher Scientific , #A11008 ) , goat anti-rat ( 1:1000; Alexa Fluor 568 , #A11077 ) and V5-tag:AlexaFluor-647 ( 1:200; Bio-Rad MCA1360A647 ) . To quantify cells immuno-labeled for GFP and AANAT1 , cells were manually counted from image stacks of the central brain near the antennal lobe and central complex regions ( excluding optic lobes ) . We chose cell bodies in this dorsal - anterior region because it routinely showed excellent immunochemical signal and good cellular resolution . Lysates for western blots were prepared at ZT0 . 5–1 . 5 from dissected adult brains in 50 μl Laemmli buffer as reported in Parinejad et al . , 2016 . 10 brains were used per lysate and incubated at 90⁰C for 5 min . 15 μl of each sample was loaded per well , run on 15% SDS-PAGE gels , blotted to nitrocellulose membrane , and probed with rabbit anti-AANAT1 ( 1:2500 ) or anti-Ebony ( 1:3000; Sean Carroll , University of Wisconsin-Madison ) , and mouse anti-actin ( 1:3000; Sigma #A4700 ) . HRP-conjugated secondary antibodies anti-rabbit ( 1:3000; Bio-Rad ) and anti-mouse ( 1:3000; Promega #W4021 ) were used for detection with chemiluminescence ( HyGLO Chemiluminescent HRP Antibody Detection Reagent , Denville Scientific ) . Mean signal intensity for AANAT1 or Ebony was quantified using Fiji and normalized to actin . We used three separate lysates for each genotype to analyze western blots . For sleep experiments , female brains were used for lysate preparation . To prepare samples for HPLC-MS , the brains of twenty female flies ( 1–2 weeks old ) for each genotype were dissected into ice-cold PBS between ZT0 . 5 and 3 . 5 . We dissected brain tissue to avoid cuticle contamination because serotonin and dopamine are intermediates in the sclerotization of Drosophila cuticle . Dissected brains were centrifuged , the PBS was removed , and samples were quickly homogenized with a motorized pestle into an aqueous solution of formic acid ( 0 . 1% ) . After centrifugation , the supernatant was collected and stored at −80°C . Preliminary analytical conditions were developed using reference standards in a solution containing either serotonin , dopamine , or octopamine . With LC-MS/MS ( Thermo-Scientific Quantiva Triple Quadrupole Mass Spectrometer ( QQQ ) ) , the absolute values for each analyte were measured in picograms ( pg ) per brain , through the addition of deuterated reference standards to sample extracts . All samples within an experiment were treated identically , and in parallel wherever possible . Prior to experimentation , flies were kept on standard food in constant conditions ( a 12 hr light/dark cycle , and 25°C ) . At least 5 days after eclosion , mated adult females were loaded into glass tubes with 5% sucrose/2% agar food for behavioral recordings . The Drosophila Activity Monitoring ( DAM ) system ( Trikinetics , Waltham , MA ) was used to quantify infrared beam breaks representing locomotor activity . Files were processed with PySolo ( Gilestro and Cirelli , 2009 ) in 1 min bins , with sleep defined as five consecutive minutes without activity , as done previously ( Hendricks et al . , 2000 ) . In SD experiments , flies were placed in DAM monitors on a vortexer that was mechanically shaken a random 2 of every 20 s over the course of the 12 hr of the dark period ( ZT12-24 ) . Recovery sleep was determined , per fly , as the difference between sleep amount in the period following deprivation and sleep amount in the same time period on the preceding baseline day in unperturbed conditions . Activity Index refers to the average number of beam crossings within an active bout . AANAT1 levels in astrocytes were quantified at 3 hr intervals between ZT12-24 with IHC , where AANAT1 fluorescence intensity in astrocyte cell bodies was measured and normalized to nuRFP intensity from 2 copies of a UAS-nuRFP transgene reporter driven by Alrm-Gal4 . At each time-point , 10 astrocytes in the antennal lobe region were measured from each of three brains .	Sleep is essential for our physical and mental health . A lack of sleep can affect our energy and concentration levels and is often linked to chronic illnesses and mood disorders . Sleep is controlled by an internal clock in our brain that operates on a 24-hour cycle , telling our bodies when we are tired and ready for bed , or fresh and alert to start a new day . In addition , the brain tracks the need for sleep and drives the recovery of sleep after periods of prolonged wakefulness – a process known as sleep-wake homeostasis . Chemical messengers in the brain such as dopamine and serotonin also play an important part in regulating our sleep drive . While dopamine keeps us awake , serotonin can both prevent us from and help us falling asleep , depending on the part of the brain in which it is released . Most research has focused on the role of different brain circuits on sleep , but it has been shown that a certain type of brain cell , known as astrocyte , may also be important for sleep regulation . So far , it has been unclear if astrocytes could be involved in regulating the need for recovery sleep after a sleep-deprived night – also known as rebound sleep . Now , Davla , Artiushin et al . used sleep-deprived fruit flies to investigate this further . The flies were kept awake over 12 hours ( from 6pm to 6am ) , using intermittent physical agitation . The researchers found that astrocytes in the brains of fruit flies express a molecule called AANAT1 , which peaked at the beginning of the night , declined as the night went on and recovered by morning . In sleep deprived flies , it inactivated the chemical messengers and so lowered the amount of dopamine and serotonin in the brain . However , in mutant flies that lacked AANAT1 , both dopamine and serotonin levels increased in the brain after sleep deprivation . When AANAT1 was selectively removed from astrocytes only , sleep-deprived flies needed more rebound sleep during the day to make up for lost sleep at night . This shows that both astrocytes and AANAT1 play a crucial role in sleep homeostasis . Molecules belonging to the AANAT family exist in both flies and humans , and these results could have important implications for the science of sleep . The study of Davla , Artiushin et al . paves the way for understanding the mechanisms of sleep homeostasis that are similar in both organisms , and may in the future , help to identify sleep drugs that target astrocytes and the molecules they express .	[ "Abstract", "Introduction", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2020	AANAT1 functions in astrocytes to regulate sleep homeostasis
Mobilization of retrotransposons to new genomic locations is a significant driver of mammalian genome evolution , but these mutagenic events can also cause genetic disorders . In humans , retrotransposon mobilization is mediated primarily by proteins encoded by LINE-1 ( L1 ) retrotransposons , which mobilize in pluripotent cells early in development . Here we show that TEX19 . 1 , which is induced by developmentally programmed DNA hypomethylation , can directly interact with the L1-encoded protein L1-ORF1p , stimulate its polyubiquitylation and degradation , and restrict L1 mobilization . We also show that TEX19 . 1 likely acts , at least in part , through promoting the activity of the E3 ubiquitin ligase UBR2 towards L1-ORF1p . Moreover , loss of Tex19 . 1 increases L1-ORF1p levels and L1 mobilization in pluripotent mouse embryonic stem cells , implying that Tex19 . 1 prevents de novo retrotransposition in the pluripotent phase of the germline cycle . These data show that post-translational regulation of L1 retrotransposons plays a key role in maintaining trans-generational genome stability in mammals . Retrotransposons are mobile genetic elements that comprise around 40% of mammalian genomes ( Beck et al . , 2011; Hancks and Kazazian , 2016; Richardson et al . , 2014a ) . Retrotransposons are a source of genetic variation that shape genome evolution and mammalian development , but their mobilization can also cause mutations associated with a variety of genetic diseases and cancers ( Beck et al . , 2011; Hancks and Kazazian , 2016; Richardson et al . , 2014a; Garcia-Perez et al . , 2016 ) . New retrotransposition events are estimated to occur in around 1 in every 20 human births , and represent around 1% of genetic disease-causing mutations in humans ( Kazazian , 1999; Hancks and Kazazian , 2016 ) . Retrotransposons can be classified into two major types depending on their genomic structure and presence of LTR ( long terminal repeat ) sequences: LINEs ( long interspersed elements ) and SINEs ( short interspersed elements ) lack LTR sequences and end in a polyA sequence , while LTR retrotransposons are similar in structure to retroviruses ( Beck et al . , 2011 ) . In humans , all new retrotransposition events are catalysed by LINE-1 ( L1 ) elements . Active L1s encode two proteins strictly required for retrotransposition ( Moran et al . , 1996 ) : ORF1p is an RNA binding protein with nucleic acid chaperone activity ( Martin and Bushman , 2001; Hohjoh and Singer , 1997 ) , and ORF2p is a multidomain protein with reverse transcriptase and endonuclease activities ( Feng et al . , 1996; Mathias et al . , 1991 ) . Both these proteins interact directly or indirectly with various cellular factors and are incorporated into ribonucleoprotein particles ( RNPs ) along with the L1 RNA ( Beck et al . , 2011; Goodier et al . , 2013; Hancks and Kazazian , 2016; Richardson et al . , 2014a; Taylor et al . , 2013 ) . While these proteins exhibit a strong cis-preference to bind to and catalyse mobilization of their encoding mRNA , they can act in trans on other RNAs , including those encoded by SINEs ( Kulpa and Moran , 2006; Wei et al . , 2001; Dewannieux et al . , 2003; Esnault et al . , 2000 ) . Some human L1s also encode a trans-acting protein , ORF0 , that stimulates retrotransposition , although its mechanism of action is currently poorly understood ( Denli et al . , 2015 ) . Host restriction mechanisms that regulate the activity of these L1-encoded proteins will impact on the stability of mammalian genomes and the incidence of genetic disease . Regulating retrotransposon activity is particularly important in the germline as de novo retrotransposon integrations that arise in these cells can be transmitted to the next generation ( Crichton et al . , 2014 ) . The mammalian germline encompasses lineage-restricted germ cells including primordial germ cells , oocytes , and sperm , and their pluripotent precursors in early embryos ( Ollinger et al . , 2010 ) . L1 mobilization may be more prevalent in pluripotent cells in pre-implantation embryos rather than in lineage-restricted germ cells ( Kano et al . , 2009; Richardson et al . , 2017 ) , and regulation of L1 activity in the pluripotent phase of the germline cycle is therefore likely to have a significant effect on trans-generational genome stability . Repressive histone modifications and DNA methylation typically suppress transcription of retrotransposons in somatic mammalian cells ( Beck et al . , 2011; Hancks and Kazazian , 2016; Richardson et al . , 2014a; Crichton et al . , 2014 ) , but many of these transcriptionally repressive marks are globally removed during pre-implantation development and during fetal germ cell development in mice ( Hajkova et al . , 2008; Popp et al . , 2010; Santos et al . , 2002; Fadloun et al . , 2013 ) . DNA methylation in particular plays a key role in transcriptionally repressing L1 in the germline ( Bourc'his and Bestor , 2004 ) , and it is not clear how L1 activity is controlled in pluripotent cells and fetal germ cells while they are DNA hypomethylated . However , evidence suggests that L1 mobilization is tightly controlled in pluripotent cells to reduce trans-generational genome instability ( Wissing et al . , 2012; Marchetto et al . , 2013 ) . In fetal germ cells , loss of DNA methylation correlates with relaxed transcriptional suppression of retrotransposons ( Molaro et al . , 2014 ) , but also induces expression of methylation-sensitive germline genome-defence genes that have roles in post-transcriptionally repressing these elements ( Hackett et al . , 2012 ) . The methylation-sensitive germline genome-defence genes include components of the PIWI-piRNA pathway . This pathway promotes de novo DNA methylation of retrotransposons in male germ cells , cleaves retrotransposon RNAs , and may also interfere with retrotransposon translation ( Fu and Wang , 2014; Xu et al . , 2008 ) . However , while mice carrying mutations in the PIWI-piRNA pathway can strongly de-repress L1-encoded RNA and protein during spermatogenesis ( Aravin et al . , 2007; Carmell et al . , 2007 ) , increased L1 mobilization has not yet been reported in these mutant models . Indeed , the level of L1 expression at different stages of the germline cycle does not completely correlate with the ability of L1 to mobilize , and post-translational control mechanisms have been proposed to restrict the ability of L1 to mobilize in the mouse germline ( Kano et al . , 2009 ) . However , the molecular identities of these post-translational L1 restriction mechanisms have not yet been elucidated . We have previously shown that programmed DNA hypomethylation in the developing mouse germline induces expression of a group of genes that are involved in suppressing retrotransposon activity ( Hackett et al . , 2012 ) . One of the retrotransposon defence genes induced in response to programmed DNA hypomethylation , Tex19 . 1 , suppresses specific retrotransposon transcripts in spermatocytes ( Ollinger et al . , 2008; Reichmann et al . , 2012 ) , however its direct mechanism of action remains unclear . Tex19 . 1 is expressed in germ cells , pluripotent cells and the placenta and is one of two TEX19 orthologs generated by a rodent-specific gene duplication ( Kuntz et al . , 2008; Wang et al . , 2001; Ollinger et al . , 2008 ) . These mammal-specific proteins have no functionally characterized protein motifs or reported biochemical activity , but mouse TEX19 . 1 is predominantly cytoplasmic in the germline ( Ollinger et al . , 2008; Yang et al . , 2010 ) . Here we show that Tex19 . 1 regulates L1-ORF1p levels and mobilization of engineered L1 elements . We show that mouse TEX19 . 1 , and its human ortholog TEX19 , physically interact with L1-ORF1p , and regulate L1-ORF1p abundance through stimulating its polyubiquitylation and proteasome-dependent degradation . We show that TEX19 . 1 likely controls L1-ORF1p abundance in concert with UBR2 , an E3 ubiquitin ligase that we show also physically interacts with and regulates L1-ORF1p levels in vivo . We also show that loss of Tex19 . 1 results in increased L1-ORF1p abundance and increased mobilization of engineered L1 constructs in pluripotent mouse embryonic stem cells , suggesting that Tex19 . 1 functions as a post-translational control mechanism to restrict L1 mobilization in the developing germline . Programmed DNA hypomethylation in the developing germline induces expression of Tex19 . 1 , which encodes a predominantly cytoplasmic protein in spermatocytes that suppresses retrotransposon expression through unknown mechanisms ( Ollinger et al . , 2008; Reichmann et al . , 2012; Yang et al . , 2010 ) . In order to define the role of TEX19 . 1 in retrotransposon regulation in more detail we investigated whether Tex19 . 1 might have post-transcriptional effects on cytoplasmic stages of the retrotransposon life cycle . Since Tex19 . 1−/− spermatocytes have defects in meiosis that induce spermatocyte death ( Ollinger et al . , 2008 ) , we analysed mouse L1 ORF1p ( mL1-ORF1p ) expression in prepubertal testes during the first wave of spermatogenesis before any increased spermatocyte death is evident ( Ollinger et al . , 2008 ) . Western blotting showed that P16 Tex19 . 1−/− testes have elevated levels of mL1-ORF1p ( Figure 1A ) , even though L1 RNA levels do not change ( Figure 1B ) , as previously shown ( Ollinger et al . , 2008; Reichmann et al . , 2012 ) . Primers designed against the active A , Gf and Tf subtypes of L1 ( de la Rica et al . , 2016 ) similarly did not detect any change in L1 RNA abundance in P16 Tex19 . 1−/− testes ( Figure 1—figure supplement 1A ) . These data suggest that Tex19 . 1 negatively regulates mL1-ORF1p post-transcriptionally in male germ cells . Immunostaining of P16 testes showed that , consistent with previous reports , mL1-ORF1p is expressed in meiotic spermatocytes in control mice ( Figure 1C ) ( Soper et al . , 2008; Branciforte and Martin , 1994 ) . However , mL1-ORF1p immunostaining is elevated approximately two fold in the same cell type in Tex19 . 1−/− mice ( Figure 1C ) . Thus , distinct from its role in transcriptional regulation of retrotransposons ( Ollinger et al . , 2008; Reichmann et al . , 2012; Crichton et al . , 2017a; Reichmann et al . , 2013 ) , Tex19 . 1 appears to have a role in post-transcriptionally suppressing mL1-ORF1p abundance in meiotic spermatocytes . 10 . 7554/eLife . 26152 . 003Figure 1 . mL1-ORF1p is post-transcriptionally regulated by Tex19 . 1 in mouse germ cells . ( A ) Western blot for mL1-ORF1p in Tex19 . 1+/− and Tex19 . 1−/− littermate P16 mouse testes . β-actin is a loading control . Data shown is representative of seven Tex19 . 1−/− animals across four litters . ( B ) qRT-PCR for L1 RNA using primers against ORF2 in testes from the same animals analyzed in panel A . Expression relative to β-actin was normalized to a Tex19 . 1+/− control animal . Error bars indicate SEM for three qPCR technical replicates from the same reverse-transcribed RNA . ( C ) Immunostaining for mL1-ORF1p ( green ) in Tex19 . 1+/− and Tex19 . 1−/− P16 mouse testis sections . Nuclei are counterstained with DAPI ( shown as red ) . Scale bar , 10 μm . Anti-mL1-ORF1p immunostaining per unit area was quantified for three animals for each genotype , and normalized to the mean for Tex19 . 1+/− animals . Means ± SEM ( 1 ± 0 . 17 and 2 . 25 ± 0 . 14 for Tex19 . 1+/− and Tex19 . 1−/− respectively ) are indicated; *p<0 . 01 ( t-test , p=0 . 005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 00310 . 7554/eLife . 26152 . 004Figure 1—figure supplement 1 . Tex19 . 1 does not inhibit L1 translation . ( A ) qRT-PCR for A , Tf and Gf active subtypes of L1 in P16 testes . L1 subtype mRNA abundance was measured relative to β-actin , and normalised to the mean Tex19 . 1+/− control level . Two animals for each genotype are shown . ( B ) Oligo ( dT ) pull-downs from P16 testes . Oligo ( dT ) cellulose beads were used to isolate poly ( A ) RNA from testis lysates , and associated proteins analysed by Western blotting with indicated antibodies . 200 or 500 µg poly ( A ) RNA was added as a competitor . The poly ( A ) RNA binding protein PABP1 was used as a positive control . TEX19 . 1 is not detectably associated with poly ( A ) RNA in testes . C . Sucrose density gradient enrichment of translation intermediates from P18 testes . The protein content of the fractions was monitored by reading absorbance at 254 nm , and peaks corresponding to messenger ribonucleoproteins ( mRNPs ) , 40S ribosomal subunits , monosomes and polysomes are indicated . Western blots for TEX19 . 1 , β-actin and PABP1 are shown for each fraction . TEX19 . 1 is not detectably associated with actively translating polysomes in testes . D . qRT-PCR for L1 mRNA in mRNP + 40S , monosome , and polysome fractions in sucrose gradients from Tex19 . 1+/− and Tex19 . 1−/− P18 testes . L1 mRNA abundance was measured relative to β-actin in each fraction , and normalized to one of the heterozygous control animals . A proportion of L1 mRNA associates with polysomes consistent with previous reports ( Tanaka et al . , 2011 ) . Meiotic arrest and increased spermatocyte death between P16 and P22 in Tex19 . 1−/− testes ( Ollinger et al . , 2008 ) may be generating some differences in testicular cell composition in these P18 samples and causing subtle differences in L1 mRNA distribution between Tex19 . 1+/− and Tex19 . 1−/− samples . However , there is no statistically significant increase in polysome-associated L1 mRNA in Tex19 . 1−/− P18 testes ( t-test , p=0 . 4 ) . Error bars indicate SEM for technical replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 004 Post-transcriptional control of protein abundance can occur through regulation of mRNA translation or protein stability . To investigate whether TEX19 . 1 might be involved in one of these processes we attempted to identify RNAs or proteins that interact with TEX19 . 1 . In contrast to the PIWI proteins MILI and MIWI ( Grivna et al . , 2006; Unhavaithaya et al . , 2009 ) , oligo ( dT ) pull-downs from mouse testicular lysate suggest that TEX19 . 1 is not physically associated with RNA in this tissue ( Figure 1—figure supplement 1B ) and neither is TEX19 . 1 enriched in testicular polysome fractions containing actively translating mRNAs ( Figure 1—figure supplement 1C ) . In addition , the increase in mL1-ORF1p abundance in Tex19 . 1−/− testes is not accompanied by an increase in L1 RNA abundance in polysomes ( Figure 1—figure supplement 1D ) . Therefore the increase in mL1-ORF1p abundance in Tex19 . 1−/− testes does not appear to reflect a direct role for TEX19 . 1 in regulating translation of L1 RNAs . We next attempted to identify TEX19 . 1-interacting proteins in order to determine how TEX19 . 1 might regulate L1-ORF1p levels . TEX19 . 1 is endogenously expressed in mouse embryonic stem cells ( ESCs ) ( Kuntz et al . , 2008 ) , and mass spectrometry of TEX19 . 1-YFP immunoprecipitates ( IPs ) from stably expressing mouse ESCs revealed co-IP of multiple components of the ubiquitin-proteasome system ( Figure 2A , Figure 2B , Supplementary file 1 , Supplementary file 2 ) . TEX19 . 1-YFP IPs contained a strong co-immunoprecipitating band of approximately stoichiometric abundance to TEX19 . 1-YFP which was identified as UBR2 , a RING domain E3 ubiquitin ligase and known interacting partner for TEX19 . 1 ( Yang et al . , 2010 ) ( Figure 2A , Figure 2B , Figure 2—figure supplement 1A , Figure 2—figure supplement 1B ) . The identification of the only known interacting partner for TEX19 . 1 in this co-IP suggests that the TEX19 . 1-YFP construct used in this experiment recapitulates interactions relevant for endogenous TEX19 . 1 . Indeed , all detectable endogenous TEX19 . 1 in ESCs co-fractionates with UBR2 in size exclusion chromatography ( Figure 2C ) , consistent with TEX19 . 1 existing in a stable heteromeric complex with UBR2 in these cells . Importantly , Ubr2 has previously been shown to be required for TEX19 . 1 protein stability in mouse testes ( Yang et al . , 2010 ) which , in combination with the co-fractionation and stoichiometric abundance of these proteins in the ESC IPs , suggests that any TEX19 . 1 protein not associated with UBR2 may be unstable and degraded . TEX19 . 1-YFP also co-IPs with additional components of the ubiquitin-proteasome system including UBE2A/B , an E2 ubiquitin-conjugating enzyme and cognate partner of UBR2 ( Kwon et al . , 2003; Xie and Varshavsky , 1999 ) , and a HECT-domain E3 ubiquitin ligase , HUWE1 ( Chen et al . , 2005; Liu et al . , 2005 ) ( Figure 2B , Supplementary file 2 ) . The physical associations between TEX19 . 1 and multiple components of the ubiquitin-proteasome system strongly suggest that the post-transcriptional increase in mL1-ORF1p abundance in Tex19 . 1−/− testes might reflect a role for TEX19 . 1 in regulating degradation of mL1-ORF1p . 10 . 7554/eLife . 26152 . 005Figure 2 . TEX19 . 1 physically interacts with components of the ubiquitin proteasome system and with L1-ORF1p . ( A ) Colloidal blue-stained cytoplasmic anti-YFP immunoprecipitates from mouse ESCs stably expressing mouse TEX19 . 1-YFP or YFP . Mass spectrometry identities of major bands are indicated , and a non-specific band marked with an asterisk . ( B ) Western blots for ubiquitin-proteasome system components in anti-YFP immunoprecipitates ( IPs ) from panel A . Anti-YFP IP inputs , IPs and IP supernatants ( S/N ) were blotted with indicated antibodies . ( C ) Size exclusion chromatography of cytoplasmic extract from ESCs showing elution of endogenous mouse TEX19 . 1 and UBR2 . PABP1 and β-actin are included as controls . Input ( IN ) sample is also shown , and eluted fraction numbers and the positions of pre-stained molecular weight markers in kD are indicated . ( D , E ) IPs from HEK293T cells co-transfected with mL1-ORF1p-T7 constructs and either mouse TEX19 . 1-YFP or YFP and Western blotted with indicated antibodies . The mutant mL1-ORF1RAp in panel E has a reduced binding affinity for RNA ( Kulpa and Moran , 2005; Martin et al . , 2005 ) . ( F ) Subcellular localization of mouse TEX19 . 1-YFP in the presence and absence of mL1-ORF1p-T7 . U2OS cells were transiently transfected with TEX19 . 1-YFP or YFP expression constructs with or without a plasmid expressing mL1-ORF1p-T7 ( pCEPL1SM-T7 ) , then stained with anti-T7 antibodies , and with DAPI to detect DNA . 49% of 51 cells examined exhibited some co-localization of mL1-ORF1p-T7 with TEX19 . 1-YFP . In 71% of these co-localizing cells mL1-ORF1p-T7 with TEX19 . 1-YFP were both present in a subset of small cytoplasmic foci ( arrows ) . In the remaining 29% of co-localizing cells , large cytoplasmic aggregates of mL1-ORF1p-T7 extensively co-localize with TEX19 . 1-YFP ( asterisks ) . Two representative images of cells transfected with either TEX19 . 1-YFP alone or TEX19 . 1-YFP in combination with mL1-ORF1p are shown . Scale bars 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 00510 . 7554/eLife . 26152 . 006Figure 2—figure supplement 1 . TEX19 orthologs interact with UBR2 and L1-ORF1p . ( A ) Co-immunoprecipitations ( co-IPs ) from stable HEK293 cell lines expressing either TEX19 . 1-YFP or YFP alone transiently transfected with FLAG-UBR2 . Anti-YFP immunoprecipitates ( IPs ) , inputs , and supernatants ( SUP ) were Western blotted with anti-FLAG and anti-YFP antibodies . ( B ) Reciprocal co-IP for panel A . HEK293T cells were transiently transfected with TEX19 . 1-YFP and either FLAG-UBR2 or FLAG alone , and anti-FLAG IPs and their inputs were Western blotted with anti-FLAG and anti-YFP antibodies . Positions of FLAG-UBR2 , TEX19 . 1-YFP , YFP alone and pre-stained molecular weight markers in kD are indicated . ( C ) Co-immunoprecipitation from HEK293T cells co-transfected with TEX19 . 1-YFP and mCherry-tagged mL1-ORF1p expression constructs and IPd for mCherry . YFP or mCherry alone were used as negative controls . Anti-mCherry IP inputs and IPs were Western blotted with anti-mCherry or anti-YFP antibodies . Positions of pre-stained molecular weight markers in kD are indicated . ( D ) Co-IPs from HEK293T cells co-transfected with epitope-tagged hL1-ORF1p and human TEX19-YFP expression constructs . YFP was used as a negative control . IP inputs and IPs were Western blotted with anti-T7 or anti-YFP antibodies . ( E ) Diagram showing the domain structure of mouse and human TEX19 orthologs . A conserved TEX19 domain is present at the N-terminus of both proteins , but the C-terminal region of mouse TEX19 . 1 is not conserved in the truncated human TEX19 protein . ( F ) Mouse L1-ORF1RAp mutants used to test for RNA-independent interactions have impaired mobilization . Plates of G418-resistant colonies from L1 retrotransposition assays in HeLa cells . Assays for mouse L1 ( pCEPL1SM-T7 ) and mouse L1 carrying the R297A and R298A mutations in the RNA binding domain of ORF1p that reduce its affinity for RNA ( Martin et al . , 2005 ) ( pCEPL1SM-T7-ORF1RA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 006 We next tested if TEX19 . 1 might also interact with mL1-ORF1p . Although we did not identify any mL1-ORF1p peptides in the mass spectrometry analysis of TEX19 . 1-YFP IPs from ESCs , we did identify a single hL1-ORF1p peptide in similar IPs from stable TEX19 . 1-YFP expressing HEK293T cells ( Reichmann et al . , 2017 ) . Since interactions between E3 ubiquitin ligases and their substrates are expected to be transient and weakly represented in IP experiments , we tested directly whether TEX19 . 1-YFP and epitope-tagged mL1-ORF1p interact by co-expressing these proteins in HEK293T cells and immunoprecipitating either TEX19 . 1-YFP or epitope-tagged mL1-ORF1p . Both IPs revealed weak reciprocal interactions between TEX19 . 1-YFP and T7 epitope-tagged mL1-ORF1p ( mL1-ORF1p-T7 ) ( Figure 2D , Figure 2—figure supplement 1C ) . Although human TEX19 is significantly truncated relative to its mouse ortholog , the physical interaction between TEX19 and L1-ORF1p is conserved in humans ( Figure 2—figure supplement 1D , Figure 2—figure supplement 1E ) . We next tested whether the biochemical interaction between TEX19 . 1-YFP and mL1-ORF1p-T7 is reflected by co-localization of these proteins . TEX19 . 1 is predominantly cytoplasmic in ES cells and in germ cells ( Ollinger et al . , 2008; Yang et al . , 2010 ) , but in the hypomethylated placenta and when expressed in somatic cell lines , TEX19 . 1 can localize to the nucleus ( Kuntz et al . , 2008; Reichmann et al . , 2013 ) . The context-dependent localization of TEX19 . 1 suggests that TEX19 . 1-interacting proteins in ES cells and germ cells could retain this protein in the cytoplasm in these cell types . L1-ORF1p has been reported to form cytoplasmic aggregates that co-localize with stress granule markers ( Doucet et al . , 2010; Goodier et al . , 2007 ) , therefore we tested whether co-expression of L1-ORF1p and TEX19 . 1 might localize TEX19 . 1 to these L1-ORF1p-containing aggregates . As expected , confocal microscopy showed that TEX19 . 1-YFP localizes to the nucleus when expressed in U2OS cells , however co-expression with mL1-ORF1p-T7 resulted in some co-localization of both these proteins in cytoplasmic aggregates in 25 of 51 cells examined . In 71% of these co-localizing cells , TEX19 . 1 and mL1-ORF1p-T7 exhibited partial co-localization in some cytoplasmic aggregates ( Figure 2F ) . In the remaining 29% co-localizing cells , more extreme co-localization was observed with expression of mL1-ORF1p-T7 re-localizing all detectable TEX19 . 1-YFP out of the nucleus and into cytoplasmic aggregates ( Figure 2F ) . In sum , these co-localization data are consistent with the co-IP data suggesting that TEX19 . 1-YFP and mL1-ORF1p-T7 physically interact , likely in a transient manner . A number of host factors have been shown to associate with L1-ORF1p , although many of these interactions are indirect and mediated by RNA , likely reflecting interactions within the L1 RNP ( Goodier et al . , 2013; Taylor et al . , 2013; Moldovan and Moran , 2015 ) . However , the interaction between host PCNA and L1-ORF2p is resistant to RNase treatment and is therefore a good candidate to be a direct interaction ( Taylor et al . , 2013 ) . We therefore tested whether the interaction between TEX19 . 1 and L1-ORF1p might be direct and independent of RNA . TEX19 . 1-YFP is able to interact with a mutant allele of mL1-ORF1p which has severely impaired binding to RNA and impaired L1 mobilization ( Kulpa and Moran , 2005; Martin et al . , 2005 ) ( Figure 2E , Figure 2—figure supplement 1F ) , suggesting that the interaction between TEX19 . 1-YFP and mL1-ORF1p is RNA-independent and could potentially be direct . We next tested whether bacterially expressed human TEX19 might interact with bacterially expressed hL1-ORF1p . Notably , co-expression of double-tagged human MBP-TEX19-GB1-His6 with Strep-tagged human L1-ORF1p ( Strep-hL1-ORF1p ) in bacteria resulted in a strong interaction between these proteins , and isolation of a stable TEX19-hL1-ORF1p complex ( Figure 3A , Figure 3B ) . This interaction required the proteins to be co-expressed ( Figure 3A ) and was resistant to micrococcal nuclease treatment ( Figure 3B ) . Furthermore , TEX19 was found to recognize the conserved and previously crystallized part of the hL1-ORF1p trimer ( Khazina et al . , 2011; Boissinot and Sookdeo , 2016 ) and the N-terminal half of hL1-ORF1p that lacks the RNA-binding domains ( Figure 3C , Figure 3E ) . In addition , the first 68 amino acids of TEX19 , which contain the conserved MCP region and a putative Zn2+-binding motif ( Bianchetti et al . , 2015 ) were found to be necessary and sufficient for the interaction ( Figure 3D , Figure 3F ) . Consequently , the MCP region of TEX19 might contact the conserved C-terminal half of the coiled-coil domain , which is present in both L1-ORF1p fragments tested for interactions , although additional contacts between the variable parts of the two proteins can not be excluded . Taken together , the co-IPs , the co-localization and the isolation of a TEX19:L1-ORF1p complex from bacterially expressed proteins suggest that TEX19 directly interacts with L1-ORF1p in a conserved manner and , to our knowledge , represents the first example of a host protein that directly binds to the retrotransposon-encoded protein L1-ORF1p from mammals . 10 . 7554/eLife . 26152 . 007Figure 3 . Direct interaction between human TEX19 and human L1-ORF1p . ( A ) Strep pull-down assays from bacterial ( Escherichia coli ) lysates . Double-tagged human TEX19 was either co-expressed with Strep-tagged human L1-ORF1p ( lane 8 ) or added after L1-ORF1p immobilization on Strep-Tactin beads ( lane 9 ) . Strep-GB1 served as a control ( lanes 6 and 7 ) . ( B ) Pull-down assays of the co-expressed proteins in the absence and presence of micrococcal nuclease ( MCN , lanes 3 and 4 ) . Strep-GST served as a control ( lanes 5 and 6 ) . ( C ) Bar diagram of human L1-ORF1p based on the crystal structure by Khazina et al . ( 2011 ) and consistent with the alignment by Boissinot and Sookdeo ( 2016 ) . Structural domains are colored and the sub-fragments used for pulldown assays are indicated below the bar with the corresponding amino acid numbers . The C-terminal fragment is sufficient for L1-ORF1p trimerization and has been crystallized . The N-terminal fragment is highly variable among mammals . ( D ) Bar diagram of human TEX19 according to the alignment by Bianchetti et al . ( 2015 ) . The conserved MCP and VPTEL regions are colored and the C-terminal extension that is present in murine TEX19 . 1 and most of the other mammalian homologs is indicated with a dotted line . Purple lines indicate a putative CHCC zinc-binding motif in the MCP region . ( E ) Strep pull-down assays with bacterially expressed sub-fragments of human L1-ORF1p and full-length human TEX19 . ( F ) Strep pull-down assays with bacterially expressed sub-fragments of human TEX19 and full-length human L1-ORF1p . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 007 The strong interaction between TEX19 and hL1-ORF1p seen with bacterially-expressed proteins contrasts with weaker interactions detected in HEK293T cells . However , it is possible that the difference in the strength of these interactions reflects the presence of UBR2 in mammalian cells , which allows a TEX19-UBR2 complex to assemble and transiently interact with hL1-ORF1p to catalyse its ubiquitylation and subsequent degradation . We therefore investigated if L1-ORF1p is ubiquitylated and degraded by the proteasome , and whether this might be stimulated by TEX19 . Endogenously expressed mL1-ORF1p in mouse testes represents a collection of protein molecules expressed from hundreds of variant copies of L1 at different genomic loci ( Chinwalla et al . 2002 ) . Therefore , to allow us to correlate the abundance of L1-ORF1p with its encoding RNA more accurately , and to detect transient polyubiquitylated intermediates that are destined for proteasome-dependent degradation , we expressed engineered epitope-tagged hL1-ORF1p constructs in HEK293T cells . HEK293T cells do not endogenously express detectable levels of TEX19 ( Reichmann et al . , 2017 ) and cell-based ubiquitylation assays show that there is basal ubiquitylation of hL1-ORF1p in these cells , detectable as a ladder of hL1-ORF1p species in his6-myc-Ub pull-downs ( Figure 4A ) . The increasing molecular weights of these bands presumably correspond to increasing ubiquitylation of hL1-ORF1p . Furthermore , treating these cells with the proteasome inhibitor MG132 showed that hL1-ORF1p abundance is negatively regulated by the proteasome in the absence of TEX19 expression ( Figure 4B ) . Interestingly , co-expression of TEX19 during the cell-based ubiquitylation assay increases polyubiquitylation of hL1-ORF1p ( Figure 4C , Figure 4—figure supplement 1A ) . TEX19 expression increases the proportion of hL1-ORF1p-T7 that has at least four ubiquitin monomers , the minimum length of polyubiquitin chain required to target proteins to the proteasome ( Thrower et al . , 2000 ) . These cell-based ubiquitylation assays were performed in the absence of proteasome inhibitor as this treatment can cause the TEX19 . 1-interacting protein UBR2 , and potentially also other regulators of L1-ORF1p , to accumulate ( An et al . , 2012 ) . Therefore , we cannot determine whether TEX19 also influences additional more extensively polyubiquitylated species of hL1-ORF1p that are more rapidly degraded by the proteasome . Nevertheless , expression of TEX19 in these cells is sufficient to reduce the abundance of the T7-tagged hL1-ORF1p protein without any change in the abundance of its encoding RNA ( Figure 4D ) . The ability of TEX19 to regulate L1-ORF1p abundance is not restricted to HEK293T cells , and expression of either mouse or human TEX19 orthologs reduces both mouse and human L1-ORF1p levels in hamster XR-1 cells ( Figure 4—figure supplement 1B , Figure 4—figure supplement 1C ) . Taken together , these gain-of-function data for TEX19 mirror the loss-of-function data obtained from Tex19 . 1−/− testes , confirm that the increased mL1-ORF1p levels in Tex19 . 1−/− testes are not a consequence of altered progression of Tex19 . 1−/− spermatocytes through meiosis ( Crichton et al . , 2017b; Ollinger et al . , 2008 ) , and strongly suggest that Tex19 . 1 orthologs function to post-translationally regulate L1-ORF1p abundance . The ubiquitylation and interaction data together suggests that , TEX19 orthologs regulate L1-ORF1p abundance by molecular recognition of L1-ORF1p and stimulation of its polyubiquitylation and proteasome-dependent degradation . 10 . 7554/eLife . 26152 . 008Figure 4 . TEX19 stimulates polyubiquitylation of hL1-ORF1p . ( A ) Cell-based ubiquitylation assay ( Ub assay ) for T7 epitope-tagged hL1-ORF1p in HEK293T cells . HEK293T cells were transfected with hL1-ORF1p-T7 and his6-myc-ubiquitin ( his6-myc-Ub ) , and his6-tagged proteins isolated using Ni2+ agarose . Inputs and Ni2+ pull-downs were analysed by Western blotting for T7 . ( B ) Western blots and quantification of hL1-ORF1p-T7 abundance in HEK293T cells after treatment with either the proteasome inhibitor MG132 ( 50 µM , 7 hr ) or DMSO as a vehicle control . HEK293T cells were co-transfected with hL1-ORF1p-T7 and RFP to control for transfection efficiency , and hL1-ORF1p-T7 abundance measured relative to RFP , then normalized to the DMSO controls for three independent transfections . MG132 treatment increases hL1-ORF1p-T7 abundance 1 . 96 ± 0 . 21 fold . p<0 . 05 ( t-test , p=0 . 04 ) . ( C ) Cell-based ubiquitylation assay ( Ub assay ) for hL1-ORF1p-T7 in HEK293T cells in the presence and absence of human TEX19 . Ni2+-pull downs were Western blotted ( WB ) with anti-T7 antibodies . Polyubiquitylated hL1-ORF1p-T7 containing four or more ubiquitin molecules ( ~100 kD band and above ) was quantified relative to monoubiquitylated hL1-ORF1p-T7 ( ~58 kD band ) and normalized to empty vector controls . Means ± SEM ( 1 ± 0 . 14 and 2 . 11 ± 0 . 31 for vector control and TEX19 respectively ) are indicated; p<0 . 05 ( t-test , p=0 . 03 ) . ( D ) Western blots of HEK293 FlpIn cells stably expressing hL1-ORF1p-T7 transfected with human TEX19 or empty vector . Abundance of hL1-ORF1p-T7 protein and its encoding RNA were measured relative to lamin B and GAPDH respectively , and normalized to empty vector controls . Means ± SEM ( 1 ± 0 . 09 and 0 . 51 ± 0 . 06 for protein abundance and 1 . 01 ± 0 . 09 and 1 ± 0 . 10 for RNA abundance for vector control and TEX19 respectively ) are indicated; p<0 . 01; ns indicates not significant ( t-test , p=0 . 0005 , 0 . 9 from left to right ) ; Pre-stained MW markers ( kD ) are indicated beside blots . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 00810 . 7554/eLife . 26152 . 009Figure 4—figure supplement 1 . TEX19 orthologs regulate L1-ORF1p abundance . ( A ) Control for cell-based ubiquitylation assay shown in Figure 4C . Ni2+ pull-downs were Western blotted ( WB ) with anti-T7 antibodies to detect the immunoprecipitated his6-myc-Ub conjugates . MW markers ( kD ) are shown beside blots . ( B , C ) Hamster XR-1 cells were transiently transfected with a synthetic mouse or human L1 construct containing T7 epitope-tagged ORF1p ( panel B: mouse , pCEPL1SM-T7; panel C , pAD2TE1 , human ) , and either empty vector , Strep-Tex19 . 1 or Strep-TEX19 expression constructs . Cells were Western blotted ( WB ) for the T7 epitope tag , and for β-actin as a loading control 72 hr post-transfection . Arrows indicate the L1-ORF1p-T7 bands ( 43 kD for mL1-ORF1p-T7 , 40 kD for hL1-ORF1p-T7 ) . The high molecular weight bands migrating more slowly than the L1-ORF1p-T7 constructs in panels B and C likely represent the anti-T7 antibody cross-reacting with cellular proteins from the host cells . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 009 L1-ORF1p has essential roles in L1 retrotransposition ( Beck et al . , 2011; Richardson et al . , 2014a; Hancks and Kazazian , 2016 ) and is strictly required for the retrotransposition of engineered L1 constructs in cultured mammalian cells ( Moran et al . , 1996 ) . Since TEX19 orthologs bind to L1-ORF1p and negatively regulate its abundance , we next investigated whether Tex19 . 1 might inhibit L1 mobilization in cultured cells . Engineered L1 retrotransposition assays with an EGFP retrotransposition indicator cassette ( Ostertag et al . , 2000; Coufal et al . , 2009 ) ( Figure 5A ) were used to measure the effect of Tex19 . 1 on the mobilization rate of active mouse L1 elements ( Goodier et al . , 2001; Han and Boeke , 2004 ) in HEK293T cells . Notably , expression of Tex19 . 1 reduced the ability of both a codon-optimized Tf type and a natural Gf type mouse L1 to mobilize in these cells , suggesting that Tex19 . 1 restricts retrotransposition of multiple active L1 subtypes ( Figure 5B ) . Control experiments verified that a mouse L1 carrying missense mutations in the EN and RT domains of ORF2 ( mouse L1mut2 ) failed to retrotranspose in this assay ( Figure 5B ) , and that retrotransposition was potently inhibited by the restriction factor APOBEC3A ( Bogerd et al . , 2006a; Bogerd et al . , 2006b ) ( Figure 5B ) . Mouse Tex19 . 1 also restricts mobilization of engineered human L1 constructs ( Figure 5—figure supplement 1A ) although less efficiently than it restricts mouse L1s . Altogether , these data show that Tex19 . 1 can function as a restriction factor for L1 mobilization in cultured cells . 10 . 7554/eLife . 26152 . 010Figure 5 . TEX19 orthologs restrict L1 mobilization . ( A ) Schematic of engineered L1 retrotransposition assay in HEK293T cells using an EGFP indicator cassette . ( B ) Flow cytometry profiles from engineered mouse L1 retrotransposition assays performed as shown in panel A . HEK293T cells were co-transfected with engineered mouse L1 retrotransposition constructs containing EGFP indicator cassettes ( 99-gfp-L1SM , 99-gfp-L1SMmut2 , 99-gfp-TGF21 ) , and either Strep-tagged mouse Tex19 . 1 , APOBEC3A ( positive control ) or empty vectors ( pBSKS for APOBEC3A , pIBA105 for Tex19 . 1 ) . EGFP fluorescence is plotted on the x-axis and side scatter on the y-axis of the flow cytometry profiles , and cells classed as EGFP-positive are shown in green . 99-gfp-L1SMmut2 carries missense mutations in the endonuclease and reverse transcriptase domains of ORF2p . p<0 . 05; *p<0 . 01 ( t-test , p=0 . 04 , 0 . 006 , 0 . 04 , 1 , 0 . 00001 , 0 . 0004 for each pairwise comparison with vector from left to right ) . ( C ) Schematic of engineered L1 retrotransposition assays in HeLa cells using a blasticidin resistance indicator cassette . ( D ) Plates stained with 0 . 1% crystal violet showing blasticidin-resistant colonies from engineered L1 retrotransposition assays performed as shown in panel C . Human ( JJ101/L1 . 3 ) and mouse ( JJL1SM ) L1 retrotransposition constructs containing blasticidin resistance indicator cassettes were co-transfected with β-ARRESTIN or APOBEC3A as negative and positive controls respectively , or with Strep-tagged mouse Tex19 . 1 , Strep-tagged human TEX19 or pIBA105 empty vector . Quantification of L1 retrotransposition was calculated relative to the β-ARRESTIN control . p<0 . 05; p<0 . 01 ( t-test , p=0 . 0004 , 0 . 02 , 0 . 002 , 0 . 0002 , 0 . 002 , 0 . 002 for each pairwise comparison with vector from left to right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 01010 . 7554/eLife . 26152 . 011Figure 5—figure supplement 1 . TEX19 orthologs restrict L1 mobilization . ( A ) Flow cytometry profiles of engineered L1 retrotransposition assays in HEK293T cells co-transfected with active and mutant human L1 constructs ( 99-gfp-LRE3 and 99-gfp-JM111 ) containing EGFP retrotransposition cassettes , and either Strep-tagged Tex19 . 1 , APOBEC3A ( positive control ) or empty vectors ( pBSKS for APOBEC3A , pIBA105 for Tex19 . 1 ) . EGFP fluorescence is plotted on the x-axis and side scatter on the y-axis of the flow cytometry profiles , and cells classed as EGFP-positive are shown in green . 99-gfp-JM111 carries the ORF1RA mutations and is severely impaired for retrotransposition ( Han and Boeke , 2004; Moran et al . , 1996 ) . p<0 . 01 ( t-test , p=0 . 0001 , 0 . 003 , 0 . 2 , 0 . 7 for each pairwise comparison with vector from left to right ) . ( B , C ) Blasticidin-resistant colonies from L1 retrotransposition assays in U2OS cells . Human ( JJ101/L1 . 3 ) and mouse ( JJL1SM ) engineered L1 constructs containing blasticidin-resistance retrotransposition cassettes were co-transfected with Strep-tagged mouse Tex19 . 1 , Strep-tagged human TEX19 , or empty vector ( B ) , or with β-ARRESTIN ( negative control ) , APOBEC3A ( positive control ) or empty vector ( C ) . Quantification of L1 retrotransposition normalized for transfection efficiency is shown . p<0 . 05; p<0 . 01 ( t-test , p=0 . 0004 , 0 . 008 , 0 . 005 , 0 . 014 for each pairwise comparison with vector from left to right for B; t-test p=0 . 2 , 0 . 008 , 0 . 3 , 0 . 002 for C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 011 Mouse Tex19 . 1 expression is activated in response to DNA hypomethylation in multiple contexts ( Hackett et al . , 2012 ) , and in humans TEX19 is a cancer testis antigen expressed in multiple types of tumor where it is associated with poor cancer prognosis ( Feichtinger et al . , 2012; Planells-Palop et al . , 2017 ) . We therefore tested whether expression of TEX19 orthologs might be sufficient to restrict L1 mobilization in multiple host cell types . L1 retrotransposition assays using a blasticidin retrotransposition indicator cassette ( Beck et al . , 2010; Goodier et al . , 2007; Morrish et al . , 2002 ) in HeLa cells ( Figure 5C ) showed that mouse Tex19 . 1 similarly restricts mobilization of mouse and human L1 constructs by ~50% in this epithelial carcinoma cell line ( Figure 5D ) . Human TEX19 also restricts mobilization of mouse and human L1 constructs by ~ 50% in HeLa cells ( Figure 5D ) . Similar effects on mobilization of L1 constructs were also observed in U2OS osteosarcoma cells ( Figure 5—figure supplement 1B , Figure 5—figure supplement 1C ) . Thus , TEX19 orthologs are host restriction factors for L1 retrotransposition in mice and humans . Importantly , although we have also shown that TEX19 orthologs promote polyubiquitylation and degradation of L1-ORF1p , since TEX19 can directly bind to L1-ORF1p it is possible that this interaction also disrupts aspects L1-ORF1p function and contributes to TEX19-dependent restriction of L1 mobilization . Moreover , there could be additional aspects of TEX19 function that may also be contributing to its ability to restrict L1 mobilization . Indeed , it is not uncommon for host restriction factors to influence multiple aspects of retrotransposon or retroviral life cycles ( Wang et al . , 2010; Burdick et al . , 2010; Goodier et al . , 2012; Holmes et al . , 2007 ) . The stoichiometric abundance of TEX19 . 1 and UBR2 in co-IPs in combination with the co-fractionation of all detectable TEX19 . 1 protein with UBR2 ( Figure 2A , Figure 2C ) suggests that TEX19-dependent polyubiquitylation of L1-ORF1p , and possibly also TEX19-dependent restriction of L1 mobilization , might be mediated by UBR2 . In contrast to Tex19 . 1 , Ubr2 is ubiquitously expressed ( Figure 6—figure supplement 1A ) and UBR2 could contribute to basal ubiquitylation of L1-ORF1p in HEK293T cells ( Figure 4A ) and other somatic cell types . Thus , TEX19 . 1 could simply stimulate this activity when transcriptionally activated by programmed DNA hypomethylation in the developing germline . A simple test of this model would be that TEX19 . 1-dependent effects on L1-ORF1p abundance or L1 mobilization ought to be abolished in a Ubr2 mutant background . However , the requirement for UBR2 to stabilize TEX19 . 1 protein ( Yang et al . , 2010 ) confounds analysis of the downstream requirement of UBR2 catalytic activity in TEX19 . 1-dependent functions: as TEX19 . 1 protein is unstable and undetectable in the absence of UBR2 ( Yang et al . , 2010 ) , TEX19 . 1 might be expected to be unable to stimulate L1-ORF1p degradation or restrict L1 mobilization regardless of whether the E3 ubiquitin ligase activity of UBR2 is required for these functions or not . Indeed , Ubr2−/− testes largely phenocopy Tex19 . 1−/− testes , including transcriptional de-repression of MMERVK10C LTR retrotransposons ( Crichton et al . , 2017a ) . To dissociate the effects of UBR2 on stability of TEX19 . 1 protein from potential effects on L1-ORF1p abundance and L1 mobilization , we tested whether UBR2 can regulate L1 in the absence of effects on TEX19 stability by using somatic HEK293T cells . Interestingly , mouse UBR2 co-IPs with mL1-ORF1p in HEK293T cells ( Figure 6A ) , a cell type that does not express any detectable TEX19 protein ( Reichmann et al . , 2017 ) . Thus , these data strongly suggest that UBR2 is able to regulate L1-ORF1p independently of any effects on TEX19 protein stability . UBR2 also interacts with mL1-ORF1RAp mutants that have reduced binding to RNA ( Figure 6B ) , suggesting that this physical interaction is not mediated by L1 RNA . Furthermore , these interactions are conserved in human L1-ORF1p ( Figure 6—figure supplement 1B , Figure 6—figure supplement 1C ) . In addition , overexpression of UBR2 alone restricts mobilization of an engineered human L1 ( Figure 6C ) . Thus , at least in overexpression experiments , UBR2 is able to physically interact with L1-ORF1p and restrict mobilization of L1 constructs in cultured cells . 10 . 7554/eLife . 26152 . 012Figure 6 . The TEX19 . 1-interacting protein UBR2 negatively regulates mL1-ORF1p abundance and L1 mobilization . ( A ) Co-immunoprecipitations ( co-IPs ) from HEK293T cells co-transfected with mL1-ORF1p-T7 and either mouse UBR2-GFP or GFP . IP inputs and IPs were Western blotted with T7 and GFP antibodies . A presumed cleavage product of UBR2-GFP running smaller than GFP itself is indicated with an asterisk . ( C ) Plates from an engineered L1 retrotransposition assay as described in Figure 5C stained with 0 . 1% crystal violet showing blasticidin-resistant colonies . Human ( JJ101/L1 . 3 ) L1 retrotransposition construct was co-transfected with β-ARRESTIN or APOBEC3A as negative and positive controls respectively , or with UBR2-Flag . p<0 . 05; **p<0 . 01 ( t-test , p=0 . 0004 , 0 . 02 from left to right ) . ( D ) Western blots of endogenous UBR2 and mL1-ORF1p in P16 Ubr2+/+ and Ubr2−/− mouse cerebellum . β-actin was used as a loading control . Quantification of mL1-ORF1p-T7 and L1 mRNA relative to β-actin and normalized to Ubr2+/+ control mice is also shown . Means ± SEM are indicated ( 1 ± 0 . 05 and 3 . 82 ± 0 . 25 for Ubr2+/+ and Ubr2−/− respectively ) p<0 . 05; ns indicates not significant ( t-test , p=0 . 048 , 0 . 9 from left to right ) ; pre-stained MW markers ( kD ) are shown beside blots . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 01210 . 7554/eLife . 26152 . 013Figure 6—figure supplement 1 . The ubiquitously-expressed E3 ubiquitin ligase UBR2 physically interacts with L1-ORF1p but does not regulate its abundance in the cerebrum . ( A ) Ubr2 transcript abundance in multiple adult tissues was determined from ENCODE RNA sequencing data GSE36025 ( Lin et al . , 2014 ) by calculating the total number of reads mapped to the Ubr2 locus per million reads mapped in the dataset , and normalising this to the length of the Ubr2 locus . ( B , C ) Co-immunoprecipitations ( co-IPs ) from HEK293T cells co-transfected with T7 epitope-tagged hL1-ORF1p and mouse UBR2-GFP expression constructs . GFP alone was used as a negative control . ( D ) Genotyping of Ubr2−/− mice . An XbaI restriction site and premature stop codon ( asterisk ) are introduced into exon 3 of Ubr2 by CRISPR/Cas9 , and mice genotyped by amplifying a region encompassing exon 3 ( primers indicated by arrows ) and digesting the PCR product with XbaI . Three Ubr2−/− mice , and Ubr2+/+ and distilled water ( dH2O ) controls are shown . ( E ) Western blots showing endogenous UBR2 and mL1-ORF1p expression in Ubr2+/+ and Ubr2−/− mouse cerebrum . β-actin was used as a loading control . Positions of epitope-tagged proteins and pre-stained molecular weight markers in kD are indicated . Quantification of endogenous mL1-ORF1p abundance and L1 RNA abundance relative to β-actin in Ubr2+/+ and Ubr2−/− mouse cerebrum is also shown . Relative abundance was normalized to the mean of the Ubr2+/+ control mice . Error bars indicate SEM , MW markers ( kD for protein , bp for DNA ) are shown beside blots and gels . No significant difference in either mL1-ORF1p or L1 RNA abundance was detected between wild-type and mutant tissue ( t-test , p=0 . 4 for mL1-ORF1p; -test , p=0 . 6 for L1 RNA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 01310 . 7554/eLife . 26152 . 014Figure 6—figure supplement 2 . Tex19 . 1 expression is not detectable in brain . qRT-PCR for Tex19 . 1 in brain from wild-type mice . Cerebrum and cerebellum were isolated at P16 . Embryonic placenta ( E12 . 5 ) was used as a positive control . RNA abundance is expressed relative to β-actin , expression in two independent animals is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 014 To investigate regulation of hL1-ORF1p abundance by UBR2 further , we generated UBR2 mutant HEK293T cell lines by CRISPR/Cas9-mediated genome editing . However , these cell lines grew slowly and poorly in culture , presumably reflecting the normal cellular roles of UBR2 in cohesin regulation , DNA repair , and chromosome stability ( Ouyang et al . , 2006; Reichmann et al . , 2017 ) . Therefore , to allow a meaningful analysis of the role of endogenous UBR2 in L1 regulation we analysed Ubr2−/− mice ( Figure 6—figure supplement 1D , Figure 6—figure supplement 1E ) which , despite having defects in spermatogenesis and female lethality , are otherwise grossly normal ( Kwon et al . , 2003 ) . Notably , mL1 , but not Tex19 . 1 ( Figure 6—figure supplement 2 ) , is expressed in the brain ( Wang et al . , 2001; Muotri et al . , 2010 ) , therefore we used this tissue to assess whether Ubr2 might have a Tex19 . 1-independent role in regulating mL1-ORF1p . Consistent with the physical interaction between UBR2 and mL1-ORF1p ( Figure 6A ) , we found that mL1-ORF1p abundance is post-transcriptionally elevated approximately four fold in the cerebellum of Ubr2−/− mice ( Figure 6D ) , suggesting that UBR2 may directly regulate polyubiquitylation and subsequent degradation of mL1-ORF1p in vivo . Interestingly , loss of Ubr2 has no detectable effect on mL1-ORF1p abundance in the cerebrum ( Figure 6—figure supplement 1E ) , which may reflect cell type specific differences in L1 regulation or genetic redundancy between UBR-domain proteins ( Tasaki et al . , 2005 ) . Nevertheless , regardless of this additional complexity in the cerebrum , the increased abundance of mL1-ORF1p in Ubr2−/− cerebellum demonstrates that endogenous Ubr2 plays a Tex19 . 1-independent role in regulating mL1-ORF1p abundance in vivo . Ubr2 has numerous endogenous cellular substrates and host functions beyond regulating mL1-ORF1p ( Ouyang et al . , 2006; Reichmann et al . , 2017; Sriram et al . , 2011 ) , but expression of Tex19 . 1 in the germline or in response to DNA hypomethylation appears to stimulate a pre-existing activity of UBR2 to regulate mL1-ORF1p , possibly at the expense of UBR2’s activity towards some endogenous cellular substrates ( Reichmann et al . , 2017 ) . As outlined earlier , L1 mobilization is thought to occur primarily in pluripotent cells within the germline cycle ( Kano et al . , 2009; Richardson et al . , 2017 ) , and regulation of L1 expression and mobilization in these cells is likely to significantly impact on the ability of L1 to influence germline mutation and genome evolution . Therefore , we tested whether Tex19 . 1 , which is expressed in pluripotent cells ( Kuntz et al . , 2008 ) , has a role in regulating L1 expression and restricting L1 mobilization in this cell type . We first investigated whether Tex19 . 1 regulates mL1-ORF1p abundance in pluripotent mouse ESCs . Biochemical isolation of polyubiquitylated proteins suggests that endogenous mL1-ORF1p is polyubiquitylated in pluripotent mouse ESCs ( Figure 7A ) . Furthermore , proteasome inhibition with lactacystin caused a ~4 fold increase in the abundance of mL1-ORF1p relative to β-actin after 6 hr of treatment ( Figure 7B ) . Taken together these data suggest that mL1-ORF1p abundance is regulated by the proteasome in pluripotent mouse ESCs . hL1-ORF1p abundance is similarly regulated by the proteasome in human ESCs and human embryonal carcinoma ( EC ) cells ( Figure 7—figure supplement 1 ) . In contrast to a previous report assessing the abundance of retrotransposon RNA in ESCs derived from heterozygous mouse crosses ( Tarabay et al . , 2013 ) , Tex19 . 1−/− mouse ESCs generated by sequential gene targeting ( Figure 7—figure supplement 2 ) in a defined genetic background , cultured in 2i conditions , and analysed at low passage number do not de-repress L1 RNA ( Figure 7C ) . These Tex19 . 1−/− mouse ESCs contain elevated levels of endogenous mL1-ORF1p , but this increase in mL1-ORF1p levels is not accompanied by increased endogenous L1 mRNA levels ( Figure 7C ) . Moreover , loss of Tex19 . 1 does not detectably affect transcription or translation of L1 reporter constructs in ESCs ( Figure 7—figure supplement 3 ) . Taken together these data suggest that , similar to male germ cells ( Figure 1 ) , Tex19 . 1 functions to post-translationally repress mL1-ORF1p in pluripotent cells . 10 . 7554/eLife . 26152 . 015Figure 7 . Tex19 . 1 negatively regulates mL1-ORF1p abundance and L1 mobilization in mouse ESCs . ( A ) Mouse ESC lysates ( input ) were incubated with polyubiquitin-binding TUBE2 beads or control agarose beads and Western blotted for endogenous mL1-ORF1p . Non-specific binding of non-ubiquitylated mL1-ORF1p is detectable ( asterisk ) , in addition to specific enrichment of polyubiquitylated mL1-ORF1p with TUBE2 . ( B ) Western blot for endogenous mL1-ORF1p after treatment with 25 µM lactacystin proteasome inhibitor for the indicated times . β-actin is a loading control . ( C ) Western blot for endogenous mL1-ORF1p in Tex19 . 1+/+ and Tex19 . 1−/− mouse ESCs . mL1-ORF1p abundance ( Western blot ) and L1 RNA abundance ( qRT-PCR using primers against ORF2 ) were quantified relative to β-actin and normalized to Tex19 . 1+/+ ESCs . Means ± SEM are indicated ( 1 ± 0 and 1 . 99 ± 0 . 36 for protein and 1 ± 0 . 19 and 1 . 07 ± 0 . 15 for RNA for Tex19 . 1+/+ and Tex19 . 1−/− respectively ) ; p<0 . 05; ns indicates not significant ( t-test , p=0 . 049 , 0 . 8 from left to right ) . ( D ) Neomycin-resistant colonies from L1 retrotransposition assays in Tex19 . 1+/+ and Tex19 . 1−/− ESCs . ESCs were transfected with LINE retrotransposition constructs carrying the mneoI indicator cassette and either synthetic mouse L1 ( pCEPL1SM ) or zebrafish LINE-2 ( Zfl2 . 2 ) sequences , the number of neomycin-resistant colonies counted , and retrotransposition frequency calculated relative to Tex19 . 1+/+ ESCs transfected with pCEPL1SM . p<0 . 05; ns indicates not significant ( t-test , p=0 . 01 , 0 . 3 from left to right ) ; error bars indicate SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 01510 . 7554/eLife . 26152 . 016Figure 7—figure supplement 1 . hL1-ORF1p abundance in human embryonal carcinoma cells and human ESCs increases in response to inhibition of the proteasome . ( A ) Western blot showing abundance of endogenous hL1-ORF1p in PA-1 human embryonal carcinoma ( EC ) cells after addition of 25 µM lactacystin to inhibit the proteasome . β-actin is shown as a loading control . ( B ) Western blot showing abundance of endogenous hL1-ORF1p in H9 human ESCs after addition of 25 µM MG132 to inhibit the proteasome . p53 is a positive control and accumulates upon MG132 treatment , β-actin is shown as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 01610 . 7554/eLife . 26152 . 017Figure 7—figure supplement 2 . Generation and validation of Tex19 . 1−/− ESCs . ( A ) Schematic diagram showing the Tex19 . 1 alleles generated in ESCs . The Tex19 . 1 locus is shown in purple , flanking DNA in grey . Introns are shown as lines , exons as rectangles , and the coding region as large rectangles . LoxP sites ( black ) and an internal ribosome entry site ( IRES , blue ) coupled to enhanced green fluorescent protein ( EGFP , green ) were introduced into the locus to generate a Tex19 . 1fl allele . This allele also contains an Frt site ( orange ) left over after excision of a neomycin resistance cassette . After treatment with Cre recombinase the entire Tex19 . 1 coding sequence is removed from the Tex19 . 1fl allele to generate Tex19 . 1- . Arrows indicate position of genotyping primers used in panel B . ( B ) PCR genotyping of Tex19 . 1−/− ESCs . Genomic DNA from Tex19 . 1+/+ ESCs , Tex19 . 1−/− ESCs or distilled water ( dH2O ) was used as a template for genotyping PCR using the primers shown in panel A . Migration of selected bands in the KB ladder ( Invitrogen ) is indicated . ( C ) Western blot for TEX19 . 1 and lamin B in Tex19 . 1+/+ and Tex19 . 1−/− ESCs . MW markers ( kD ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 01710 . 7554/eLife . 26152 . 018Figure 7—figure supplement 3 . Loss of Tex19 . 1 does not affect L1 promoter or L1 translation reporter activity in mouse ESCs . ( A ) Schematic diagram showing promoter-luciferase constructs containing indicated control or L1-derived promoters . Luciferase activity ( relative light units per second ) of these constructs after transfection into Tex19 . 1+/+ and Tex19 . 1−/− ESCs is shown . Luciferase activity was corrected for transfection efficiency and normalized to the SV40 promoter construct in control ESCs . Error bars indicate SEM for technical replicates of luciferase assays from the same cell lysates . ( B ) Schematic diagram showing translation-luciferase constructs . Regions of L1 ( A: 400 bp upstream of ORF1p covering the 5' UTR; D: 200 bp upstream of ORF2p covering the intergenic region; 3: 312 bp from the 3' UTR ) inserted in the pRF dicistronic reporter construct ( Li et al . , 2006 ) were transfected into Tex19 . 1+/− and Tex19 . 1−/− ESCs . The pRFD construct contains the ORF2p internal ribosome entry site that binds hnRNPL and nucleolin , cellular factors that restrict L1 ( Peddigari et al . , 2013 ) . Luciferase acivity for these translation-luciferase constructs in Tex19 . 1+/− and Tex19 . 1−/− ESCs is shown . Firefly luciferase ( FLUC ) was measured relative to Renilla luciferase ( RLUC ) . Data represents three replicate transfections for each construct , error bars represent SEM . There is no statistically significant difference in luciferase activity between Tex19 . 1−/− ESCs and controls for any of the pRF , pRFA , pRFD or pRF3 constructs ( t-test , p=0 . 2 , 0 . 8 , 0 . 9 , 0 . 5 respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 01810 . 7554/eLife . 26152 . 019Figure 7—figure supplement 4 . Tex19 . 1 restricts mobilization of engineered L1 constructs in mouse ESCs . ( A ) Plates stained with 0 . 1% crystal violet showing G418-resistant colonies from L1 retrotransposition assays in Tex19 . 1+/+ and Tex19 . 1−/− ESCs . ESCs were co-transfected with a synthetic mouse L1 construct and either empty vector or the L1 restriction factor APOBEC3A . Retrotransposition frequency was calculated relative to Tex19 . 1+/+ ESCs transfected with empty vector . **p<0 . 01 ( t-test , p=0 . 005 , 0 . 003 from left to right ) ; error bars indicate SEM . ( B ) Additional control for L1 retrotransposition assays in mouse ESCs . Plates stained with 0 . 1% crystal violet showing G418-resistant colonies from L1 retrotransposition assays . Engineered L1 constructs ( pCEPL1SMN21A ) carrying the N21A mutation in the endonuclease domain of ORF2p that impairs L1 mobilization ( Alisch et al . , 2006 ) have greatly reduced retrotransposition in both Tex19 . 1+/+ and Tex19 . 1−/− ESCs relative to codon-optimized L1 ( pCEPL1SM ) . ( C ) Transfection efficiency controls for L1 retrotransposition assays in mouse ESCs . Tex19 . 1+/+ and Tex19 . 1−/− ESCs are able to form similar numbers of colonies when transfected with a control plasmid conferring G418 resistance ( pU6ineo ) . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 019 Next we tested whether loss of Tex19 . 1 also results in increased mobilization of mouse L1 constructs in pluripotent ESCs . Although L1 retrotransposition assays have previously been performed in pluripotent human cells ( Wissing et al . , 2011; Garcia-Perez et al . , 2007 , 2010 ) , this assay has not yet been adapted to mouse ESCs and , to our knowledge , no restriction factor has been shown to restrict mobilization of L1 constructs in mouse pluripotent cells or germ cells . Therefore we optimized the L1 retrotransposition assay in mouse ESCs ( García-Cañadas et al . , manuscript in preparation ) using a neomycin retrotransposition indicator cassette ( Freeman et al . , 1994 ) . Notably , the optimized assay routinely resulted in the appearance of hundreds of G418-resistant colonies when mouse ESCs were transfected with an active mouse Tf L1 construct ( Han and Boeke , 2004 ) ( Figure 7D ) . Controls verified that co-transfection of the L1 restriction factor APOBEC3A ( Bogerd et al . , 2006b ) severely reduces mL1 retrotransposition in mouse ESCs ( Figure 7—figure supplement 4A ) , and that an allelic mL1 containing the N21A missense mutation in the EN domain of ORF2p ( Alisch et al . , 2006 ) retrotransposes at low levels in mouse ESCs ( Figure 7—figure supplement 4B ) . Thus , the adapted L1 retrotransposition assay appears to reflect bone fide mobilization of L1 constructs in mouse ESCs . We next used the optimized assay to investigate the role of Tex19 . 1 in controlling L1 retrotransposition in pluripotent mouse ESCs . Interestingly , mobilization of an active mouse Tf L1 is reproducibly elevated around 1 . 5-fold in Tex19 . 1−/− ESCs relative to Tex19 . 1+/+ wild-type ESCs ( Figure 7D , Figure 7—figure supplement 4B ) . Control experiments revealed that both Tex19 . 1+/+ and Tex19 . 1−/− ESCs could generate similar numbers of G418-resistant foci when transfected with a plasmid carrying a neomycin resistance cassette ( Figure 7—figure supplement 4C ) . Thus , these data strongly suggest that Tex19 . 1 controls L1 retrotransposition in mouse pluripotent ESCs , presumably at least in part by promoting proteasome degradation of mL1-ORF1p . To further test this , we analysed whether Tex19 . 1 could restrict retrotransposition of an active zebrafish LINE-2 element that naturally lacks ORF1p but can efficiently retrotranspose in cultured human cells ( Sugano et al . , 2006; Garcia-Perez et al . , 2010 ) , and in cultured chicken cells that lack endogenous L1-ORF1p ( Suzuki et al . , 2009 ) . Remarkably , loss of Tex19 . 1 does not influence the rate of retrotransposition of the ORF1p-independent engineered zebrafish LINE-2 construct in mouse ESCs ( Figure 7D ) . Thus , these data suggest that one role of endogenously expressed Tex19 . 1 in mouse pluripotent cells is to restrict L1 mobilization , and thereby promote genome stability in the cells that can transmit new L1 integrations to the next generation . This study identifies Tex19 . 1 as a host restriction factor for L1 mobilization in the mammalian germline . We have previously reported that Tex19 . 1 plays a role in regulating the abundance of retrotransposon RNAs ( Ollinger et al . , 2008; Reichmann et al . , 2012 , 2013 ) , which appears to reflect transcriptional de-repression of specific retrotransposons ( Crichton et al . , 2017a ) . Although loss of Tex19 . 1 results in de-repression of L1 RNA in placenta ( Reichmann et al . , 2013 ) , L1 RNA abundance is not affected by loss of Tex19 . 1 in male germ cells ( Ollinger et al . , 2008 ) or , in contrast to a previous report ( Tarabay et al . , 2013 ) , in mouse ESCs ( Figure 7 ) . Indeed here we show that Tex19 . 1 has a role in the post-translational regulation of L1-ORF1p steady-state levels in these cells . Thus , Tex19 . 1 appears to regulate retrotransposons at multiple stages of their life cycle . It is possible that Tex19 . 1 is affecting different E3 ubiquitin ligases , or different E3 ubiquitin ligase substrates , in order to repress different stages of the retrotransposon life cycle . However , loss of Tex19 . 1 results in a 1 . 5-fold increase in the rate of mobilization of L1 constructs in pluripotent cells . Since L1 mobilization mostly takes place in the pluripotent phase of the germline cycle , and new L1-dependent mobilization events are thought to be inherited by one in every twenty human births ( Kazazian , 1999 ) , TEX19 activity could be having a significant impact on L1-derived mutations during evolution . Retrotransposons appear to provide functions that are advantageous for mammalian development and evolution ( Garcia-Perez et al . , 2016 ) , and the activity of restriction mechanisms like the TEX19-dependent mechanism we have described here , that control the ability of retrotransposons to mobilize , rather than eliminate their transcriptional activity altogether , could potentially allow retrotransposons to participate in and drive the evolution of key gene regulatory networks in pluripotent cells while minimising their mutational load on the germline genome . Our data suggests that L1-ORF1p is post-translationally modified by ubiquitylation in somatic and germline cells . Phosphorylation of L1-ORF1p has been previously reported in somatic tissues and is required for L1 retrotransposition in these cells ( Cook et al . , 2015 ) . However , we are not aware of any previous reports that post-translational modifications of L1-ORF1p are present in the germline , particularly in the pluripotent phase of the germline cycle when L1 retrotransposition is thought to primarily occur ( Kano et al . , 2009 ) . There are 32 lysine residues in human L1-ORF1p that could act as potential ubiquitylation sites ( deHaro et al . , 2014 ) , and 42 , 47 and 39 lysines in mouse Tf , Gf and A subtypes of L1-ORF1p respectively that could act as potential ubiquitylation sites . It will be of interest to determine which of these lysines are ubiquitylated in somatic and germline tissues , and how variant these residues are between retrotransposition-competent L1s . Post-translational regulation of L1 potentially provides an additional layer of genome defence that could be particularly important during periods of epigenetic reprogramming in early embryogenesis or in the developing primordial germ cells when transcriptional repression of retrotransposons might be more relaxed ( Molaro et al . , 2014; Fadloun et al . , 2013 ) . Indeed , the sensitivity of Tex19 . 1 expression to DNA hypomethylation ( Hackett et al . , 2012 ) will allow post-translational suppression of L1 to be enhanced during these stages of development . Post-translational regulation of L1s is also likely important to limit the activity of L1 variants that evolve to escape transcriptional repression by the host and will provide a layer of genome defence while the host adapts its KRAB zinc-finger protein repertoire to these new variants ( Jacobs et al . , 2014 ) . Analysis of L1 evolution shows that regions within L1-ORF1p are under strong positive selection suggesting that host restriction systems are targeting L1-ORF1p post-translationally and impacting on evolution of these elements ( Boissinot and Furano , 2001; Sookdeo et al . , 2013 ) . Although this evidence for post-translational restriction factors acting on L1-ORF1p has been known for over 15 years , to our knowledge no host factors have been identified that directly bind to L1-ORF1p and restrict L1 mobilization in germline cells . It is possible that the physical interactions between L1-ORF1p and TEX19:UBR2 that we describe here are contributing to these selection pressures acting on L1-ORF1p . While UBR2 is able to target L1-ORF1p in the absence of TEX19 , evolution of a less constrained TEX19 adapter to provide a further link between UBR2 and L1-ORF1p could potentially resolve the contradictory pressures on UBR2 to maintain interactions with some endogenous cellular substrates while targeting a rapidly evolving retrotransposon protein for degradation . Our data strongly suggest that TEX19 . 1 likely exists in a complex with UBR2 in ESCs , and that TEX19 . 1 stimulates a basal activity of UBR2 to promote polyubiquitylation of L1-ORF1p ( Figure 8 ) . Ubr1 , a yeast ortholog of UBR2 , has different binding sites for different types of substrate ( Xia et al . , 2008 ) . Ubr1 participates in the N-end rule pathway that degrades proteins depending on their N-terminal amino acids , and can bind to and ubiquitylate proteins containing specific residues at their N-termini ( N-end rule degrons ) . Ubr1 also binds to and catalyses ubiquitylation of proteins that have more poorly defined non-N-terminal internal degrons ( Xia et al . , 2008; Sriram et al . , 2011; Kim et al . , 2014 ) . Full-length human L1-ORF1p does not have a potential N-end rule degron at its N-terminus ( Kim et al . , 2014; Sriram et al . , 2011 ) , and we speculate the interaction between UBR2 and L1-ORF1p likely reflects an internal degron in the retrotransposon protein . One of the known internal degron substrates of yeast Ubr1 is CUP9 , a transcription factor that regulates expression of a peptide transporter ( Turner et al . , 2000 ) . Binding and polyubiquitylation of CUP9 by Ubr1 is allosterically activated by specific dipeptides binding to the N-end rule degron binding sites in Ubr1 ( Du et al . , 2002; Xia et al . , 2008; Turner et al . , 2000 ) . The effect of these dipeptides on Ubr1 activity in yeast strongly resonates with the effects of TEX19 orthologs on UBR2 activity in mammals: TEX19 orthologs binds to UBR2 and inhibits its activity towards N-end rule substrates ( Reichmann et al . , 2017 ) , but stimulate polyubiquitylation of L1-ORF1p . The direct interaction between TEX19 orthologs and L1-ORF1p could further enhance L1-ORF1p binding to UBR2 by stabilizing the highly flexible L1-ORF1p trimers ( Khazina et al . , 2011 ) in a conformational state that exposes an internal degron and favors their ubiquitylation . Thus , TEX19 orthologs appear to function , at least in part , by re-targeting UBR2 away from N-end rule substrates and towards a retrotransposon substrate . However , the direct interaction between TEX19 orthologs and L1-ORF1p means that it is possible that TEX19 orthologs are interfering with L1-ORF1p function in multiple ways in order to restrict L1 mobilization . Thus , while one outcome of this interaction appears to be increased polyubiquitylation and degradation of L1-ORF1p , the interaction between TEX19 orthologs and L1-ORF1p could also interfere with the nucleic acid chaperone activity of L1-ORF1p ( Martin et al . , 2005 ) , or its interactions with either L1-encoded or host-encoded molecules ( Taylor et al . , 2013; Goodier et al . , 2013 ) . 10 . 7554/eLife . 26152 . 020Figure 8 . Model For UBR2 and TEX19 . 1-mediated polyubiquitylation of mL1-ORF1p . In methylated somatic cells , the RING domain E3 ubiquitin ligase UBR2 and its cognate E2 ubiquitin conjugating enzyme UBE2A/B can interact with mL1-ORF1p and catalyse ubiquitylation and proteasome-dependent turnover of this protein . TEX19 . 1 in hypomethylated cells , including pluripotent cells and germ cells , interacts with both UBR2 and mL1-ORF1p , stimulating further polyubiquitylation and proteasome-dependent turnover of mL1-ORF1p . The interaction between TEX19 . 1 and UBR2 concomitantly inhibits the activity of UBR2 towards N-end rule substrates ( Reichmann et al . , 2017 ) . This model does not exclude additional factors and/or mechanisms contributing to the effects of UBR2 and TEX19 . 1 on the stability of mL1-ORF1p . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 02010 . 7554/eLife . 26152 . 021Figure 8—figure supplement 1 . Model for retrotransposon regulation during epigenetic reprogramming in lineage-restricted germ cells . Schematic diagram illustrating how post-translational control mechanisms can contribute to retrotransposon control and genomic stability in hypomethylated male germ cells . RNAs are indicated by wavy lines , proteins by small solid polygons . DOI: http://dx . doi . org/10 . 7554/eLife . 26152 . 021 The constellation of L1 sequences in the genome ( Chinwalla et al . 2002 ) makes it difficult to quantitatively determine how much each L1 locus contributes to the cellular pool of L1 RNAs , and how much each L1 RNA contributes to the amount of L1-encoded proteins in the cell . We have been unable to detect effects on bulk transcription of L1 in the absence of Tex19 . 1 , and Tex19 . 1 could be potentially regulating endogenous L1-ORF1p abundance in testes and ES cells entirely post-transcriptionally . However , we cannot rule out the possibility that transcriptional or translational de-repression of specific variant copies of L1 are contributing to the increase in the abundance of L1-ORF1p species detected in Tex19 . 1−/− ES cells and testes . Our data using tagged copies of L1-ORF1p have allowed us to link transcription and protein abundance from a single defined L1 sequence suggesting that Tex19 . 1 can act , at least in part , at post-transcriptional level to regulate endogenous L1-ORF1p abundance in the germline . Our data are consistent with TEX19 . 1 playing a role in promoting polyubiquitylation of mL1-ORF1p in mouse ESCs , thereby reducing the steady-state abundance of mL1-ORF1p in these cells ( Figure 7 ) . Quantifying the amount of the transient heterogeneous mixture of polyubiquitylated mL1-ORF1p endogenously expressed in control and Tex19 . 1−/− mESCs cells is technically challenging . This is partly due to the activity of deubiquitylases present in the ES cell lysates , partly due to the heterogeneous nature of endogenously expressed mL1-ORF1p , which may be recognised with multiple different affinities by anti-mL1-ORF1p antibodies , particularly when present in different ubiquitylated states , and partly because this experiment would likely require ESCs to be treated with proteasome inhibitor to allow polyubiquitylated species to accumulate . This treatment can stabilize E3 ubiquitin ligases like UBR2 itself ( An et al . , 2012 ) , or other proteins that can regulate L1-ORF1p abundance independently of TEX19 . 1 . Cell-based ubiquitylation assays ( Figure 4 ) circumvent these challenges by assessing the effect of TEX19 on a single epitope-tagged copy of L1-ORF1p in the absence of intervention with proteasome inhibitors and under denaturing conditions that inactivate deubiquitylases in the lysate . Taken together , the protein interactions , gain-of-function cell-based ubiquitylation data , and loss-of-function phenotyping in ESCs and in mouse testes indicate that TEX19 . 1 plays a role in regulating the polyubiquitylation and stability of mL1-ORF1p . The data presented here suggests that programmed DNA hypomethylation in the mouse germline extends beyond activating components of the PIWI-piRNA pathway ( Hackett et al . , 2012 ) to include enhancing the activity of the ubiquitin-proteasome system towards retrotransposon substrates . Recent data has suggested that TEX19 . 1 physically interacts with components of the PIWI-piRNA pathway ( Tarabay et al . , 2017 ) , although it is not clear whether these proposed interactions have functional consequences for retrotransposon suppression in vivo . Activation of post-translational genome-defence mechanisms may allow mammalian germ cells to safely transcribe retrotransposons by preventing these transcripts from generating RNPs that can mutate the germline genome ( Figure 8—figure supplement 1 ) . The retrotransposon transcripts can then potentially be processed into piRNAs and used to identify retrotransposon loci where epigenetic silencing needs to be established . De novo establishment of epigenetic silencing at retrotransposons in the Arabidopsis germline involves transfer of small RNAs between a hypomethylated vegetative cell and a germ cell ( Slotkin et al . , 2009 ) , whereas these processes happen sequentially in the same germ cell in mammals ( Figure 8—figure supplement 1 ) . Therefore the ability to enhance post-translational control of retrotransposons may be a key feature of epigenetic reprogramming in the mammalian germline that limits the trans-generational genomic instability caused by retrotransposon mobilization . Tex19 . 1 mutant mice ( RRID:MGI:4453205 ) on a C57BL/6J genetic background ( RRID:IMSR_JAX:000664 , obtained from Charles River ) were maintained and genotyped as described ( Ollinger et al . , 2008; Reichmann et al . , 2012 ) . Tex19 . 1+/− heterozygotes have no detectable testis phenotype and indistinguishable sperm counts from wild-type animals ( Ollinger et al . , 2008 ) , and prepubertal Tex19 . 1−/− homozygotes were typically compared with heterozygous littermates to control for variation between litters . Ubr2−/− mice were generated by CRISPR/Cas9 double nickase-mediated genome editing in zygotes ( Ran et al . , 2013 ) . Complementary oligonucleotides ( Supplementary file 3 ) targeting exon 3 of UBR2 were annealed and cloned into plasmid pX335 ( Cong et al . , 2013 ) , amplified by PCR , then in vitro transcribed using a T7 Quick High Yield RNA Synthesis kit ( NEB ) to generate paired guide RNAs . RNA encoding the Cas9 nickase mutant ( 50 ng/µl , Tebu-Bio ) , paired guide RNAs targeting exon 3 of UBR2 ( each at 25 ng/µl ) , and 150 ng/µl single-stranded DNA oligonucleotide repair template ( Supplementary file 3 ) were microinjected into the cytoplasm of B6CBAF1/J × B6CBAF1/J zygotes ( RRID:IMSR_JAX:100011 , obtained from Charles River ) . The repair template introduces an XbaI restriction site and mutates cysteine-121 within the UBR domain of UBR2 ( Uniprot Q6WKZ8-1 ) to a premature stop codon . The zygotes were then cultured overnight in KSOM ( Millipore ) and transferred into the oviduct of pseudopregnant recipient females . Pups were genotyped for the presence of the XbaI restriction site . The Ubr2−/− male mice generated in this way have no overt phenotypes except testis defects and infertility and Ubr2−/− females are born at sub-Mendelian ratios , all as previously described for Ubr2−/− mice generated by gene targeting in ESCs ( Kwon et al . , 2003 ) . The day of birth was designated P1 , and mice were culled by cervical dislocation . Mouse experiments were performed in accordance with local ethical guidelines and under authority of UK Home Office Project Licence PPL 60/4424 . For mouse experiments , a sample size of three mutant animals was typically used and alongside littermate controls to allow consistent phenotypic changes in retrotransposon expression to be associated with genotype . Each animal was considered a biological replicate . We used cell lines that were previously shown to support retrotransposition of engineered L1 constructs or Tex19 . 1−/− models generated in this study . Cell lines were maintained at 37°C in 5% CO2 . HEK293T and U2OS cells were obtained from ATCC ( ATCC Cat# CRL-3216 , RRID:CVCL_0063; ATCC Cat# HTB-96 , RRID:CVCL_0042 ) and HeLa cells were provided by John V . Moran ( University of Michigan , US ) . These cell lines were grown in Dulbecco’s Modified Eagle's Media ( DMEM ) supplemented with 10% foetal calf serum , 1% penicillin-streptomycin , and 1% L-glutamine . E14Tg2a mouse ESCs ( RRID:CVCL_9108 ) were obtained from Julia Dorin ( MRC Human Genetics Unit , UK ) and cultured on gelatinized flasks in 2i culture conditions ( 1:1 DMEM/F12 media:neurobasal media supplemented with N2 and B27 , 10% foetal calf serum , 1% L-glutamine , 0 . 1% β-mercaptoethanol , 1 µM PD0325901 ( StemMACS ) , and 3 µM CHIR99021 ( StemMACS ) . Hamster XR-1 cells ( RRID:CVCL_K253 ) ( Stamato et al . , 1983 ) were provided by Thomas D . Stamato ( The Lankenau Institute fro Medical Research , US ) and grown in DMEM low glucose medium containing 10% foetal calf serum , 1% L-glutamine , 1% penicillin-streptomycin and 0 . 1 mM non-essential amino acids . Human PA-1 cells ( Zeuthen et al . , 1980 ) were obtained from ATCC ( ATCC Cat# CRL-1572 , RRID:CVCL_0479 ) and grown in Minimal Essential Media ( MEM ) supplemented with 10% heat-inactivated foetal calf serum , 1% L-glutamine , 1% penicillin-streptomycin and 0 . 1 mM non-essential amino acids . H9 human ESCs ( Thomson et al . , 1998 ) were obtained from Wicell ( RRID:CVCL_9773 ) and cultured and passaged as previously described using conditional media ( CM ) ( Garcia-Perez et al . , 2007 ) . To prepare CM , human foreskin fibroblasts obtained from ATCC ( ATCC Cat# SCRC-1041 , RRID:CVCL_3285 ) were mitotically inactivated with 3000–3200 rads γ-irradiation , seeded at 3 × 106 cells/225 cm2 flask and cultured with hESC media ( KnockOut DMEM supplemented with 4 ng/ml bFGF , 20% KnockOut serum replacement , 1 mM L-glutamine , 0 . 1 mM β-mercaptoethanol and 0 . 1 mM non-essential amino acids ) for at least 24 hr before media harvesting . We collected CM 24 , 48 and 72 hr after seeding . H9 human ESCs ( Wicell , RRID:CVCL_9773 ) were maintained on Matrigel ( BD Biosciences ) -coated plates in human foreskin fibroblast-conditioned media . The absence of Mycoplasma in cultured cells was confirmed once a month using a PCR-based assay ( Minerva ) . Single tandem repeat genotyping was done at least once a year to ensure the identity of the human cell lines used . The identity of parental mouse ESCs was confirmed by generation of chimeric mice and germline transmission , and parental and targeted mouse ESCs were confirmed to contain forty chromosomes by karyotyping . The identity of hamster XR-1 cells was confirmed using an endonuclease-independent retrotransposition assay ( Morrish et al . , 2002 ) . Independent wells , plates or transfections were used as biological replicates . ESCs and HEK293 cell lines stably expressing TEX19 . 1-YFP or YFP alone were generated by transfecting E14Tg2a ESCs or HEK293 cells with linearized pCAG-TEX19 . 1-YFP and pCAG-YFP expression plasmids ( Supplementary file 4 ) containing the CAG promoter for expression ( Niwa et al . , 1991 ) , and selecting for the G418 resistance cassette . Stable cell lines were flow sorted to select for YFP expression . For pCAG-YFP transfection , the cell lines were flow sorted to select for cells expressing YFP at similar levels to the pCAG-TEX19 . 1-YFP cell lines . Stable Flp-In-293 cells ( Invitrogen ) expressing T7-tagged hL1-ORF1p from a CMV promoter at the Flp-In locus were generated using the pcDNA5⁄FRT Flp-In vector , and selected using 100 μg/ml hygromycin and 100 μg/ml Zeocin according to the supplier’s instructions . Tex19 . 1−/− ESCs were generated by sequential targeting of E14Tg2a ESCs . The Tex19 . 1 targeting vector was generated by inserting an IRES-GFP cassette into position chr11:121147942 ( mm10 genome assembly ) in the 3' untranslated region of Tex19 . 1 in a bacterial artificial chromosome ( BAC ) by BAC recombineering ( Liu et al . , 2003 ) . A 13 kb region ( chr11:121143511–121156687 ) containing Tex19 . 1 was gap-repaired into PL253 ( Liu et al . , 2003 ) , then a LoxP site from PL452 was recombined upstream of the coding exon at position chr11:121146376 , and an Frt-flanked neomycin-resistance cassette and second LoxP site from PL451 ( Liu et al . , 2003 ) recombined downstream of the coding exon at chr11:121148877 . E14Tg2a ESCs were electroporated with the resulting targeting vector , selected for neomycin resistance , and correct integrants identified by PCR . The Tex19 . 1 coding exon in the targeted allele was removed by transfection with a Cre-expressing plasmid , and the resulting cells electroporated with the targeting vector again , selected for neomycin resistance , and correct integrants on the second Tex19 . 1 allele identified by PCR . ESCs were then transiently transfected with a Flp-expressing plasmid to generate a conditional Tex19 . 1fl allele . This was subsequently converted to a Tex19 . 1− allele by transient transfection with a Cre-expressing plasmid to remove the Tex19 . 1 coding exon . ESCs were cultured in gelatinized flasks in LIF+serum ( Glasgow Modified Eagle's Media , 10% foetal calf serum , 1% non-essential amino acid , 1% sodium pyruvate , 1% penicillin-streptomycin , 1% L-glutamine , 0 . 001% β-mercaptoethanol , and 0 . 2% leukaemia inhibitory factor-conditioned media ) during the generation of Tex19 . 1−/− ESCs , then low passage Tex19 . 1−/− ESCs with a euploid karyotype were used for experiments after transitioning to 2i culture conditions for at least 14 days . RNA was isolated from cells or tissues using TRIzol reagent ( Life Technologies ) , treated with DNAse ( DNAfree , Ambion ) and used to generate random-primed cDNA ( First Strand cDNA Kit , Life Technologies ) as described by the suppliers . qPCR was performed on the cDNA using the SYBR Green PCR System ( Stratagene ) and a CFX96 Real-Time PCR Detection System ( Bio-Rad ) . Control qRT-PCR reactions were performed in the absence of either reverse transcriptase or qPCR template to verify the specificity of any qRT-PCR signals obtained . Primers were validated to perform at >90% efficiency in the qRT-PCR assay , and expression quantified using the 2-∆∆Ct method ( Livak and Schmittgen , 2001 ) . Alternatively , qPCR was performed using SYBR Select Master Mix ( Applied Biosystems ) and a Light Cycler 480 II ( Roche ) , and expression quantified using the relative standard curve method as described by the suppliers . Sequences of oligonucleotide primers used for qRT-PCR are listed in Supplementary file 3 . Tissue or cells were homogenized in 2× Laemmli SDS sample buffer ( Sigma ) with a motorized pestle , then boiled for 2–5 min and insoluble material pelleted in a microcentrifuge . Protein samples were resolved on pre-cast Bis-Tris polyacrylamide gels in MOPS running buffer ( Invitrogen ) , or Tris-Acetate polyacrylamide gels in Tris-Acetate SDS running buffer ( Invitrogen ) and Western blotted to PVDF membrane using a GENIE blotter ( Idea Scientific ) or the iBlot Transfer system ( Invitrogen ) . Pre-stained molecular weight markers ( Thermo Fisher ) were used to monitor electrophoresis and blotting . Membranes were blocked with 5% non-fat skimmed milk powder in PBST ( PBS , 0 . 1% Tween-20 ) , then incubated with primary antibodies ( Supplementary file 5 ) diluted in blocking solution . Membranes were then washed with PBST and , if required , incubated with peroxidase-conjugated secondary antibody in blocking solution . Membranes were washed in PBST and bound secondary antibodies detected using West Pico Chemiluminescent Substrate ( Thermo Scientific ) . Western blots were quantified using ImageJ ( Schneider et al . , 2012 ) . For simultaneous two-color detection and quantification , proteins were transferred to nitrocellulose membranes , rabbit L1-ORF1p antibodies were used at a 1:1000 dilution and mouse β-actin at 1:2000 , and IRDye-conjugated secondary antibodies ( LI-COR ) detected using an Odyssey imager ( LI-COR ) . Immunostaining on P16 testes was performed by fixing decapsulated P16 testes in 4% paraformaldehyde in PBS , embedding the tissue in paraffin wax , and cutting 6 μm sections on a microtome . Sections were de-waxed in xylene , rehydrated , and antigen retrieval was performed by boiling slides in a microwave for 15 mins in 10 mM sodium citrate pH 6 . Sections were blocked in PBS containing 10% goat serum , 3% BSA , 0 . 1% Tween-20 , then incubated in 1:300 rabbit anti-mL1-ORF1p primary antibody ( Martin and Branciforte , 1993; Soper et al . , 2008 ) diluted in blocking solution . Sections were then washed with PBS , incubated in 1:500 Alexa Fluor-conjugated secondary antibodies ( Life Technologies ) , washed with PBS again , then mounted under a coverslip using Vectashield mounting media containing DAPI ( Vector Laboratories ) . Slides were imaged on a Zeiss Axioplan II fluorescence microscope equipped with a Hamamatsu Orca CCD camera . Anti-mL1-ORF1p fluorescence intensity was measured per unit area , with slides immunostained with non-specific rabbit IgG and secondary antibodies used to calculate and subtract background . Polysome gradients were prepared as described ( Gillian-Daniel et al . , 1998 ) . In brief , P18 testes were homogenized in 200 μl lysis buffer ( 20 mM HEPES pH 7 . 4 , 150 mM KCl , 5 mM DTT , 5 mM MgCl2 , 100 U/mL RNasein , Complete protease inhibitors ( Roche ) , 10 nM calyculin A , 150 μg/mL cycloheximide ) then NP-40 added to 0 . 5% and the samples incubated on ice for 10 min . After centrifugation at 12 , 000 g for 5 min at 4°C the soluble supernatant was layered onto an 11 mL 10–50% linear sucrose prepared in gradient buffer ( 20 mM HEPES pH 7 . 4 , 250 mM KCl , 5 mM DTT , 10 mM MgCl2 , 1 μg/μL heparin ) , then centrifuged in a SW41Ti rotor ( Beckman ) for 120 min at 38 , 000 rpm at 4°C . 1 mL fractions were collected and absorbance of RNA at 254 nm was recorded by using a UV monitor . To isolate RNA , fractions were digested with 20 μg/μL proteinase K in presence of 1% SDS and 10 mM EDTA for 30 min at 37°C then RNAs recovered using Trizol LS reagent ( Invitrogen ) . To isolate proteins , fractions were precipitated with methanol/chloroform and pellets resuspended by boiling in Laemmli SDS sample buffer . P16 testes were homogenized with a motorized pestle in lysis buffer ( 20 mM HEPES pH 7 . 4 , 150 mM KCl , 5 mM DTT , 5 mM MgCl2 ) supplemented with 100 U/mL RNasein , Complete protease inhibitors ( Roche ) and insoluble debris removed by centrifugation ( 12 , 000 g , 5 min at 4°C ) . Oligo ( dT ) -cellulose beads ( Ambion ) were blocked in lysis buffer containing 5% BSA for 1 hr at 4°C , then incubated with lysate for 1 hr at 4°C . Oligo ( dT ) -cellulose beads were washed three times with lysis buffer , and bound proteins eluted by boiling in Laemmli SDS sample buffer and analysed by Western blotting . For competition assays , 200 μg of a 25-mer poly ( A ) oligonucleotide ( Sigma Genosys ) was incubated with the oligo ( dT ) -cellulose beads for 30 min at 4°C before the addition of lysates . Poly ( A ) binding protein PABP1 was used as a positive control ( Burgess et al . , 2011 ) . Cytoplasmic extracts were prepared as described ( Wright et al . , 2006 ) . Briefly , stable YFP or TEX19 . 1-YFP ESCs growin in LIF+serum conditions were resuspended in three volumes buffer A ( 10 mM HEPES pH 7 . 6 , 15 mM KCl , 2 mM MgCl2 , 0 . 1 mM EDTA , 1 mM DTT , 0 . 2 mM PMSF , Complete protease inhibitors ( Roche ) ) and incubated on ice for 30 mins . Cells were lysed in a Dounce homogenizer , one-tenth volume buffer B ( 50 mM HEPES pH 7 . 6 , 1 M KCl , 30 mM MgCl2 , 0 . 1 mM EDTA , 1% NP-40 , 1 mM DTT , 0 . 2 mM PMSF ) added , then the lysate centrifuged twice for 15 min at 3400 g at 4°C to deplete nuclei . Glycerol was added to a final volume of 10% , the extracts centrifuged at 12 , 000 g for 5 min at 4°C , pre-cleared with protein A agarose beads ( Sigma ) then with blocked agarose beads ( Chromotek ) , before incubation with GFP-Trap agarose beads ( ChromoTek Cat# gta-20 RRID:AB_2631357 ) for 90 min at 4°C . Beads were collected by centrifugation at 2700 g for 2 min at 4°C , washed three times with 9:1 buffer A:buffer B , and protein eluted by boiling in 2× Laemmli SDS sample buffer for 3 min . Protein samples were separated on pre-cast Bis-Tris polyacrylamide gels ( Invitrogen ) and stained with Novex colloidal blue staining kit ( Invitrogen ) . Lanes were cut into seven regions according to migration of molecular weight markers and in-gel digestion with trypsin , and mass spectrometry using a 4800 MALDI TOF/TOF Analyser ( ABSciex ) equipped with a Nd:YAG 355 nm laser was performed by St . Andrews University Mass Spectrometry and Proteomics Facility . Mass spectrometry data was analysed using the Mascot search engine ( Matrix Science ) to interrogate the NCBInr database using tolerances of ± 0 . 2 Da for peptide and fragment masses , allowing for one missed trypsin cleavage , fixed cysteine carbamidomethylation and variable methionine oxidation . Superdex 200 10/300 GL ( GE Healthcare Life Sciences ) was calibrated with molecular weight markers for gel filtration ( Sigma-Aldrich ) in BC200 buffer ( 25 mM HEPES pH 7 . 3 , 200 mM NaCl , 1 mM MgCl2 , 0 . 5 mM EGTA , 0 . 1 mM EDTA , 10% glycerol , 1 mM DTT , and 0 . 2 mM PMSF ) . 2 mg cytoplasmic extract from ESCs grown in LIF+serum were diluted in 500 µl buffer A/B ( 15 mM HEPES pH7 . 6 , 115 mM KCl , 3 mM MgCl2 , 0 . 1 mM EDTA , 1 mM DTT , 0 . 2 mM PMSF , Complete protease inhibitors ( Roche ) ) containing 20 µg RNase Inhibitor ( Promega ) , centrifuged ( 12 , 000 g , 10 min at 4°C ) , then loaded on the column . The column was run isocratically in BC200 buffer for 1 . 4 column volumes and 0 . 5 ml fractions were collected . Fractions were precipitated with trichloroacetic acid and resuspended in Laemmli SDS sample buffer . Data shown is representative of two replicates . HEK293T cells were transfected with plasmids ( pCAG-Tex19 . 1-YFP , pCAG-TEX19-YFP , pEGFP3N1-Ubr2 , pCMV5-hORF1-T7 , pCMV5-mORF1-T7 , pCMV5-mORF1-mCherry , pCMV5-hORF1RA-T7 , pCMV5-mORF1RA-T7 , Supplementary file 4 ) using Lipofectamine 2000 ( Invitrogen ) according to the manufacturer's instructions and incubated for 24 hr before harvesting . GFP-Trap agarose beads ( Chromotek ) were used to immunoprecipitate YFP- or GFP-tagged proteins following manufacturer’s instructions . RFP-Trap agarose beads ( ChromoTek Cat# rta-20 RRID:AB_2631362 ) was similarly used to immunoprecipitate mCherry-tagged proteins ( Shaner et al . , 2005 ) , with the addition of a pre-clearing step using binding control agarose beads ( Chromotek ) . The ORF1RA mutants contain two mutations in the RNA binding domain of L1-ORF1p ( R260A and R261A in hL1-ORF1p , R297A and R298A in mL1-ORF1p ) that reduce the ability of L1-ORF1p to bind RNA and form a RNP ( Kulpa and Moran , 2005; Martin et al . , 2005 ) . These mutations abolish the ability of engineered L1 constructs to retrotranspose ( Figure 2—figure supplement 1F ) ( Moran et al . , 1996 ) . For anti-FLAG immunoprecipitation , cell pellets were lysed for 20 min on ice in lysis buffer ( 10 mM Tris pH 7 . 5 , 150 mM NaCl , 0 . 5 mM EDTA , 0 . 5% NP-40 , 1 mM PMSF , Complete Protease Inhibitors ( Roche ) ) , and insoluble material removed by centrifugation at 12 , 000 g for 10 min at 4°C . Supernatants were diluted 1:4 in lysis buffer without NP-40 , then combined with washed anti-FLAG M2 affinity gel ( Sigma-Aldrich Cat# A2220 RRID:AB_10063035 ) , and rotated at 4°C for 1 hr . The anti-FLAG gel was washed three times in lysis buffer without NP-40 , then protein eluted in 2× Laemmli SDS sample buffer for Western blot analysis . For all co-immunoprecipitation data , data shown is representative of at least two replicates . HEK293T cells were cotransfected with equal amounts of the indicated plasmids ( pCMV-TEX19 , pCMV-His6-myc-ubiquitin ( Ward et al . , 1995 ) , and pCMV5-hORF1-T7 , Supplementary file 4 ) using Lipofectamine 2000 ( Invitrogen ) . Cells were harvested 72 hr after transfection and lysed in 6 M guanidinium-HCl , 0 . 1 M Na2HPO4 , 0 . 1 M NaH2PO4 , 0 . 01 M Tris-HCl pH 8 . 0 , 5 mM imidazole and 10 mM β-mercaptoethanol . Following sonication , samples were rotated with washed Ni-NTA agarose ( Qiagen ) at room temperature for 4 hr . The agarose beads were washed as described ( Rodriguez et al . , 1999 ) and ubiquitylated proteins eluted with 200 mM imidazole , 0 . 15 M Tris-HCl pH 6 . 7 , 30% glycerol , 0 . 72 M β-mercaptoethanol and 5% SDS then analysed by Western blotting . Data shown is representative of three replicates . E14Tg2a ES cells were lysed ( 50 mM Tris pH 7 . 5 , 0 . 15 M NaCl , 1 mM EDTA , 1% NP-40 , 10% glycerol , 5 mM N-ethylmaleimide , Complete Protease Inhibitors ( Roche ) ) on ice for 20 min . Cell lysates were centrifuged at 12 , 000 g for 10 min at 4°C and soluble supernatant incubated at 4°C overnight with TUBE2 or control agarose ( LifeSensors ) prepared according to manufacturer’s instructions . Agarose beads were washed three times in 50 mM Tris pH 7 . 4 , 150 mM NaCl , 0 . 1% Tween and protein eluted with 2× Laemmli SDS sample buffer . Data shown is representative of three replicates . For the Strep pull-down assays , hL1-ORF1p and human TEX19 were either co-expressed or separately expressed overnight at 20°C in Escherichia coli BL21 ( DE3 ) Star cells . Expression plasmids ( Diebold et al . , 2011 ) , including a GB1 solubility tag ( Cheng and Patel , 2004 ) for TEX19 , are described in Supplementary file 4 . The cells were lysed in a binding buffer ( 50 mM Hepes pH 7 . 0 , 200 mM NaCl , 2 mM DTT ) containing DNase I , lysozyme and protease inhibitors . For proteins expressed separately , 200 µl of the Strep tagged binding partner ( hL1-ORF1p or GB1 ) were incubated with 50 µl Strep-Tactin Sepharose beads ( IBA ) in a total volume of 1 ml of binding buffer for 45 min at 4°C . After centrifugation ( ~1500 g ) and two washes with 700 µl of binding buffer , 1 ml of TEX19 lysate was added to the beads , followed by an additional incubation for 45 min at 4°C . For co-expressed proteins , 1 ml of the lysate was added to 50 µl Strep-Tactin Sepharose beads ( IBA ) and incubated for 45 min at 4°C . In the end , the beads were washed five times with 700 µl of binding buffer . The bound proteins were eluted with 100 µl of the binding buffer supplemented with 2 . 5 mM biotin . The eluted proteins were then precipitated by trichloroacetic acid , resuspended in 1x SDS-PAGE sample buffer and analyzed by SDS-PAGE . For pull-downs with fragments of hL1-ORF1p and TEX19 , proteins were always co-expressed as described above . Gel loading volumes were adjusted to obtain approximately equal amounts of bait protein in each lane . For the treatment with micrococcal nuclease , co-expressed hL1-ORF1p and TEX19 were lysed in binding buffer ( 50 mM Hepes pH 7 . 0 , 150 mM NaCl , 2 mM DTT ) containing DNase I , lysozyme and protease inhibitors . After centrifugation for 30 min at 14000 g at 4°C , CaCl2 was added to the final concentration of 2 . 5 mM to the lysate . To one half of the lysate micrococcal nuclease was added to the final concentration of 4 × 103 gel units/ml . The lysate was incubated for 15 min at 4°C , then centrifuged for 15 min at 18000 g at 4°C . The supernatant was then added to Strep-Tactin beads ( IBA ) as described above . Under these conditions , 4000 gel units/ml MNase entirely degrades 50 ug/ml oligo ( A ) 27 RNA . Luciferase activity was measured 24 hr post-transfection using the Dual-Luciferase Reporter Assay system ( Promega ) following manufacturer’s instructions and as described previously ( Heras et al . , 2013 ) . We used three different L1 retrotransposition assays in HEK293T , U2OS , HeLa and mouse ESCs . In all retrotransposition assays , we confirmed that overexpression of human TEX19 or mouse Tex19 . 1 is not toxic to cultured HeLa , HEK293T or U2OS cells . Where indicated , transfection efficiency controls were used to calculate rates of engineered retrotransposition as described ( Garcia-Perez et al . , 2010; Kopera et al . , 2016 ) , and engineered L1 retrotransposons were co-transfected with a second expression plasmid for TEX19 orthologs or controls . For mneoI and mblastI-based assays , we included a plasmid containing a neomycin or blasticidin resistance expression cassette respectively , to control for cytotoxicity ( Kopera et al . , 2016; Richardson et al . , 2014b ) when over-expressing TEX19 orthologs . Retrotransposition assays with mneoI or mblastI tagged L1 constructs in cultured HeLa and U2OS cells were performed as described ( Kopera et al . , 2016; Morrish et al . , 2002; Richardson et al . , 2014b; Wei et al . , 2000 ) . L1 constructs used in these assays were derived from active human L1 elements ( Brouha et al . , 2002; Kimberland et al . , 1999; Sassaman et al . , 1997; Moran et al . , 1996 ) , active mouse L1 elements ( Goodier et al . , 2001 ) , or a synthetic codon-optimized mouse L1 element ( Han and Boeke , 2004 ) , and are described in Supplementary file 4 . HeLa cells were transfected with Fugene6 ( Promega ) using 1 μg plasmid DNA per 35 mm diameter well and OptiMEM ( Invitrogen ) according to the manufacturer instructions . 400 μg/ml G418 selection for 12 days was initiated 72 hr post-transfection for mneoI constructs , or 10 μg/ml blasticidin S selection was initiated 120 hr post-transfection for 7 days for mblastI constructs . Drug-resistant foci were then fixed ( 2% formaldehyde , 0 . 2% glutaraldehyde in PBS ) and stained ( 0 . 1% crystal violet ) . Retrotransposition assays with mneoI tagged L1 constructs in mouse ESCs were conducted by plating 4 × 105 cells per 35 mm diameter well onto gelatin-coated tissue culture plates and transfecting 18 hr later with Lipofectamine 2000 ( Invitrogen ) using 1 μg plasmid DNA per well and OptiMEM ( Invitrogen ) according to the manufacturer instructions . Media was replaced after 8 hr and transfected mouse ESCs passaged into a gelatin-coated 100 mm tissue culture plate 24 hr later . 200 μg/ml G418 selection for 12 days was initiated after an additional 24 hr , and drug-resistant foci fixed , stained and counted as described for HeLa cells . Independent transfections were used as biological replicates , and assays using mneoI or mblastI constructs were performed in duplicate to allow clear and consistent effects on retrotransposition rate to be detected . Retrotransposition assays with mEGFPI tagged L1 constructs in cultured HEK293T cells were performed as described ( Goodier et al . , 2013; Wei et al . , 2000 ) . 2 × 105 HEK293T cells were plated in a 35 mm diameter well , then transfected with Lipofectamine 2000 ( Invitrogen ) and 1 μg plasmid DNA per well using OptiMEM ( Invitrogen ) following the manufacturer instructions 20 hr later . After a further 24 hr , fresh media was added and 48 hr later media containing 5 μg/ml puromycin ( Sigma ) was added daily for 7 days to select for transfected cells . Cells were collected by trypsinization and the percentage of EGFP-expressing cells determined using a FACSCanto II flow cytometer ( BD Biosciences ) . Transfection with mutant L1 plasmid ( 99-gfp-JM111 or 99-gp-L1SMmut2 ) allowed a threshold to be established for background fluorescence . Independent transfections were used as biological replicates , and assays using mEGFPI constructs were performed in triplicate to allow clear and consistent effects on retrotransposition rate to be detected . 1 × 105 U2OS cells were plated in 35 mm diameter wells , then 20 hr later transfected with Fugene6 ( Promega ) and 1 μg plasmid DNA per well using OptiMEM ( Invitrogen ) following the manufacturer instructions . Media was replaced 20 hr after transfection and cells allowed to grow for a total of 36 hr . Next , the transfected cells were trypsinized and 25–50% plated on a 15 mm diameter sterile circular polysterene coverslip in a 35 mm diameter well . 12 hr later , cells were fixed with 4% paraformaldehyde at room temperature for 30 min , permeabilized with PBS containing 0 . 1% ( v/v ) Triton X-100 , then incubated with blocking solution ( 10% normal goat serum , 0 . 5% Triton-X-100 in PBS ) for 30 min . After two washes in PBS containing 0 . 1% goat serum and 0 . 05% Triton X-100 , coverslips were incubated with 1:1000 mouse anti-T7 primary antibody ( Millipore Cat# 69522–3 RRID:AB_11211744 ) diluted in PBS containing 1% normal goat serum and 0 . 5% Triton-X-100 at 4°C overnight in a humidified chamber . Coverslips were then washed three times with PBS containing 1% normal goat serum and incubated with 1:1000 Alexa Fluor-conjugated goat anti-mouse secondary antibodies ( Life Technologies ) for 30 min at room temperature . Coverslips were then washed twice and mounted with SlowFade Gold antifade with DAPI ( ThermoFisher ) and sealed with nail polish . Slides were imaged using a Zeiss LSM-710 confocal microscope ( Leica ) , an Axio Imager A1 Microscope ( Zeiss ) and captured images analyzed with ZEN lite software ( Zeiss ) .	Around half of the DNA in a human cell is made of stretches of genetic code that were copied from one part of the human genome and pasted into new locations . Such moveable pieces of genetic code are known as retrotransposons , most which are no longer active . However , one type can still move around: the ‘long interspersed element class 1’ , called LINE-1 for short . There are several hundred thousand copies of LINE-1 in the human genome , and each copy encodes two proteins that work together to insert new copies of LINE-1 into the genome . Each time LINE-1 is pasted into a new location , there is a risk that it will disrupt a gene , creating a mutation . If this happens in the cells that make sperm or eggs – known as germline cells – the mutation can be passed on to the next generation . Human cells have some defence against LINE-1 . They commonly modify the DNA at the start of the LINE-1 genes , which stops the LINE-1 proteins from being made . However , germline cells temporarily remove these DNA modifications at certain stages of development , and previous work in mice suggests that this is when LINE-1 moves . When mouse germline cells remove DNA modifications , they activate a gene called Tex19 . 1 . This led MacLennan , García-Cañadas et al . to ask whether this gene plays a role in regulating LINE-1 activity in germline cells . When mice were genetically engineered to inactivate the Tex19 . 1 gene in developing sperm cells , levels of one of the LINE-1 proteins , called L1-ORF1p , increased . This indicates that Tex19 . 1 most likely acts to keep the levels of this protein down . To find out how Tex19 . 1 does this , a technique called immunoprecipitation was used to pull the the protein encoded by the Tex19 . 1 gene out of mouse cells to see which other proteins came along with it . The interacting proteins included L1-ORF1p and components of a molecular machine that identifies and marks undesired proteins for destruction . Furthermore , the levels of L1-ORF1p in mouse cells increased when this molecular machine ( which is known as the ubiquitin system ) was blocked . This suggests that cells use Tex19 . 1 to keep LINE-1 in check by detecting its proteins and promoting their destruction . The findings reveal that germline cells have another layer of defence that kicks in when DNA modifications are removed during development . In this situation , LINE-1 proteins are detected and destroyed before they can copy and paste the retrotransposon . Since LINE-1 retrotransposons have the potential to cause mutations in around one in every twenty people , if these findings are transferrable to humans , they could open new avenues for research into inherited mutations .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "genetics", "and", "genomics" ]	2017	Mobilization of LINE-1 retrotransposons is restricted by Tex19.1 in mouse embryonic stem cells
High-throughput assays of three-dimensional interactions of chromosomes have shed considerable light on the structure of animal chromatin . Despite this progress , the precise physical nature of observed structures and the forces that govern their establishment remain poorly understood . Here we present high resolution Hi-C data from early Drosophila embryos . We demonstrate that boundaries between topological domains of various sizes map to DNA elements that resemble classical insulator elements: short genomic regions sensitive to DNase digestion that are strongly bound by known insulator proteins and are frequently located between divergent promoters . Further , we show a striking correspondence between these elements and the locations of mapped polytene interband regions . We believe it is likely this relationship between insulators , topological boundaries , and polytene interbands extends across the genome , and we therefore propose a model in which decompaction of boundary-insulator-interband regions drives the organization of interphase chromosomes by creating stable physical separation between adjacent domains . Beginning in the late 19th century , cytological investigations of the polytene chromosomes of insect salivary glands implicated the physical structure of interphase chromosomes in their cellular functions ( Balbiani , 1881; Balbiani , 1890; Heitz and Bauer , 1933; King and Beams , 1934; Painter , 1935 ) . Over the next century plus , studies in the model insect species Drosophila melanogaster were instrumental in defining structural features of animal chromatin . Optical and electron microscopic analysis of fly chromosomes produced groundbreaking insights into the physical nature of genes , transcription and DNA replication ( Benyajati and Worcel , 1976; Laird and Chooi , 1976a; Laird et al . , 1976b; McKnight and Miller , 1976; McKnight and Miller , 1977; Vlassova et al . , 1985; Belyaeva and Zhimulev , 1994 ) . Detailed examination of polytene chromosomes in Drosophila melanogaster revealed a stereotyped organization , with compacted , DNA-rich ‘bands’ alternating with extended , DNA-poor ‘interband’ regions ( Bridges , 1934; Rabinowitz , 1941; Lefevre , 1976; Benyajati and Worcel , 1976; Laird and Chooi , 1976a ) , and it appears likely that this structure reflects general features of chromatin organization shared by non-polytene chromosomes . While these classical studies offered extensive structural and molecular characterization of chromosomes in vivo , the question of what was responsible for organizing chromosome structure remained unanswered . A critical clue came with the discovery of insulators , DNA elements initially identified based on their ability to block the activity of transcriptional enhancers when located between an enhancer and its targeted promoters ( Kellum and Schedl , 1991; Holdridge and Dorsett , 1991; Geyer and Corces , 1992; Kellum and Schedl , 1992 ) . Subsequent work showed that these elements could also block the spread of silenced chromatin states ( Roseman et al . , 1993; Sigrist and Pirrotta , 1997; Mallin et al . , 1998; Recillas-Targa et al . , 2002; Kahn et al . , 2006 ) and influence the structure of chromatin . Through a combination of genetic screens and biochemical purification , a number of protein factors have been identified that bind to Drosophila insulators and modulate their function , including Su ( Hw ) , BEAF-32 , mod ( mdg4 ) , CP190 , dCTCF , GAF , Zw5 , and others ( Lindsley and Grell , 1968; Lewis , 1981; Parkhurst and Corces , 1985; Parkhurst and Corces , 1986; Spana et al . , 1988; Parkhurst et al . , 1988; Zhao et al . , 1995; Gerasimova et al . , 1995; Bell et al . , 1999; Gaszner et al . , 1999; Scott et al . , 1999; Büchner et al . , 2000; Pai et al . , 2004; Melnikova et al . , 2004; Moon et al . , 2005 ) . Except for CTCF , which is found throughout bilateria , all of these proteins appear to be specific to arthropods ( Heger et al . , 2013 ) . Staining of polytene chromosomes with antibodies against such insulator proteins showed that many of them localize to polytene interbands ( Belyaeva and Zhimulev , 1994; Zhao et al . , 1995; Byrd and Corces , 2003; Eggert et al . , 2004; Pai et al . , 2004; Gortchakov et al . , 2005; Gilbert et al . , 2006; Berkaeva et al . , 2009; Vatolina et al . , 2011b ) , with some enriched at interband borders . Further , some , though not all , insulator protein mutants disrupt polytene chromosome structure ( Roy et al . , 2007 ) . Together , these data implicate insulator proteins , and the elements they bind , in the organization of the three-dimensional structure of fly chromosomes . Several high-throughput methods to probe three-dimensional structure of chromatin have been developed in the last decade ( Lieberman-Aiden et al . , 2009; Fullwood et al . , 2009; Rao et al . , 2014; Beagrie et al . , 2017 ) . Principle among these are derivatives of the chromosome conformation capture ( 3C ) assay ( Dekker et al . , 2002 ) , including the genome-wide ‘Hi-C’ ( Lieberman-Aiden et al . , 2009 ) . Several groups have performed Hi-C on Drosophila tissues or cells and have shown that fly chromosomes , like those of other species , are organized into topologically associated domains ( TADs ) , regions within which loci show enriched 3C linkages with each other but depleted linkages with loci outside the domain . Disruption of TAD structures by gene editing in mammalian cells has been shown to disrupt enhancer-promoter interactions and significantly alter transcriptional activity ( Guo et al . , 2015; Lupiáñez et al . , 2015 ) . Although TADs appear to be a common feature of animal genomes , the extent to which TAD structures are a general property of a genome or if they are regulated as a means to control genome function remains unclear , and the question of how TAD structures are established remains largely open . Previous studies have implicated a number of features in the formation of Drosophila TAD boundaries , including transcriptional activity and gene density , and have reached differing conclusions about the role played by insulator protein binding ( Sexton et al . , 2012; Hou et al . , 2012; Van Bortle et al . , 2014; Ulianov et al . , 2016; Li et al . , 2015 ) . Tantalizingly , Eagen et al . , using 15 kb resolution Hi-C data from D . melanogaster have shown that there is a correspondence between the distribution of large TADs and polytene bands ( Eagen et al . , 2015 ) . We have been studying the formation of chromatin structure in the early D . melanogaster embryo because of its potential impact on the establishment of patterned transcription during the initial stages of development . We have previously has shown that regions of ‘open’ chromatin are substantially remodeled at enhancers and promoters during early development ( Harrison et al . , 2011; Li et al . , 2014 ) and were interested in the role three-dimensional chromatin structure plays in spatial patterning . We therefore generated high-resolution Hi-C datasets derived from nuclear cycle 14 Drosophila melanogaster embryos ( Foe and Alberts , 1983 ) , and from the anterior and posterior halves of hand-dissected embryos at the same developmental stage . We show that high-resolution chromatin maps of anterior and posterior halves are nearly identical , suggesting that chromatin structure neither drives nor directly reflects spatially patterned transcriptional activity . However , we show that stable long-range contacts evident in our chromatin maps generally involve known patterning genes , implicating chromatin conformation in transcriptional regulation . To investigate the origins of three-dimensional chromatin structure , we carefully map the locations of the boundaries between topological domains using a combination of manual and computational annotation . We demonstrate that these boundaries resemble classical insulators: short ( 500–2000 bp ) genomic regions that are strongly bound by ( usually multiple ) insulator proteins and are sensitive to DNase digestion . Additionally , we find that boundaries share the molecular features of polytene interband regions . Finally , we show that for a region in which the fine polytene banding pattern has been mapped to genomic positions , boundaries show precise colocalization with interband regions that separate compacted bands corresponding to TADs . We propose that this relationship between insulators , TADs and polytene interbands extends across the genome , and suggest a model in which the decompaction of these regions drives the organization of interphase fly chromosomes by creating stable physical separation between adjacent domains . We prepared and sequenced in situ Hi-C libraries from two biological replicates of hand-sorted cellular blastoderm ( mitotic cycle 14; mid-stage 5 ) embryos using a modestly adapted version of the protocol described in Rao et al . , 2014 . To examine possible links between chromatin maps and transcription , we sectioned hand-sorted mitotic cycle 14 embryos along the anteroposterior midline , and generated Hi-C data from the anterior and posterior halves separately , also in duplicate . In total , we produced ~452 million informative read pairs ( see Supplementary file 1 ) . We assessed the quality of these data using metrics similar to those described by ( Lieberman-Aiden et al . , 2009; Rao et al . , 2014 ) . Specifically , the strand orientations of our reads were approximately equal in each sample ( as expected from correct Hi-C libraries but not background genomic sequence; see Supplementary file 1 ) , the signal decay with genomic distance was similar across samples , and , critically , visual inspection of heat maps prepared at a variety of resolutions showed these samples to be very similar both to each other and to previously published data prepared using similar methods ( Sexton et al . , 2012 ) . We conclude that these Hi-C are of high quality and reproducibility . We next sought to ascertain the general features of the data at low resolution . We examined heatmaps for all D . melanogaster chromosomes together using 100 kb bins , as shown in Figure 1 . Several features of the data are immediately apparent . The prominent ‘X’ patterns for chromosomes 2 and 3 , which indicate an enrichment of linkages between chromosome arms , reflects the known organization of fly chromosomes during early development known as the Rabl configuration ( Csink and Henikoff , 1998; Wilkie et al . , 1999; Duan et al . , 2010 ) : telomeres are located on one side of the nucleus , centromeres are located on the opposite side , and chromosome arms are arranged roughly linearly between them . Centromeres and the predominantly heterochromatic chromosome 4 cluster together , as , to a lesser extent , do telomeres , reflecting established cytological features that have been detected by prior Hi-C analysis ( Sexton et al . , 2012 ) and fluorescence in situ hybridization ( FISH ) ( Lowenstein et al . , 2004 ) . These features were evident in all replicates , further confirming both that these datasets are reproducible and that they capture known features of chromatin topology and nuclear arrangement . Because we used a 4-cutter restriction enzyme and deep sequencing , and because the fly genome is comparatively small , we were able to resolve features at high resolution . We visually inspected genome-wide maps of a number of genomic regions constructed using bins of 500 bp , and were able to see a conspicuous pattern of TADs across a wide range of sizes , some smaller than 5 kb ( Figure 2 , Figure 2—figure supplements 1–5 ) . When we compared maps for several of these regions with available functional genomic data from embryos , we observed that the boundaries between these domains showed a remarkably consistent pattern: they were formed by short regions of DNA ( 500–2000 bp ) that are nearly always associated with high chromatin accessibility , measured by DNase-seq ( Li et al . , 2011 ) , strong occupancy by known insulator proteins as measured by chromatin immunoprecipitation ( ChIP ) ( Nègre et al . , 2010 ) ( Figure 2 , Figure 2—figure supplements 1–5 ) properties characteristic of classical Drosophila insulator elements . To confirm this visually striking association , we systematically called TAD boundaries by visual inspection of panels of raw Hi-C data covering the entire genome . Critically , these boundary calls were made from Hi-C data alone , and the human caller lacked any information about the regions being examined , including which region ( or chromosome ) was represented by a given panel . In total , we manually called 3122 boundaries in the genome for nc14 embryos . Taking into account the ambiguity associated with intrinsically noisy data , the difficulty of resolving small domains , and the invisibility of sections of the genome due to repeat content or a lack of MboI cut sites , we consider 4000–5500 to be a reasonable estimate for the number of boundaries in the genome . To complement these manual calls , we developed a computational approach for calling boundaries that is similar to methods used by other groups ( Lieberman-Aiden et al . , 2009; Sexton et al . , 2012; Rao et al . , 2014; Crane et al . , 2015 ) . In brief , we assigned a directionality score to each genomic bin based on the number of Hi-C reads linking the bin to upstream versus downstream regions , and then used a set of heuristics to identify points of transition between regions of upstream and downstream bias . We adjusted the parameters of the directionality score and the boundary calling to account for features of the fly genome , specifically the relatively small size of many topological domains . Attempts to exhaustively and definitively identify features within genomic data are necessarily variable due to differences in the choice of algorithm , parameters , cutoffs , and unavoidable tradeoffs between sensitivity and accuracy . We therefore sought a representative set of TAD boundaries with which to analyze features of these elements . Of our top 1000 computationally-identified domain boundaries , we found that 952 were matched by a manually-called boundary within 1 kb . This high level of agreement suggested that the computational approach robustly identified the domain features that are apparent by eye . By taking the union of our computational calls , applied with a stringent cutoff , and our manual calls , we developed a very conservative set of exceptionally high confidence boundaries . We emphasize that this set represents only a subset of the boundaries identified by manual and computational approaches , the complete lists of which are provided in Supplementary file 1 . Comparing these 952 boundaries to other genomic datasets confirms our initial observations and reveals a highly stereotyped pattern of associated genomic features . Most strikingly , boundaries are enriched for the binding of the known insulator proteins CP190 , BEAF-32 , mod ( mdg4 ) , dCTCF , and to a lesser extent GAF and Su ( Hw ) ( Figure 3 ) . CP190 and BEAF-32 show the strongest enrichment , and indeed , virtually all ( 95 . 1% ) of the examined boundaries appear to be associated with CP190 binding ( Figure 3—figure supplement 1 ) . Domains of H3K27 trimethylation , a marker of polycomb silencing , showed a strong tendency to terminate at boundaries , and the enhancer mark H3K4me1 showed an interesting pattern of depletion at boundaries but enrichment immediately adjacent to boundary locations ( Figure 3 ) . Boundaries also exhibit peaks of DNase accessibility and nucleosome depletion ( Figure 3 ) , as well as marks associated with promoters , including the general transcription factors TFIIB and the histone tail modification H3K4me3 . Despite the presence of promoter marks , we find that RNA polII is present at only a subset ( 45 . 1% ) of stage 5 boundaries ( Figure 3 , Figure 3—figure supplement 1 ) . It is striking that we observe that not only are sites of combinatorial insulator protein binding enriched at TAD boundaries , but they are highly predictive . Of our representative set of boundaries , 95 . 1% are are enriched >2 fold for CP190 binding within a 1 . 5 kb window . Conversely , of the strongest 1000 CP190 peaks , 75 . 2% are within 2 kb of a manual or computationally-called boundary ( compared to 37 . 4% of the top 1000 RNAPII peaks ) . It is important to note that we do identify a small subset of boundaries that are not apparently associated with sites of insulator binding ( ~1–2% show no enrichment for CP190 , BEAF-32 , or dCTCF , depending on thresholds used ) , suggesting that there are multiple phenomena that can create topological boundaries in flies ( e . g . , see Figure 6 ) . However , the overwhelming majority of topological boundaries identified in this study coincide with elements that match the properties of CP190-associated insulators . An important confounding factor in sorting out the nature of topological boundaries is the strong tendency , observed by multiple authors , of insulator proteins to bind near promoters specifically between divergent promoters ( Nègre et al . , 2010; Ramirez et al . , 2017; Schwartz et al . , 2012 ) . Indeed , we find that boundary elements , as identified from Hi-C , are often found proximal to promoters and show a general enrichment of promoter-associated marks ( Figure 3 ) , raising the possibility that transcriptional activity at promoters may drive topological boundary formation . However , a number of features of the data argue against this possibility . First and most critically , many of the topological boundaries ( 54 . 9% ) we identify are not associated with RNAPII binding in nc14 embryos . Similarly , there are many active promoters that do not appear to form topological boundaries ( e . g . , Figure 8 and supplements ) . Hug et al . , 2017 pharmacologically inhibited transcription in early embryos and observe that TADs remain intact . Finally , topological boundaries are invariant between anterior and posterior sections of embryos despite substantial differences in the transcriptional profiles of these regions ( see below ) . We further examined the distributions of the same genomic features around the top 1000 peaks of H3K4me3 , a marker of active promoters , in data from stage 5 embryos ( Figure 3—figure supplement 2 ) ( Li et al . , 2014 ) . While these sites show enrichments for insulator proteins , these enrichments are substantially weaker than those observed at topological boundaries , while RNA polII enrichment is much stronger at promoters than boundaries . The tendency for polycomb domains to terminate at promoters is also much less pronounced at promoters than boundaries . Together , these data argue that boundaries constitute a distinct class of genetic elements that are not formed by promoter transcription , but are instead frequently located near promoters , possibly as a result of selective pressure to insulate these proximal promoters from distal regulatory elements . While we cannot rule out any role for promoter-bound transcription machinery in the formation of topological boundaries ( notably , TFIIB is enriched at 69 . 1% of boundaries ) , we think it is unlikely that transcriptional activity plays a major role in establishing the topological domains of interphase fly chromosomes . Finally , we examined the sequence composition of boundary elements by comparing the frequency of DNA words of up to seven base pairs in the set of high confidence boundaries to flanking sequence . The most enriched sequences correspond to the known binding site of BEAF-32 and to a CACA-rich motif previously identified as enriched in regions bound by CP190 ( Nègre et al . , 2010; Yang and Corces , 2012 ) , both of which show strong association with the set of boundary sequences as a whole ( Figure 4 ) . The examination of these boundary elements led us to consider the physical basis of topological domain separation . Chromosome conformation capture is a complex assay ( Gavrilov et al . , 2013; Gavrilov et al . , 2015 ) , and inferring discrete physical states of the chromatin fiber from Hi-C signals generally requires orthogonal experimental data . To address this problem , we sought to leverage information from polytene chromosomes to draw associations between features of Hi-C data and physical features of chromosomes . The Zhimulev laboratory has extensively studied the nature and composition of polytene bands and interbands for decades . Using a combination of approaches , they have identified interbands as a set of ~5700 short decompacted regions that tend to be located near divergent promoters and are characterized by DNase hypersensitivity and the binding of characteristic proteins , including insulator proteins ( Zhimulev et al . , 2014 ) . It was immediately apparent to us that these elements bore significant similarity to the topological boundary elements we identified . We thus sought to compare our Hi-C data to known polytene chromosome structures . There is surprisingly little data mapping features of polytene structure to specific genomic coordinates at high resolution . Vatolina et al . , 2011a used exquisitely careful electron microscopy to identify the fine banding pattern of the 65 kb region between polytene bands 10A1-2 and 10B1-2 , revealing that this region , which appears as a single interband under a light microscope , actually contains six discrete , faint bands and seven interbands . The region is flanked by two large bands , whose genomic locations have been previously mapped and refined by FISH ( Vatolina et al . , 2011a ) . Vatolina et al . then used available molecular genomic data to propose a fine mapping of these bands and interbands to genomic coordinates . Figure 5 shows the correspondence between Vatolina et al . ’s proposed polytene map from this region and our high-resolution Hi-C data , along with measures of early embryonic DNase hypersensitivity from ( Li et al . , 2011 ) and the binding of six insulator proteins ( Nègre et al . , 2010 ) . There is a striking correspondence between the assignments of Vatolina et al . and our Hi-C data: faint polytene bands correspond to TADs , and interbands correspond to the boundary elements that separate the TADs . This correspondence is not perfect . Specifically , the evidence in our Hi-C data for the separation between the major band 10A1-2 and the minor band 10A3 is weak , though that may be partly explained by the absence of MboI cut sites obscuring much of this region . This minor band is barely detectable in polytene spreads , and the combination of this and weak support in Hi-C data may suggest that this band is not real or is perhaps only present in a minority of nuclei . Similarly , the Vatolina et al . report that they only rarely observe the interband between bands 10A6 and 10A7 , and we indeed observe substantial contact between these two putative bands in Hi-C maps ( the light orange region near the peak of the ‘pyramid’ formed by 10A6 and 10A7 in Figure 5 ) , though each shows stronger intra- than inter-domain interactions . One possible explanation for this observation is that the interband separating these two domains is not constitutive but rather is formed in only a fraction of nuclei . The pattern exhibited by these two domains--adjacent domains that show a clear separation but also a substantial interaction signal--is one we observe frequently in our early embryonic Hi-C data , suggesting that variable boundaries may be common features of the fly genome . Overall , the alignment between polytene band mapping and Hi-C data in this region supports a strong correspondence between these two types of data . For five interbands which were easily visible in polytene spreads ( 10A3/4-5 , 10A4-5/10A6 , 10A7/10A8-9 , 10A8-9/10A10 , 10A10-11/10B1 ) , we observe strong domain boundaries in Hi-C data . For two interbands supported by weaker evidence in polytene analysis , we observe in Hi-C maps a weak or non-existent boundary ( 10A1-2/10A3 ) and a boundary with significant interaction across it , possibly representing heterogeneity between nuclei matching heterogeneity in polytenes ( 10A6/10A7 ) . The 5’ region of the Notch gene has also been carefully mapped . Rykowski et al . used high-resolution in situ hybridization to determine that the coding sequences of Notch lies within polytene band 3C7 , while the sequences upstream of the transcription start site ( TSS ) lie in the 3C6-7 interband . Examining the Notch locus in our Hi-C data , we see that the gene body is located within an ~20 kb TAD , and the TSS directly abuts a TAD boundary that is strongly bound by CP190 and dCTCF ( Figure 5—figure supplement 1 ) , an arrangement consistent with the correspondence of boundaries and interbands . The chromodomain-containing protein Chriz has been suggested as the strongest diagnostic feature of polytene interbands ( Zhimulev et al . , 2014 ) . Using publicly available ChIP datasets from Kc167 cells ( derived from late embryonic tissue ) , we observed a strong enrichment of Chriz binding at our representative boundaries ( 87 . 9% >2 fold enriched within 1 . 5 kb , Figure 5—figure supplement 2A ) . Further , Hi-C directionality around Chriz peaks shows the characteristic pattern of boundary formation , and Chriz profiles at representative loci show substantial correspondence between boundary regions and Chriz binding ( Figure 5—figure supplement 2B and C ) , offering further support for the association between boundaries and interbands . Eagen et al . previously identified a broad correspondence between polytene interbands and inter-TAD regions from Hi-C data at 15 kb resolution ( Eagen et al . , 2015 ) . Our Hi-C data allows the detection of fine structure within these inter-TAD regions , down to individual boundary elements . Owing to the dearth of finely mapped polytene regions , the association between topological boundaries and interband regions is necessarily based on a limited number of example loci . However , the combination of data from these loci with the close agreement of the molecular composition of these regions , specifically the strong localization of the interband marker Chriz to topological boundaries , leads us to propose that the precise relationship between topological boundaries , insulator elements , and decompacted interband regions we observe is a general one , and that it extends across the genome . The association between boundary elements and interbands suggests a simple model for insulator function . A key feature that distinguishes polytene interbands from bands is their low compaction ratio: they span a larger physical distance per base pair . The association between insulator binding and genomic regions with low compaction ratios suggests insulators may function by simply increasing the physical distance between adjacent domains via the unpacking and extension of intervening chromatin . Figure 5 ( top ) shows a representation of the conversion of genomic distance to physical distance for the 10A1-B1 region , as measured by Vatolina et al . Any model for insulator function must explain several features of insulator function , including the ability to organize chromatin into physical domains , block interactions between enhancers and promoters exclusively when inserted between them , protect transgenes from position effect variegation and block the spread of chromatin silencing states . This chromatin extension model for fly insulator function can potentially explain these defining characteristics via simple physical separation . We reasoned that if our Hi-C data is capable of resolving fine banding patterns such as that at the 10A1-B1 locus , we should be able to resolve the borders of major bands with precision . We focused on a region of chromosome 2L that had previously been shown by Eagen et al . to appear as a single ~500 kb TAD using Hi-C at 15 kb resolution , but contains a faint interband in Bridge’s map . Our Hi-C data reveal an intricate structure at this locus ( Figure 6A ) . There are two large TADs on the left and right , divided by a series of smaller domains in the center . We suspect that this middle region accounts for the interband in Bridge’s map , in a manner similar to the 10A1-1/10B1-2 region: a complex region consisting of several minor bands bounded by decompacted boundary regions appears as a single interband region under optical microscopy . This region provides examples of a number of interesting features that we observe in our Hi-C data . First , the large TADs are bounded on both sides by gene-rich regions consisting of a number of smaller topological domains ( Figure 6B , D ) . The boundaries of large and small domains in this region nearly all share the common features of boundary elements: DNase hypersensitivity and binding of diagnostic insulator ( e . g . CP190 ) and interband ( CHRIZ ) proteins . This region also contains a prominent example of an exception to this pattern: a loop is formed that appears to generate boundaries not associated with these characteristic protein binding events ( Figure 6C , indicated by dotted yellow lines and loop ) . This example highlights a critical point: while the description we provide of the association between TAD boundaries , insulator elements , and decompacted interbands appears to describe the overwhelming majority of cases , there are counter-examples . Indeed given the extraordinary capacity of nature to innovate with respect to gene regulation and structures , we expect that animal genomes will provide no shortage novel chromosome topological and structural features for future investigations . We next asked whether the boundaries we identified as boundary elements represent constitutive features of chromatin organization or whether their function might be regulated in a cell-type specific or developmental manner . We reasoned that , since different sets of patterning genes are transcribed in the anterior and posterior portions of the pre-gastrula D . melanogaster embryos , a comparison of chromatin interaction maps between anterior and posterior regions would reveal whether context , especially transcriptional state , affects the TAD/boundary structure of the genome . To this end , we performed two separate biological replicates of an experiment in which we sectioned several hundred mid stage 5 embryos along the anteroposterior midline , and produced deep-sequenced Hi-C libraries from the anterior and posterior halves in parallel . Resulting Hi-C signals at boundaries are virtually identical in the two halves , despite substantially different gene expression profiles in these two embryonic regions ( Figure 7A ) . Indeed , overall Hi-C signals are remarkably similar , with anterior and posterior samples correlating as strongly as replicates . Examination of individual loci at high resolution reveal consistent profiles and boundaries , notably including genes expressed differentially in the anterior or posterior ( Figure 7B ) . The correspondence of insulator boundary elements and interbands , and the chromatin extension model , implies that the chromatin accessibility of insulator regions will be a useful proxy for their functionality in structurally organizing the genome . Intriguingly , ( Van Bortle et al . , 2014 ) found that DNase accessibility of insulator protein-bound regions tracked with the ability of these sequences to block enhancer-promoter interactions in a cell-culture assay . We again sectioned embryos into anterior and posterior halves and performed ATAC-seq ( Buenrostro et al . , 2013 ) on pools of 20 embryo halves . ATAC-seq is a technique in which intact chromatin is treated with Tn5 transposase loaded with designed DNA sequences which are preferentially inserted into open , accessible chromatin regions . These insertions can be used to generate high-throughput sequencing libraries , producing data that is largely analogous to DNase-seq data . Analysis of ATAC-seq signal at insulator boundary elements in anterior and posterior halves showed that these elements have nearly identical accessibility in these two samples ( Figure 7C ) . Additionally , DNase-seq data from later embryonic stages that feature substantial tissue differentiation , transcription , and chromatin changes show highly consistent profiles at boundaries ( Figure 7C , Figure 7—figure supplement 1 ) . It is also striking that we observe significant enrichment of insulator proteins and Chriz at boundaries , despite the fact that boundaries were identified from Hi-C data from carefully-staged nc14 embryos ( 2–3 hr ) , whereas these ChIP datasets are derived from 0 to 12 hr old embryos or late embryonic cultured cells ( Chriz ) . Together , these results are consistent with a model in which insulator-mediated chromatin organization is a constitutive feature of interphase chromatin of D . melanogaster embryos . Many models of insulator function invoke physical contact between insulators to form ‘looped’ chromatin domains ( Fujioka et al . , 2009; Yang and Corces , 2012; Kyrchanova and Georgiev , 2014; Kravchenko et al . , 2005 ) , and a substantial literature exists demonstrating that many insulator proteins are able to interact with each other and to self-associate ( Büchner et al . , 2000; Gause et al . , 2001; Ghosh et al . , 2001; Blanton et al . , 2003; Pai et al . , 2004; Mohan et al . , 2007; Golovnin et al . , 2007; Vogelmann et al . , 2014 ) . In general , we do not observe looping interactions between domain boundaries in our Hi-C data . However , during manual calling of topological boundaries for the entire genome , we noted 46 prominent examples of interactions between non-adjacent domains ( Figure 8 and Figure 8—figure supplements 1–10 , Supplementary file 1 ) , in addition to the previously noted clustering of PcG-regulated Hox gene clusters ( Sexton et al . , 2012 ) . Because the interactions we observed were not of a uniform character , we did not attempt to computationally search for all such phenomena in our data , nor do we claim that this list is necessarily complete . It is merely the union of two sets of ‘interesting’ loci identified in two independent rounds of visual inspection Hi-C maps for the entire genome , and we feel it is informative with respect to the nature and significance of distal interaction in the fly embryo . The most visually striking locus , which we emphasize was identified in an unbiased manner without knowing its identify , is the locus containing the Scr , ftz , and Antp genes ( Figure 8A ) . This locus has been extensively studied , and a number of regulatory elements have been identified that reside between the ftz and Antp genes but ‘skip’ the ftz promoter to regulate Scr ( Calhoun et al . , 2002; Calhoun and Levine , 2003 ) . Consistent with this , we observe enriched contacts between the region containing the Scr promoter and a domain on the other side of ftz that contains the known Scr-targeting cis regulatory elements , while the ftz-containing domain makes minimal contact with its neighboring domains . Critically , we observe hot spots of apparent interaction between two sets of boundary elements ( Figure 8A: 1 and 4 , 2 and 3 ) , suggesting that physical association of boundary elements ( or their associated proteins ) may play a role in this interaction . Curiously , we detected a similar situation on the other side of Scr , where a domain containing the hox gene Dfd is ‘skipped’ over by the Ama locus to interact with a short element 3’ of the Scr transcription unit ( Figure 8—figure supplement 1 ) . We also observe a similar arrangement near the eve locus ( Figure 8—figure supplement 2 ) . In these cases , a plausible topology is that the skipped domain is ‘looped out’ , preventing interaction with neighbors , while the adjacent domains are brought into proximity . In addition to these domain-skipping events , we observe a small number of looping interactions , where two distal loci show high levels of interaction , without the associated enriched interactions between the domains flanking the loop . In every case we observe , the loop forms between two domain boundaries . As shown in Figure 8B , one of these loops brings together the promoters of knirps and the related knrl ( knirps-like ) genes . Other loops connect the achaete and scute genes ( Figure 8—figure supplement 3 ) , sloppy paired 1 and sloppy paired 2 ( Figure 8—figure supplement 4 ) , and the promoter of Ultrabithorax with an element in its first intron ( Figure 8—figure supplement 5 ) . These loci demonstrate that looping and domain-skipping events can be detected in our Hi-C data , but it appears that such interactions are rare and that looping does not occur between the overwhelming majority of insulator boundary elements . Nevertheless , it is striking that of the limited number of distal interactions we observed , many of them involve genes that are transcriptionally active during stage 5 of embryogenesis . This raises the possibility that these interactions may be stage or tissue-specific regulatory phenomena , and that more may be present in other tissues , developmental time points , or conditions . The data presented here offer a picture of the structure of the interphase chromatin of Drosophila that attempts to unify years of studies of polytene chromosomes with modern genomic methods ( Figure 9 ) . In this picture , interphase chromatin consists of alternating stretches of compacted , folded chromatin domains separated by regions of decompacted , stretched regions . The compacted regions vary in size from a few to hundreds of kilobases and correspond to both polytene band regions and TADs in Hi-C data . Decompacted regions that separate these domains are short DNA elements that are defined by the strong binding of insulator proteins and correspond to polytene interbands and TAD boundaries ( insulators ) . An intuitive view of this structure in a non-polytene context might resemble the well-worn ‘beads on a string’ , in which insulator/interband regions are the string and bands/TADs form beads of various sizes . Future work , including experimental manipulation of the sequences underlying these structures , will focus on validating and refining this model , exploring how it fits into hierarchical levels of genome organization , and understanding its implications for genome function . OregonR strain D . melanogaster ( RRID:FlyBase_FBst1000080 ) embryos were collected on molasses plates seeded with fresh yeast paste from a population cage and aged to appropriate developmental stages , all at 25°C . Embryos were washed into nitex meshes , dechorionated by treatment with dilute bleach for 2 min , dipped briefly ( 15–20 s ) in isopropanol , and gently rocked in fixative solution of ( 76 . 5% hexanes , 5% formaldehyde in 1x PBS ) for 28–30 min . Embryos were then thoroughly washed in PBS with 0 . 5% triton and stored for no more than fivethree days at 4°C . For sample HiC-2/4 , embryos were inspected under a light microscope to confirm that the vast majority corresponded to early cellularized blastoderm , and approximately 4000 embryos were used in the Hi-C protocol . For samples HiC-10 , 12 , 13–16 , fixed embryos were hand-sorted under a light microscope as described in ( Harrison et al . , 2011 ) , using morphological markers to identify early cellularized embryos ( nc14 , stage 5 ) . For whole embryo experiments , sorted embryos were placed directly into the Hi-C protocol , with no more than 3 days having elapsed since fixation . For sectioned embryos , hand-sorted embryos of precise developmental stages were first arranged in rows on a block of 1% agarose with bromophenol blue in a shared anterior-posterior orientation , with between 20–40 embryos per block . Aligned embryos were then transferred to the bottom of a plastic embedding mold ( Sigma Aldrich E6032 ) , the bottom of which had previously been coated with hexane glue , carefully keeping track of the anterior-posterior orientation of embryos by marking the cup with marker . Embryos were covered with clear frozen section compound ( VWR 95057–838 ) and frozen at −80°C for up to two months . Frozen blocks wer4 ) e retrieved from the freezer and embryos rapidly sliced at approximately the mid-point by hand using a standard razor blade under a dissecting microscope . Anterior and posterior halves were separately transferred to microcentrifuge tubes containing ~200 µL PBS with 0 . 5% triton using an embryo pick ( a tool of mysterious provenance that appears to be a clay sculpting tool ) . Successful transfer was confirmed visually by the presence of blue embryos which had absorbed bromophenol blue from the agarose block . Between transferring anterior and posterior halves , the pick was washed thoroughly with water and ethanol , and rubbed vigorously with kimwipes . We note that anterior and posterior half samples are precisely matched: samples HiC-13 and 14 contain the anterior and posterior halves ( respectively ) of the same embryos , and the same is true for HiC-15 and 16 . No explicit statistical method was used to compute sample size . All unique experiments were prepared in duplicate .	The DNA inside a cell is packaged into threaded structures called chromosomes . Early studies of chromosomes using insect larvae revealed a pattern of dark- and light-colored bands on the chromosomes that was unique to every region . For decades , it remained unclear if the bands had a specific role . More advanced techniques have shown that chromosomes are organized into a series of compact domains that contain separate regions of genes . Each region can be turned on or off at different times , depending on the needs of different cells . This allows cells to specialize into different cell types and tissues . Until now , it was unclear how these different regions are formed and controlled , and how they relate to the chromosome bands . Here , Stadler , Haines and Eisen used a specific technique to map the structure of chromosomes in early fly embryos , by using a chemical trap to capture closely located DNA pieces . The results showed that the chromosome domains corresponded to the banding patterns seen in the early studies . This suggest that light bands represent extended DNA that act as spacers between the dark gene regions . This study adds to the view that the way the DNA is organized influences gene activity . Creating a high-resolution model of the chromosomes will help us to better understand how their structure can influence the activity of genes . In future , scientists may be able to identify diseases that are caused by errors in the chromosome organization .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2017	Convergence of topological domain boundaries, insulators, and polytene interbands revealed by high-resolution mapping of chromatin contacts in the early Drosophila melanogaster embryo
The Parkinson’s disease ( PD ) -associated gene leucine-rich repeat kinase 2 ( LRRK2 ) has been studied extensively in the brain . However , several studies have established that mutations in LRRK2 confer susceptibility to mycobacterial infection , suggesting LRRK2 also controls immunity . We demonstrate that loss of LRRK2 in macrophages induces elevated basal levels of type I interferon ( IFN ) and interferon stimulated genes ( ISGs ) and causes blunted interferon responses to mycobacterial pathogens and cytosolic nucleic acid agonists . Altered innate immune gene expression in Lrrk2 knockout ( KO ) macrophages is driven by a combination of mitochondrial stresses , including oxidative stress from low levels of purine metabolites and DRP1-dependent mitochondrial fragmentation . Together , these defects promote mtDNA leakage into the cytosol and chronic cGAS engagement . While Lrrk2 KO mice can control Mycobacterium tuberculosis ( Mtb ) replication , they have exacerbated inflammation and lower ISG expression in the lungs . These results demonstrate previously unappreciated consequences of LRRK2-dependent mitochondrial defects in controlling innate immune outcomes . Mutations in leucine-rich repeat kinase 2 ( LRRK2 ) are a major cause of familial and sporadic Parkinson’s disease ( PD ) , a neurodegenerative disease characterized by selective loss of dopaminergic neurons in the substantia nigra pars compacta region of the midbrain ( Cookson , 2017; Kim and Alcalay , 2017; Martin et al . , 2014; Schulz et al . , 2016 ) . Despite LRRK2 having been implicated in a variety of cellular processes , including cytoskeletal dynamics ( Civiero et al . , 2018; Kett et al . , 2012; Pellegrini et al . , 2017 ) , vesicular trafficking ( Herbst and Gutierrez , 2019; Sanna et al . , 2012; Shi et al . , 2017 ) , calcium signaling ( Bedford et al . , 2016; Calì et al . , 2014 ) , and mitochondrial function ( Ryan et al . , 2015; Singh et al . , 2019; Yue et al . , 2015 ) , its precise mechanistic contributions to triggering and/or exacerbating PD and other disease pathologies are not known . Of all the cellular pathways affected by LRRK2 mutations , dysregulation of mitochondrial homeostasis has emerged as a centrally important mechanism underlying PD pathogenesis and neuronal loss ( Cowan et al . , 2019; Panchal and Tiwari , 2019 ) . Indeed , other PD-associated genes , such as PARK2 ( Parkin ) , PINK1 , and DJ1 , all play crucial roles in mitochondrial quality control via mitophagy . LRRK2 has been implicated in mitophagy directly through interactions with the mitochondrial outer membrane protein MIRO ( Hsieh et al . , 2016 ) , and several lines of evidence support roles for LRRK2 in controlling mitochondrial network dynamics through interactions with the mitochondrial fission protein DRP1 ( Wang et al . , 2012 ) . Accordingly , a number of different cell types , including fibroblasts and iPSC-derived neurons from PD patients harboring mutations in LRRK2 exhibit defects in mitochondrial network integrity as well as increased reactive oxygen species ( ROS ) and oxidative stress ( Sison et al . , 2018; Smith et al . , 2016 ) . In spite of these well-appreciated links , LRRK2’s contribution to mitochondrial health in cells outside of the brain remains vastly understudied . There is mounting evidence that mutations in LRRK2 , as well as in other genes related to PD including PARK2 and PINK1 , contribute to immune outcomes both in the brain and in the periphery . For example , mutations in LRRK2 impair NF-κB signaling pathways in iPSC-derived neurons and render rats prone to progressive neuroinflammation in response to peripheral innate immune triggers ( López de Maturana et al . , 2016 ) , and chemical inhibition of LRRK2 attenuates inflammatory responses in microglia ex vivo ( Moehle et al . , 2012 ) . In addition to these strong connections between LRRK2 and inflammatory responses in the brain , numerous genome-wide association studies suggest that LRRK2 is an equally important player in peripheral immune responses . Single nucleotide polymorphisms ( SNPs ) in LRRK2 are associated with susceptibility to mycobacterial infection , inflammatory colitis ( Umeno et al . , 2011 ) , and Crohn’s disease ( Van Limbergen et al . , 2009 ) . Consistent with a role for LRRK2 in pathogen defense and autoimmunity , it is abundant in many immune cells ( e . g . B cells , dendritic cells , monocytes , macrophages ) , and expression of LRRK2 is induced in human macrophages treated with IFN-γ ( Gardet et al . , 2010 ) . Loss of LRRK2 reduces IL-1β secretion in response to Salmonella enterica infection in macrophages ( Liu et al . , 2017 ) and enhances expression of pro-inflammatory cytokines in response to Mycobacterium tuberculosis ( Mtb ) infection at early time points of mouse infection ( Härtlova et al . , 2018 ) . However , the precise mechanistic contributions of LRRK2 to controlling immune responses in the periphery remain poorly understood . Here , we provide evidence that LRRK2’s ability to influence inflammatory gene expression in macrophages is directly linked to its role in maintaining mitochondrial homeostasis . Specifically , we demonstrate that mitochondrial stress and hyper-activation of DRP1 in Lrrk2 KO macrophages leads to the release of mitochondrial DNA ( mtDNA ) , chronic engagement of the cGAS-dependent DNA sensing pathway , and abnormally elevated basal levels of type I IFN and ISGs . These high basal levels of type I IFN appear to completely reprogram Lrrk2 KO macrophages , rendering them refractory to a number of distinct innate immune stimuli , including infection with Mtb . While Mtb-infected Lrrk2 KO mice did not exhibit significant differences in bacterial burdens compared to controls , we did observe exacerbated pathology and lower expression of ISGs in the lungs at early infection timepoints . Collectively , these results demonstrate that LRRK2’s role in maintaining mitochondrial homeostasis is critical for proper induction of type I IFN gene expression in macrophages and for downstream inflammatory responses during in vivo infection . To begin to implicate LRRK2 in the peripheral immune response , we took an unbiased approach and asked how loss of LRRK2 impacts innate immune gene expression during Mtb infection of macrophages ex vivo . Briefly , primary murine bone marrow-derived macrophages ( BMDMs ) derived from littermate heterozygous ( HET ) and knockout ( KO ) Lrrk2 mice were infected with Mtb at MOI of 10 . RNA-seq analysis was performed on total RNA collected from uninfected and infected cells 4 hr post-infection ( Lrrk2 KO n = 4 , Lrrk2 HET n = 3 ) . Previous studies have identified 4 hr as a key innate immune time point during Mtb infection , corresponding to the peak of transcriptional activation downstream of several pattern recognition receptors ( PRRs ) , including the cytosolic DNA sensor cGAS ( Manzanillo et al . , 2012; Watson et al . , 2015; Watson et al . , 2012 ) . Following analysis with CLC Genomics Workbench , we first asked whether we could detect gene expression differences in uninfected Lrrk2 HET and KO macrophages . Surprisingly , we identified hundreds of genes whose expression was significantly higher in Lrrk2 KO macrophages ( blue genes , Figure 1A ) . Taking a closer look at the most affected genes ( zoom-in , right ) , we noted that a number of well-characterized ISGs ( e . g . Mx1 , Ifit1 , Irf7 , Rsad2 , etc . ) were expressed several times higher in macrophages lackingLRRK2 ( p<0 . 05 ) . These trends persisted when we compared Lrrk2 WT vs . KO or Lrrk2 HET vs . KO ( Figure 1—figure supplement 1B-C ) . Unbiased canonical pathway analysis confirmed a global upregulation of ISGs , identifying ‘Interferon signaling’ and ‘Activation of IRF by cytosolic PRRs’ as the top enriched pathways in uninfected Lrrk2 KO vs . HET BMDMs ( Figure 1B ) . We next looked at gene expression differences in Lrrk2 KO vs . HET BMDMs at 4 hr post-infection with Mtb . Mtb is a potent activator of type I IFN expression , thought to occur mostly through perturbation of the Mtb-containing phagosome and release of bacterial dsDNA into the cytosol , where it is detected by DNA sensors like cGAS , activating the STING/TBK1/IRF3 axis ( Collins et al . , 2015; Wassermann et al . , 2015; Watson et al . , 2015; Wiens and Ernst , 2016 ) . Curiously , many of the same ISGs whose expression was statistically higher at baseline in Lrrk2 KO BMDMs failed to induce to the same levels following Mtb infection ( e . g . Ifit , Cmpk2 , Gbp2 , Rsad2; Figure 1C , orange genes , zoom-in , left ) . This blunted global type I IFN response was also evident via qualitative assessment of genes whose expression was measurably lower in Mtb-infected Lrrk2 KO BMDMs but failed to demonstrate statistical significance ( Figure 1—figure supplement 1A ) . Again , canonical pathway analysis identified an enrichment for immune genes whose expression was impacted by loss of LRRK2 in response to Mtb ( Figure 1D ) . RT-qPCR analysis confirmed higher baseline expression and lower induction during Mtb infection of several ISGs: Rsad2 , Gbp2 , Cmpk2 , Stat2 , and Ifit1 in Lrrk2 KO BMDMs ( Figure 1E; see ‘Statistical analysis’ section in Materials and methods for details regarding the statistical analysis of baseline and induced gene expression ) . We also measured high basal levels of Ifnb and Isg15 , although the differences in induction of these genes between Lrrk2 KO and HET macrophages were more modest in this particular experiment ( Figure 1E ) . Increased basal expression and decreased induction of IFN and ISGs was also detected during Mtb infection in the human monocyte cell line U937 ( Figure 1—figure supplement 1D ) and in RAW 264 . 7 murine macrophages when Lrrk2 expression was knocked down by shRNA ( Lrrk2 KD ) ( Figure 1—figure supplement 1E ) . Importantly , blunted expression was not observed for all immune genes; for example , loss of Lrrk2 had no effect on the NFκB gene Tnfa despite the transcript being dramatically induced upon Mtb infection ( Figure 1F ) . Interestingly , expression of several non-ISG , non-immune genes was reduced in uninfected Lrrk2 KO BMDMs , including ApoE , which has been repeatedly linked to inflammatory and neurodegenerative diseases , and Ldhb , a critical metabolic gene involved in post-glycolytic energy production ( Figure 1G ) . Collectively , these transcriptome-focused analyses revealed that Lrrk2 KO macrophages have a high baseline IFN signature but generally fail to induce the type I IFN response to the same level as control cells when infected with Mtb . This phenotype is unusual and suggests that Lrrk2 KO macrophages are somehow fundamentally reprogrammed . Typically , high resting IFN levels potentiate type I IFN responses , leading to a hyperinduction of ISGs following innate immune stimuli ( West et al . , 2015; Yang et al . , 2018 ) . We next wanted to define the nature of the innate immune stimuli that would elicit a blunted type I IFN response in Lrrk2 KO macrophages . We began by infecting macrophages with Mycobacterium leprae ( Mlep ) , which shares a virulence-associated ESX-1 secretion system with Mtb and also induces type I IFN through cytosolic nucleic acid sensing ( de Toledo-Pinto et al . , 2016 ) . We measured a significant defect in ISG expression 8 hr post-infection in Lrrk2 KO BMDMs and Lrrk2 KO RAW 264 . 7 macrophages compared to control cells ( Figure 2A and Figure 2—figure supplement 1A ) . We next treated primary macrophages and macrophage cell lines with a panel of agonists designed to elicit type I IFN expression downstream of a variety of PRRs . Transfection of immunostimulatory dsDNA ( ISD ) , which is recognized by cGAS and stimulates the STING/TBK1/IRF3 axis , induced blunted Ifnb expression in Lrrk2 KO BMDMs ( Figure 2B ) , Lrrk2 KO peritoneal macrophages ( PEM ) ( significant differences in Ifnb expression were measured between Lrrk2 KO and HET at baseline but induction differences failed to reach statistical significance via 2-way ANOVA Tukey post-hoc testing ) ( Figure 2C ) , Lrrk2 KO RAW 264 . 7 macrophages ( Figure 2D ) , Lrrk2 KO mouse embryonic fibroblasts ( MEFs ) ( Figure 2E ) , and Lrrk2 KD RAW 264 . 7 macrophages ( Figure 2F ) . Consistent with the BMDM phenotype from Figure 1 , we observed higher basal expression of Ifnb/ISGs and a blunted response to ISD in all the Lrrk2 KO/KD cell lines tested ( Figure 2A–E and Figure 2—figure supplement 1C ) . We also found that Lrrk2 KO BMDMs failed to fully induce Ifnb if we bypassed cGAS and stimulated the DNA sensing adapter STING directly using the agonist DMXAA ( Figure 2G ) . In support of a defect in cytosolic nucleic acid sensing and IFNAR signaling , western blot analysis of IRF3 ( phospho-Ser396 ) and STAT1 ( phospho-Tyr701 ) activation showed a significant defect in the ability of Lrrk2 KO macrophages to signal through IFNAR ( phospho-STAT1 ) and a modest defect in cytosolic DNA sensing ( phospho-IRF3 ) over the course of 6 hr following ISD transfection ( Figure 2H , quantitation below , and Figure 2—figure supplement 1B ) . Collectively , these results suggest that type I IFN-generating pathways are chronically activated in cells lacking LRRK2 , but their induction is muted compared to controls when faced with agonists of the cytosolic DNA sensing pathway . We next tested whether loss of LRRK2 impacted the ability of cells to respond to activators of the type I IFN response outside of the cytosolic DNA sensing cascade . To this end , we treated Lrrk2 KO and HET BMDMs with transfected poly ( I:C ) ( to activate cytosolic RNA sensing ) , LPS ( to stimulate TRIF/IRF3 downstream of TLR4 ) , and CpG and CL097 ( to stimulate nucleic acid sensing via TLR9 and TLR7 , respectively ) . Interestingly , while we observed a defect in Ifnb induction in Lrrk2 KO BMDMs stimulated with poly ( I:C ) , we saw no difference in the ability of Lrrk2 KO BMDMs to express type I IFNs following treatment with LPS , CL097 , or CpG ( Figure 2I and Figure 2—figure supplement 1C ) , suggesting that TLR responses are intact in the absence of LRRK2 but cytosolic DNA and RNA sensing pathways are perturbed . We observed similar phenotypes for PEMs and MEFs treated with LPS ( Figure 2—figure supplement 1D–E ) and poly ( I:C ) ( Figure 2—figure supplement 1E ) . Lrrk2 KO BMDMs were , however , defective in ISG expression following recombinant bioactive IFN-β treatment ( which directly engages with IFNAR ) ( Figure 2J and Figure 2—figure supplement 1F ) . Because Lrrk2 KO BMDMs failed to induce ISG expression following IFN-β treatment , we hypothesized that the elevated basal levels of type I IFN transcripts prevented Lrrk2 KO macrophages from inducing a response at the level of IFNAR . To begin to test this prediction , we wanted to see if blocking IFN-β engagement with IFNAR could break this loop and ‘reset’ basal ISGs in Lrrk2 KO macrophages . Indeed , when HET and KO Lrrk2 BMDMs were treated with an IFN-β neutralizing antibody , there was a reduction in basal levels of Isg15 and Irf7 in Lrrk2 KO cells ( Figure 2K and Figure 2—figure supplement 1G ) . We next tested if loss of IFNAR signaling could similarly rescue the Lrrk2 KO phenotype by crossing Lrrk2 KO mice to Ifnar KO mice . In Lrrk2/Ifnar double KO BMDMs , we also observed a significant reduction in basal ISG levels ( Figure 2L and Figure 2—figure supplement 1H ) . These results provide additional evidence that the type I IFN program is chronically engaged in Lrrk2 KO macrophages . Because both IFN-β blockade and loss of Ifnar normalized basal ISG expression in Lrrk2 KO macrophages , we hypothesized that Lrrk2 contributes to basal type I IFN expression upstream of cytosolic RNA ( MAVS/RIG-I ) or DNA ( cGAS/STING ) sensing , two nucleic acid sensing pathways that are interconnected between positive and negative feedback loops ( Zevini et al . , 2017 ) . To directly test the involvement of cGAS in generating elevated resting levels of type I IFN in Lrrk2 KO macrophages , we crossed Lrrk2 KO and cGas KO mice and compared type I IFN transcript levels in double KO BMDMs with those of littermate controls . Although basal Isg15 expression differences between Lrrk2 KO and HET BMDMs were more modest in this experiment , loss of cGAS significantly reduced basal ISG expression in Lrrk2 KO BMDMs ( Figure 3A and Figure 3—figure supplement 1A ) . With lowered resting type I IFN levels , cGas/Lrrk2 double KOs were able to respond normally to IFN/ISG-generating innate immune stimuli like DMXAA , which bypasses cGAS and stimulates STING directly ( Diner et al . , 2013 ) , and poly ( I:C ) transfection ( Figure 3A and Figure 3—figure supplement 1A ) . Consistent with the ability of cGas ablation to rescue Lrrk2 KO baseline and induction defects , western blot analysis showed that levels of STAT1 phosphorylation were restored in cGas/Lrrk2 double KOs ( Figure 3B ) . Together , these results support a model where high basal levels of type I IFN and ISGs in Lrrk2 KO macrophages are due to chronic engagement of the cGAS-dependent DNA sensing pathway . We next sought to identify the source of the chronic cGAS-activating signal . Mitochondrial DNA ( mtDNA ) has been shown to be a potent activator of type I IFNs downstream of cGAS ( West et al . , 2015 ) , and LRRK2 is known to influence mitochondrial homeostasis ( Singh et al . , 2019 ) , albeit through mechanisms that are not entirely clear . To begin implicating mtDNA in the dysregulation of type I IFNs in Lrrk2 KO cells , we first investigated the status of the mitochondrial network in Lrrk2 HET and KO MEFs . As previously described for cells overexpressing wild-type or mutant alleles of LRRK2 ( Yang et al . , 2014 ) , Lrrk2 KO MEFs had a more fragmented mitochondrial network , especially around the cell periphery , as evidenced by punctate TOM20 staining ( Figure 3C ) . We hypothesized that this fragmentation was a sign of mitochondrial damage that could allow mitochondrial matrix components , including mtDNA , to leak into the cytosol . Therefore , we isolated the cytosolic fraction of control and Lrrk2 KD RAW 264 . 7 macrophages ( Figure 3D–F ) and Lrrk2 HET and KO MEFs ( Figure 3—figure supplement 1B–C ) and measured cytosolic mtDNA abundance . We found that although Lrrk2 KD cells had only slightly higher total mtDNA compared to controls ( Figure 3D ) , they had ~3 fold more cytosolic mtDNA ( Figure 3E ) . A similar phenotype was seen with Lrrk2 KO MEFs ( Figure 3—figure supplement 1B-C ) . This increase in cytosolic mtDNA was not simply an artifact of fragmented mitochondria contaminating cytosolic fractions as neither TFAM , an abundant mitochondrial transcription factor , nor VDAC , a mitochondrial outer membrane protein , were detectable in the cytosolic fraction by western blot ( Figure 3F ) . We hypothesized that cytosolic mtDNA results in activation of cGAS/IFNAR signaling , which ultimately limits the ability of Lrrk2 KO macrophages to respond to additional cytosolic nucleic acid agonists by downregulating canonical IFNAR signaling ( consistent with a reduction in STAT1 phosphorylation ( Figure 2H ) ) . To exacerbate the proposed mitochondrial defect , we crossed Lrrk2 KO mice with Tfam HET mice . Tfam HET mice are deficient in the mitochondrial transcription factor required for maintaining the mtDNA network and thus have high levels of cytosolic mtDNA ( Kasashima et al . , 2011; West et al . , 2015 ) . Depleting TFAM in Lrrk2 KO BMDMs led to even higher basal ISG expression ( Figure 3G ) , further suggesting that release of mtDNA into the cytosol in Lrrk2 KO cells contributes to their elevated type I IFN expression . We next sought to rescue type I IFN defects in Lrrk2 KO macrophages by depleting mtDNA using ddC ( 2' , 3'-dideoxycytidine ) , an inhibitor of mtDNA synthesis ( Leibowitz , 1971; Meyer and Simpson , 1969 ) . Treating Lrrk2 KO RAW 264 . 7 cells with ddC substantially reduced mtDNA copy number ( Figure 3H ) and resulted in similar basal expression of type I IFN and ISGs in resting Lrrk2 HET and KO cells ( Figure 3I and Figure 3—figure supplement 1D ) . Importantly , when mtDNA-depleted Lrrk2 KO RAW 264 . 7 macrophages were stimulated with ISD , their ability to induce Ifnb was restored to that of wild-type; Lrrk2 KO macrophages induced Ifnb approximately 500-fold in the absence of ddC but approximately 5000-fold following ddC treatment while WT macrophages induced Ifnb between 4000–5000-fold +/- ddC treatment ( Figure 3I and Figure 3—figure supplement 1D-E ) . These results demonstrate a critical role for mtDNA in driving both the high basal levels of type I IFN and the inability to properly induce type I IFN expression in Lrrk2 KO macrophages . Previous studies of microglia have shown that LRRK2 contributes to mitochondrial homeostasis through interaction with the mitochondrial fission protein DRP1 ( Ho et al . , 2018 ) . Thus , we hypothesized that the loss of LRRK2 may compromise mitochondrial stability via misregulation of DRP1 activity , leading to fragmented mitochondria and spillage of mtDNA into the cytosol . To assess gross defects in DRP1 distribution in the absence of LRRK2 , we performed immunofluorescence microscopy and did not observe any obvious , qualitative changes to the expected distribution of DRP1 at the ends of mitochondrial tubules in Lrrk2 KO MEFs , although TOM20 staining again revealed extensive fragmentation of the peripheral network ( Figure 4—figure supplement 1A ) . We next asked whether DRP1 activity was impacted by loss of LRRK2 . Because DRP1 is known to be positively regulated via phosphorylation at Ser616 ( Taguchi et al . , 2007 ) , we performed flow cytometry with an antibody specific for phospho-S616 DRP1 and observed significantly higher levels of phospho-S616 DRP1 in Lrrk2 KD RAW 264 . 7 cells , Lrrk2 KO BMDMs , and Lrrk2 KO MEFs compared to controls ( Figure 4A , C , D ) . Western blot analysis of phospho-S616 DRP1 confirmed a modest increase in Lrrk2 KD cells , while total DRP1 protein levels remained unchanged ( Figure 4B and Figure 4—figure supplement 1B ) . Accumulation of phospho-S616 DRP1 was enhanced in MEFs by the addition of H2O2 , which induces DRP1-dependent mitochondrial fission , and was eliminated with the addition of Mdivi-1 , a specific inhibitor of DRP1 ( Figure 4—figure supplement 1C ) . Next , to test if DRP1 activity was linked to high basal type I IFN/ISG expression in Lrrk2 deficient cells , we chemically inhibited DRP1 with Mdivi-1 and measured basal gene expression levels . In Lrrk2 KD RAW 264 . 7 macrophages and Lrrk2 KO BMDMs , DRP1 inhibition returned ISG expression to control levels ( Figure 4E and Figure 4—figure supplement 1D ) . DRP1 inhibition also restored the cytosolic mtDNA levels in Lrrk2 KD RAW 264 . 7 cells and Lrrk2 KO MEFs to those of control cells ( Figure 4F and Figure 4—figure supplement 1E ) . Together , these data indicate that dysregulated ISG expression in Lrrk2 KO macrophages is caused by leakage of mtDNA into the cytosol , which occurs downstream of excessive DRP1-dependent mitochondrial fission . Given that cytosolic mtDNA contributes to type I IFN defects in Lrrk2 KO macrophages , we predicted that mitochondria in Lrrk2 KO cells may be more damaged and/or more prone to damage . To better understand the health of the mitochondrial network in Lrrk2 KO vs . HET BMDMs , we first used the carbocyanine dye JC-1 , which accumulates in mitochondria to form red fluorescent aggregates . Upon loss of mitochondrial membrane potential , JC-1 diffuses into the cytosol where it emits green fluorescence as a monomer . Thus , a decrease in red fluorescence ( aggregates ) and increase in green fluorescence ( monomers ) signifies mitochondrial depolarization , making JC-1 dye a highly sensitive probe for mitochondrial membrane potential . Flow cytometry analysis of resting Lrrk2 HET and KO cells revealed lower levels of JC-1 dye aggregation ( i . e . , lower mitochondrial membrane potential ) in Lrrk2 KO BMDMs ( Figure 5A–B ) , Lrrk2 KD RAW 264 . 7 macrophages ( Figure 5—figure supplement 1A ) , and primary Lrrk2 KO MEFs ( Figure 5—figure supplement 1B ) , compared to control cells . A baseline reduction in membrane potential was also detected using TMRE ( tetramethylrhodamine , ethyl ester , perchlorate ) , a cell-permeable dye that is readily sequestered by active ( positively charged ) mitochondria , in Lrrk2 KO BMDMs , Lrrk2 KO MEFs , and Lrrk2 KD RAW 264 . 7 cells ( Figure 5—figure supplement 1D ) . In addition , Lrrk2 KO BMDMs were more sensitive to the mitochondrial damaging and depolarizing agents , rotenone/ATP and FCCP , as measured by both JC-1 ( Figure 5C; RAW 264 . 7 Lrrk2 KDs and Lrrk2 KO MEFs in Figure 5—figure supplement 1A and B , respectively ) and TMRE ( Figure 5D and Figure 5—figure supplement 1C , E ) . Interestingly , the mitochondrial membrane potential of Lrrk2 KO BMDMs was normalized after treatment with Mdivi-1 to inhibit DRP1 ( Figure 5E–F ) , suggesting that misregulation of DRP1 occurs upstream of LRRK2-dependent defects in mitochondrial membrane potential . Previous reports have indicated that LRRK2 dysfunction alters ROS levels ( Pereira et al . , 2014; Russo et al . , 2019 ) . To test whether ROS could contribute to the defective type I IFN signature in Lrrk2 KO cells , we treated Lrrk2 HET and KO BMDMs with mitoTEMPO ( mitoT ) , a mitochondrially-targeted scavenger of superoxide ( Liang et al . , 2010 ) . Consistent with oxidative stress driving misregulation of the type I IFN response in the absence of LRRK2 , we observed a dramatic rescue of basal ISG expression in Lrrk2 KO cells treated with mitoT ( Figure 5G ) . Together these data provide strong evidence that Lrrk2 KO cells harbor a baseline defect in mitochondrial membrane potential , likely due to DRP1 activation , that results in chronic type I IFN induction due to increased levels of cytosolic mtDNA . Metabolic reprogramming is becoming increasingly appreciated as a critical contributor to macrophage polarization and transcriptional output ( Angajala et al . , 2018; Sancho et al . , 2017 ) . We hypothesized that mitochondrial defects may render Lrrk2 KO macrophages incapable of meeting metabolic demands . To test this idea , we manipulated levels of sodium pyruvate , an intermediate metabolite of glycolysis and the TCA cycle , in the media of Lrrk2-deficient cells plus their respective controls . The presence of pyruvate places additional demands on the mitochondria by pushing cells towards oxidative metabolism rather than glycolysis . Remarkably , addition of as little as 1 mM sodium pyruvate to the growth media increased the already high basal levels of type I IFN in macrophages and MEFs lacking LRRK2 ( Figure 6A–B and Figure 6—figure supplement 1A ) , suggesting that increasing metabolic demands on mitochondria promotes further leakage of mtDNA into the cytosol in these cells . To better understand how Lrrk2 KO macrophages may be defective in meeting the energy needs of the cell , we used the Agilent Seahorse Metabolic Analyzer to measure cellular respiration . In this assay , oxidative phosphorylation ( OXPHOS ) and glycolysis are assayed by oxygen consumption rate ( OCR ) and extracellular acidification rate ( ECAR ) , respectively . We found that mitochondria in Lrrk2 KO BMDMs were defective in both maximal and reserve respiratory capacity ( Figure 6C , top panels ) , indicating reduced OXPHOS . Lrrk2 KO macrophages were also defective in non-glycolytic acidification and had reduced glycolysis . ( Figure 6C , bottom panels ) . This result was surprising as macrophages typically switch from OXPHOS to glycolysis when activated ( Kelly and O'Neill , 2015 ) , but Lrrk2 KO macrophages have reduced utilization of both energy producing pathways . Remarkably , co-treatment of Lrrk2 KO BMDMs with mitoT and IFN-β neutralizing antibody completely restored OXPHOS and glycolysis . This rescue was greater than treatment of either IFN-β blockade or mitoT alone ( Figure 6C–D ) , indicating that mitochondrial ROS and constitutive IFNAR signaling independently contribute to the metabolic defects in Lrrk2 KO macrophages . Conversely , when Lrrk2 KO BMDMs were cultured in the presence of increasing concentrations of sodium pyruvate ( 1 and 2 mM ) , we observed exacerbated metabolic defects in OXPHOS and glycolysis ( Figure 6—figure supplement 1B-C ) . Collectively , these data demonstrate that loss of LRRK2 in macrophages has a profound impact on the mitochondria , not only promoting their fragmentation , but also rendering them less capable of utilizing different carbon sources and meeting the energy needs of the cell . To better understand possible molecular changes driving or resulting from damaged mitochondria in Lrrk2 KO macrophages , we performed an unbiased query of metabolites using LC/MS/MS ( Table S2 ) . In Lrrk2 KO BMDMs , we found lower levels of inosine monophosphate ( IMP ) and hypoxanthine , two intermediates in the purine biosynthesis pathway , which we validated using pure molecular weight standards ( Figure 7A and Figure 7—figure supplement 1A-B ) . Interestingly , purine metabolism is tightly associated with generation of antioxidant compounds , and several metabolites in this pathway are well-characterized biomarkers of Parkinson's disease ( Figure 7B; Zhou et al . , 2012 ) . Consistent with lower levels of antioxidants , we detected increased oxidized glutathione and glutamate metabolism compounds in Lrrk2 KO macrophages ( Table S2 ) . Moreover , in accordance with lower levels of purine metabolites , we observed significantly fewer de novo biosynthesis puncta containing formylglycinamidine ribonucleotide synthase ( FGAMS , also known as PFAS ) , a core purinosome component , per Lrrk2 KO MEF cell compared to HET controls ( Figure 7C–D ) . Because depleted antioxidant pools and concomitant accumulation of ROS can lead to mitochondrial damage , we hypothesized that ROS might contribute to the mitochondrial and type I IFN defects we observed in Lrrk2 KO macrophages . To test this , we first supplemented cells with antioxidants in order to rescue the type I IFN defect in Lrrk2 KO macrophages . Addition of urate , a free radical scavenger and major breakdown product of purine metabolism ( Figure 7B ) , reduced basal ISG expression in a dose-dependent fashion in Lrrk2 KO BMDMs and in Lrrk2 KD RAW 264 . 7 cells ( Figure 7E–F , respectively ) . Furthermore , treatment with urate or mitoT restored the resting mitochondrial membrane potential of Lrrk2 KO BMDMs ( Figure 7G–H ) , suggesting that radical oxygen species contribute to mitochondrial depolarization in the absence of LRRK2 . Neither urate nor mitoT was sufficient to alter DRP1 activation in Lrrk2 KO BMDMs , suggesting the antioxidant defects are either downstream or independent of LRRK2-dependent DRP1 misregulation ( Figure 7I ) . Collectively , these results suggest that the depletion of antioxidant pools in Lrrk2 KO macrophages from defective purine metabolism contributes to their mitochondrial dysfunction and aberrant type I IFN expression . Previous reports have linked SNPs in LRRK2 with susceptibility to mycobacterial infection and excessive inflammation in humans ( Fava et al . , 2016; Wang et al . , 2015; Zhang et al . , 2009 ) . Our studies demonstrate that LRRK2 plays a key role in homeostasis of macrophages , the first line of defense and replicative niche of Mtb . Therefore , we sought to understand how LRRK2 deficiency influences Mtb pathogenesis in macrophages ex vivo and during an in vivo infection . Lrrk2 HET and KO BMDMs were infected with Mtb ( Erdman strain; MOI = 1 ) , and colony forming units ( CFUs ) were measured over the course of five days . We observed a significant increase in CFUs recovered at 5 days ( 120 hr ) following infection ( Figure 8A ) , suggesting that while Lrrk2 KO macrophages can control Mtb replication early after infection , they are more permissive once the bacteria have established a niche . These results are consistent with a recent report demonstrating that defective IFNAR signaling in BMDMs leads to increased Mtb replication ( Banks et al . , 2019 ) . To test whether this replication phenotype impacted Mtb pathogenesis in vivo , we infected Lrrk2 HET and KO mice with ~150 CFUs via aerosol chamber delivery . We observed no difference in the survival time of Mtb-infected Lrrk2 KO mice compared to HET controls ( Figure 8B ) , and at Days 7 , 21 , 63 , and 126 days post-infection , we observed no significant differences in bacterial burdens in the lungs or spleens of infected mice ( Figure 8C ) . We also measured serum cytokines and found no major differences ( Figure 8D ) . We next examined if local inflammation in the lungs was impacted by loss of LRRK2 . While major NF-κB pathway inflammatory cytokines ( e . g . Tnfa , Il6 , Il1b ) were expressed at similar levels in the lungs of Mtb-infected Lrrk2 KO and HET mice at Day 21 ( Figure 8E ) , we observed lower levels of several canonical ISGs including Mx1 , Irf9 , and Gbp8 ( Figure 8F ) , consistent with trends we observed in Lrrk2 KO macrophages ex vivo ( Figure 1E ) . These results began to suggest that during Mtb infection , most of the effects of LRRK2 ablation occur at the local site of infection . To better understand the nature of this local inflammatory phenotype , we inspected lung tissues via hematoxylin and eosin ( H&E ) staining . We observed significantly more inflammatory granulomatous nodules in the lungs , specifically in the perivascular region ( Figure 8G–H ) , indicating that more macrophages had infiltrated infected lungs of Lrrk2 KO mice at Day 21 post-infection . Consistent with increased inflammation , we observed more total neutrophils ( polymorphonuclear leukocytes , PMNs ) as well as more PMNs undergoing cell death ( degenerate PMNs ) in the lungs of Mtb-infected Lrrk2 KO mice compared to controls ( Figure 8I–J ) . These results are consistent with a recently published study that reported enhanced inflammatory innate immune responses to Mtb infection in Lrrk2 KO mice compared to wild-type controls ( Härtlova et al . , 2018 ) . Despite being repeatedly associated with susceptibility to mycobacterial infection and inflammatory disorders in genome-wide association studies , very little is known about how LRRK2 functions outside of the central nervous system . Here , we provide evidence that loss of LRRK2 in macrophages alters type I IFN and ISG expression due to elevated levels of cytosolic mtDNA and chronic cGAS signaling . During Mtb infection , loss of LRRK2 dysregulates type I IFN production and enhances local neutrophil and macrophage infiltration and cell death in the lung . These data help explain why LRRK2 missense mutations are associated with exacerbated inflammation and poor disease outcomes in leprosy patients ( Fava et al . , 2016 ) . They also hint at a previously unappreciated but potentially crucial role for LRRK2 in regulating the central nervous system immune milieu in PD patients ( Patrick et al . , 2019 ) via alteration of mitochondrial homeostasis in brain-resident glial cells . Our observations connect LRRK2’s role in innate immune dysregulation with its requirement for maintaining mitochondrial homeostasis and are consistent with numerous recent studies linking mitochondrial metabolism and energy production to immune outcomes ( Angajala et al . , 2018; Bird , 2019; Walker et al . , 2014 ) . There are several unique aspects of the Lrrk2 KO macrophage phenotype that reveal new insights into how mitochondrial stress impacts type I IFN expression and innate immune cell priming . Lrrk2 KO macrophages fail to activate normal levels of phospho-STAT1 following innate immune stimuli , resulting in blunted ISG induction , but this defect can be rescued by depleting mtDNA , reducing mitochondrial ROS , and deleting cGas . In this way , Lrrk2 KO macrophages are phenotypically distinct from other cells in which mitochondrial stress and increased cytosolic mtDNA have been shown to dramatically increase phospho-STAT1 , further amplifying the IFN response ( West et al . , 2015 ) . We propose that cytosolic DNA and increased ROS— which perhaps together generate oxidized cytosolic DNA that is resistant to the exonuclease TREX-1 ( Gehrke et al . , 2013 ) —drive this unique refractory phenotype in Lrrk2 KO macrophages . Alternatively or additionally , this phenomenon could be driven via by upregulation of one or more unidentified negative regulators of these key signaling pathways . Interestingly , Lrrk2 KO macrophages also challenge the general paradigms of type I IFN priming . Typically , tonic or basal IFN levels are thought to ‘rev the engine’ so that cells can rapidly induce type I IFN expression after receiving a stimulus ( Taniguchi and Takaoka , 2001 ) . Using this metaphor , although the engine is revved in Lrrk2 KO macrophages , these cells still lose the race . Additional studies will be needed to define the molecular mechanisms driving this puzzling phenotype . The metabolic phenotypes of Lrrk2 KO macrophages are also unique . It is curious that they suffer from defects in both glycolysis and oxidative phosphorylation , since generally , these two energy-producing pathways compensate for each other . Defects in both pathways is indicative of a more quiescent cellular metabolic state consistent with a reduced capacity for IFNAR signaling . Importantly , treatment of Lrrk2 KO macrophages with IFN-β neutralizing antibody was sufficient to rescue glycolysis ( as measured by extracellular acidification rate ( ECAR ) ) but not OXPHOS ( as measured by oxygen consumption rate ( OCR ) ) ( Figure 6C ) , which suggests chronic IFNAR signaling alters the glycolytic rate in Lrrk2 KO macrophages . Rescue of the OXPHOS defect required treatment with both IFN-β neutralizing antibody and mitoTEMPO , indicating that a more complex defect drives changes to mitochondrial respiration , perhaps linked to the depolarization defect . It will be important moving forward to understand the precise molecular contributions of LRRK2 to specific aspects of mitochondrial health ( e . g . energy production , morphology , fission/fusion , etc . ) and to link these defects with outcomes in diverse cell types ( e . g . neurons and macrophages ) . We propose that dysregulation of type I IFN expression in Lrrk2 KO macrophages is the result of two distinct cellular defects conferred by loss of LRRK2 . First , in the absence of LRRK2 , decreased levels of purine metabolites and urate contribute to oxidative stress , leading to damage of the mitochondrial network . This idea is supported by our experiments showing that urate and mitoTEMPO treatments could rescue defects in mitochondrial polarization and return elevated basal type I IFNs to normal in Lrrk2 KO macrophages ( Figures 5E and 7E–G ) . A recent human kinome screen identified LRRK2 as a kinase involved in dynamics of the purinosome , a cellular body composed of purine biosynthetic enzymes that assembles at or on the mitochondrial network ( French et al . , 2016 ) . Specifically , shRNA knockdown of LRRK2 in HeLa cells inhibited purinosome assembly and disassembly . As purinosomes are posited to form in order to protect unstable intermediates and increase metabolic flux through the de novo purine biosynthetic pathway ( An et al . , 2008; Schendel et al . , 1988 ) , we propose that the lower levels of IMP and hypoxanthine we measure in Lrrk2 KO macrophages results from LRRK2-dependent defects in purinosome assembly ( Figure 7A ) . Lower levels of purine nucleotide intermediates are especially notable in the context of PD; the plasma of PD patients ( both LRRK2 mutant-associated and idiopathic ) has been shown to contain significantly less hypoxanthine and uric acid ( Johansen et al . , 2009 ) , and patients with higher plasma urate levels , despite carrying LRRK2 mutations , are less likely to develop PD ( Bakshi et al . , 2019 ) . Furthermore , urate is currently being investigated as a potential therapeutic of PD , highlighting the importance of purine biosynthesis in this disease . Second , we propose that loss of LRRK2 contributes to type I IFN dysregulation through defects associated with the mitochondrial fission protein DRP1 . Previous reports have shown that LRRK2 can physically interact with DRP1 and that LRRK2 mediates mitochondrial fragmentation through DRP1 ( Wang et al . , 2012 ) . Overexpression of both wild type LRRK2 and the PD-associated G2019S allele of LRRK2 have been shown to cause mitochondrial fragmentation ( Wang et al . , 2012 ) . Curiously , we observe a similar phenotype in macrophages lacking LRRK2 ( Figure 3C ) . Previous studies have linked LRRK2 and LRRK2 kinase activity to DRP1 activation via phosphorylation at several sites including T595 in neurons ( Su and Qi , 2013 ) and S616 in a neuron-like carcinoma cell line ( Esteves et al . , 2015 ) . Our observation that Lrrk2 KO macrophages accumulate phospho-S616 DRP1 and exhibit increased fragmentation of the mitochondrial network indicates that LRRK2 is not required for DRP1 phosphorylation or activation of mitochondrial fission in macrophages . Indeed , other kinases have been shown to phosphorylate DRP1 at S616 in other cell types , including ERK2 ( Kashatus et al . , 2015 ) and CDK1 ( Taguchi et al . , 2007 ) . We propose that loss of LRRK2 could alter DRP1 phosphorylation indirectly through changing serine accessibility or protein-protein interactions , or by modifying other pathways that control mitochondrial turnover or lysosome homeostasis . It will be important for future studies to compare the molecular mechanisms driving the DRP1-dependent mitochondrial fission defects in Lrrk2 KO cells and in cells harboring the PD-associated ‘gain of function’ G2019S allele . Mtb is a potent activator of cytosolic DNA sensing ( Manzanillo et al . , 2012; Watson et al . , 2012 ) , and type I IFNs are important biomarkers of Mtb infection associated with poor outcomes in humans and in mouse models of infection ( Berry et al . , 2010 ) . New insights into the requirement of IFNAR signaling for nitric oxide production in macrophages ex vivo suggest critical roles for type I IFN induction in cell-intrinsic control of Mtb replication ( Banks et al . , 2019 ) . However , the degree to which these macrophage phenotypes translate to mouse models of infection remains poorly understood . Although we observed a striking type I IFN defect ( both higher basal levels and blunted induction ) in a number of macrophage primary cells and cell lines , we did not find major differences in infection outcomes in Lrrk2 HET vs . KO mice . Our previous experiments demonstrated that while loss of cGAS almost completely abrogates type I IFN expression in macrophages , it has only minor effects in vivo ( serum IFN-β levels and lung type I IFN/ISG expression levels ) ( Collins et al . , 2015; Watson et al . , 2015 ) , suggesting that Mtb infection can elicit type I IFN expression in important cGAS-independent ways in vivo that we do not yet fully understand . Another recent publication that investigated the role of LRRK2 in controlling Mtb infection does report a significant decrease in CFUs in Lrrk2 KO mice at very early infection time points ( Day 7 and 14 ) , which correlates with increased inflammation in the lungs ( as we also report ) ( Härtlova et al . , 2018 ) . It is likely that minor discrepancies between our data and that reported by Härtlova et al . are the consequence of differences in mouse and Mtb strains and the fact that we compared Lrrk2 KO and HET littermate controls as opposed to WT controls . It will be crucial moving forward to more directly interrogate the molecular drivers of inflammation and Mtb pathogenesis in Lrrk2 KO mice as well as in mouse genotypes associated with human disease susceptibility , for example LRRK2 G2019S . Because LRRK2 inhibitors are a major area of drug development for the treatment of PD , it is crucial to understand how both loss of and mutations in this protein might impact the ability of patients receiving such therapies to respond to and clear infection . Bone marrow derived macrophages ( BMDMs ) were differentiated from bone marrow cells isolated by washing mouse femurs with 10 ml DMEM . Cells were then centrifuged for 5 min at 1000 rpm and resuspended in BMDM media ( DMEM , 20% FBS , 1 mM Sodium pyruvate , 10% MCSF conditioned media ) . BM cells were counted and plated at 5 × 106 in 15 cm non-TC treated dishes in 30 ml complete media and fed with an additional 15 ml of media on Day 3 . On Day 7 , cells were harvested with 1x PBS-EDTA . Mouse embryonic fibroblasts ( MEFs ) were isolated from embryos . Briefly , embryos were dissected from yolk sacs , washed two times with cold 1x PBS , decapitated , and peritoneal contents were removed . Headless embryos were disaggregated in cold 0 . 05% trypsin-EDTA and incubated on ice for 20 min , followed by incubation at 37°C for an additional 20 min and DNase treatment ( 20 min , 37°C , 100 ug/ml ) . Supernatants were removed and spun down at 1000 rpm for 5 min . Cells were resuspended in DMEM , 10% FBS , 1 mM sodium pyruvate , and plated in 15 cm TC treated dishes , three dishes per embryo . MEFs were allowed to expand for 2–3 days before harvest with Trypsin 0 . 05% EDTA . Peritoneal macrophages ( PEMs ) , were elicited by intraperitoneal injection of 1 ml 3% Thioglycollate broth ( BD Biosciences ) for 4 days prior to harvest . For harvest , PEMs were isolated from mice by lavage ( 1x PBS 4°C ) and resuspended in RPMI 1640 media with 20% FBS , 1 mM sodium pyruvate and 2 mM L-Glutamine . Following overnight incubation at 37°C , cells were washed twice ( 1x PBS 37°C ) to remove non-adherent cells . RAW 264 . 7 and U937 cell lines were each purchased from ATCC . All our cell lines are minimally passaged to maintain genomic integrity and new cell lines are generated from these low passage stocks . Cell lines were passaged no more than 10 times . Our cell lines stocks have all tested negative for mycoplasma contamination . RAW 264 . 7 Lrrk2 KO cells ( ATCC SC-6004 ) generated by the MJFF , were obtained from the ATCC and used with wild type control Lrrk2 parental RAW 264 . 7 ( ATCC SC-6003 ) . To deplete mtDNA , RAW 264 . 7 cells were seeded at 2 × 106 cells/well in 10 cm non-TC treated dishes and cultured for 4 days in complete media ( DMEM , 10% FBS , 1 mM sodium pyruvate ) with 10 μM ddC . Cells were split and harvested with 1x PBS-EDTA . Prior to treatment/stimulation , BMDMs were plated in 12 well plates at 5 × 105 cells/well , or 6-well plates at 1 × 106 cells/well . MEFs were plated in 12 well dishes at 3 × 105 cells/well . PEMs were plated in 24-well flat-bottomed plates at 1 × 106 cells/ well . RAW 264 . 7 cells were plated at 7 . 5 × 105 cells/well . Cells were stimulated for 4 hr with 1 μM CLO97 , 100 ng/ml LPS , or transfected 1 μg/ml ISD , 1 μg/ml poly ( I:C ) , 1 μg/ml cGAMP with lipofectamine . Cells were transfected for 4 hr with 1 μM CpG 2395 with Gene Juice . Cells were stimulated for 2–4 hr with 10 μM DMXAA ( RAW 264 . 7 ) or 200 IU IFN-β ( BMDMs ) . Lrrk2 KO mice ( C57BL/6-Lrrk2tm1 . 1Mjff/J ) stock #016121 , and Ifnar KO mice ( B6 ( Cg ) -Ifnar1tm1 . 2Ees/J ) stock #028288 were purchased from Jackson Laboratories ( Bar Harbor , ME ) . Tfam HET ( An et al . , 2008; Schendel et al . , 1988; Zhao et al . , 2013 ) and Mb21d1 ( cGas ) KO ( B6 ( C ) -Cgastm1d ( EUCOMM ) Hmgu/J ) mice were provided by A . Phillip West at Texas A&M Health Science Center . Lrrk2 KO mice used in experiments were backcrossed once onto C57BL6/NJ by the Jackson labs and then maintained with filial breeding . ( N1F8 ) . The Lrrk2 KO strain has been maintained with filial breeding on a C57BL6/NJ background for five more generations . When breeding Lrrk2 KO mice to Ifnar KO , cGas KO and Tfam HET strains all of which are on a C57BL6/J background , mice were backcrossed for two generations onto the C57BL6/NJ and then were maintained with filial breeding ( currently F3 ) . All mice used in experiments were compared to age- and sex- matched controls . In order to ensure littermate controls were used in all experiments Lrrk2 KO crosses were made with ( KO ) Lrrk2-/- x ( HET ) Lrrk2+/- mice . Mice used to generate BMDMs and PEMs were between 8 and 12 weeks old . Mice were infected with Mtb at 10 weeks . Embryos used to make primary MEFs were 14 . 5 days post coitum . All animals were housed , bred , and studied at Texas A&M Health Science Center under approved Institutional Care and Use Committee guidelines . The Erdman strain was used for all Mtb infections ( Watson et al . , 2015; Watson et al . , 2012 ) . Low passage lab stocks were thawed for each experiment to ensure virulence was preserved . Mtb was cultured in roller bottles at 37°C in Middlebrook 7H9 broth ( BD Biosciences ) supplemented with 10% OADC , 0 . 5% glycerol , and 0 . 1% Tween-80 or on 7H11 plates . All work with Mtb was performed under Biosafety Level 3 ( BSL3 ) containment using procedures approved by the Texas A and M University Institutional Biosafety Committee . Prior to infection , BMDMs were seeded at 1 . 2 × 106 cells/well ( 6-well dish ) or 3 × 105 cells/well ( 12-well dish ) , RAW cells at 5 × 105 cells/well ( 12-well dish ) , and U937s at 1 × 106 cells/well . U937s were cultured with 10 ng/ml phorbol 12-myristate 13-acetate ( PMA ) for 48 hr to induce differentiation and then recovered in fresh media for an addition 24 hr prior to infection . To prepare the inoculum , bacteria grown to log phase ( OD 0 . 6–0 . 8 ) were spun at low speed ( 500 g ) to remove clumps and then pelleted and washed with 1x PBS twice . Resuspended bacteria were briefly sonicated and spun at low speed once again to further remove clumps . The bacteria were diluted in DMEM + 10% horse serum and added to cells , MOI = 10 . Cells were spun with bacteria for 10 min at 1000 g to synchronize infection , washed twice with PBS , and then incubated in fresh media . RNA was harvested from infected cells using 0 . 5–1 . 0 ml Trizol reagent 4 hr post-infection unless otherwise indicated . M . leprae was cultivated in the footpads of nude mice and generously provided by the National Hansen’s Disease Program . Bacilli were recovered overnight at 33°C , mixed to disperse clumps and resuspended in DMEM + 10% horse serum . Cells were infected as with Mtb but with an MOI of 50 . All infections were performed using procedures approved by Texas A&M University Institutional Care and Use Committee . The Mtb inoculum was prepared as described above . Age- and sex-matched mice were infected via inhalation exposure using a Madison chamber ( Glas-Col ) calibrated to introduce 100–200 CFUs per mouse . For each infection , approximately five mice were euthanized immediately , and their lungs were homogenized and plated to verify an accurate inoculum . Infected mice were housed under BSL3 containment and monitored daily by lab members and veterinary staff . At the indicated time points , mice were euthanized , and tissue samples were collected . Blood was collected in serum collection tubes , allowed to clot for 1–2 hr at room temperature , and spun to separate serum . Serum cytokine analysis was performed by Eve Technologies . Organs were divided to maximize infection readouts ( CFUs: left lobe lung and ½ spleen; histology: two right lung lobes and ¼ spleen; RNA: one right lung lobe and ¼ spleen ) . For histological analysis organs were fixed for 24 hr in either neutral buffered formalin and moved to ethanol ( lung , spleen ) . Organs were further processed as described below . For cytokine transcript analysis , organs were homogenized in Trizol Reagent , and RNA was isolated as described below . For CFU enumeration , organs were homogenized in 5 ml PBS + 0 . 1% Tween-80 , and serial dilutions were plated on 7H11 plates . Colonies were counted after plates were incubated at 37°C for 3 weeks . Lungs and spleens were fixed with paraformaldehyde , subjected to routine processing , embedded in paraffin , and 5 μm sections were cut and stained with hematoxylin and eosin ( H and E ) or acid-fast stain . A boarded veterinary pathologist performed a masked evaluation of lung sections for inflammation using a scoring system: score 0 , none; score 1 , up to 25% of fields; score 2 , 26–50% of fields; score 3 , 51–75% of fields; score 4 , 76–100% of fields . To quantify the percentage of lung fields occupied by inflammatory nodules , scanned images of at least 2 sections of each lung were analyzed using Fiji Image J ( Johansen et al . , 2009 ) to determine the total cross-sectional area of inflammatory nodules per total lung cross sectional area . Total neutrophil scores were determined using digital images of H and E slides divided into 500 × 500 um grids and counting the percentage of squares containing neutrophils ( total PMN ) or degenerate neutrophils . RNA-seq represents analysis of 16 samples ( biological quadruplicates of Lrrk2 HET uninfected , Lrrk2 HET +Mtb , Lrrk2 KO uninfected , and Lrrk2 KO +Mtb; one sample from the Lrrk2 HET +Mtb group was removed from the analysis due to poor quality sequencing ) . Briefly , RNA was isolated from BMDMs using PureLink RNA mini kits ( Ambion ) and quantified on an Agilent Bioanalyzer 2100 . PolyA+ PE 100 libraries were sequenced on a HiSeq 4000 at the UC Davis Genome Center DNA Technologies and Expression Analysis Core . Raw reads were processed with expHTS ( Streett et al . , 2015 ) to trim low-quality sequences and adapter contamination , and to remove PCR duplicates . Trimmed reads for each sample were mapped to the Mus musculus Reference genome ( RefSeq ) using CLC Genomics Workbench 8 . 0 . 1 . Relative transcript expression was calculated by counting Reads Per Kilobase of exon model per Million mapped reads ( RPKM ) . Differential expression analyses were performed using CLC Genomics Workbench EDGE test . Differentially expressed genes were selected as those with p-value threshold <0 . 05 in the heatmaps represented . Heatmaps were generated using GraphPad Prism software ( GraphPad , San Diego , CA ) . RNA was isolated using Direct-zol RNAeasy kits ( Zymogen ) . cDNA was synthesized with BioRad iScript Direct Synthesis kits ( BioRad ) per manufacturer’s protocol . qRT-PCR was performed in triplicate wells using PowerUp SYBR Green Master Mix . Data were analyzed on a QuantStudio 6 Real-Time PCR System ( Applied Biosystems ) . 3 × 106 MEFs or 1 × 107 RAW 264 . 7 cells were plated in 10 cm dishes . The next day , confluent plates were treated as indicated with inhibitors . To harvest , cells were lifted with 1x PBS-EDTA . To determine total DNA content , 1% of the input was saved and processed by adding NaOH to 50 mM , boiling 30 min , and neutralizing with 1:10 1M Tris pH 8 . 0 . To isolate cytosolic DNA , the cells were pelleted and resuspended in digitonin lysis buffer ( 150 mM HEPES pH 7 . 4 , 50 mM NaCl , 10 mM EDTA , 25–50 μg/ml digitonin ) . Cells were incubated for 15 min at 4°C on an end-over-end rotator . Cells were spun at 980 x g for 3 min , and the supernatant was collected and spun again at 15 , 000 x g for 3 min to remove any remaining organelle fragments . DNA from the cleared supernatant ( cytosolic fraction ) was then extracted via phenol:chloroform ( 1:1 supernatant:phenol/chloroform ) . The DNA from the aqueous layer was precipitated in 0 . 3 M sodium acetate , 10 mM magnesium chloride , 1 μg/ml glycogen , and 75% ethanol . After freezing overnight at −20°C , the DNA was pelleted , washed in 70% ethanol , dried , resuspended in TE , and solubilized at 50°C for 30 min . qPCR was performed on the input ( 1:50 dilution ) and cytosolic ( 1:2 dilution ) samples using nuclear ( Tert ) and mitochondrial ( 16s and cytB ) genes . The total and cytosolic mitochondrial DNA was normalized to nuclear DNA in order to control for variation in cell number . Cells were washed with PBS and lysed in 1x RIPA buffer with protease and phosphatase inhibitors , with the addition of 1 U/ml Benzonase to degrade genomic DNA . Proteins were separated by SDS-PAGE and transferred to nitrocellulose membranes . Membranes were blocked for 1 hr at RT in LiCOR Odyssey blocking buffer . Blots were ( Licor ) and incubated overnight at RT with the following antibodies: IRF3 ( Cell Signaling , 1:1000 ) ; pIRF3 Ser396 ( Cell Signaling , 1:1000 ) ; STAT1 ( Cell Signaling , 1:1000 ) ; pSTAT1 Tyr701 ( Cell Signaling , 1:1000 ) ; Beta Actin ( Abcam , 1:2000 ) , β-Tubulin ( Abcam , 1:5000 ) ; DRP1 ( Cell Signaling , 1:1000 ) ; pDRP1 Ser616 ( Cell Signaling , 1:1000 ) , TFAM ( Millipore , 1:1000 ) , VDAC ( Protein Tech , 1:1000 ) . Membranes were incubated with appropriate secondary antibodies for 2 hr at RT prior to imaging on a LiCOR Odyssey Fc Dual-Mode Imaging System . Seahorse XF mito stress test kits and cartridges were prepared per manufacturers protocol as described in An et al . ( 2008 ) ; Schendel et al . ( 1988 ) ; Zhao et al . ( 2013 ) and analyzed on an Agilent Seahorse XF 96-well Analyzer . BMDMs were seeded at 5 × 104 cells/well overnight and treated with 200 μM mitoTEMPO , IFN-β neutralizing Ab , or sodium pyruvate at 0 , 1 , or 2 mM final concentration . Sera was analyzed by Eve Technologies: Mouse Cytokine Array/Chemokine Array 13-plex Secondary Panel ( MD13 ) . Briefly , sera was isolated following decapitation in Microtainer serum separator tubes ( BD Biosciences ) followed by 2x sterile filtration with Ultrafree-MC sterile filters , 10 min at 10 , 000 rpm ( Millipore Sigma ) . For analysis sera was prediluted 1:1 to a final volume of 100 μl in 1x PBS pH 7 . 4 and assayed/analyzed in duplicate . MEFs were seeded at 1 × 105 cells/well on glass coverslips in 24-well dishes . Cells were fixed in 4% paraformaldehyde for 10 min at RT and then washed three times with PBS . Coverslips were incubated in primary antibody diluted in PBS + 5% non-fat milk + 0 . 1% Triton-X ( PBS-MT ) for 3 hr . Cells were then washed three times in PBS and incubated in secondary antibodies and DAPI diluted in PBS-MT for 1 hr . Coverslips were washed twice with PBS and twice with deionized water and mounted on glass slides using Prolong Gold Antifade Reagent ( Invitrogen ) . All data are representative of two or more independent experiments with n = 3 or greater . In the majority of experiments involving induction or infection , the independent variables were heavily skewed ( i . e . the variable of Mtb infection vs . variable of genotype ) . Therefore to avoid a type II error , a log transformation was performed on the data prior to analysis . Data were then analyzed by two-way , or three-way ANOVA , followed by Tukey’s post-hoc test to determine significance . Experiments involving one independent variable were analyzed without a log transformation . Here significance was determined using a Student’s two-tailed T test or a one-way ANOVA followed by a Tukey’s multiple comparisons test . For LCM/MS MS significance was determined with a one-way ANOVA followed by a Tukey HSD post-hoc test . A Benjamini-Hochberg correction was used for the false discovery rate . Graphs were generated using Prism ( GraphPad ) . In order to depict baseline and induced gene expression on the same graph , we have broken the y-axis into segments where needed . Error bars represent SEM . For mouse experiments , we estimated that detecting a significant effect requires two samples to differ in CFUs by 0 . 7e^10 . Using a standard deviation of 0 . 35e^10 for each population , we calculated that a minimum size of 5 age- and sex-matched mice per group per time point is necessary to detect a statistically significant difference by a t-test with alpha ( 2-sided ) set at 0 . 05 and a power of 80% . Therefore , we used a minimum of 5 mice per genotype per time point to assess infection-related readouts . For statistical comparison , each experimental group was tested for normal distribution . Data were tested using a Mann-Whitney test .	Parkinson’s disease is a progressive nervous system disorder that causes tremors , slow movements , and stiff and inflexible muscles . The symptoms are caused by the loss of cells known as neurons in a specific part of the brain that helps to regulate how the body moves . Researchers have identified mutations in several genes that are associated with an increased risk of developing Parkinson’s . The most common of these mutations occur in a gene called LRRK2 . This gene produces a protein that has been shown to be important for maintaining cellular compartments known as mitochondria , which play a crucial role in generating energy . It remains unclear how these mutations lead to the death of neurons . Mutations in LRRK2 have also been shown to make individuals more susceptible to bacterial infections , suggesting that the protein that LRRK2 codes for may help our immune system . Weindel , Bell et al . set out to understand how this protein works in immune cells called macrophages , which ‘eat’ invading bacteria and produce type I interferons , molecules that promote immune responses . Mouse cells were used to measure the ability of normal macrophages and macrophages that lack the mouse equivalent to LRRK2 ( referred to as Lrrk2 knockout macrophages ) to make type I interferons . The experiments showed that the Lrrk2 knockout macrophages made type I interferons even when they were not infected with bacteria , suggesting they are subject to stress that triggers immune responses . It was possible to correct the behavior of the Lrrk2 knockout macrophages by repairing their mitochondria . When mice missing the gene equivalent to LRRK2 were infected with the bacterium that causes tuberculosis , they experienced more severe disease . The protein encoded by the LRRK2 gene is considered a potential target for therapies to treat Parkinson’s disease , and several drugs that inhibit this protein are being tested in clinical trials . The findings of Weindel , Bell et al . suggest that these drugs may have unintended negative effects on a patient’s ability to fight infection . This work also indicates that LRRK2 mutations may disrupt immune responses in the brain , where macrophage-like cells called microglia play a crucial role in maintaining healthy neurons . Future studies that examine how mutations in LRRK2 affect microglia may help us understand how Parkinson’s disease develops .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2020	LRRK2 maintains mitochondrial homeostasis and regulates innate immune responses to Mycobacterium tuberculosis
Interactions between epithelial cells and neurons influence a range of sensory modalities including taste , touch , and smell . Vertebrate and invertebrate epidermal cells ensheath peripheral arbors of somatosensory neurons , including nociceptors , yet the developmental origins and functional roles of this ensheathment are largely unknown . Here , we describe an evolutionarily conserved morphogenetic mechanism for epidermal ensheathment of somatosensory neurites . We found that somatosensory neurons in Drosophila and zebrafish induce formation of epidermal sheaths , which wrap neurites of different types of neurons to different extents . Neurites induce formation of plasma membrane phosphatidylinositol 4 , 5-bisphosphate microdomains at nascent sheaths , followed by a filamentous actin network , and recruitment of junctional proteins that likely form autotypic junctions to seal sheaths . Finally , blocking epidermal sheath formation destabilized dendrite branches and reduced nociceptive sensitivity in Drosophila . Epidermal somatosensory neurite ensheathment is thus a deeply conserved cellular process that contributes to the morphogenesis and function of nociceptive sensory neurons . The innervation patterns of cutaneous receptors determine our responses to external stimuli . Many types of cutaneous receptors form specialized terminal structures with epithelial cells that contribute to somatosensation ( Owens and Lumpkin , 2014; Zimmerman et al . , 2014 ) . For example , some low threshold mechanoreceptor afferents form synapse-like contacts with Merkel cells ( Mihara et al . , 1979 ) , which directly respond to mechanical stress and tune gentle touch responses ( Maksimovic et al . , 2014; Woo et al . , 2014 ) . Similarly , afferent interactions with radially packed Schwann cell-derived lamellar cells in Pacinian corpuscles facilitate high frequency sensitivity ( Loewenstein and Skalak , 1966 ) . By contrast , although various types of free nerve endings , including nociceptive C-fibers , course over and insert into keratinocytes , much less is known about the anatomy of keratinocyte-sensory neuron coupling , or the mechanisms by which keratinocytes modulate sensory neuron structure and function . Recent findings that keratinocytes express sensory channels ( Peier et al . , 2002; Bidaux et al . , 2015; Chen et al . , 2016 ) , respond to sensory stimuli ( Koizumi et al . , 2004; Xu et al . , 2006; Moehring et al . , 2018 ) , release compounds that modulate sensory neuron function ( Woolf et al . , 1997; Koizumi et al . , 2004; Moehring et al . , 2018 ) , and can drive sensory neuron firing ( Baumbauer et al . , 2015; Pang et al . , 2015 ) , underscore the importance of understanding the coupling of keratinocytes to sensory neurons . Anatomical studies have demonstrated that peripheral arbors of some mammalian somatosensory neurons insert into keratinocytes , not just intercalate between them ( Munger , 1965; Cauna , 1973 ) . Several factors have hindered characterization of sensory neuron-keratinocyte interactions in mammalian systems , including region-specific differences in sensory neuron-epidermis interactions ( Kawakami et al . , 2001; Liu et al . , 2014 ) , a still-growing inventory of neuronal cell types that innervate the epidermis ( Usoskin et al . , 2015; Nguyen et al . , 2017 ) , and a shortage of markers that label discrete populations of sensory neurons . Peripheral arbors of somatosensory neurons are likewise inserted into epidermal cells in invertebrate and non-mammalian vertebrate model systems , making these promising settings for characterizing epithelial cell-neurite interactions . Notably , portions of Drosophila melanogaster larval nociceptive class IV dendrite arborization ( da ) neuron dendrites and Danio rerio ( zebrafish ) larval trigeminal and Rohon-Beard ( RB ) sensory axons become ensheathed by epidermal cells ( Han et al . , 2012; Kim et al . , 2012; O'Brien et al . , 2012 ) , and studies in these systems have provided insight into the structure and possible function of this epidermal ensheathment of free nerve endings . Drosophila and zebrafish epidermal cells wrap sensory neurites by extending membranes around the entire circumference of the sensory neurite . The wrapping epidermal membranes are tightly apposed to one another and the ensheathed neurites , embedding them inside a mesaxon-like structure ( Whitear and Moate , 1998; Han et al . , 2012; Kim et al . , 2012; O'Brien et al . , 2012 ) . A similar structure has been documented for ensheathed somatosensory neurites in Caenorhabditis elegans and humans ( Cauna , 1973; Chalfie and Sulston , 1981 ) , suggesting that ensheathment by epidermal cells is a conserved feature of sensory endings . The most extensive ultrastructural analysis of these structures suggests that the sensory neurites can be continuously ensheathed over extended lengths of the arbor , stretching several micrometers or more ( O'Brien et al . , 2012 ) . Structurally , the interaction between keratinocytes and somatosensory neurites is reminiscent of ensheathment of peripheral axons by nonmyelinating Schwann cells in Remak bundles , suggesting that keratinocyte ensheathment may likewise regulate sensory neuron structure ( Chen et al . , 2003 ) and function ( Orita et al . , 2013; Faroni et al . , 2014 ) . Although the extent and distribution of sensory neurite-epidermal ensheathment have not been systematically analyzed , many of the documented instances involve highly branched mechanosensory and/or nociceptive neurons . In Drosophila , epidermal ensheathment has been linked to control of branching morphogenesis in two ways . First , nociceptive class IV dendrite arborization ( c4da ) neurons are largely restricted to a two-dimensional plane along the basal surface of epidermal cells to potentiate contact-dependent repulsion and hence tiling ( Han et al . , 2012; Kim et al . , 2012 ) . However , portions of c4da neurons are apically shifted and ensheathed inside the epidermis , allowing for dendrites of other da neurons to innervate the unoccupied basal space and hence ‘share’ the territory ( Tenenbaum et al . , 2017 ) . Second , epidermal ensheathment appears to regulate dendrite branching activity , as mutation of the microRNA bantam , which regulates dendrite-epidermis interactions ( Jiang et al . , 2014 ) , or knockdown of coracle ( cora ) , which encodes a band 4 . 1-related protein required for sheath formation ( Tenenbaum et al . , 2017 ) , each increase dendrite branching . Although these studies provide the first signs that epidermal ensheathement plays key roles in somatosensory neuron development , the cellular basis and functional consequences of this sensory neuron-epidermis coupling remain to be determined . Here , we characterized the cellular events involved in formation of epidermal ensheathment of somatosensory neurites in Drosophila and zebrafish . First , we identified a series of reporters that accumulate at epidermal sites of somatosensory dendrite ensheathment in Drosophila , demonstrating that sheaths form at specialized membrane domains and providing markers for in vivo tracking of the sheaths . Remarkably , epidermal sheaths are labeled by similar markers in zebrafish , suggestive of a conserved molecular machinery for ensheathment . Using these reporters , we found that epidermal sheaths in Drosophila and zebrafish wrap different types of neurons to different extents and that somatosensory neurons are required for formation and maintenance of epidermal sheaths . Finally , we found that blocking epidermal sheath formation led to exuberant dendrite branching and branch turnover , as well as reduced nociceptive sensitivity in Drosophila . Altogether , these studies demonstrate that ensheathment of somatosensory neurons by epidermal cells is a deeply conserved cellular process that plays key roles in the morphogenesis and function of nociceptive sensory neurons . Recent studies have demonstrated that large portions of Drosophila c4da dendrite arbors are ensheathed by the epidermis ( Tenenbaum et al . , 2017; Jiang et al . , 2018 ) . To gain a high resolution view of ensheathment over extended length scales , we subjected Drosophila third instar larvae to serial block-face scanning electron microscopy ( SBF-SEM ) ( Denk and Horstmann , 2004 ) . Consistent with prior TEM studies that provided a snapshot of these sheath structures ( Han et al . , 2012; Kim et al . , 2012; Jiang et al . , 2014 ) , in individual sections we observed dendrites embedded inside epithelial cells and connected to the basal epithelial surface by thin , tubular invaginations formed by close apposition of epidermal membranes ( Figure 1A ) . To determine whether c4da dendrites were continuously ensheathed in these mesaxon-like structures , we followed individual dendrites from the site of insertion into the epidermis through EM volumes of abdominal segments cut into 60-nm sections along the apical-basal axis . We found that dendrites were embedded in epithelial cells over extended distances ( often several microns or more ) , that dendrites were continuously embedded in these mesaxon-like structures with elongated tubular invaginations , and that the epidermal membranes comprising the walls of these tubular invaginations were tightly juxtaposed and electron-dense along their entire length ( Figure 1B and C ) . Each of these structural elements was previously described for the ensheathment of peripheral axons by keratinocytes in zebrafish ( O'Brien et al . , 2012 ) , suggesting that the mechanism of epidermal somatosensory neuron ensheathment may be conserved between invertebrates and vertebrates . We hypothesized that formation of dendrite sheaths likely involves recruitment of factors that create specialized membrane domains . To identify epithelial membrane-associated markers that preferentially localize to sites of dendrite ensheathment , we used the Gal4-UAS system to selectively express GFP-tagged markers in the epidermis of Drosophila larvae also expressing the c4da-specific marker ppk-CD4-tdTomato and assayed for GFP enrichment at sites of dendrite-epidermis apposition . Whereas the single-pass transmembrane marker CD4-GFP broadly labeled epithelial membranes and showed no obvious enrichment at sites of dendrite contact ( Figure 1D and E ) , our screen of ~90 GFP-tagged membrane- and cytoskeleton-associated proteins identified several markers enriched in basal domains of epithelial cells adjacent to c4da dendrites ( Figure 1—figure supplement 1A , Supplementary file 1 ) . First , we screened a collection of membrane markers to determine whether ensheathment occurs at specialized membrane domains . Among these markers , the phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) probe PLCδ-PH-GFP ( Várnai and Balla , 1998; Verstreken et al . , 2009 ) exhibited the most robust enrichment at sites of epidermal dendrite ensheathment . In epithelial cells of third instar larvae , PLCδ-PH-GFP accumulated at epithelial cell-cell junctions , punctate patches , and elongated filamentous membrane microdomains adjacent to c4da dendrites ( Figure 1F–1K ) . These PIP2 microdomains were also labeled by antibodies to the Drosophila 4 . 1 protein cora ( Figure 1—figure supplement 1B ) , a previously described marker of epidermal dendrite sheaths ( Kim et al . , 2012; Tenenbaum et al . , 2017 ) , demonstrating that these PIP2 microdomains correspond to epidermal dendrite sheaths . In addition to labeling epidermal sheaths , anti-cora immunostaining labels glial sheaths , which wrap axons , cell bodies , and proximal dendrite segments of sensory neurons; however , epidermal PLCδ-PH-GFP was not enriched at these sites of glial ensheathment . PLCδ-PH-GFP labeled membrane domains that often appeared wider than c4da dendrites ( Figure 1H–1I ) , suggesting that PIP2 labels the entire sheath structure , including the convoluted tubular extensions to the basal surface of the epidermis . To more systematically analyze whether epidermal PIP2 microdomains mark sites of dendrite ensheathment by epidermal cells , we monitored staining intensity for a surface-exposed neuronal antigen ( HRP ) ( Kim et al . , 2012 ) and the epidermal PIP2 marker PLCδ-PH-GFP simultaneously . The intensity of HRP labeling along dendrites was inversely related to GFP labeling intensity , further suggesting that PLCδ-PH-GFP and hence PIP2 marks sites of neurite ensheathment ( Figure 1—figure supplement 2 ) . Because many of the sheath structures are smaller than the axial resolution of a standard confocal microscope , we used expansion microscopy ( ExM ) to gain a 3-dimensional view of epidermal PLCδ-PH-GFP localization adjacent to c4da dendrites ( Jiang et al . , 2018 ) . We found that PLCδ-PH-GFP labeled epidermal structures that extend from the most apical extent of dendrite insertion to the basal surface of individual epithelial cells ( Figure 1J–1K ) , suggesting that PLCδ-PH-GFP indeed labels the entire sheath structure . PLCδ-PH-GFP was locally depleted at branch points ( Figure 1J , white arrows; Figure 1—figure supplement 1D ) , consistent with prior observations that dendrite branch points are less extensively ensheathed than dendrite shafts ( Tenenbaum et al . , 2017 ) . These sheath structures often appeared to terminate at epidermal cell-cell junctions , where dendrites were displaced to occupy domains basal to junctional domains . Point mutations in the PH domain of PLCδ-PH-GFP that abrogate PIP2 binding ( Várnai and Balla , 1998; Verstreken et al . , 2009 ) prevented accumulation of PLCδ-PH-GFP at sites of ensheathment ( Figure 1—figure supplement 1C ) . Other PIP2-binding proteins , including OSH2-PH-GFP ( Figure 1—figure supplement 1E ) , which binds phosphatidylinositol 4-phosphate and PIP2 with similar affinities ( Hardie et al . , 2015 ) , exhibited similar patterns of accumulation at sheaths . Altogether , these observations demonstrate that epithelial sites of dendrite ensheathment are enriched in PIP2 . PIP2 is a negatively charged phospholipid that recruits a variety of proteins to the plasma membrane to regulate vesicular trafficking and actin remodeling ( De Craene et al . , 2017 ) . We therefore examined whether endocytic , cytoskeletal , and/or phagocytic markers also accumulated at sites of epidermal ensheathment . Although we observed no enrichment of mature phagocytic markers prior to sheath formation or in mature sheaths , we identified a number of PIP2-linked markers that together provide a framework for sheath assembly ( Supplementary file 1 ) . First , we found that a GFP-tagged version of the endocytic adaptor Arf51F/dArf6 was enriched at sites of dendrite ensheathment ( Figure 1L–1M ) . Arf6 regulates clathrin-dependent endocytosis as well as trafficking of recycling endosomes to the plasma membrane ( D'Souza-Schorey and Chavrier , 2006 ) , and the Arf6 effector phosphatidylinositol4-monophosphate 5-kinase catalyzes plasma membrane synthesis of PIP2 ( Honda et al . , 1999 ) . Thus , dArf6 and endocytosis may contribute to PIP2 accumulation at sites of sheath formation . Second , we found that a GFP-tagged version of the GTPase Rho1 , which promotes filamentous actin ( F-actin ) assembly , and the F-actin probe GMA-GFP accumulated at sites of epidermal sheath formation ( Figure 1N–1Q ) , consistent with the fact that PIP2 stimulates actin assembly ( Yin and Janmey , 2003 ) . Finally , in addition to the septate junction marker cora ( Figure 1R–1S ) , which was previously identified as a component of epidermal sheaths ( Kim et al . , 2012; Tenenbaum et al . , 2017 ) , other septate junction markers , including GFP-Neurexin-IV and Neuroglian-GFP , as well as adherens junction markers , including Armadillo-GFP and Shotgun-GFP , Drosophila homologues of β-catenin and E-cadherin , respectively , accumulated at epidermal dendrite sheaths ( Figure 1T–1U , Figure 1—figure supplement 1F , Supplementary file 1 ) . PIP2 binding regulates membrane association of 4 . 1R ( An et al . , 2006 ) and the maturation of adherens junctions via exocyst-dependent recruitment of E-cadherin ( Xiong et al . , 2012 ) , thus PIP2 may promote sheath maturation via recruitment of these proteins . Sensory axon terminals in the epidermis of zebrafish larvae and adults are ensheathed by the apical membranes of epidermal keratinocytes ( Figure 2A ) ( O'Brien et al . , 2012 ) , and ensheathment channels have also been seen in adult fish ( Whitear and Moate , 1998; Rasmussen et al . , 2018 ) . These axonal ensheathment channels are remarkably similar at the ultrastructural level to the sheaths wrapping somatosensory dendrites in Drosophila larvae . To determine whether zebrafish and Drosophila epidermal sheaths are similar at the molecular level , we examined the localization of fluorescent reporters for the membrane , cytoskeleton , and cell junctions in basal zebrafish epidermal cells . At early stages , before sensory axons have grown into the skin , a reporter for PIP2 ( PLCδ-PH-GFP ) localized at cell-cell junctions and sparse microdomains near the apical surface ( Figure 2B ) . After axons grew into the skin , PIP2 was enriched in continuous , linear apical microdomains , closely associated with axons of both larval zebrafish somatosensory neuron cell types , trigeminal and Rohon-Beard neurons ( Figure 2C , H ) . Farnesylated GFP ( CaaX-GFP ) similarly localized to microdomains below axons , consistent with the notion that axons are associated with specialized membrane domains in skin cells ( Figure 2—figure supplement 1A–C ) . Reporters for F-actin ( LifeAct-GFP and Utrophin-GFP ) were also enriched at these axon-associated domains ( Figure 2D and E and data not shown ) . Electron microscopy of zebrafish epidermal sheaths revealed that autotypic junctions appear to seal the ‘neck’ of these sheaths ( O'Brien et al . , 2012 ) . To determine the molecular nature of these junctions , we used α-catenin and E-cadherin in-frame , functional gene traps ( Trinh et al . , 2011; Cronan et al . , 2018 ) ; transiently expressed C-terminal-tagged Desmocolin-like two and Desmoplakin BAC reporters to visualize desmosomes; and a gene trap of Jupa [a . k . a . Plakoglobin/γ-catenin] ( Trinh et al . , 2011 ) , a protein found in both types of junctions . Reporters for both adherens junction and desmosome proteins localized to apical domains directly above axons , suggesting that both types of junctions associate with epidermal sheaths ( Figure 2F–G , I; Figure 2—figure supplement 1D–O ) . Consistent with the observation that autotypic junctions are only visible in some TEM images ( O'Brien et al . , 2012 ) , some of the fluorescent junctional reporters ( α-catenin , Dspa , Jupa ) appeared as dotted lines along the length of axons ( Figure 2G , Figure 2—figure supplement 1J–O ) , suggesting that they form spot junctions , rather than continuous belts . Taken together , our results demonstrate similarity in ultrastructure and molecular composition of Drosophila and zebrafish epidermal sheaths , suggesting that these structures form via an evolutionarily conserved pathway . To determine whether epidermal sheaths are specific to somatosensory neurons in zebrafish , or can occur at any site of axon-basal skin cell contact , we mislocalized axons of another sensory neuron type to the skin . Axons of posterior Lateral Line neurons ( pLL ) are usually separated from the skin by ensheathing Schwann cells , forming a nerve just internal to the epidermis . Treating animals with an inhibitor of the Neuregulin receptor Erbb3b , which is required for Schwann cell development , causes the entire bundle of pLL axons to directly contact the basal membrane of basal skin cells ( Raphael et al . , 2010 ) . This treatment created a notable indentation in the basal membrane , but PLCδ-PH-GFP was not enriched in these domains ( Figure 2J and K ) , indicating either that somatosensory axons can uniquely promote the formation of PIP2-rich microdomains , or that only the apical membranes of basal keratinocytes are competent to form these domains . Next , we examined whether PIP2-rich microdomains formed around all somatosensory neurons or preferentially around particular subsets of somatosensory neurons . In Drosophila larvae , the vast majority of PIP2-positive sheath structures ( 94 . 8 ± 7 . 8% , n = 8 abdominal hemisegments ) were present at sites occupied by sensory dendrites of da neurons ( Figure 3A–3C ) . The few sheaths that were not apposed by dendrites were located directly adjacent to dendrites , suggesting that these sheaths may persist after ensheathed dendrites retracted . Next , to investigate whether different classes of da neurons were differentially ensheathed , we expressed membrane-targeted RFP in different classes of somatosensory neurons and visualized sheaths via epidermal expression of UAS-PLCδ-PH-GFP or anti-cora antibody staining . Among the multi-dendritic da neurons , we found that nociceptive c4da neurons exhibited the most extensive ensheathment , mechanosensitive and thermosensitive c3da and c2da neurons exhibited an intermediate level of ensheathment , and proprioceptive c1da neurons exhibited very little ensheathment ( Figure 3D–3F , Figure 3—figure supplement 1 ) . Thus , different morphological and functional classes of somatosensory neurons are ensheathed by the epidermis to different extents . Although zebrafish somatosensory neurons have not been as clearly categorized into subtypes as Drosophila da neurons , similar to Drosophila , different individual sensory neurons in zebrafish were ensheathed to different degrees ( Figure 3G–3K ) . The degree of ensheathment appeared to correlate with axon arbor complexity: axons with fewer branches associated with α-catenin along a greater proportion of their length ( up to ~80% axon length ) than did highly complex axons ( <30% axon length ) . This observation implies that the degree of axon ensheathment may be a subtype-specific feature in zebrafish , like in Drosophila . As epidermal sheaths occur almost exclusively at sites occupied by sensory neurites , we investigated whether an epidermal pre-pattern dictates sites of sheath formation or , alternatively , whether neuronal signals induce epidermal sheath formation . To differentiate between these possibilities , we first monitored the timing of arrival and distribution of epidermal sheath markers throughout Drosophila larval development . Whereas c4da dendrites tile the larval body wall by ~36 h after egg laying ( AEL ) ( Parrish et al . , 2009 ) , PIP2 first accumulated in isolated patches adjacent to dendrites at 48 h AEL ( Figure 4A–4C and G ) . Epidermal PIP2 did not co-occur with large portions of the dendrite arbor until after 96 hAEL ( Figure 4D–4F and G ) , a time point at which dendrites are internalized in epithelial cells ( Jiang et al . , 2014; Jiang et al . , 2018 ) . Furthermore , time-lapse imaging demonstrated that PIP2 enrichment at sheaths is not transient; once formed , PIP2-positive epidermal sheaths persist or grow , but rarely retract ( Figure 4H , Figure 4—figure supplement 1 ) . Finally , although PIP2 markers and cora extensively co-localized and labeled a nearly identical population of sheaths by the end of larval development ( 95 . 7 ± 5 . 8% of cora-positive sheaths are PIP2-positive; 88 . 7 ± 7 . 4% of PIP2-positive sheaths are cora-positive; n = 8 hemisegments ) , cora accumulation lagged behind PLCδ-PH-GFP ( Figure 4G , Figure 4—figure supplement 2 ) . Thus , although PIP2 accumulation marks an earlier stage in sheath formation than does cora recruitment , we found no evidence that a pre-pattern predicts the site of ensheathment . In the course of our imaging we occasionally observed hemisegments lacking a c4da neuron . In such cases , epidermal PIP2-positive sheath structures were largely absent , although PIP2 accumulation at epithelial cell-cell junctions was comparable to neighboring segments containing c4da neurons ( Figure 4—figure supplement 3 ) . This observation suggested that dendritic signals induce formation of epidermal sheaths . To test the requirement for sensory neurons in epidermal sheath formation , we used a genetic cell-killing assay in Drosophila to eliminate all c4da neurons and assayed for sheath formation using anti-cora immunostaining . Expressing the pro-apoptotic gene reaper ( rpr ) in c4da neurons with two copies of the c4da-specific ppk-Gal4 Gal4 driver ( Grueber et al . , 2003 ) resulted in fully penetrant death and clearance of c4da neurons but not other sensory neurons by the end of the first larval instar , prior to appearance of epithelial sheaths . Anti-cora staining of these larvae revealed that although the overall extent of ensheathment was significantly reduced , levels of ensheathment in c1/c2/c3da neurons were unaffected by this treatment ( cora-positive sheath length per mm2 of body wall: 2 . 72 ± 0 . 64 mm following c4da rpr expression; 11 . 44 ± 1 . 81 mm in sibling controls without rpr; 3 . 18 ± 1 . 16 mm for c1/c2/c3da neurons from sibling controls; mean ±sd , n = 8 ) ( Figure 4I–4K ) . These results demonstrate that dendrite-derived signals induce sheath formation; such signals are likely short-range signals , as sheaths form at sites directly apposed to dendrites . These results further suggest that modality-specific levels of ensheathment do not reflect competitive interactions between c4da and other da neurons for sheath formation , as the absence of c4da neurons did not potentiate sheath formation in spared neurons . Next , we investigated the temporal requirement for dendrite-derived signals in epidermal sheath formation . Using a focused laser beam we ablated Drosophila c2da , c3da , and c4da neurons at 48 h AEL , prior to appreciable accumulation of sheath markers or appearance of sheaths in TEM sections ( Jiang et al . , 2014 ) , and assayed for sheath formation at 120 h AEL using anti-cora immunostaining . Following this treatment , cora-positive sheaths did not form ( Figure 4L–4N ) , suggesting that dendrite signals initiate sheath formation after 48 h AEL , the same timeframe at which PIP2 markers first accumulate at sites of dendrite contact . These results further demonstrate that different neuron classes have different capacities for ensheathment , because removing all of the da neurons that are normally ensheathed did not potentiate c1da neuron ensheathment . To examine whether dendritic signals are likewise required for sheath maintenance , we used a focused laser beam to sever the dorsal-anterior dendrites from a c4da neuron at 108 h AEL , after epidermal sheaths had formed , and used time-lapse confocal microscopy to monitor effects on sheath maintenance in larvae expressing the sheath marker UAS-PLCδ-PH-GFP ( Figure 4O–4Q ) . By 12 h post-severing , both the c4da dendrites distal to the cut site and the epidermal sheaths that wrapped them had disappeared ( Figure 4R–4T ) . By contrast , sheaths wrapping the spared dorsal-posterior portion of the c4da dendrite arbor , as well as sheaths that wrapped c2da/c3da neurons in both the lesioned and unlesioned half of the hemisegment persisted . Therefore , short-range dendrite-derived signals are required both for the formation and maintenance of epidermal sheaths . To determine whether , as in Drosophila , axons are required for formation of epidermal sheaths in zebrafish , we examined sheath-associated reporters in larvae injected with a morpholino targeting neurogenin 1 ( neurog1 ) , a manipulation that blocks somatosensory neuron development ( Andermann et al . , 2002; Cornell and Eisen , 2002; O'Brien et al . , 2012 ) . Basal cells in neurog1 MO-treated animals lacked coherent PIP2-rich microdomains , apical accumulations of F-actin , and α-catenin-containing autotypic junctions , demonstrating that epidermal sheaths are initiated by axons in zebrafish larvae ( Figure 4U–4Z ) . As in Drosophila , axons were also required to maintain sheaths , as PIP2-rich microdomains disappeared soon after laser axotomy and axon degeneration ( Figure 4AA ) . To determine the order of assembly of these sheath-associated proteins , we conducted a series of double-labeling and genetic epistasis analyses in Drosophila larvae . We simultaneously expressed the PIP2 marker UAS-PLCδ-PH-Cerulean together with either the endocytic marker UAS-dArf6-GFP or the F-actin marker UAS-GMA-GFP in the epidermis of larvae additionally expressing the c4da neuron marker ppk-CD4-tdTomato and monitored the timing of arrival of each marker at epidermal sheaths . From the earliest time-point that PIP2 enrichment was detectable at sheaths , we also detected dArf6-GFP enrichment , albeit at a subset of PIP2-positive sheaths , suggesting that dArf6 is recruited to sheaths shortly after PIP2 enrichment ( Figure 5A–5B ) . By contrast , F-actin labeling lagged behind PIP2 ( Figure 5C–5D ) , appearing on a comparable timescale as cora . To directly visualize the stepwise recruitment of sheath components , we labeled epidermal sheaths with the PIP2 marker UAS-PLCδ-PH-Cerulean and assayed for recruitment of GFP-tagged sheath components using time-lapse microscopy . Consistent with our time-lapse imaging of PIP2-positive sheaths ( Figure 4—figure supplement 1 ) , we found that PLCδ-PH-Cerulean labeling persisted at the vast majority of sheaths over a 12 h time-lapse ( Figure 5E–5F ) . However , sheaths that were initially labeled by UAS-PLCδ-PH-Cerulean but not UAS-GFP-cora1-383 , a GFP-tagged fusion protein that mimics endogenous cora localization at epidermal sheaths and septate junctions ( Figure 5—figure supplement 1C ) , were positive for both markers following a 12 h time-lapse ( Figure 5E–5F ) . Similarly , we found that GMA-GFP and dArf6-GFP were recruited to sheaths that were initially PIP2-positive but GFP-negative ( Figure 5F , Figure 5—figure supplement 1A–1B ) . Epidermal sheath assembly therefore appears to proceed via separable steps . Examining ensheathment channel-associated markers at four stages of zebrafish development revealed a similar sequence of events . As in Drosophila , we found that membrane reporters appeared near zebrafish axons before F-actin or junctional reporters ( Figure 5G ) . PIP2-rich microdomains frequently apposed axons by 32 hpf , before ensheathment channels were evident ultrastructurally ( O'Brien et al . , 2012 ) . This observation suggests that the formation of PIP2-positive membrane microdomains is an early step in sheath morphogenesis in zebrafish , as in Drosophila . Indeed , time-lapse confocal microscopy demonstrated that these domains formed during development just minutes after an axonal grown cone passed through that region ( Figure 5—figure supplement 2 ) . To assess the relationship between these sheath-associated proteins , we knocked down lipids or proteins associated with sheaths in Drosophila . Specifically , to deplete phosphatidylinositol 4-phosphate and PIP2 , we expressed RNAi targeting the phosphatidylinositol 4-kinase gene PI4KIIIα; to block endocytosis , we expressed a dominant negative version of shibire ( shiDN ) , which is defective in GTP binding/hydrolysis ( Damke et al . , 2001 ) ; to block septate junction formation , we expressed cora ( RNAi ) in the epidermis . We found that epidermal PI4K ( RNAi ) and shiDN expression severely attenuated PIP2 accumulation at sheaths ( Figure 5H , Figure 5—figure supplement 3 ) . As PLCδ-PH-GFP accumulation precedes dArf6 accumulation at the onset of sheath formation , PIP2 accumulation and endocytic events may engage in feed-forward signaling to promote epidermal sheath formation . By contrast , epidermal cora ( RNAi ) had no effect on PLCδ-PH-GFP accumulation , suggesting that cora accumulation is a downstream event in sheath assembly . Consistent with this notion , both epidermal PI4K ( RNAi ) and shiDN expression blocked cora accumulation at sheaths ( Figure 5H , Figure 5—figure supplement 3 ) , suggesting that cora recruitment to sheaths depends on PIP2 accumulation . PIP2 accumulation and cora accumulation therefore mark genetically separable steps in sheath assembly that we subsequently refer to as initiation and maturation , respectively ( Figure 5I ) . What are the functions of epidermal sheaths that wrap somatosensory neurons ? Prior studies suggested a role for epidermal ensheathment in restricting dendrite branching in Drosophila larvae ( Jiang et al . , 2014; Tenenbaum et al . , 2017 ) . We therefore assayed the requirement in dendrite growth of each of the sheath assembly components we identified in this study . We expressed PI4K ( RNAi ) to reduce epidermal PIP2 levels and monitored effects on c4da dendrite morphogenesis . Compared to controls , epidermis-specific expression of PI4K ( RNAi ) significantly increased the number and decreased the average length of terminal dendrites ( Figure 6A–6B and G–H ) . PLCδ-PH-GFP can function as a competitive inhibitor of PIP2 signaling ( Raucher et al . , 2000 ) , and epidermal PLCδ-PH-GFP expression increased terminal dendrite branch number and decreased dendrite branch length in a dose-dependent manner ( Figure 6—figure supplement 1 ) . Similarly , blocking epidermal endocytosis via constitutive epidermal expression of shiDN or expressing temperature sensitive shits and using it to conditionally blocking epidermal endocytosis specifically in the time window during which dendrites are normally ensheathed led to severe terminal dendrite branching defects qualitatively similar to PI4K ( RNAi ) ( Figure 6C–6D and G–H ) . Finally , epidermal expression of cora ( RNAi ) induced growth of short terminal dendrites ( Figure 6E and G–H ) , as has been previously reported ( Tenenbaum et al . , 2017 ) , as did epidermal expression of shg ( RNAi ) ( Figure 6F–6H ) . Thus , blocking the early or late events of epidermal sheath formation deregulates branching morphogenesis of Drosophila nociceptive c4da neurons . To identify the cellular basis of these dendrite growth defects , we monitored dendrite dynamics in control or sheath-defective larvae using time-lapse microscopy during the time window when sheaths normally form . Over an 18 h time-lapse beginning at 96 h AEL more than 80% of terminal dendrites persisted in control larvae , with the vast majority of those dendrites elongating ( Figure 6I and L ) . By contrast , using epidermis-specific expression of PI4K ( RNAi ) or cora ( RNAi ) to block sheath initiation or maturation , respectively , led to significant alterations in branch dynamics ( Figure 6J–6L ) . First , a larger fraction of terminal dendrites exhibited dynamic growth behavior . Second , the relative levels of growth and retraction were altered; whereas growth predominated in controls , growth and retraction occurred with comparable frequency in PI4K ( RNAi ) and cora ( RNAi ) larvae . Third , the average change in terminal dendrite length was reduced in PI4K ( RNAi ) and cora ( RNAi ) larvae ( Figure 6M ) . These results suggest that epidermal ensheathment alters dendrite growth properties by stabilizing existing terminal dendrites and promoting their elongation . To further test this possibility , we simultaneously labeled epidermal sheaths ( Epi > PLCδ-PH-GFP ) and c4da dendrite arbors ( ppk-CD4-tdTomato ) and monitored terminal dendrite dynamics in ensheathed and unensheathed arbors . Whereas >65% of terminal dendrites were present only transiently during a 12 h time lapse at the onset of ensheathment ( 72–84 h AEL ) , most terminal dendrites persisted during a 12 h time lapse after arbors were extensively ensheathed ( 108–120 h AEL ) ( Figure 6N ) . In this latter time window ( 108–120 h AEL ) we compared the growth dynamics of ensheathed and unensheathed terminal dendrites and found that a significantly higher proportion of ensheathed terminal dendrites were growing or stable over the 12 h time-lapse ( Figure 6O ) . Altogether , our time-lapse imaging results strongly suggest that epidermal sheaths contribute to stabilization of somatosensory dendrites . What is the relationship between epidermal ensheathment and dendrite branching ? While dendrite branch points are occasionally ensheathed ( Figure 1B ) and new branches can be initiated from ensheathed dendrites ( Han et al . , 2012 ) , we found that sheath formation is first initiated on long-lived dendrite shafts in proximal portions of the dendrite arbor rather than the more dynamic distal portions of the dendrite arbor ( Figure 6—figure supplement 2 ) , and that terminal dendrites in general and newly formed terminal dendrites in particular were less extensively ensheathed than other portions of the dendrite arbor ( Figure 6P , Figure 6—figure supplement 3 ) . We therefore monitored the frequency of dendrite branching from ensheathed and unensheathed portions of dendrite arbors during a 12 h time-lapse . Consistent with prior observations ( Han et al . , 2012 ) , we occasionally observed new branch initiation from ensheathed portions of dendrite arbors ( Figure 6Q ) . However , these events were rare and usually occurred at the ends of existing sheaths ( Figure 6—figure supplement 3 ) ; the majority of new branch initiation occurred on unensheathed portions of dendrites . Intriguingly , a large proportion of new branches was formed in the vicinity of epithelial intercellular junctions; whether this is simply a result of discontinuities in sheaths at intercellular junctions or reflects the function of non-autonomous branch-promoting activities associated with junctions remains to be determined . Given that epidermal ensheathment constrains terminal dendrite dynamics in Drosophila , we next examined whether epidermal ensheathment limits structural plasticity of dendrite arbors , as has been suggested ( Parrish et al . , 2009; Jiang et al . , 2014 ) . Embryonic ablation of c4da neurons leads to exuberant dendrite growth in spared neurons beyond their normal boundaries to fill vacated territory ( Grueber et al . , 2003; Sugimura et al . , 2003 ) . This capacity of c4da neurons to expand their dendrite arbors beyond normal boundaries is progressively limited during development , concomitant with the increase in epidermal dendrite ensheathment ( Parrish et al . , 2009; Jiang et al . , 2014 ) . Following ablation of a single c4da neuron at 72 h AEL , the spared neighboring neurons extend their dendrite arbors to cover 13% of the vacated territory , on average ( Figure 6R and U ) . If epithelial ensheathment limits the structural plasticity of c4da dendrite arbors , we reasoned that blocking epithelial sheath formation should potentiate the invasive growth activity of c4da neurons following ablation of their neighbors . Indeed , epidermis-specific PI4K ( RNAi ) or cora ( RNAi ) resulted in a significant potentiation of dendrite invasion ( Figure 6S and U ) . In addition to regulating the growth dynamics and elongation of individual terminal dendrites , these results suggest that epidermal ensheathment contributes to the fidelity of receptive field coverage by coupling dendrite and epidermis expansion . What role , if any , does epidermal ensheathment play in somatosensation ? Having found that nociceptive c4da neurons and proprioceptive c1da neurons were the most extensively and least ensheathed da neurons , respectively , we investigated whether blocking sheath formation affected sensory-evoked behavioral responses regulated by these neurons . Harsh touch activates c4da nociceptive neurons to elicit stereotyped nocifensive rolling responses ( Zhong et al . , 2010 ) , so we monitored touch-evoked rolling responses and rates of larval locomotion in control or sheath-defective larvae as a measure for sheath influence on c4da neuron function . Stimulation with a 78 nM von Frey filament induced nociceptive rolling behavior in >60% of control larvae , whereas c4da-specific expression of the inward rectifying potassium channel Kir2 . 1 strongly attenuated this rolling response ( Figure 7A ) . Compared to controls , epidermal expression of either PI4KIIIa ( RNAi ) to block PIP2 accumulation or PIS ( RNAi ) to reduce phophoinositol biosynthesis , or feeding larvae the cell permeant polyphosphoinositide-binding peptide PBP10 to antagonize PIP2 signaling during the time window of sheath formation significantly attenuated mechanonociceptive behavior ( Figure 7A , Figure 7—figure supplement 1 ) . Epidermal expression of shiDN to block epidermal endocytosis and cora ( RNAi ) to block sheath maturation similarly attenuated mechanonociception . We additionally found that previously reported treatments that block ensheathment including overexpressing α- and β-integrin in c4da neurons to tether dendrites to the ECM ( Han et al . , 2012; Jiang et al . , 2014 ) and mutation of the miRNA bantam ( Jiang et al . , 2014 ) displayed reduced rolling rates in response to von Frey stimuli . Finally , we assayed for effects of ensheathment on larval locomotion . Input from proprioceptive c1da neurons is required for coordinated larval locomotion , and perturbing c1da neuron function severely attenuates larval crawling speed ( Song et al . , 2007 ) . Treatments that reduced epidermal sheath formation did not reduce larval stride length or crawling speed as would be expected for disruption of proprioceptor function , but instead led to increased larval crawling speed ( Figure 7B and data not shown ) . This increased crawling speed was largely the result of reduced turning frequency and a concomitant increase in forward-directed locomotion ( Figure 7C ) , similar to defects in crawling trajectory induced by perturbing c4da function ( Ainsley et al . , 2003; Gorczyca et al . , 2014 ) , further suggesting that ensheathment modulates c4da function . Thus , epidermal ensheathment potentiates nociceptive mechanosensory responses and is apparently dispensable for proprioceptor function , consistent with our observation that nociceptive c4da but not proprioceptive c1da neurons exhibit extensive epidermal ensheathment . Although the signals are not yet known , our studies define key features of the signaling system that drives sheath formation . First , epidermal sheath formation likely relies on short-range , contact-mediated signals involving neuron-expressed ligands and epidermal receptors , as sheaths form exclusively at sites occupied by peripheral sensory neurites . Such a signaling system bears similarity to the C . elegans epidermal SAX-7/L1CAM and MNR-1/Menorin co-ligand complex that interacts with neuronal DMA-1 to regulate patterning of PVD dendrites ( Dong et al . , 2013; Salzberg et al . , 2013 ) . However , whereas PVD dendrites are positioned according to a hypodermal grid of SAX-7/L1CAM expression ( Liang et al . , 2015 ) , the location of epidermal sheaths is dependent on neuron-derived signals rather than an epidermal pre-pattern . Second , different types of neurons have different capacities to induce epidermal sheath formation; in zebrafish , only somatosensory neurons are capable of inducing sheath formation on the apical membranes of basal keratinocytes , and different classes of somatosensory neurons are ensheathed to different degrees in Drosophila and zebrafish . The epidermal sheaths that wrap different types of somatosensory neurons are structurally similar , thus it seems likely that different levels of the sheath-inducing ligand determine the extent of ensheathment much as Nrg1 levels can drive the extent of Schwann cell ensheathment ( Michailov et al . , 2004 ) . Based on the conservation in the molecular machinery of sheath formation , such a ligand and its epidermal receptor are likely conserved in fish and flies . Third , sheath formation is temporally regulated . In both Drosophila and zebrafish , somatosensory neurites innervate the epidermis more than a day prior to sheath formation ( Parrish et al . , 2009; O'Brien et al . , 2012 ) . This may reflect a lack of competence by epithelial cells to ensheath somatosensory neurites as accelerating developmental progression in the Drosophila epidermis leads to precocious dendrite ensheathment ( Jiang et al . , 2014 ) . Finally , our laser severing experiments suggest that peripheral neurites are required to maintain epidermal sheaths . Whether maintenance of sheaths is dependent on a dedicated maintenance signal or simply reflects the absence of morphogenetic signals that would remodel sheaths , for example the exposure by neurites to phosphatidylserine or other engulfment-promoting signals , remains to be determined . Epidermally embedded dendrites that lack identifiable sheath-like structures have been previously described ( Han et al . , 2012 ) ; whether such structures represent cases in which sheaths have been lost or form via a distinct developmental mechanism remains to be determined . The earliest epidermal morphogenetic event we identified downstream of neurite-derived ensheathment signals is the appearance of PIP2-enriched membrane microdomains . How might neurite-derived signals trigger local accumulation of epidermal PIP2 ? Two prominent mechanisms exist to form localized pools of PIP2 in the plasma membrane ( Kwiatkowska , 2010 ) , and each can be triggered by cell-cell contacts . First , PIP2 can be locally clustered via electrostatic interactions with polybasic proteins such as myristoylated alanine-rich C-kinase substrate ( MARCKS ) ( Glaser et al . , 1996; Gambhir et al . , 2004; McLaughlin and Murray , 2005 ) , which additionally binds and cross-links filamentous actin ( Myat et al . , 1997 ) . Protocadherins regulate cortical dendrite morphogenesis in part by maintaining a membrane-associated pool of active MARCKS ( Garrett et al . , 2012 ) , thus protocadherin-based adhesion provides one potential mechanism for localizing MARCKS and hence PIP2 in epidermal cells . Neuronal signals could likewise trigger PIP2 localization via engagement of transmembrane receptors with intracellular domains that electrostatically interact with and cluster PIP2 ( McLaughlin and Murray , 2005 ) or via membrane recruitment of other polybasic proteins such as adducins or GAP43 ( Kwiatkowska , 2010 ) . Second , PIP2 can be locally synthesized , most commonly via phosphorylation of phosphatidylinositol 4-phosphate , and type I phosphatidylinositol 4-phosphate five kinase ( PIP5KI ) can associate with N-cadherin to locally produce PIP2 at sites of N-cadherin adhesion ( El Sayegh et al . , 2007 ) . PIP5KIγ associates with the exocyst via direct interaction with Exo70 to promote membrane targeting of E-cadherin ( Xiong et al . , 2012 ) , thus cadherin-based adhesion can be both a cause and effect of localized PIP2 synthesis . Although we have not found evidence for an epidermal PIP2 pre-pattern that determines sites of ensheathment , PIP5K additionally localizes to focal adhesions to provide a local source of PIP2 ( Ling et al . , 2002 ) . Thus , it will be intriguing to determine whether integrin-based adhesions contribute to epidermal sheath formation by generating local asymmetries in PIP2 levels that get amplified by neuron-derived signals . Plasma membrane enrichment of epidermal PIP2 serves as a critical control point for a variety of cellular processes ( Sun et al . , 2013 ) . Among these , we note remarkable similarities between epidermal sheath formation and the early events of phagocytosis . First , sheath formation and the early stages of phagocytosis appear to involve similar cellular rearrangements , with ensheathing cells and engulfing cells wrapping their targets with membrane protrusions . Second , sheath formation and phagocytosis share a common set of molecular mediators as PIP2 accumulates in nascent epidermal sheaths and in the phagocytic cup of engulfing cells ( Botelho et al . , 2000 ) , as does a network of F-actin ( Scott et al . , 2005 ) . Third , many types of ensheathing cells additionally exhibit phagocytic activity , including Drosophila and zebrafish keratinocytes ( Han et al . , 2014; Rasmussen et al . , 2015 ) , Drosophila ensheathing glia ( Doherty et al . , 2009 ) , and astrocytes ( Chung et al . , 2013 ) . However , whereas PIP2 levels persist at sheaths , PIP2 disappears from the phagosomal membrane during the late stages of phagocytosis ( Botelho et al . , 2000 ) , leading to disassembly of the associated actin network ( Scott et al . , 2005 ) . Similarly , transient accumulation of PIP2 is a feature of endocytosis , cell migration , and other PIP2 regulated morphogenetic events . Thus , reducing PIP2 levels may facilitate phagocytic engulfment of neurites , providing a mechanism for rapid conversion of the epidermal ensheathment channels to engulfment channels . Consistent with prior reports , we found that epidermal ensheathment limits dendrite branching of Drosophila nociceptive c4da neurons ( Jiang et al . , 2014; Tenenbaum et al . , 2017 ) . We also found that the extent of ensheathment is inversely related to peripheral axon branch number in zebrafish somatosensory neurons , suggesting that epidermal ensheathment could similarly regulate neurite branching in vertebrates . This epidermal growth control of peripheral sensory arbors appears to involve two related mechanisms . First , epidermal ensheathment limits dendrite branching; dendrite branching events rarely occur on ensheathed dendrites , and blocking epidermal ensheathment potentiates dendrite branching . This dendrite branching control may reflect a masking of dendrite arbors from substrate-derived signals that promote branching or a steric hindrance of branching . Second , epidermal ensheathment stabilizes existing neurites; blocking epidermal ensheathment potentiates dynamic growth behavior and structural plasticity in Drosophila sensory neurons . Determining whether ensheathment similarly regulates structural plasticity in zebrafish will require development of more and better tools for effectively blocking sheath formation in zebrafish . However , given that the timing of epidermal sheath formation correlates with the developmental restriction in structural plasticity in both Drosophila and zebrafish ( O'Brien et al . , 2012; Jiang et al . , 2014 ) , developmental control of ensheathment appears to be a likely mechanism to stabilize receptive fields of somatosensory neurons . Different types of somatosensory neurons appear to be ensheathed to different degrees . What would be the purpose of such an arrangement ? Many different types of somatosensory neurons innervate overlapping territories , and one recent study suggests that selective ensheathment of particular sensory neuron types facilitates coexistence of different types of sensory neurons in a given territory ( Tenenbaum et al . , 2017 ) . Differential levels of ensheathment may additionally allow for differential coupling of somatosensory neurons to epidermal growth-promoting signals . Likewise , differential ensheathment of somatosensory neuron types may allow different levels of functional coupling of sensory neurons and epidermis . Our finding that nociceptive c4da neurons are the most extensively ensheathed Drosophila somatosensory neurons , and that ensheathment regulates nociceptive sensitivity , suggests that epidermal ensheathment may play a particularly important role in tuning responses to noxious stimuli . Intriguingly , mutations that block ensheathment impair the function of a subset of C . elegans mechanosensory neurons ( Chen and Chalfie , 2014 ) ; whether these mechanosensory impairments are a consequence of ensheathment defects or other effects of the mutations remains to be determined . How might epidermal sheaths influence nociceptive sensitivity ? First , epidermal sheaths may potentiate the functional coupling of epidermal cells to somatosensory neurons . Recent studies suggest that sensory-evoked responses of keratinocytes may modulate sensory neuron function ( Koizumi et al . , 2004; Baumbauer et al . , 2015; Pang et al . , 2015; Moehring et al . , 2018 ) , and epidermal sheaths could provide sites for vesicle release from keratinocytes or direct electrical coupling between keratinocytes and somatosensory neurons . Merkel cells provide a precedent for the former possibility ( Maksimovic et al . , 2013 ) , but whether keratinocytes possess presynaptic release machinery and which neurotransmitters they express remain to be determined . Alternatively , epidermal ensheathment could potentiate nociceptor sensitivity by increasing proximity to stimulus source , by clustering sensory channels , or by some other means . Regardless of the mechanism , our findings that epidermal ensheathment modulates nociceptive sensitivity suggest that defects in epidermal ensheathment could contribute to sensory deficits in human disease . Intriguingly , some forms of peripheral neuropathy exhibit loss of unmyelinated intraepidermal nerves ( Weis et al . , 2011; Üçeyler et al . , 2013 ) ; whether defects in epithelial ensheathment play a role in these sensory neuropathies remains to be determined . Flies were maintained on standard cornmeal-molasses-agar media and reared at 25° C under 12 halternating light-dark cycles . Alleles used in this study are detailed in the Key Resources Table , results from the reporter screen are detailed in Supplementary file 1 and Figure 1—figure supplement 1 . Experimental genotypes are listed in Supplementary file 2 . Larvae were transferred at 72 h AEL to 35 mm dishes containing unmodified cornmeal-molasses agar ( mock ) or cornmeal-molasses agar supplemented with 20 μM PBP10 and assayed for behavior responses and cora immunoreactivity at 120 h AEL . Zebrafish ( Danio rerio ) were grown at 28 . 5°C on a 14 h/10 h light/dark cycle . The following previously described transgenic strains were used: TgBAC ( tp63:GAL4FF ) la213 , Tg ( isl1[ss]: LEXA-VP16 , LEXAop:tdTomato ) la215 ( Rasmussen et al . , 2015 ) , Tg ( isl1:GAL4-VP16 , UAS:EGFP ) zf154 ( Sagasti et al . , 2005 ) , Tg ( isl1:GAL4-VP16 , UAS:RFP ) zf234 ( O'Brien et al . , 2009a ) , Gt ( ctnna-citrine ) ct3a ( Trinh et al . , 2011 ) , Gt ( jupa-citrine ) ct520a ( Trinh et al . , 2011 ) , Tg ( UAS:lifeact-GFP ) mu271 ( Helker et al . , 2013 ) , Gt ( cdh1-tdtomato ) xt18 ( Cronan et al . , 2018 ) , and Tg ( UAS:GFP-CAAX ) pd1025 ( Ellis et al . , 2013 ) . All experimental procedures were approved by the Chancellor’s Animal Research Care Committee at UCLA . To generate BAC reporters for dsc2l and dspa , the corresponding stop codons in BACs CH73-316A13 and CH211-120J4 , respectively , were replaced by a GFP-KanR cassette as previously described ( Suster et al . , 2011 ) . To generate pME-EGFP-PH-PLC , the PH domain of rat PLC1δ1 was PCR amplified from pAA173 ( Kachur et al . , 2008 ) and cloned into pME-EGFP ( Kwan et al . , 2007 ) using the restriction enzymes XhoI and BglII . The pDEST-4xUASnr-EGFP-PH-PLC-pA plasmid was created by Gateway cloning of p5E-4xUASnr ( Akitake et al . , 2011 ) , pME-EGFP-PH-PLC , and p3E-pA ( Kwan et al . , 2007 ) . pDEST-krtt1c19e-EGFP-CAAX-pA was assembled by Gateway cloning of p5E-krtt1c19e ( Rasmussen et al . , 2015 ) , pME-EGFP-CAAX , and p3E-pA ( Kwan et al . , 2007 ) . To create a stable line , one cell stage embryos were injected with pDEST-4xUASnr-EGFP-PH-PLC-pA and tol2 mRNA , raised to adulthood and screened for transgene transmission to the F1 generation . To label lateral line axons , one to four-cell stage zebrafish embryos were injected with 25 pg of a neurod:mTangerine plasmid ( gift from Alex Nechiporuk , Oregon Health and Science University , Portland , OR ) . 200 pg of BAC reporters for dsc2l and dspa were injected at the one to four-cell stage . To block somatosensory neuron development , one cell stage embryos were injected with 1 nl of injection mixture containing an antisense morpholino oligonucleotide targeting neurog1 ( 5’-ACGATCTCCATTGTTGATAACCTGG-3’ ) at a concentration of 0 . 7 mM ( Andermann et al . , 2002; Cornell and Eisen , 2002 ) . Loss of response to touch was monitored to confirm efficacy of the treatment . As a control , embryos were injected with 1 nl of an antisense morpholino that targets an intron of the human beta-globin gene ( 5’-CCTCTTACCTCAGTTACAATTTATA-3’ ) at a concentration of 0 . 7 mM . Antisense morpholino oligonucleotides were synthesized by GeneTools ( Philomath , OR ) . The ErbB receptor antagonist AG1478 was used to perturb repositioning of the pLLn below the epidermis ( Raphael et al . , 2010 ) . Embryos were bathed in embryonic medium containing either 4 μM AG1478/1% DMSO or 1% DMSO as a control . Immunostaining was as above with the following antibodies: mouse anti-GFP , clone 3E6 ( Invitrogen #A11120 , 1:100 ) , rabbit anti-dsRed ( Clonetech #632496 , 1:50 ) , goat anti-mouse Alexa488 ( Thermofisher A31561 , 1:100 ) , donkey anti-rabbit ATTO 565 ( Vaughan lab , 1:10 ) . Following immunostaining , samples were mounted on lysine-coated #1 . 5 cover glass in polydimethylsiloxane wells and incubated in monomer solution ( 2 M NaCl , 8 . 625% sodium acrylate , 2 . 5% acrylamide , 0 . 15% bisacrylamide in PBS ) for 1 h at 4° C prior to gelation . A stock of 4-hydroxy-2 , 2 , 6 , 6-tetramenthylpiperidin-1-oxyl ( 4-hydroxy-TEMPO ) at 1% ( wt/wt ) in water was added to the incubation solution and diluted to a concentration of 0 . 01% . Concentrated stocks of tetramethylethylenediamine ( TEMED ) and ammonium persulfate ( APS ) at 10% ( wt/wt ) in water were added sequentially to the incubation solution and diluted to concentrations of 0 . 2% ( wt/wt ) . The tissues were then incubated at 37°C for 3–4 h . After gelation , the gels were cut and placed in a small 12-well chamber and 1 unit/ml ( 5 mg/ml ) of chitinase in PBS ( pH 6 . 0 ) was used to digest the cuticles for ~4 d at 37°C . Chitinase-treated samples were incubated with 1000 units/ml collagenase solution ( prepared with buffer 1x HBSS lacking calcium , magnesium , and phenol red ) with 0 . 01 M CaCl2 and 0 . 01 M MgCl2 overnight in a 37°C shaking incubation chamber . Samples were then rinsed with PBS twice for 5 min and digested in 8 units/ml proteinase K solution in digestion buffer ( 40 mM Tris pH 8 . 0 , 1 mM EDTA , 0 . 5% Triton , 0 . 8 M Guanidine HCl ) for 1 h at 37°C . Subsequently , samples were removed from the digestion solution and were allowed to expand overnight in a large excess of deionized water . After expansion , the expanded gel was trimmed to fit onto the coverglass , excess water was removed , and the gel was mounted on a lysine-coated cover glass for imaging . Confocal microscopy was performed on a Leica SP5 inverted confocal scanning microscope using a 63 × 1 . 2 NA water lens . Third instar larva were perforated with insect pins and cut open on ice in freshly made fixative ( 2 . 5% glutaraldehyde , 4% paraformaldehyde , 0 . 1 M sodium cacodylate ) . Samples were centrifuged at 15000 x rpm in a microcentrifuge for 1 h and then incubated at 4° C overnight to achieve thorough fixation . Next , samples were washed five times for 5 min each in 0 . 1 M sodium cacodylate and then post-fixed in osmium ferrocyanide for 1 h on ice . The tissues were then washed five times for 5 min each in ddH2O at room temperature and incubated in a 1% thiocarbohydrazide solution for 20 min at room temperature . The samples were washed five times for 5 min each in ddH2O at room temperature and then incubated in 2% osmium tetroxide for 30 min at room temperature . Following another five washes for 5 min each in ddH2O at room temperature , samples were stained en bloc in 1% uranyl acetate at 4° C overnight . The following day , tissues were washed five times for 5 min each in ddH2O at room temperature and stained en bloc in Walton’s lead aspartate for 30 min at 60° C . The samples were then washed five times for 5 min each in ddH2O and dehydrated in an ice cold ethanol series ( 30% , 50% , 70% , and 95% EtOH ) , then transferred to room temperature for 5 min . This was followed by two changes of 100% EtOH and two changes of propylene oxide for 5 min each . The tissues were then infiltrated in a 1:1 mixture of propylene oxide: Durcupan resin , for 2 h at room temperature followed by overnight infiltration in fresh Durcupan . The following day , tissues were given a fresh change of Durcupan for 2 h at room temperature and then placed in flat embedding molds and polymerized in a 60° C oven for 2 days . The blocks were trimmed and imaged using a Zeiss Sigma scanning electron microscope with a Gatan 3-view system at 2 . 5–1 . 7 KV . Stacks ( 1000 sections ) were collected with a 60 nm step size . All image analysis was performed using Fiji ( Schindelin et al . , 2012 ) . The Simple Neurite Tracer plugin ( Longair et al . , 2011 ) was used to trace neurites , ensheathment channels , and cell borders . Only basal cells for which the entire perimeter of the cell was visible were traced . R ( https://www . r-project . org/ ) was used to generate plots and perform statistical tests . Harsh Touch . Larvae were placed in a plastic petri dish with enough water , so larvae remained moist , but did not float in the dish . von Frey filaments made from fishing line and affixed to glass capillaries were applied to the dorsal side of the larvae between segments A3 and A6 until the filament buckled , exhibiting a pre-determined force ( ~78 mN ) . A positive response was scored if one complete nocifensive roll occurred within 10 s of the mechanical stimulus . Datasets were tested for normality using Shapiro-Wilks goodness of fit tests . Details on statistical tests are provided in figure legends . Sample genotypes were blinded for both data acquisition and analysis .	Humans and other animals perceive and interact with the outside world through their sensory nervous system . Nerve cells , acting as the body’s ‘telegraph wires’ , convey signals from sensory organs – like the eyes – to the brain , which then processes this information and tells the body how to respond . There are different kinds of sensory nerve cells that carry different types of information , but they all associate closely with the tissues and organs they connect to the brain . Human skin contains sensory nerve cells , which underpin our senses of touch and pain . There is a highly specialized , complex connection between some of these nerve cells and cells in the skin: the skin cells wrap tightly around the nerve cells’ free ends , forming sheath-like structures . This ‘ensheathment’ process happens in a wide range of animals , including those with a backbone , like fish and humans , and those without , like insects . Ensheathment is thought to be important for the skin’s nerve cells to work properly . Yet it remains unclear how or when these connections first appear . Jiang et al . therefore wanted to determine the developmental origins of ensheathment and to find out if these were also similar in animals with and without backbones . Experiments using fruit fly and zebrafish embryos revealed that nerve cells , not skin cells , were responsible for forming and maintaining the sheaths . In embryos where groups of sensory nerve cells were selectively killed – either using a laser or by making the cells produce a toxin – ensheathment did not occur . Further studies , using a variety of microscopy techniques , revealed that the molecular machinery required to stabilize the sheaths was similar in both fish and flies , and therefore likely to be conserved across different groups of animals . Removing sheaths in fly embryos led to nerve cells becoming unstable; the animals were also less sensitive to touch . This confirmed that ensheathment was indeed necessary for sensory nerve cells to work properly . By revealing how ensheathment first emerges , these findings shed new light on how the sensory nervous system develops and how its activity is controlled . In humans , skin cells ensheath the nerve cells responsible for sensing pain . A better understanding of how ensheathments first arise could therefore lead to new avenues for treating chronic pain and related conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	A conserved morphogenetic mechanism for epidermal ensheathment of nociceptive sensory neurites
Female mosquitoes need a blood meal to reproduce , and in obtaining this essential nutrient they transmit deadly pathogens . Although crucial for the spread of mosquito-borne diseases , blood feeding remains poorly understood due to technological limitations . Indeed , studies often expose human subjects to assess biting behavior . Here , we present the biteOscope , a device that attracts mosquitoes to a host mimic which they bite to obtain an artificial blood meal . The host mimic is transparent , allowing high-resolution imaging of the feeding mosquito . Using machine learning , we extract detailed behavioral statistics describing the locomotion , pose , biting , and feeding dynamics of Aedes aegypti , Aedes albopictus , Anopheles stephensi , and Anopheles coluzzii . In addition to characterizing behavioral patterns , we discover that the common insect repellent DEET repels Anopheles coluzzii upon contact with their legs . The biteOscope provides a new perspective on mosquito blood feeding , enabling the high-throughput quantitative characterization of this lethal behavior . Blood feeding is essential for the reproduction of many mosquito species , and in the process , mosquitoes transmit myriad pathogens to their ( human ) host . Yet , despite being the focal point of pathogen transmission , many aspects of blood feeding remain ill understood . The initial step in obtaining a blood meal , flying toward a host , is relatively well characterized ( Dekker and Cardé , 2011; McMeniman et al . , 2014; van Breugel et al . , 2015 ) . The steps that unfold after a mosquito has landed on a host , however , are much less understood . Once landed , mosquitoes exhibit exploratory bouts during which the legs and proboscis frequently contact the skin ( Jones and Pilitt , 1973; De Jong and Knols , 1995; Clements , 2013 ) . An increasing body of literature reports the presence of receptors involved in contact-dependent sensing on the legs and proboscis ( Sparks et al . , 2013; Matthews et al . , 2019; Dennis et al . , 2019 ) , suggesting that these appendages evaluate the skin surface and thus serve an important role in bite-site selection . Yet , the role and mechanism of contact-dependent sensing in blood feeding is largely unclear ( Benton , 2017 ) . In addition to the body parts that come in contact with the skin surface , the skin piercing labrum also serves as a chemosensory organ , guiding blood feeding in currently unknown ways ( Lee , 1974; Werner-Reiss et al . , 1999; Jove et al . , 2020 ) . In addition to external cues , an animal’s ( internal ) physiology may also affect its behavior . Nutrition , hydration , and pathogen infections , for instance , have been hypothesized to affect blood feeding behavior , for example by altering feeding avidity ( i . e . number of feeding attempts ) or the size of the meal taken ( Rossignol et al . , 1984; Choumet et al . , 2012; Cator et al . , 2013; Vantaux et al . , 2015; Hagan et al . , 2018 ) . These topics , however , remain a matter of debate , due to a lack of ( standardized ) assays to measure mosquito behavior ( Stanczyk et al . , 2017 ) . Quantitative mapping of Drosophila behavior provides an important perspective , suggesting that innovative experimental approaches and computational tools can fuel the acquisition of new insights ( e . g . Branson et al . , 2009; Kain et al . , 2013; Berman et al . , 2014; Corrales-Carvajal et al . , 2016; Robie et al . , 2017; Moreira et al . , 2019 ) . Yet , apart from olfactometers and other flight chambers , very few assays to characterize the blood-feeding behavior of mosquitoes exist ( Geier and Boeckh , 1999; Verhulst et al . , 2011; McMeniman et al . , 2014; van Breugel et al . , 2015; Murray et al . , 2020 ) . Due to this paucity of assays , studies often expose human subjects to quantify the number of landings and/or bites , or the time it takes to complete a blood meal , and score experimental outcomes by hand ( Jones and Pilitt , 1973; Ribeiro , 2000; Moreira et al . , 2009; DeGennaro et al . , 2013; Dennis et al . , 2019; Hughes et al . , 2020 ) . The use of humans as bait constrains the number and type of experiments that can be done ( e . g . prohibiting the use of infected mosquitoes ) and limits the type , detail , and throughput of measurements that can be made . Furthermore , the opaque nature of skin prevents the visualization of the stylets after piercing the skin leaving this aspect of blood feeding almost entirely unstudied , except for one notable study using intravital imaging of dissected mouse skin ( Choumet et al . , 2012 ) and two much earlier descriptions ( Gordon and Lumsden , 1939; Griffiths and Gordon , 1952 ) . To overcome these limitations , we developed the biteOscope , an open platform that allows the high-resolution and high-throughput characterization of surface exploration , probing , and engorgement by blood-feeding mosquitoes . The biteOscope consists of a rudimentary skin mimic: a substrate that attracts mosquitoes to its surface , induces them to land , pierce the surface , and engage in blood feeding . The bite substrate can be mounted in the wall of a mosquito cage allowing freely behaving mosquitoes access . By virtue of its transparent nature , the substrate facilitates imaging of mosquitoes interacting with it , including the visualization of the skin piercing mouthparts of the mosquito . We developed a suite of computational tools that automates the extraction of behavioral statistics from image sequences , and use machine learning to track the individual body parts of behaving mosquitoes . These capabilities enable a detailed characterization of blood-feeding mosquitoes . We demonstrate that the biteOscope is an effective instrument to study the behavior of several medically relevant species of mosquito and describe behavioral patterns of the two main vectors of dengue , Zika , and chikungunya virus ( Aedes aegypti and Aedes albopictus ) , and two important malaria vectors ( Anopheles coluzzii and Anopheles stephensi ) . The biteOscope allows detailed tracking of the complex interactions of mosquitoes with a substrate and can be used to characterize behavioral alterations in the presence of chemical surface patterns . Using this capability , we provide evidence that DEET repels Anopheles coluzzii upon contact with their legs , demonstrating the utility of body part tracking to understand behaviors mediated by contact-dependent sensing . We anticipate that the biteOscope will enable studies that increase our understanding of the sensory biology and genetics of blood feeding , and the effects external ( environmental ) and internal ( physiology ) variables have on this behavior . Given its relevance for pathogen transmission , dissecting the interplay between the mosquito sensory system and host-associated cues during blood feeding is of clear interest , and may suggest new avenues to interfere with blood feeding , and eventually curb pathogen transmission . To allow mosquitoes to engage in blood feeding and feed to full repletion , a device needs to attract mosquitoes , allow them to explore and pierce the surface , and subsequently imbibe a blood meal . To design a tool that can easily be used in a variety of ‘mosquito labs’ ( including ( semi- ) field settings ) , we sought to recapitulate this behavioral sequence using readily available and low-cost laboratory materials . Heat is a dominant factor in short-range mosquito attraction and can be used to attract mosquitoes to a surface and elicit probing behavior ( Healy et al . , 2002; Corfas and Vosshall , 2015; Zermoglio et al . , 2017; Greppi et al . , 2020 ) . We constructed a bite substrate using an optically clear flask filled with water as a controllable heat source ( see Figure 1A ) . An artificial blood meal is applied on the outside of the flask and covered using Parafilm ( a commonly used membrane in laboratory blood feeders ) creating a thin fluid cell on which mosquitoes can feed ( see Figure 1—figure supplement 1 ) . To elicit blood feeding in a transparent medium , we use adenosine triphosphate ( ATP ) as a strong phagostimulant , which , together with an osmotic pressure similar to that of blood and the presence of sodium ions , is sufficient to induce Aedes mosquitoes to feed to full engorgement ( Galun et al . , 1963; Duvall et al . , 2019 ) . Anopheles also require sodium ions and a tonicity similar to blood to feed to full engorgement , but interestingly their feeding rate on artificial meals is independent of ATP ( Galun et al . , 1985 ) . To allow freely behaving mosquitoes access to the bite substrate , we constructed acrylic cages having an opening in the wall or floor where the bite substrate can be mounted . The bite substrate is transparent , facilitating imaging with a camera mounted outside the cage ( Figure 1A shows a schematic of the set up ) . For the majority of data presented here , we used a 4 . 3 × 4 . 3 cm field of view ( see Figure 1C ) which allows up to 15 mosquitoes to explore and feed simultaneously while providing images at a resolution where small body parts like the stylets can easily be resolved . Depending on experimental requirements , the field of view ( and correspondingly assay throughput ) can be much larger at the expense of resolution . Figure 1B , for example , shows a 13 × 13 cm field of view . Individual mosquitoes can be easily tracked at that resolution , yet the visualization of small body parts is challenging . Experiments on Ae . aegypti and Ae . albopictus , both active during the day , were performed using white light illumination; we used an infrared ( IR ) LED array as light source during experiments on An . coluzzii and An . stephensi which were performed in the dark , corresponding to their peak activity during the night . Figure 1B demonstrates that Ae . aegypti mosquitoes show strong attraction to the bite substrate ( surface indicated using a dashed line ) and spend more time on its surface compared to the surrounding wall . Figure 1C–F shows Ae . aegypti undertaking the full blood feeding trajectory on the substrate: starting with surface exploration ( Figure 1C and G ) , piercing of the membrane and insertion of the stylet into the artificial meal ( Figure 1D–F ) , and feeding to full engorgement , as evidenced by the expanded abdomen ( Figure 1E ) . Videos 1 , 2 , 3 and 4 show blood feeding Ae . albopictus , Ae . aegypti , An . stephensi , and An . coluzzii , respectively . Imaging the stylet ( Videos 1 and 5 ) as it evaluates the artificial meal reveals the striking dexterity of the organ as it rapidly bends , extends , and retracts—aspects of feeding that normally remain hidden inside the skin . We created a computational pipeline to extract behavioral statistics from image sequences ( see Figure 1—figure supplement 2 for an overview and Materials and methods for details ) . The position of individual mosquitoes is tracked over time to yield locomotion statistics ( see Figure 1G and Video 6 ) , and select all time slices that make up a single behavioral trajectory ( e . g . landing , exploration , feeding , and take off ) . The error rate of tracking was 0 . 045 ( 5 errors in a validation data set of n=111 tracks , see Materials and methods for details ) with the majority of errors arising from erroneously assigned identities when two mosquitoes cross . Validation videos ( see Video 7 for an example ) make it straightforward to manually correct such errors yielding near-perfect tracking . To determine a mosquito’s engorgement status , we take advantage of the dilation of the mosquito abdomen when it takes a blood meal ( Figure 1E ) . We determine a mosquito’s body shape ( excluding appendages ) using an active contour model to quantify feeding dynamics and engorgement status at each timepoint of a trajectory , and detect full engorgement with a sensitivity of 81% and a specificity of 100% ( see Figure 1 G1-3 , Video 8 , and Materials and methods for details ) . Together with locomotion statistics , engorgement data provides a high-level description of the behavioral trajectory . To assess the capability of the biteOscope to characterize the behavior of different species of mosquito , we performed experiments with the two most important vectors of arboviral diseases ( Ae . aegypti and Ae . albopictus ) and two dominant malaria vectors ( An . stephensi and An . coluzzii , formerly known as Anopheles gambiae M molecular form ) . Figure 2 and Figure 2—figure supplement 1 show locomotion and feeding statistics for the four species . All species land readily on the bite substrate and undertake exploratory bouts leading to full engorgement in 18% , 7% , 4% , and 14% of all trajectories and 46% , 22% , 10% , and 31% of all >10 second trajectories , for Ae . aegypti , Ae . albopictus , An . stephensi , and An . coluzzii , respectively , when offered a meal consisting of 1 mM ATP in phosphate buffered saline ( PBS ) . Figure 2A–D shows summary statistics of 349 behavioral trajectories of An . coluzzii obtained from a total of 1 hr and 15 min of imaging data ( five 15-min experiments with 15 females per experiment ) , demonstrating the throughput of the biteOscope . Figure 2E shows the time spent on the surface versus the distance covered for trajectories that did ( large opaque circles ) and did not ( small transparent dots ) lead to full engorgement for the four species . As expected , rather short trajectories do not lead to engorgement , yet less intuitive is the observation that exploratory trajectories that do not lead to engorgement rarely exceed the duration of successful feeding trajectories ( 8% of non-feeding trajectories takes longer than the mean time to engorge ) . This suggests that a mosquito’s search for blood has a characteristic timescale that is independent of success , and when blood is not found within the time a typical meal takes , the search is aborted . We further explored this observation using individual Ae . albopictus which were offered a bite substrate with a meal of PBS with or without ATP . As PBS alone does not lead to engorgement , mosquitoes offered the PBS only feeder never engorged whereas mosquitoes interacting with the PBS + ATP feeder engorged to full repletion in the majority of cases ( 55% ) . High-resolution trajectory analysis enables us to dissect behavioral patterns that lead to ( non- ) feeding; a trajectory here is defined as landing , the ensuing behavioral sequence , followed by leaving the bite substrate by walking or flying ( see Videos 9 and 10 for two example trajectories ) . The velocity of a mosquito’s centroid can be used to classify locomotion behaviors ( stationary , walking , flight ) with high accuracy ( 89% see Figure 3—figure supplement 1 and Materials and methods for details ) . Figure 3 presents ethograms of Ae . albopictus on these two bite substrates , and in agreement with the data in Figure 2E , shows that trajectories on feeders without ATP ( non-feeding ) have an approximately equal maximum duration as trajectories leading to full engorgement on the feeder with ATP . While mosquitoes do not increase the duration of exploratory trajectories when not feeding to repletion , the number of exploratory bouts mosquitoes undertook on the PBS only substrate was significantly higher compared to the PBS + ATP case ( Wilcoxon rank-sum test p < 0 . 05 ) , resulting in a slightly longer total exploration time ( Figure 3C ) . This suggests that mosquitoes not finding their desired resource increase the frequency with which they initiate searches rather than the duration of individual searches . This observation may be interpreted in the context of the dangers associated with blood-feeding: while on a host , a mosquito runs the risk of being noticed and subsequently killed . When not finding blood , it may therefore be beneficial to abort the search and evacuate from a risky , yet unproductive situation to try elsewhere . The trade-off between exploiting a potential resource versus exploring other options has been shown to depend on the internal state of individuals in other insects ( Katz and Naug , 2015; Corrales-Carvajal et al . , 2016 ) , it is possible that such mechanisms play a role here too . Figure 3 furthermore shows a strong behavioral heterogeneity between individual mosquitoes . While all individuals are from the same mosquito population ( and raised and maintained under identical conditions ) and interact with the same bite substrate , there is a clear heterogeneity in the number of times a mosquito visits the surface ( Figure 3C , middle panel ) , the amount of time she spends exploring the surface ( Figure 3C , left panel ) , and the behaviors they engage in . Automatic classification of locomotion behaviors , shows that some individuals often land on the surface to engage in short interactions , while other individuals undertake much longer trajectories . These long trajectories , in turn , vary in the amount of stationary versus locomotion behaviors . The richness of these data highlight the potential of the biteOscope to quantitatively characterize the intricacy of individual behaviors hidden in population averages . We next turned to body part tracking to acquire a more detailed description of behavioral trajectories . Body part tracking is powerful to address a variety of questions , for example by determining points of surface contact of specific appendages , or to estimate the pose of an animal , which when tracked over time can be translated into a behavioral sequence . We used a recently developed deep learning framework , DeepLabCut ( Mathis et al . , 2018 ) , to train a convolutional neural network ( CNN ) to detect the head , proboscis , abdomen , abdominal tip , and six legs of Ae . aegypti and Ae . albopictus . Due to their morphological similarity , the same CNN can be used to track the body parts of both Aedes species with a mean accuracy of 11 pixels ( 275 micrometer , see Materials and methods for details ) in a 4 . 3 × 4 . 3 cm field of view . Tracking stylet insertions into the artificial meal during probing and feeding using DeepLabCut was challenging , and therefore not included . Figure 4A–C shows body part tracking results of Ae . albopictus and reveals the choreography of three distinct behaviors . Anterior grooming is characterized by circular motion of the forelegs followed by the proboscis , while the middle legs remain stationary ( see Figure 4—video 1 ) . During walking , the tips of all six legs oscillate along the body axis while the proboscis explores laterally ( see Figure 4—video 2 ) , while during probing , the fore and middle legs pull toward the body and the proboscis remains stationary ( see Figure 4—video 3 ) . Inference is done on raw images and the obtained coordinates thus subject to movement of the mosquito . To correct for this , the coordinates are translated and rotated to align along the body axis taking the abdominal tip as the origin . Figure 4D–I shows time series of the obtained egocentric coordinates and their corresponding wavelet transforms . The three behaviors each are associated with distinct periodic movements: smooth periodic motion of the forelegs during anterior grooming ( x , and y coordinates ) , punctuated oscillations along the body axis during walking ( x coordinate ) , and faster jerky movement during probing ( x , and y coordinate of forelegs , y coordinate of middle legs ) . These trajectories can be used in concert with locomotion and body-shape features as inputs for behavioral classification algorithms . The data outputted by our computational pipeline is ideally suited for classification in either a supervised ( e . g . Kain et al . , 2013; Kabra et al . , 2013 ) or unsupervised ( e . g . Berman et al . , 2014; Marques et al . , 2018; Calhoun et al . , 2019; Tao et al . , 2019 ) approaches ( see Figure 4—figure supplement 1 ) . The biteOscope provides an alternative for current methods using human subjects or mice to study mosquito blood feeding . The elimination of the need for a human subject opens new avenues of research , for example allowing blood-feeding studies with pathogen-infected mosquitoes , enabling precise surface manipulations and characterization of the associated behavior , and facilitates the use of high-resolution imaging and machine-learning-based image analysis . Through these innovations , the biteOscope increases experimental throughput and expands the type of experiments that can be performed and measurements that can be made . We developed computational tools that allow the behavioral monitoring of mosquitoes at an unprecedented level of detail . Behavioral research on other animals , including fruit flies ( Werkhoven et al . , 2019; Pereira et al . , 2019 ) and zebrafish ( Marques et al . , 2018; Johnson et al . , 2020 ) shows that high spatiotemporal resolution data describing the posture of animals can be very informative to dissect behavioral trajectories and compare behavioral statistics across individuals and experimental treatments . While the details of computational approaches differ , a common theme is the two dimensional embedding of a high-dimensional representation of an animal at a given time point ( e . g . body part coordinates and derived features ) , data points in two dimensions can subsequently be clustered to reveal behavioral classes ( see Figure 4—figure supplement 1 for an illustration of this concept using tSNE to embed the data represented in Figure 4 ) . Translating such advances in computational ethology to mosquito research is a very promising avenue for future research . We used the biteOscope to describe behavioral patterns of four medically relevant mosquito species and anticipate that such datasets will provide a useful ‘behavioral baseline’ for future studies quantifying the effect of a mosquito’s physiology on blood feeding behavior . The role of pathogen infections is particularly interesting in this respect , as infections may alter feeding behavior , for example by affecting the structural integrity of the salivary glands or other tissues , or inducing systemic change through the immune system or infection of neural tissues ( Rossignol et al . , 1984; Girard et al . , 2007; Cator et al . , 2013; Turley et al . , 2009 ) . A quantitative understanding of such behavioral alterations , however , is lacking . Gaining such insights is of high epidemiological relevance , as mathematical models suggest that ( pathogen induced ) changes in bite behavior can have important implications for pathogen transmission ( Cator et al . , 2014; Abboubakar et al . , 2016 ) . In addition to pathogen-induced behavioral changes , there are many other promising lines of inquiry , including the behavioral influence of the microbiome ( Dickson et al . , 2017 ) , which , in other insects such as Drosophila , influences locomotor behavior ( Schretter et al . , 2018 ) and food choice ( Leitão-Gonçalves et al . , 2017; Wong et al . , 2017 ) . Drosophila research furthermore shows interesting examples of collective behaviors mediated by for example olfaction or direct contact between animals ( Schneider et al . , 2012; Ramdya et al . , 2015; Lihoreau et al . , 2016; Ramdya et al . , 2017 ) , it would be interesting to explore if mosquitoes also take advantage of collective intelligence when searching for food or avoiding noxious stimuli . Tools enabling high-throughput behavioral monitoring may also be useful to characterize population intervention strategies aimed at curbing pathogen transmission , such as Wolbachia infected Ae . aegypti , or Anopheles genetically engineered to be refractory to P . falciparum infection . Quantifying the behavioral effects of such interventions is an important step toward assessing the competitiveness of engineered mosquitoes in the field . As the biteOscope enables novel high-throughput experiments with a variety of mosquito species , we anticipate that it will prove useful for the characterization of various behaviors relevant to pathogen transmission . By tracking the individual body parts of An . coluzzii , we discovered that they are repelled by DEET upon leg contact—a mechanism that may work in concert with other ways in which DEET prevents anopheline mosquitoes to locate humans . Our findings regarding An . coluzzii are in agreement with observations in Ae . aegypti which are also repelled by DEET upon leg contact ( DeGennaro et al . , 2013; Dennis et al . , 2019 ) . However , in contrast to An . coluzzii , olfactory neurons of Ae . aegypti are activated by volatile DEET ( Davis and Rebert , 1972; Boeckh et al . , 1996; Stanczyk et al . , 2010 ) and Ae . aegypti has been reported to avoid volatile DEET in recent studies ( Stanczyk et al . , 2013; Afify and Potter , 2020 ) ( in contrast , an earlier study reported attraction of Ae . aegypti by DEET [Dogan et al . , 1999] ) . Together , these observations suggest that contact-based repellency may be conserved across Anopheles and Aedes mosquitoes and thus may be a potentially interesting target for the design of new repellents . It is less clear , however , what degree of conservation exists for the olfactory modes of action , as the only study comparing the olfactory effects of volatile DEET on Anopheles and Aedes mosquitoes in the same assay , suggests that the former is not repelled at all by volatile DEET , while the latter showed moderate repulsion ( these behavioral responses may be concentration dependent ) ( Afify and Potter , 2020 ) . This observation , together with the observation that volatile DEET activates olfactory neurons in Ae . aegypti while it does not seem to do this in An . coluzzii , suggest that volatile DEET may modulate the response of olfactory neurons to attractive stimuli ( ‘scrambling of the odor code’ [Pellegrino et al . , 2011] ) and/or trigger repulsion in Ae . aegypti , while these mechanisms seem less appropriate for An . coluzzii . In addition to effects on olfactory signaling , DEET has also been suggested to decrease the amount of volatile odorants emanating from hosts through chemical interactions between DEET and the odorants resulting in the masking of a host ( Afify et al . , 2019 ) . As in this scenario the amount of attractive odorants reaching a mosquito is reduced , it may affect the behavior of a variety of species . The observation that both Ae . aegypti and An . coluzzii avoid DEET upon leg contact , while the effects of volatile DEET may partly overlap and partly differ , may guide efforts aimed at uncovering the underlying molecular mechanisms . Our results highlight the use of body part tracking in assigning roles to the various sensory appendages the mosquito body has . The recent surge in genetic tools available to manipulate mosquitoes is shedding light on the genetic elements that mediate pathogen transmission relevant behaviors ( Matthews et al . , 2019; Ingham et al . , 2020; Raji et al . , 2019; Greppi et al . , 2020 ) . Combining such molecular level insights with detailed behavioral tracking and chemical surface patterning , may enable a deep understanding of how contact-dependent sensing drives blood feeding , and other important phenotypes such as insecticide resistance and egg laying preferences . When studying animal behavior in the lab a trade-off exists between the level of experimental control and detail of observation on the one hand , and an accurate representation of natural conditions and behaviors on the other . In case of the biteOscope , an engineered bite substrate opens up a variety of possibilities including surface modifications and high-resolution imaging impossible on human skin , yet the bite substrate does not offer the full set of cues ( and thus behavioral responses ) a human host would . It would therefore be interesting to add more human-associated cues , for instance using materials that resemble the texture of skin , or by coating the bite substrate with attractive human odorants ( Okumu et al . , 2010 ) . In addition to more closely mimicking human hosts by presenting olfactory stimuli , surface coatings could be used to dissect the role of contact-dependent gustatory behaviors on the skin surface in bite site selection . It is important to note that many of the factors that may change behavior mentioned above ( e . g . infections/nutritional status or components of the microbiome ) are best assessed in a relative manner , for example comparing non-infected to infected individuals . Comparing cohorts of mosquitoes undergoing different experimental treatments puts less emphasis on the absolute attractiveness of the bite substrate and thus mitigates potential issues related to the fact that a synthetic bite substrate is likely less attractive than a real live host . We took advantage of the possibility to elicit engorgement on a transparent meal to facilitate imaging . It seems feasible to add a dye to the meal to provide visual cues to the mosquito without interfering with image quality . Using whole blood , however , is challenging in the current system . It would therefore be worthwhile to explore the use of microfluidics to incorporate blood flow into the bite substrate while maintaining optical access . A recent study took advantage of the biteOscope to quantify stylet contact with artificial meals ( Jove et al . , 2020 ) , combining such efforts with artificial vasculature presents exciting opportunities to characterize the role of the stylets in the search for blood . The biteOscope is designed with a variety of possible users in mind . It has a relatively modest price tag ( 900–3500 USD depending on the configuration ) , uses readily available materials and components , and when disassembled fits in a backpack—characteristics we hope will facilitate adoption . Beyond the lab , we foresee interesting applications of the behavioral tracking of mosquitoes in ( semi- ) field settings , and expect that innovative tools that provide high-quality quantitative data will enable discoveries in this space . We anticipate that the techniques and computational tools presented here will provide a fresh perspective on mosquito behaviors that are relevant to pathogen transmission , and enable researchers to gain a detailed understanding of blood feeding without having to sacrifice their own skin . The mosquito species/strains used in this study are described in Key resources table . Larvae were hatched and reared in water at a density of approximately 200 larvae per liter on a diet of fish food . Adult mosquitoes were maintained at 28⁢°C , 75% relative humidity , and a photoperiod of 12 hr light : 12 hr dark in 30 × 30 × 30 cm screened cages having continuous access to 10% sucrose . Prior to experiments , mosquitoes were deprived of sucrose for 6–12 hr while having access to water . Mosquitoes aged 6–25 days old were used for behavioral experiments . Experiments using Ae . aegypti and Ae . albopictus were performed during light hours , while experiments with An . stephensi and An . coluzzii were performed during dark hours . Mosquitoes had no access to water during experiments . A full list of components necessary to build the biteOscope is available in Appendix 1—table 1 . Depending on the experimental requirements , several components can be easily adapted ( e . g . cage geometry or bite substrate ) or replaced by more economical alternatives ( e . g . imaging components ) . All image processing and downstream analysis code was written in Python 3 and is available from Github ( https://github . com/felixhol ) . Raw images were background subtracted , thresholded , and subjected to a series of morphological operations to yield binary images representing mosquito bodies of which the center of mass was determined using SciPy ( Virtanen et al . , 2019 ) . The Crocker–Grier algorithm ( Crocker and Grier , 1996 ) was used to link the obtained coordinates belonging to an individual mosquito in time using trackPy ( Allan et al . , 2016 ) . The obtained tracking data is used to select all images that make up a single behavioral trajectory ( e . g . landing , exploration , feeding , and take off ) and store cropped image sequences centered on the focal mosquito . In addition to the computationally extracted data described below , such image sequences can also be used for the manual annotation of other events ( e . g . stylet insertion as done in Jove et al . , 2020 ) . We verified the tracking results of 111 individual trajectories across 12293 images resulting in an error rate of 0 . 045 ( 5/111 ) . The validation dataset includes data from both Aedes and Anopheles experiments and consists of images having a variety of densities ranging from 0 . 05 to 0 . 4 mosquitoes per cm2 . The most common error ( 4/5 ) is caused by wrongly assigning the identity of two mosquitoes that cross ( e . g . an individual moving over another one and thus overlapping in the image ) . Interestingly , the validation videos ( e . g . Video 7 ) make it straightforward to correct such errors by manually re-assigning the correct identity to the track . A rather minor amount of manual interventions therefore results in nearly perfect tracking .	Scientists often sacrifice their own skin to study how mosquitos drink blood . They allow mosquitos to bite them in laboratory settings so they can observe the insects’ feeding behavior . By observing blood feeding , scientists hope to find ways to prevent deadly diseases like malaria , which is transmitted by bites from mosquitos carrying the malaria parasite . These studies are not only unpleasant for the volunteers , they also have important limitations . For example , it is too risky to use pathogen-infected mosquitos that could make the volunteers sick . A device called the biteOscope developed by Hol et al . may give scientists and their skin a reprieve . The device has a transparent skin-like covering that attracts mosquitos and supplies them an artificial blood meal when they bite . The device captures high-resolution images of the insects’ behavior . It is small enough to fit in a backpack when disassembled , costs about $900 to $3 , 500 US dollars , and is suitable for use in the laboratory or in the field . Using machine-learning techniques , Hol et al . also developed an automated system for analyzing the images . The researchers tested the device on four types of disease-transmitting mosquitos . In one set of experiments , Anopheles mosquitos were recorded interacting with a biteOscope partially coated with an insect repellent called DEET . The images captured by the biteOscope showed that the mosquitos are attracted to the warm surface and land on the part coated with DEET . But when their legs come in contact with the repellent , they leave . The biteOscope provides scientists a new way to study blood feeding , even in mosquitos infected with dangerous pathogens . It might also be used to test new ways to prevent mosquitos from biting and spreading disease . Because the device is portable and relatively inexpensive , it may enable larger studies in a variety of settings .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "tools", "and", "resources", "neuroscience" ]	2020	BiteOscope, an open platform to study mosquito biting behavior
Eukaryotes have two types of spliceosomes , comprised of either major ( U1 , U2 , U4 , U5 , U6 ) or minor ( U11 , U12 , U4atac , U6atac; <1% ) snRNPs . The high conservation of minor introns , typically one amidst many major introns in several hundred genes , despite their poor splicing , has been a long-standing enigma . Here , we discovered that the low abundance minor spliceosome’s catalytic snRNP , U6atac , is strikingly unstable ( t½<2 hr ) . We show that U6atac level depends on both RNA polymerases II and III and can be rapidly increased by cell stress-activated kinase p38MAPK , which stabilizes it , enhancing mRNA expression of hundreds of minor intron-containing genes that are otherwise suppressed by limiting U6atac . Furthermore , p38MAPK-dependent U6atac modulation can control minor intron-containing tumor suppressor PTEN expression and cytokine production . We propose that minor introns are embedded molecular switches regulated by U6atac abundance , providing a novel post-transcriptional gene expression mechanism and a rationale for the minor spliceosome’s evolutionary conservation . Small nuclear RNAs ( snRNAs ) are a class of non-coding RNAs ( ncRNAs ) that function in pre-mRNA splicing ( Wahl et al . , 2009 ) and in mRNA length regulation ( telescripting ) ( Kaida et al . , 2010; Berg et al . , 2012 ) . The vast majority of introns ( >200 , 000 ) in complex eukaryotes are spliced by the major spliceosome ( Wahl et al . , 2009 ) , consisting of U1 , U2 , U4 , U5 and U6 snRNPs . However , ∼700 minor introns , also known as U12 introns , are spliced by the much less abundant ( <1% ) minor spliceosome , consisting of U11 , U12 , U4atac , U5 and U6atac snRNPs ( Hall and Padgett , 1996; Tarn and Steitz , 1996; Patel and Steitz , 2003 ) . Despite their low occurrence , minor introns , typically one amidst many major introns , are found in genes essential for diverse cellular processes ( Burge et al . , 1998; Sheth et al . , 2006 ) . Their splice site sequences , position and the genes in which they reside are all highly conserved ( Basu et al . , 2008; Turunen et al . , 2012 ) . However , previous data suggested that minor introns are generally poorly spliced , and therefore the purpose of the minor splicing pathway has been a long-standing mystery , as it has been difficult to rationalize their remarkable conservation and function in splicing alone . The importance of the minor spliceosome has been recently underscored by reports that mutations in U4atac cause microcephalic osteodysplastic primordial dwarfism type I ( MOPD I ) or Taybi-Linder syndrome ( TALS ) ( Abdel-Salam et al . , 2011; Edery et al . , 2011; He et al . , 2011 ) . We have previously shown that perturbations in the repertoire of both major and minor snRNPs cause widespread transcriptome and splicing defects in spinal muscular atrophy ( Zhang et al . , 2008 ) . Using a ncRNA microarray that we devised and mRNA sequencing ( RNA-Seq ) , we discovered that U6atac , the minor spliceosome’s low abundance catalytic snRNA , also has a very short half-life and can be rate limiting . We found a wide variation of minor intron dependence on U6atac level and that mRNA production from ∼200 minor intron-containing genes are strongly suppressed in HeLa cells under normal growth conditions . Importantly , we show that U6atac level can be rapidly up-regulated by activated p38MAPK , which results in enhanced splicing of minor introns and increased expression of these mRNAs . Among these is the tumor suppressor , PTEN , which we show buffers cell response to stress by modulating TNF-α and IL-8 cytokine production . We propose that minor introns are control switches that are regulated by U6atac abundance , providing a novel gene expression regulation mechanism and explaining the retention of the minor splicing pathway during evolution . A deeper understanding of ncRNAs’ diverse and fundamental functions requires quantitative , high-throughput , and systematic profiling in various cell types and physiological conditions . Here , we designed a custom array with probes complementary to 129 human ncRNAs , including snRNAs , rRNAs , snoRNAs , scaRNAs , tRNAs and others ( Supplementary file 1A ) , but excluding miRNAs for which commercial microarrays exist ( Davison et al . , 2006 ) . Total RNAs were directly labeled , alleviating the need for amplification and reducing the distortion often caused by this step ( Hiley et al . , 2005 ) . Hybridization conditions were optimized and the reproducibility between biological replicates is shown in Figure 1—figure supplements 1 and 2 . Our method allows for the rapid measurement of many ncRNAs over a large dynamic range with high sensitivity and specificity , making it a powerful tool for global analysis of ncRNA levels . The ncRNA microarray was used here to monitor global changes in the levels of ncRNAs after transcription and translation inhibition and cell cycle arrest . Interestingly , Pol II transcription inhibition by Actinomycin D ( ActD ) , as well as by the structurally and chemically unrelated Pol II inhibitor 5 , 6-Dichloro-1-β-D-ribofuranosylbenzimidazole ( DRB ) , led to a rapid and strong decrease in the levels of three ncRNAs transcribed by Pol III: U6atac and U6 snRNAs , and vault RNAs ( Figure 1A ) . Previous studies have shown co-dependence of Pol III transcription on Pol II , which binds upstream of Pol III at the promoter of these ncRNAs ( Listerman et al . , 2007 ) . The decrease in vault RNAs after ActD treatment has been previously observed in mouse cells ( Kickhoefer et al . , 2001 ) , and U6 was noted to have a shorter half-life compared to other major snRNAs ( Fury and Zieve , 1996 ) . However , this is the first report of U6atac’s fast turnover . We have not found significant sequence similarity among these ncRNAs to explain their fast turnover . Real-time qPCR validated the short half-life of U6 and U6atac , but not U5 , for which the microarray did not predict a change ( Figure 1B ) . In comparison , HeLa cells arrested in mitosis using nocodazole , which causes transcription by all three types of polymerases to cease , showed a moderate U6atac decrease , whereas translation inhibition with cycloheximide ( CHX ) caused no significant changes in the levels of the ncRNAs tested ( Figure 1A ) . In cells , U6atac mostly exists as part of U4atac/U6atac di-snRNP or U4atac/U6atac/U5 tri-snRNP complexes . To test whether U6atac instability is rate limiting for U4atac/U6atac di-snRNP formation , we immunoprecipitated the di-snRNP specific protein p110/SART3 and measured the associated U4atac/U6atac by real-time PCR in control or ActD-treated cells . This showed that ActD decreased U4atac level by ∼20% in total RNA , whereas the U4atac/U6atac di-snRNP formation was reduced by >80% ( Figure 1—figure supplement 3 ) . These results demonstrate the fast turnover and vulnerability of U6atac to transcriptional inhibition , which makes it rate limiting for di-snRNP formation . 10 . 7554/eLife . 00780 . 003Figure 1 . U6atac is an extremely unstable ncRNA . ( A ) The heat map summarizes data for HeLa cells treated with inhibitors of several cellular processes , including the transcription inhibitors Actinomycin D ( ActD ) for 2–6 hr or DRB for 1–8 hr , the cell cycle inhibitor nocodazole for 18 hr , and the translation inhibitor cycloheximide ( CHX ) for 16 hr . A color scale bar for heat maps ( −10 to +10 fold change ) is indicated . ( B ) Real time qPCR verification of U5 , U6 and U6atac snRNA level after ActD treatment . Absolute snRNA levels were measured in triplicates and normalized to 5S rRNA . Error bars represent standard deviations of three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 00310 . 7554/eLife . 00780 . 004Figure 1—figure supplement 1 . Hybridization of total RNA and in vitro transcribed RNA standards . Hybridization of total RNA and in vitro transcribed RNA standards . ( A ) Representative microarray image of total RNA from HeLa cells , in vitro transcribed U11 and U12 snRNA . Total RNA or in vitro transcribed snRNAs of the same RNA sample , labeled with Alexa647 ( red ) and Alexa546 ( green ) , were hybridized at 48°C . Yellow signifies an overlap of green and red fluorescence . Dotted circles indicate the probe locations for U11 and U12 snRNAs . ( B ) Total RNA labeled with Alexa647 was hybridized at six different amounts: 50 , 100 , 250 , 500 , 1000 and 2500 ng . The heat maps are shown for all probes on the array with signals twofold above background . The 250 ng sample signals were set at 1 ( black ) while the range covered −10 fold ( green ) to +10 fold ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 00410 . 7554/eLife . 00780 . 005Figure 1—figure supplement 2 . ncRNA microarray is highly reproducible . Total RNA from three HeLa cell cultures were labeled and hybridized as described in ‘Materials and methods’ . The average normalized signals from the three experiments were plotted along with their standard deviations that are presented as error bars . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 00510 . 7554/eLife . 00780 . 006Figure 1—figure supplement 3 . U6atac , not U4atac , is rate limiting for di-snRNP association . ( A ) U4atac level was measured by real time PCR ( absolute quantification and normalized to 5S ) in total RNA extracted from cells treated with actinomycin D or anisomycin for 4 hr . ( B ) U4atac and ( C ) U6atac association with di-snRNP specific protein p110/SART3 was measured after immunoprecipitation of SART3 or control IgG from cells treated as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 006 To determine the effects of U6atac inactivation on the transcriptome , we used an antisense morpholino oligonucleotide ( AMO ) that targets the 5′ end of U6atac snRNA to rapidly inhibit its activity in cells ( Konig et al . , 2007 ) . RNase H protection using an antisense DNA probe complementary to the same sequence confirmed the binding of the AMO to U6atac snRNP and its efficient interference following an 8 hr transfection with 15 nmole of U6atac AMO ( Figure 2A ) , which was used for all subsequent experiments . The specificity and efficiency of the AMOs were verified by RT-PCR showing that U6atac AMO , but not control AMO or U6 AMO , resulted in inhibition of minor intron splicing ( Figure 2B ) , whereas U6 AMO inhibited splicing of only the major introns ( Figure 2C ) . 10 . 7554/eLife . 00780 . 007Figure 2 . U6atac inactivation shows a wide range of minor intron splicing dependence on U6atac level . ( A ) RNase H protection of U6atac snRNA in total cellular RNA after 8 hr treatment with U6atac AMO showing efficient U6atac protection at 15 nmole U6atac AMO . The same RNA sample ( 15 nmole U6atac AMO ) was used for RT-PCR of representative minor ( B ) and major ( C ) introns . The diagrams to the right of the gel indicate spliced or unspliced RNA . Arrows show position of the primers used . ( D ) Representative view on UCSC genome browser for two genes containing highly efficient minor introns ( red reads ) . Numbers on the left represent peak reads count . Gene structures are depicted in blue boxes ( exons ) and lines ( introns ) . ( E ) Same as in ( D ) for low efficiency minor intron . ( F ) The splicing index is plotted for each expressed minor intron in HeLa cells . The dashed arrows point to minor introns with the lowest ( <twofold ) and highest ( >fourfold ) splicing indices . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 00710 . 7554/eLife . 00780 . 008Figure 2—figure supplement 1 . Functional knockdown of U6atac affects a large number of genes . ( A ) Scatter plot comparing the FPKM values representing differential expression analysis from Cufflinks between U6atac AMO and Control AMO . ( B ) Volcano plot of fold change vs significance for genes in U6atac AMO and Control AMO samples . Red dots indicate significant genes ( p<0 . 05 ) , whereas blue dots show genes having no changes . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 008 To identify transcriptome and splicing changes that would be most readily detected in nascent transcripts , we metabolically labeled RNAs in HeLa cells with 4-thiouridine for the last 2 hr of AMO treatment , then selected and sequenced only nascent polyadenylated transcripts produced in the time window during which U6atac was functionally inhibited . More than 120 million RNA-seq reads were obtained for each sample , with ∼70% of these mapping to 9799 and 8397 genes in control and U6atac AMO-treated cells , respectively ( Table 1 ) . Cufflinks ( Trapnell et al . , 2012 ) was used to assemble transcripts , estimate their abundance , and score differential expression . 10 . 7554/eLife . 00780 . 009Table 1 . Summary of the RNA-seq statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 009Control AMOU6atac AMOPlatformIllumina HiSeq2000Illumina HiSeq2000Read length100 base100 baseSingle/paired endssingle endsingle endTotal number of reads120 , 274 , 499121 , 794 , 728Mapped reads81 , 198 , 95385 , 645 , 962Expressed genes ( FPKM > 1 ) 97998397Upregulated genes ( > fourfold ) ND3Downregulated genes ( < fourfold ) ND2088Expressed U12-dependent genes ( FPKM > 1 ) 429350 Focusing first on predicted minor introns ( ∼700 ) in the minor intron database ( U12DB ) ( Alioto , 2007 ) , we detected 429 that passed the threshold for expressed genes of fragments per kilobase per million mapped reads ( FPKM ) > 1 in control AMO . Interestingly , 206 ( 48% ) of these were expressed at low levels in control and severely down-regulated ( FPKM < 1 ) in U6atac AMO , suggesting that limiting amount of U6atac suppresses splicing and expression of these genes even under normal growth conditions , and that their unspliced pre-mRNAs are very unstable . On the other hand , unspliced , yet stable pre-mRNAs for the remaining 223 introns were detected . To rank these minor introns according to their relative dependence on U6atac level , we calculated the splicing index for each intron as the ratio of Y/X whereas: X = the reads count in a given intron normalized to the reads count in the surrounding exons in control AMO . Y = the reads count for the same intron normalized to the reads in the surrounding exons in U6atac AMO . This showed that 44 ( 20% ) out of these 223 have little to no intronic reads in control but significant intron accumulation in U6atac AMO ( Figure 2D; CHD4 , ARPC5 ) . For these minor introns , the level of U6atac snRNP is sufficient for splicing catalysis under normal growth conditions ( control AMO ) , but they are extremely sensitive to U6atac decrease . In contrast , 81 ( 36% ) minor introns were poorly spliced , with their unspliced pre-mRNAs clearly detectable in control and showing minimal additional response to U6atac inactivation ( Figure 2E; MTBP ) . For these introns , U6atac snRNP is limiting under normal growth conditions and they are predicted to splice more efficiently when U6atac level increases . Taken together , these data show a strict dependence on U6atac abundance for minor intron splicing , which actually varied over a wide range ( >20-fold; Figure 2F ) and thus challenges the accepted view that minor introns typically splice poorly . This wide range could not be explained by analysis of the splice sites ( AT-AC vs GT-AG ) or their surrounding sequences ( data not shown ) . A list of minor introns with highest and lowest U6atac-dependent splicing indices is shown in Supplementary file 1B . Comparison of mRNA levels ( Table 1 and Figure 2—figure supplement 1 ) showed that U6atac AMO caused >twofold down-regulation in 2088 genes , including all 429 expressed minor intron genes . Only three genes were up-regulated . Splicing analysis revealed 657 alternative splicing changes resulting from U6atac inactivation ( Figure 3A ) , including all 223 minor introns that showed various levels of splicing inhibition . Importantly , the major introns in these same transcripts were spliced normally . In a small number of minor intron-containing genes ( <20 ) U6atac inactivation elicited alternative major intron splicing from cryptic 5′ and 3′ splice sites around the minor introns ( e . g . , C11ORF10 and ZNF207; Figure 3B ) . This resulted in mRNA isoform switching without overall transcript level down-regulation . Interestingly , while these effects are likely due to misplicing as a result of U6atac inhibition , they nevertheless resemble the previously described twintrons that utilize either the minor or major spliceosome by switching between adjacent U11-U12 or U1-U2 splice sites , and play an important role in Drosophila development ( Scamborova et al . , 2004 ) , but has not been previously identified in human cells . U12 AMO , tested on select minor introns ( Figure 3 ) , had the same effects as U6atac AMO , confirming that the U6atac AMO effects we observed result from the minor spliceosome inhibition . Although the effect of U6atac inactivation on expression level and alternative splicing in ∼2000 genes that have no minor introns could be secondary , it nevertheless reflects a significant and immediate biological phenomenon . We conclude that U6atac decrease down-regulates or causes splicing change in all minor intron genes and that minor spliceosome inhibition reverberates widely throughout the transcriptome . 10 . 7554/eLife . 00780 . 010Figure 3 . Genome-wide analysis of the minor splicing pathway shows widespread transcriptome changes . ( A ) A summary of the splicing effects in the U6atac AMO treated cells identified using MISO . ( B ) Representative minor intron-containing genes showing alternative splicing pattern changes after U6atac knockdown . Exon junction reads are presented on top . Red junction reads are unique to U6atac AMO sample . RT-PCR confirmations for U6atac AMO as well as U6 and U12 knockdown for comparison are also shown . Arrows show position of the primers used , which are located in the exons . ( C ) Representative genes that do not contain a minor intron showing alternative splicing pattern changes after U6atac inactivation . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 010 As many of the minor intron-containing genes that were significantly affected by U6atac level change have key roles in growth regulation and cell stress signaling ( Supplementary file 1B ) , we explored the possibility that U6atac level might change in response to specific cell stress . For this we tested the effect of anisomycin , a potent activator of stress-induced protein kinase , p38MAPK ( Bunyard et al . , 2003; Yong et al . , 2010 ) , and observed a ∼twofold increase in U6atac level within 4 hours ( Figure 4A ) . Because anisomycin activates both p38MAPK and JNK , we pre-treated cells with either SB203580 , an inhibitor of p38MAPK activation ( Davies et al . , 2000 ) , or SP600125 , an inhibitor of JNK ( Heo et al . , 2004 ) . While SB203580 blocked the anisomycin-mediated U6atac increase , SP600125 had only a small effect ( Figure 4A ) , indicating that U6atac increase is predominantly due to p38MAPK activation . To assess whether this increase is transcriptional or post-transcriptional , cells were co-treated with ActD and anisomycin . In contrast to the drastic U6atac decrease after transcription inhibition with ActD ( 82% ) , co-treatment with anisomycin resulted in only about half that amount of U6atac decrease ( ∼40% ) ( Figure 4B ) , suggesting that cell stress-induced U6atac increase is due , at least in part , to U6atac stabilization . Interestingly , the U6atac increase correlated with enhanced minor intron splicing and gene expression of a splicing reporter ( LucI-minor ) whose expression is strictly dependent on splicing of a minor intron , showing a twofold increase of its expression after anisomycin treatment ( Figure 4C ) . In contrast , intronless and major intron-containing reporters ( Younis et al . , 2010 ) were not affected , indicating that the increased expression is due to enhanced minor intron splicing and not an effect on transcription or translation . A similar U6atac increase was observed in Jurkat T-cells and THP-1 monocytes following anisomycin treatment ( Figure 4D ) , indicating that U6atac increase by activated p38MAPK is a general phenomenon and not limited to HeLa cells . 10 . 7554/eLife . 00780 . 011Figure 4 . p38MAPK cell signaling up-regulates U6atac level and minor intron splicing . ( A ) Real time qPCR of U6atac snRNA after treatment with 1 μg/ml anisomycin ( 4 hr ) , a potent p38MAPK activator , in HeLa cells . Cells were treated with 10 μM of p38MAPK inhibitor , SB203580 , or JNK inhibitor , SP600125 , 30 min prior to anisomycin . ( B ) HeLa cells were treated with 5 μg/ml Actinomycin D ( ActD ) or ActD plus 1 μg/ml anisomycin ( Aniso ) or DMSO as control for 4 hr , followed by measurement of U6atac level . ( C ) Luciferase activity of the various splicing reporters after cell treatment with DMSO or anisomycin for 4 hr . ( D ) Real time qPCR of U6atac snRNA after treatment with anisomycin ( 4 hr ) in Jurkat T cells and THP-1 monocytes . Error bars represent the standard deviation of at least three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 011 Similarly , p38MAPK activation with anisomycin increased minor intron splicing and mRNA levels of several endogenous genes that contain minor introns ( e . g . , PTEN and E2F2; Figure 5A ) . Splicing and mRNA levels of eIF3K , whose minor intron splices with high efficiency in control cells , as well as GAPDH and actin , used as controls because they have no minor introns , were unaffected ( Figure 5 ) . Importantly , anisomycin also enhanced splicing and expression from genes whose levels are suppressed in HeLa under normal growth conditions ( Figure 5B; e . g . , BRMS1L and E2F6 ) . These data are consistent with the idea that limiting U6atac suppresses the expression of several hundred minor intron-containing genes , which can be up-regulated by U6atac increase . 10 . 7554/eLife . 00780 . 012Figure 5 . p38MAPK activation increases expression of minor intron-containing genes . ( A ) Real time qPCR for a representative set of minor introns with low splicing indices with or without 1 μg/ml anisomycin treatment . All primer sets span the splice junction and thus measure spliced mRNAs . ( B ) Real time qPCR for representative minor intron-containing genes that are normally suppressed in HeLa cells . Error bars represent the standard deviation of three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 012 To determine if the p38MAPK–induced up-regulation of minor intron splicing depends on U6atac increase , we examined , as an example , the splicing of PTEN’s minor intron , which increased by up to fivefold after 4 hr of anisomycin treatment ( Figure 6A ) . This increase was completely blocked after co-treatment with U6atac AMO , even at a low amount ( 1 nmole ) that only partially inactivates U6atac ( Figure 6A ) , confirming that p38MAPK up-regulation of minor intron-containing mRNAs expression is strictly reliant on its ability to increase U6atac level . 10 . 7554/eLife . 00780 . 013Figure 6 . U6atac level regulates PTEN minor intron splicing and buffers cytokine production in response to p38MAPK activation . ( A ) Real time qPCR for PTEN in HeLa cells transfected with control or U6atac AMO followed by treatment with DMSO or 1 μg/ml anisomycin for 4 hr . ( B ) and ( C ) Real time qPCR for TNF-α and IL-8 after 4 hr of anisomycin treatment in HeLa cells . ( D ) Real time qPCR for PTEN in HeLa cells transfected with control or PTEN siRNA followed by treatment with DMSO or anisomycin for 4 hr . ( E ) and ( F ) Real time qPCR for TNF-α and IL-8 after 4 hr of anisomycin treatment in HeLa cells transfected with control or PTEN siRNA for 48 hr . Error bars represent the standard deviation of three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 013 We next asked if U6atac level plays a role in specific physiological downstream effects of p38MAPK activation . Whereas p38MAPK activation increases transcription of several key cell physiology modulators , including up-regulation of the cytokines TNF-α and IL-8 ( Figure 6B , C ) , which do not contain minor introns ( Chang et al . , 2006 ) , production of these cytokines has been shown to be suppressed by PTEN , a minor intron-containing gene ( Furumoto et al . , 2006; Furgeson et al . , 2010 ) . We therefore determined if U6atac inactivation affects p38MAPK-induced TNF-α and IL-8 . As shown in Figure 6B , C , U6atac inactivation increased the levels of p38MAPK-mediated TNF-α and IL-8 production to a much higher level , >twofold compared to control AMO , suggesting that a minor intron-containing gene ( s ) suppresses production of these cytokines . To test if this is due to loss of PTEN suppression , PTEN was knocked down by siRNA ( Figure 6D ) . This resulted in the same effect on p38MAPK-induced TNF-α and IL-8 production as U6atac AMO ( Figure 6 , compare panels 6E and 6F to 6B and 6C ) , indicating that PTEN expression , which is regulated by U6atac level , plays a role in p38MAPK-induced cytokine production . We note , however , that while PTEN mRNA level was increased by 2–5 fold we have only been able to detect a small ( ∼20% ) increase of PTEN protein level by western blot . Nevertheless , previous reports have shown that even very small changes in PTEN level are sufficient to have a strong phenotype ( Alimonti et al . , 2010 ) . Importantly , our findings establish a role for U6atac in regulation of major signaling pathways . The findings we describe reveal a novel mechanism for gene regulation as well as have important implications for cell signaling , and they provide a new perspective to explain the evolutionary conservation of the minor spliceosome . Previous studies using a small number of model minor introns have led to the perception that minor introns generally splice at a slower rate than major introns and could therefore be rate limiting for the production of mRNAs that contain them ( Patel et al . , 2002; Singh and Padgett , 2009 ) . However , there has been no evidence that minor introns could be regulated . Here we discovered that U6atac is highly unstable relative to other snRNAs and its level can be rapidly up- or down-regulated . Furthermore , modulation of U6atac amount produces a corresponding change in the amount of mRNA from many minor intron-containing genes , thus providing a novel post-transcriptional mechanism for regulating minor intron splicing and the expression of the genes that contain them , as depicted in Figure 7 . Interestingly , in growing HeLa cells ∼50% of minor intron-containing genes are suppressed by the limiting U6atac level and produce very low amounts of mRNA . Many others are strongly down-regulated with U6atac decrease because a retained minor intron causes the pre-mRNA to degrade or switch alternative splicing to make a different isoform . However , their splicing and expression can be strongly and rapidly enhanced by an increase in U6atac level , at least in part due to its stabilization by p38MAPK ( Figure 7 ) . Thus , the apparent sluggishness observed for some minor introns can be explained by the rarity of U6atac , rather than an intrinsic inefficiency of minor introns . Furthermore , the rapid turnover of U6atac after attenuation of either RNA polymerases II or III makes the minor spliceosome a real time sensor of transcriptional activity in cells . 10 . 7554/eLife . 00780 . 014Figure 7 . Minor introns are embedded molecular switches regulated by U6atac abundance . Splicing of minor introns , typically one amidst several major introns , is dependent on the limiting level of U6atac snRNP in cells . Low abundance and fast turnover of U6atac results in incompletely spliced pre-mRNAs that are either degraded or switch into spliced isoforms that do not require the minor spliceosome . While transcription attenuation rapidly lowers U6atac level and limits minor intron-containing gene’s expression , activated p38MAPK rapidly stabilizes U6atac , increasing its level and enhancing minor splicing and production of full-length mRNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 00780 . 014 The increased amount of U6atac following p38MAPK activation indicates that U6atac and the many minor intron-containing genes whose expression depends on sufficient U6atac are unexpected downstream targets of p38MAPK . These genes , including PTEN , E2F2 , E2F6 and MTBP , play important roles in cell stress physiology such as regulating cytokine production . PTEN , phosphatase and tensin homolog , is a tumor suppressor that plays key roles in cell growth and apoptosis ( Tamguney and Stokoe , 2007 ) . We show that U6atac and the minor splicing pathway , through regulation of PTEN expression , play an antagonistic role that buffers p38MAPK-induced production of TNF-α and IL-8 , two major inflammatory response mediators . TNF-α also functions in a wide range of processes including cell proliferation , differentiation , apoptosis , lipid metabolism and coagulation . IL-8 is a chemo-attractant and a potent angiogenic factor . While activated p38MAPK has been shown to phosphorylate the transcription factor ATF2 , which in turn activates PTEN transcription ( Shen et al . , 2006; Qian et al . , 2012 ) , our data show that PTEN expression can also be regulated post-transcriptionally by the efficiency of splicing of its minor intron . This illustrates a physiological role for minor splicing pathway in regulating cell signaling response . It has been difficult to rationalize the conservation of minor introns and the minor spliceosome on the basis of splicing alone , as with slight sequence variation this function could have simply been relegated to the major spliceosome . Indeed , despite their early evolutionary origin , minor introns and snRNPs have been lost at multiple points during eukaryotic evolution ( Bartschat and Samuelsson , 2010 ) , suggesting that , in contrast to the major spliceosome , they are not absolutely essential . We propose a new perspective on the conservation of minor introns and their splicoesome , that they function as a specialized post-transcriptional mechanism to regulate expression of their host genes . Minor intron-containing genes that typically contain a single minor intron amidst many major ones function in diverse cellular processes that are critical for cell growth and organism development ( Otake et al . , 2002; Alioto , 2007; Abdel-Salam et al . , 2011; Edery et al . , 2011; He et al . , 2011 ) . Fine-tuning the level of the catalytic snRNP , U6atac , allows for a circuit design based on the capacity to completely shut off or rapidly up-regulate the production of the full-length mRNAs from a pool of pre-mRNAs in which all the other major introns have been spliced , without having to affect their transcription ( Figure 7 ) . Indeed , as our RNA-seq show , a single minor intron is sufficient to regulate the expression of an entire pre-mRNA . The conservation of this design principle over >500 million years ( Burge et al . , 1998; Shukla and Padgett , 1999; Russell et al . , 2006 ) demonstrates the effectiveness of this gene regulation mechanism . Thus , minor introns function as control switches that are embedded in hundreds of genes and regulated by U6atac abundance , providing a rationale for evolutionary conservation of the minor spliceosome . Total RNAs were prepared from HeLa cells using Trizol ( Invitrogen , Grand Island , NY ) . RNA standards ( U11 and U12 ) were transcribed using T7 MegaScript Kit ( Ambion , Grand Island , NY ) and gel purified . A modified protocol of the ULYSIS nucleic acid labeling kits ( Invitrogen ) was used to label RNAs with Alexa546 or Alexa647 . Total RNA ( 1 μg ) was combined with 3 pmol each of standard E . coli tRNALys , Escherichia coli tRNAVal , and Saccharomyces cerevisiae tRNAPhe and resuspended in ULYSIS labeling buffer ( 5 mM Tris , 1 mM EDTA , pH 8 . 0 ) . The RNAs were denatured at 95°C for 5 min , added to either Alexa546 or Alexa647 ULYSIS dye and incubated at 90°C for 10 min . G25 spin columns ( GE Healthcare , Pittsburgh , PA ) were used to remove excess dye . The labeled RNAs were then precipitated with ethanol for use in microarray hybridization . Two to four 60-mer DNA probes complementary to different regions of the ncRNAs were designed with a GC content between 35% and 55% and limited secondary structure . The probes were spotted on glass slides in quadruplicate at 200 μM on Matrix II slides ( Full Moon Biosystems , Sunnyvale , CA ) and cross-linked by UV irradiation . Before use slides were pre-hybridized in 2X SSC , 0 . 2% SDS , 0 . 1% BSA for 20–30 min , rinsed in water twice and then dried . Labeled total RNA samples were resuspended in Oligo Hyb Buffer ( Full Moon Biosystems , Sunnyvale , CA ) containing 10 μg polyA RNA and 20 μg salmon sperm DNA . Samples were applied to the hybridization chamber of the HS4800 Pro hybridzation station ( Tecan , Morrisville , NC ) . The following program for hybridization was used: Wash 1 ( 54°C , wash 30 s , soak 30 s , 3 runs ) , two-step Sample Injecion ( 54°C ) , Hybridization ( 54°C , 16 hr ) , Wash 1 ( 54°C , wash 30 s , soak 30 s ) , Wash 1 , 2 and 3 ( 23°C , wash 30 s , soak 30 s ) , Slide drying ( 30°C , 5 min ) . Wash buffers are Wash 1 ( 2X SSC , 0 . 2% SDS ) , Wash 2 ( 2X SSC ) , and Wash 3 ( 0 . 2X SSC ) . Microarray slides were imaged using GenePix 4000b scanner ( Axon Instruments , Molecular Devices , Sunnyvale , CA ) . The fluorescence intensities were quantified and background was subtracted using the GenePix Pro 6 . 0 software . Values for the 4 replicate spots on each slide were averaged . Several hybridization temperatures were tested using total RNA and in vitro transcribed snRNAs showing reliable detection of 5 ng of spiked-in U11 and U12 snRNAs without significant cross-hybridization at temperatures ranging from 48°C to 54°C ( Figure 1—figure supplement 1 ) . Fluorescence from non-specific probes ( red dots ) was <10% of the signal of the specific probes , demonstrating the specificity of this array . Similar results were obtained for the measurements of other in vitro transcribed ncRNAs , including U1 , U2 , U4 , U5 , U6 , U7 , U85 , U90 , U93 and hTR ( human telomerase RNA ) . The reproducibility between biological replicates was determined on total RNA isolated from three separate HeLa cell cultures , which were labeled with both Alexa546 and Alexa647 and hybridized on the array . For those probes with detectable signals at least twofold above background , the intensity for each probe normalized to the total signal on the chip did not show significant variation among the three biological replicates ( Figure 1—figure supplement 2 , error bars ) , indicating that the array can reliably discern ncRNA abundance between replicate samples . The dynamic range of the array was determined by testing the linearity of the signals from the different probes with varying amounts of RNA . Figure 1—figure supplement 1B shows a range of 50–2500 ng of total RNA labeled with Alexa647 that was hybridized in the presence of 200 ng of Alexa546-labeled total RNA . Most probes exhibited linearity over the range in which experiments are typically performed ( 100–500 ng total RNA ) . snRNA-specific primers and 5S rRNA-specific primers ( endogenous control ) were used to generate cDNAs using Transcriptor First-strand cDNA synthesis kit ( Roche Applied Sciences , Indianapolis , IN ) from an input of 100 ng of total RNA according to manufacturer’s instructions . Two and a half percent of the cDNA generated was used for each qPCR reaction ( Applied Biosystems 7500 Fast Real-time PCR system ) using SYBR Green dye chemistry . The same reverse primers were used for both RT and qPCR . Each reaction was performed in triplicate . For real-time qPCRs and RT-PCRs used to measure splicing of minor and major introns ( Figures 2–6 ) , 1 μg total RNA was converted to cDNA using the VILO kit ( Invitrogen ) according to manufacturers recommendations . cDNA was then diluted to 10 ng/μl and 20–50 ng cDNA was then used as input for qPCR ( with SYBR Green as described above ) or regular PCR using platinum Taq ( Invitrogen ) . For qPCR , all the primer sets for various genes span the splice junction and thus measure only spliced mRNAs . For RT-PCR , the primers are located in the surrounding exons and thus amplify both spliced and unspliced mRNAs . HeLa cells were grown in DMEM supplemented with 10% fetal bovine serum ( FBS ) , 1% antibiotics and maintained at 37°C in a 5% CO2 humidified atmosphere . Actinomycin D ( 5 μg/ml ) , cycloheximide ( 10 μg/ml ) , DRB ( 100 μM ) , nocodazole ( 200 nM ) , or anisomycin ( 1 μg/ml ) were added to the media for the indicated time , followed by RNA extraction as indicated above . Luciferase splicing reporters were generated as previously described ( Younis et al . , 2010 ) . Briefly , the minor intron of the CHD4 gene was inserted at nucleotide position 571 of firefly luciferase gene . The luciferase protein is destabilized using both a PEST protein degradation sequence as well as a CL1 sequence , and mRNA is destabilized by adding five tandem AUUUA repeats into the 3′UTR . Both intron-containing and intronless luciferase were transcribed from a CMV promoter . Control ( 5′-CCT CTT ACC TCA GTT ACA ATT TAT A-3′ ) and U6atac ( 5′-AAC CTT CTC TCC TTT CAT ACA ACA C-3′ ) antisense morpholinos ( AMOs ) were transfected into cells using the Neon system ( Invitrogen ) . 6 hr post transfection , cells were labeled with 200 μM 4-thiouridine ( 4-SU ) for an additional 2 hr . Total RNA was extracted using Trizol ( Invitrogen ) followed by extraction of polyA-containing RNA on oligo-dT columns using Oligotex kit ( Qiagen , Germantown , MD ) . For isolation of 4-SU-labeled nascent transcripts , EZ-Link biotin-HPDP ( Thermo Scientific , Waltham , MA ) was reacted with the 4-SU and purified on streptavidin Dynabeads ( Invitrogen ) as previously described ( Dolken et al . , 2008 ) . Nascent transcripts prepared as described above were used to prepare cDNA libraries using Encore NGS Library System I ( Nugen , San Carlos , CA ) according to the manufacturer’s recommendations . Briefly , 100 ng RNA was converted to cDNA and amplified using the Ribo-SPIA technology . The cDNA was then fragmented , end repaired , ligated to Illumina adaptors and bead purified according to instructions in the Ovation RNA-seq System ( Nugen ) . Sequencing was performed at the University of Pennsylvania Core Facility on Illumina Hi-Seq2000 platform to generate single end 100 base reads . Raw and processed data is available on GEO under the accession number GSE48263 . This link can be used to retrieve the data: http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE48263 . RNA-seq reads were aligned to the reference genome ( UCSC , hg19 ) using Tophat by default parameters ( Trapnell et al . , 2009 ) . Uniquely mapped reads and the best of multiple hits were kept for downstream analysis . BEDTools software ( Quinlan and Hall , 2010 ) was used to generate reads per exon , from which FPKM ( read fragments per kilobase of exon model per million mapped reads ) values were calculated ( Mortazavi et al . , 2008 ) . The analysis of differential gene expression was performed using Cufflinks ( Trapnell et al . , 2010 ) . Significant genes were detected based on FDR level at 0 . 05 using Benjamini-Hochberg correction for multiple-testing ( Trapnell et al . , 2012 ) . To identify differentially expressed splicing isoforms across samples and quantify the expression level of those alternative spliced genes , MISO ( version 0 . 4 . 1 with default parameters ) was applied to the aligned RNA-seq data ( Katz et al . , 2010 ) . Briefly , MISO estimates expression at alternative splicing event level by computing PSI ( Percent Spliced Isoform ) and measures the differential expression by Bayes factors . The MISO results were filtered for alternative splicing events using the following criteria: ( a ) at least 15 inclusion read , ( b ) 15 exclusion read , such that ( c ) the sum of inclusion and exclusion reads is at least 30 , and ( d ) the ΔΨ is at least 0 . 30 , and ( e ) the Bayes factor is at least 10 , and that ( a ) – ( e ) are true in one of the samples . Total cell extract was prepared from AMO-transfected cells using 10 mM Tris-HCl pH 7 . 5 , 2 . 5 mM MgCl2 , 100 mM NaCl and 0 . 1% NP-40 . RNase H along with 5 μM antisense DNA oligonucleotide for U6atac ( 5′-TCA TAC AAC AC-3′ ) was added for 25 min at 30°C , and RNA was purified and analyzed by northern blotting with a U6atac snRNA probe ( 5′-CCG TAT GCG TGT TGT CAG GCC CGA GGG CCT-3′ ) .	The central dogma of biology states that genetic material , DNA , is transcribed into RNA , which is then translated into proteins . However , the genes of many organisms have stretches of non-coding DNA that interrupt the sequences that code for protein . These non-coding sequences , which are called introns , must be removed , and the remaining sequences—which are called exons—must then be joined together to produce a messenger RNA ( mRNA ) transcript that is ready to be translated into protein . The process of removing the introns and joining the exons is called splicing , and it is carried out by a molecular machine called the spliceosome . However , in addition to containing typical ( ‘major’ ) introns , several hundred human genes also contain a single ‘minor’ intron , and a minor spliceosome is needed to remove it . Minor introns occur in many highly conserved genes , but they are often inefficiently spliced . This means that the resulting mRNA transcripts may not be translated into proteins—which is puzzling given that these proteins perform important roles within the cell . The major and minor spliceosomes are composed of proteins and small non-coding RNA molecules ( which , as their name suggests , are never translated in cells ) . Now Younis et al . shed new light on the minor spliceosome by showing that a small non-coding RNA molecule known as U6atac , which catalyzes the removal of introns by the minor spliceosome , is highly unstable in human cells . This means that U6atac is a limiting factor for the splicing of minor introns—a process that is already limited by the very low abundance of the minor spliceosome under normal conditions . However , Younis et al . found that this bottleneck could be relieved by halting the degradation of U6atac . Experiments showed that U6atac can be stabilized by a key signaling molecule , a protein kinase ( called p38MAPK ) , which is activated in response to stress . The resulting higher levels of U6atac promoted splicing of the introns in its target mRNA transcripts , and also modulated various signaling pathways in the cells . Together , these results imply that the minor spliceosome is used as a valve that can help cells to adapt to stress and other changes . Moreover , by helping to translate mRNA transcripts that are already present in cells , it enables proteins to be produced rapidly in response to stress , bypassing the need for a fresh round of transcription .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2013	Minor introns are embedded molecular switches regulated by highly unstable U6atac snRNA
Eukaryotic cells modulate their metabolism by organizing metabolic components in response to varying nutrient availability and energy demands . In rat axons , mitochondria respond to glucose levels by halting active transport in high glucose regions . We employ quantitative modeling to explore physical limits on spatial organization of mitochondria and localized metabolic enhancement through regulated stopping of processive motion . We delineate the role of key parameters , including cellular glucose uptake and consumption rates , that are expected to modulate mitochondrial distribution and metabolic response in spatially varying glucose conditions . Our estimates indicate that physiological brain glucose levels fall within the limited range necessary for metabolic enhancement . Hence mitochondrial localization is shown to be a plausible regulatory mechanism for neuronal metabolic flexibility in the presence of spatially heterogeneous glucose , as may occur in long processes of projection neurons . These findings provide a framework for the control of cellular bioenergetics through organelle trafficking . Cellular metabolism comprises an intricate system of reactions whose fine-tuned control is critical to cell health and function . A number of quantitative studies have focused on metabolic control through modulating reactant and enzyme concentrations and turnover rates ( Grima and Schnell , 2006; Amar et al . , 2008 ) . However , these studies generally neglect the spatial organization of metabolic components within the cell . By localizing specific enzymes in regions of high metabolic demand ( Laughton et al . , 2007; Zecchin et al . , 2015 ) , as well as clustering together consecutively acting enzymes ( O'Connell et al . , 2012 ) , cells have the potential to substantially enhance their metabolism . Spatial organization is particularly critical in highly extended cells , such as mammalian neurons , whose axons can grow to lengths on the meter scale . Metabolic demand in neurons is spatially and temporally heterogeneous , with especially rapid ATP turnover found in the presynaptic boutons ( Rangaraju et al . , 2014 ) , and ATP requirements peaking during synaptic activity and neuronal firing ( Shulman et al . , 2004; Ferreira et al . , 2011; Weisová et al . , 2009 ) . Neurons rely primarily on glucose as the energy source for meeting these metabolic demands ( Peppiatt and Attwell , 2004 ) . Due to the long lengths of neural processes , the glucose supply can vary substantially over different regions of the cell ( Ferreira et al . , 2011; Weisová et al . , 2009; Hall et al . , 2012 ) . In myelinated neurons , for instance , it has been speculated that glucose transport into the cell is localized primarily to narrow regions around the nodes of Ranvier ( Magnani et al . , 1996; Harris and Attwell , 2012; Rosenbluth , 2009 ) , which can be spaced hundreds of microns apart ( Ibrahim et al . , 1995; Butt et al . , 1998 ) . Glucose transporters in neurons have also been shown to dynamically mobilize to active synapses , providing a source of intracellular glucose heterogeneity ( Ashrafi et al . , 2017 ) . Furthermore , varying levels of activity in the mammalian brain may lead to varying extracellular glucose levels , resulting in spatially heterogeneous nutrient access ( Hawkins et al . , 1979 ) . Individual axons have been shown to span across multiple regions of the brain ( Matsuda et al . , 2009 ) , enabling them to encounter regions with different glucose concentrations . Most ATP production in neurons occurs within mitochondria: motile organelles that range from interconnected networks to individual globular structures that extend throughout the cell . As energy powerhouses and metabolic signaling centers of the cell , mitochondria are critical for neuronal health ( Nunnari and Suomalainen , 2012 ) . Their spatial organization within the neuron plays a pivotal role in growth and cell physiology ( Li et al . , 2004 ) . Defects in mitochondrial transport are involved in the pathologies of several neurological disorders such as peripheral neuropathy and Charcot-Marie-Tooth disease ( Baloh , 2008; Baloh et al . , 2007 ) . A number of studies have shown that mitochondria are localized preferentially to regions of high metabolic demand , such as the synaptic terminals ( Li et al . , 2004; Chang and Reynolds , 2006 ) . Such localization can occur via several molecular mechanisms , mediated by the Miro-Milton mitochondrial motor adaptor complex that links mitochondria to the molecular motors responsible for transport ( Mishra and Chan , 2016 ) . Increased Ca2+ levels at active synapses lead to loading of calcium binding sites on Miro , releasing mitochondria from the microtubule and thereby halting transport ( Wang and Schwarz , 2009; Macaskill et al . , 2009 ) . High glucose levels can also lead to stalling , through the glycosylation of motor adaptor protein Milton by the glucose-activated enzyme O-GlcNAc transferase ( OGT ) ( Pekkurnaz et al . , 2014 ) . This mechanism has been shown to lead to mitochondrial accumulation at glucose-rich regions in cultured neurons ( Pekkurnaz et al . , 2014 ) . It is postulated to regulate mitochondrial spatial distribution , allowing efficient metabolic response to heterogeneous glucose availability . Mitochondrial positioning relies on an interplay between heterogeneously distributed diffusive signaling molecules ( such as Ca2+ and glucose ) , their consumption through metabolic and other pathways , and their effect on motor transport kinetics . While the biochemical mechanisms and physiological consequences of mitochondrial localization have been a topic of much interest in recent years ( MacAskill and Kittler , 2010; Mishra and Chan , 2016 ) , no quantitative framework for this phenomenon has yet been developed . In this work we focus on glucose-mediated regulation of mitochondrial transport , developing quantitative models to examine the consequences of this phenomenon for metabolism under spatially varying glucose conditions . Our approach relies on a reaction-diffusion formalism , which describes the behavior of species subject to both consumption and diffusion . Reaction-diffusion systems have been applied to describe the spatial organization of a broad array of cellular processes ( Kondo and Miura , 2010 ) , ranging from protein oscillations in E . coli ( Howard et al . , 2001 ) , to coordination of mitotic signalling ( Chang and Ferrell , 2013 ) , to pattern formation in developing embryos ( Bunow et al . , 1980; Gregor et al . , 2005 ) . The response of actively moving particles to spatially heterogeneous , diffusive regulators has also been extensively investigated in the context of chemotaxis ( Van Haastert and Devreotes , 2004 ) . In contrast to most chemotactic cells , however , mitochondria have no currently known mechanism for directly sensing glucose gradients . Instead , they are expected to accumulate in response to local glucose concentration only . Our goal is to delineate the regimes in which such a crude form of chemotaxis can lead to substantial spatial organization and enhancement of metabolism . Specifically , we model the modulation of mitochondrial density with glucose concentration in a tubular axonal region , focusing on two forms of spatial heterogeneity . In one case , we consider an axonal domain between two localized regions of glucose entry , representing the internodal region between nodes of Ranvier in myelinated neurons ( Figure 1a ) . The second case focuses on an unmyelinated cellular region with continuous glucose permeability , embedded in an external glucose gradient ( Figure 1b ) . In both cases , we show that mitochondrial accumulation and enhanced metabolic flux is expected to occur over a limited range of glucose concentrations , which overlaps with physiological brain glucose levels . Our simplified quantitative model allows identification of a handful of key parameters that govern the extent to which glucose-mediated mitochondrial halting can modulate metabolism . We establish the region of parameter space where this mechanism has a substantial effect , and highlight its potential importance in neuronal metabolic flexibility and ability to respond to spatially varying glucose . We begin by formulating a quantitative model to describe the spatial localization of mitochondria that halt in a glucose-dependent manner , in the presence of localized sources of glucose . This situation arises in myelinated neurons , which have glucose transporters enriched at the nodes of Ranvier , leading to highly localized sources of glucose spaced hundreds of micrometers apart within the cell ( Saab et al . , 2013 ) . Neuronal glucose transporters are known to be bidirectional ( Simpson et al . , 2007 ) , allowing glucose concentration within the cell to equilibrate with external glucose . For simplicity , we assume rapid transport of glucose through these transporters , so that the internal concentration of glucose at the nodes where transporters are present is assumed to be fixed . The cellular region between two glucose sources is modeled as a one-dimensional interval of length L with glucose concentration fixed to a value c0 at the interval boundaries ( Figure 1a ) . Glucose diffuses throughout this interval with diffusivity D , while being metabolized by hexokinase enzyme in the first step of mammalian glucose utilization ( Figure 1c ) ( Wilson , 2003 ) . The concentration of glucose is thus governed by the reaction-diffusion equation , ( 1 ) dGdt=D∂2G∂x2−k ( x ) G ( x ) where k⁢ ( x ) describes the spatial distribution of the hexokinase enzyme as well as the rate of consumption . In the case of spatially uniform , linear consumption [k⁢ ( x ) =k , a constant] this equation can be solved directly , yielding a distribution of glucose that falls exponentially from each source boundary , with a decay length λ=D/k ( Kholodenko , 2006 ) . Hexokinase 1 ( HK1 ) , the predominant form of hexokinase expressed in neurons , is known to localize preferentially to mitochondria ( John et al . , 2011 ) , which in mammalian axons can form individual organelles approximately 1 µm in length ( Fawcett , 1981 ) . We carry out numerical simulations of Equation 1 where consumption is limited to locations of individual discrete mitochondria , represented by short intervals of length Δ . Specifically , we define the mitochondria density as M⁢ ( x ) =n⁢ ( x ) / ( π⁢r2⁢Δ ) , where n⁢ ( x ) is the number of mitochondria overlapping position x , and r is the axon radius . The phosphorylation of glucose by mitochondrial hexokinase is assumed to follow Michaelis-Menten kinetics , described byk ( x ) =kgM ( x ) G ( x ) +KMwhere KM is the saturation constant and kg is the turnover rate of glucose ( per unit time per mitochondrion ) . The turnover rate kg incorporates both the catalytic rate of hexokinase and the number of hexokinase enzymes per mitochondrion . This expression reduces to the case of constant linear consumption when glucose concentration is low ( G≪KM ) and mitochondria are uniformly distributed throughout the region . In general , glucose consumption depends on the location of mitochondria within the domain . Mitochondrial distribution in neurons is known to be mediated through regulation of their motor-driven motility ( Chang and Reynolds , 2006; Pekkurnaz et al . , 2014 ) . Individual mitochondria switch between processively moving and paused states , modulated by the interplay between kinesin and dynein motors and the adaptor proteins that link these motors to the mitochondria ( Schwarz , 2013 ) . In our model , we simulate mitochondria as stochastically switching between a processive walking state that moves in either direction with velocity v and a stationary state . The rate of initiating a walk ( kw ) is assumed to be constant , while the halting rate ( ks⁢ ( x ) ) can be spatially heterogeneous . For simplicity , we assume the mitochondria are equally likely to move in the positive ( + ) or negative ( - ) direction each time they initiate a processive walk ( Figure 1b ) . It has recently been demonstrated that the key motor adaptor protein ( Milton ) is sensitive to glucose levels , halting mitochondrial motility when it is modified through O-GlcNAcylation by the OGT enzyme ( Pekkurnaz et al . , 2014 ) . Our model employs a highly simplified description of mitochondrial dynamics , which assumes that all pauses are associated with such an O-GlcNAcylation event . Recovery from the pause at the constant rate kw corresponds to removal of the modification through the activity of the complementary enzyme O-GlcNAcase ( OGA ) . Although there is evidence indicating long-term glucose deprivation can reduce OGA expression ( Zou et al . , 2012 ) , for simplicity we assume in our model that OGA activity is independent of glucose levels . In vivo axonal mitochondria have been observed to undergo short-lived sporadic pausing while continuing to move processively in their previous anterograde or retrograde direction ( Russo et al . , 2009; Wang and Schwarz , 2009 ) . Such pauses are subsumed into an effective processive velocity v in our model . Other sources of pausing , such as Ca2+-regulated motor disengagement , PINK1/Parkin-mediated detachment of motors , and anchoring to the microtubules by syntaphilin ( Schwarz , 2013 ) , are not considered here in order to focus specifically on the effect of glucose-dependent mitochondrial spatial organization . Upon entry into the cell , the first rate-limiting step of glucose metabolism is its conversion into glucose-6-phosphate by hexokinase . Further downstream metabolic pathways split , with much of the flux going to glycolysis while a small fraction is funneled into the pentose phosphate pathway and the hexosamine biosynthetic pathway ( HBP ) . The HBP produces UDP-GlcNAc , the sugar substrate for O-GlcNAcylation ( Figure 1c ) ( Hart et al . , 2011 ) . In our model , we assume that the rate of UDP-GlcNAc production equals the rate of glucose conversion by hexokinase , scaled by the fraction of G6P that is channeled into the hexosamine biosynthetic pathway . This assumption is valid if , at each point of pathway branching , the Michaelis-Menten saturation constants for the two branches are similar . This in fact appears to be the case for both the branching of the pentose phosphate pathway and glycolysis from the hexosamine biosynthetic pathway which is the focus of this work ( see Appendix 2 ) . Consequently , saturation of the initial glucose conversion step will imply saturation of the entire hexosamine biosynthetic pathway . We therefore model the kinetics of Milton modification using the same Michaelis-Menten form as for hexokinase activity , with the pathway flux leading to Milton modification subsumed within a rate constant for mitochondrial stopping ( ks ) . We note that the subcellular organization of the intermediates in the conversion from glucose into O-GlcNAcylated Milton is largely unknown . In our model , we make the extreme case assumption that all intermediates are localized to mitochondria , with only the initial glucose substrate capable of diffusing through the cytoplasm . We note that cytoplasmic diffusion of any of the pathway intermediates would attenuate the effect on mitochondrial localization . Our simplified model thus gives an upper limit on the extent to which mitochondria can localize at high glucose regions through the Milton modification mechanism . Following these simplified assumptions , we treat the kinetics of mitochondrial halting as dependent only on the local glucose concentration , according to the functional formks ( x ) =ksG ( x ) G ( x ) +KMwhere KM is the Michaelis-Menten constant of hexokinase . We proceed to evolve the simulation forward in time , with glucose consumption localized to regions within ±Δ/2 of each discrete mitochondrial position ( details in Materials and methods ) . A snapshot of one simulation run is shown in Figure 2a , highlighting the accumulation of stationary mitochondria in the high glucose regions near the ends of the domain . We are interested primarily in investigating the steady-state distribution of mitochondria and glucose in this system , averaged over all possible mitochondrial trajectories . We thus proceed to coarse-grain our model by treating the distribution of mitochondria as a continuous field M⁢ ( x ) =W+⁢ ( x ) +W-⁢ ( x ) +S⁢ ( x ) , where W+⁢ ( x ) is the distribution of mitochondria walking in the positive direction , W-⁢ ( x ) is the distribution of those walking in the negative direction , and S⁢ ( x ) is the distribution of stationary mitochondria . We can then write down the coupled differential equations governing the behavior of the mitochondrial distributions as: ( 4 ) dW+dt=−v∂W+∂x−ks ( x ) W++kwS2dW−dt=v∂W−∂x−ks ( x ) W−+kwS2dSdt=ks ( x ) [W++W−]−kwS . The glucose distribution evolves according to Equation 1 with consumption rate k⁢ ( x ) given by Equation 2 . The boundary conditions at the ends of the domain are assumed to be reflective for the mitochondrial distributions , and to have a fixed glucose concentration c0 . The stationary state for this system can be calculated numerically ( see Materials and methods ) . The formulation with a continuous mitochondrial density faithfully represents the behavior of simulations with discrete mitochondria , as illustrated in Figure 2b . The steady-state spatial distribution of mitochondria and glucose in the continuous system depend on six parameters: ks/kw , KM , c0 , D , L , kg⁢M¯ where M¯ is the average mitochondrial density in the axon ( number of mitochondria per unit volume ) . Estimates of physiologically relevant values are provided in Table 1 . Dimensional analysis indicates that three of these parameters can be used to define units of time , length , and glucose concentration , leaving three dimensionless groups . We choose to use the following three dimensionless parameters , each of which has an intuitive physical meaning: ( 5 ) λ^=DKMkgM¯L2 , c^0=c0KM , k^s=kskw Here λ^ is the length-scale of glucose decay relative to the domain length , c^0 is the boundary glucose concentration relative to the saturation constant KM , and k^s is the ratio of stopped to walking mitochondria at high glucose levels . We proceed to explore the steady-state distribution of mitochondria and glucose as a function of these three parameters . In order for mitochondria to preferentially accumulate at the source of glucose via a glucose-dependent stopping mechanism , three criteria must be met . First , the glucose concentration needs to be higher at the source than in the bulk of the cell , as occurs when the decay length due to consumption is much smaller than the size of the domain ( λ^≪1 ) . Second , if glucose levels become too high ( c^0≫1 ) then both glucose consumption rates and stopping rates of the mitochondria become saturated , leading to a flattening of glucose and mitochondrial distributions ( Figure 3 ) . There is thus an upper limit on the possible external glucose concentrations that will yield mitochondrial localization at the edges of the domain . Finally , the mitochondria must spend a substantial amount of time in the stationary state , since walking mitochondria will be broadly distributed throughout the domain . Because the stopping rate is itself dependent on the glucose concentration , this criterion implies that very low concentrations will also not allow mitochondrial localization . Figure 3 shows the distribution of glucose and mitochondria at different values of the external glucose c^0 , illustrating that accumulation of mitochondria at the edges requires intermediate glucose levels . To characterize the distribution of mitochondria along the interval , we introduce an accumulation metric A , defined byA=6σ2/L2−0 . 5where σ2 is the variance in the mitochondrial distribution . This metric scales from A=0 for a uniform distribution to A=1 for two narrow peaks at the domain edges . Mitochondrial distributions with several different values of the accumulation metric are shown in Figure 3a . We use a cutoff of A=0 . 2 to define distributions where the mitochondria are localized at the glucose source . We explore the dependence of the mitochondrial accumulation on the three dimensionless parameters defining the behavior of the system: the stopping rate constant k^s , the glucose decay length λ^ , and the external concentration c^0 . Because only the stopped mitochondria localize near the glucose sources , increasing the fraction of mitochondria in the stopped state ( increased k^s ) inevitably raises the overall accumulation ( Figure 4a ) . The fraction of mitochondria in the stopped state will depend on both k^s and the overall levels of glucose , as dictated by c^0 ( Figure 4b ) . Experimental measurements indicate that at high glucose concentrations , approximately 95% of mitochondria are in the stationary state ( Pekkurnaz et al . , 2014 ) . We are thus interested primarily in the parameter regime of high stopping rates: k^s≳10 . The limited range of concentrations that lead to mitochondrial accumulation at the edges of the domain can be seen in Figure 4a . For a high stopping rate ( k^s=10 ) , we then calculate the mitochondrial accumulation as a function of the remaining two parameters: λ^ , c^0 . Here , again , we note that only intermediate glucose concentrations result in accumulation , with the range of concentrations becoming narrower as the decay length λ^ becomes comparable to the domain size ( Figure 4c ) . We can establish the concentration range within which substantial accumulation is expected , by setting a cutoff A=0 . 2 on the accumulation metric and calculating the resulting phase diagram ( Figure 4d ) . Below the lower concentration cutoff , insufficient mitochondria are in the stationary state and so no localization is seen . This lower cutoff disappears in the limit of infinite k^s . At intermediate concentrations , mitochondria are localized near the domain edges . Above the upper concentration cutoff , no localization is observed due to saturation of the Michaelis-Menten kinetics . Using empirically derived approximations for the rate of glucose consumption by mitochondria and the diffusivity of glucose in cytoplasm ( see Table 1 ) , we estimate the decay length parameter as λ^≈0 . 03 . The mitochondria are then expected to localize near the glucose source only if c^0<66 . Because the saturation concentration for hexokinase is quite low ( KM≈0 . 03mM ) ( Wilson , 2003 ) , we would expect mitochondrial accumulation for glucose concentrations below about 2 mM . We note that physiological brain glucose levels have been measured at 0 . 7 − 1 . 3 mM , depending on the brain region ( McNay et al . , 2001 ) , implying that glucose-dependent halting of mitochondrial transport would be expected to result in localization of mitochondria at nodes of Ranvier . Localizing mitochondria to the glucose entry points is expected to increase the flux of glucose entering the cell , thereby potentially enhancing the overall metabolic rate . We calculate the overall effect of transport-based regulation on the net metabolic flux within the simplified model with localized glucose entry . Figure 5 shows the effect of increasing mitochondrial stopping rates ( k^s ) on the total rate of glucose consumption in the interval between nodes of glucose influx . At low k^s values , mitochondria are distributed uniformly throughout the interval . At high k^s values and at sufficiently low glucose concentrations , the mitochondria cluster in the regions of glucose entry , increasing the overall consumption rate by up to 40% at physiologically relevant glucose levels ( c0 = 1 mM ) . We note that in hypoglycemic conditions , glucose levels can drop to 0 . 1 mM ( Silver and Erecińska , 1994 ) , further increasing the magnitude of this effect . In the case of limited glucose transport into the cell , intracellular glucose levels could be significantly below the concentrations outside the cell . Measurements of intracellular glucose in a variety of cultured mammalian cell types indicate internal concentrations within the range of 0 . 07 − 1mM , up to an order of magnitude lower than glucose concentrations in the medium ( John et al . , 2008 ) . However , neuronal cells are known to express a particularly efficient glucose transporter ( GLUT3 ) ( Simpson et al . , 2008 ) , and these transporters have been shown to be highly concentrated near the nodes of Ranvier ( Magnani et al . , 1996; Rosenbluth , 2009 ) . We therefore assume that glucose import into the nodes is not rate limiting for myelinated neurons in physiological conditions . Introducing a finite rate of glucose transport would effectively decrease the intracellular glucose concentration at the nodes c0 , increasing the enhancement in metabolic flux due to mitochondrial localization . In subsequent sections , we explore the role of limited glucose import in unmyelinated axons with spatially uniform glucose permeability . Extracellular brain glucose levels exhibit substantial regional variation , particularly under hypoglycemic conditions where more than ten-fold differences in local glucose concentrations have been reported ( Paschen et al . , 1986 ) . Because individual neurons can traverse multiple different brain regions ( Matsuda et al . , 2009 ) , a single axon can be subjected to heterogeneous glucose levels along its length . This raises the possibility that glucose-dependent mitochondrial localization can play a role in neuronal metabolic flexibility even in the case where glucose entry into the cell is not localized to distinct nodes . We thus extend our model to quantify the distribution of mitochondria in an axon with limited but spatially uniform glucose permeability that is subjected to a gradient of external glucose . This situation is relevant , for instance , to unmyelinated neurons in infant brains , as well as to in vitro experiments with neurons cultured in a glucose gradient ( Pekkurnaz et al . , 2014 ) . In this model , the extracellular environment provides a continuous source of glucose whose influx is limited by the permeability of the cell membrane . Intracellular glucose dynamics are then defined by the reaction-diffusion equationdGdt=D∂2G∂x2−k ( x ) G+P ( x ) ( Gext ( x ) −G ) where the first term corresponds to diffusive glucose spread , the second to a spatially varying metabolism of glucose , and the third to the entry of glucose into the cell . Here , Gext is the external glucose concentration , and P⁢ ( x ) is the membrane permeability to glucose , which we assume to depend in a Michaelis-Menten fashion on the difference between external and internal glucose concentration:P ( x ) = ( 2/r ) PKMPKMP+\|Gext ( x ) −G ( x ) \|where P is the spatially uniform permeability constant in units of length per time . This functional form incorporates two known features of glucose transporters: ( 1 ) they are bidirectional , so that the overall flux through the transporter at low glucose levels should scale linearly with the difference between external and internal glucose ( Carruthers , 1990 ) ; ( 2 ) neuronal glucose transporters saturate at high glucose levels ( GLUT3 KMP≈3mM ( Maher et al . , 1996 ) , with an even higher saturation constant for GLUT4 ( Nishimura et al . , 1993 ) . When the difference in glucose levels is low , the overall flux of glucose entering the cell reduces to P ( Gext ( x ) −G ( x ) ) . Mitochondria dynamics are defined as before ( Equation 4 ) , and we again assume Michaelis-Menten kinetics for glucose metabolism by hexokinase localized to mitochondria ( Equation 2 ) . We note that the dynamics in Equation 6 are governed by three time-scales: the rate of glucose transport down the length of the axon , rate of glucose consumption , and rate of glucose entry . The first of these rates becomes negligibly small in the limit L≫D⁢ ( G+KM ) / ( kg⁢M¯ ) . Because internal glucose levels can never exceed the external concentrations , in the range where Gext<10mM , the rate of diffusive transport should become negligible for L≫150μm . In the limit where intracellular glucose is much less than KM , this criterion reduces to λ^≪1 , indicating that glucose diffuses over a very small fraction of the interval before being consumed . The interval length L in this model represents an axonal length which can range over many orders of magnitude . We focus on axon lengths above several hundred microns , allowing us to neglect the diffusive transport of intracellular glucose ( see Appendix 3 ) . The steady-state glucose profile can then be determined entirely by the local concentration of mitochondria and external glucose . For a given mitochondrial density M⁢ ( x ) and external glucose profile Gext ( x ) , the corresponding intracellular glucose concentration can be found directly by solving the quadratic steady-state version of Equation 6 without the diffusive term . However , the steady-state mitochondrial distribution cannot be solved locally , because the limited number of mitochondria within the axon couples the mitochondrial density at different positions . We thus employ an iterative approach to numerically compute the steady-state solution for both glucose and mitochondrial density under a linear external glucose gradient Gext=Gmin+ ( Gmax−Gmin ) xL ( see Materials and methods ) . For parameter combinations where intracellular glucose concentrations are above KM but well below Gext , the entry and consumption processes for glucose are both saturated . There is then a steep transition between two different regimes . In one regime , glucose entry exceeds consumption and internal glucose levels approach the external concentrations . In the other , consumption dominates and glucose levels drop below saturating concentrations . The key dimensionless parameter governing this transition can be defined as the ratio of entry to consumption rates:γ=2PKMPG¯extkgM¯r ( KMP+G¯ext ) This ratio can be modulated in the cell either by recruiting varying amounts of glucose transporters ( adjusting P ) or changing the total amount of active hexokinase ( adjusting kg⁢M¯ ) . The remaining dimensionless parameters determining the behavior of this simplified model are the external glucose concentration relative to the hexokinase saturation constant ( G^ext=G¯ext/KM ) , the relative magnitude of the glucose gradient , ΔG^ext= ( Gmax−Gmin ) /G¯ext , the ratio of stopped to walking mitochondria ks^=ks/kw , and the saturation constant for glucose transporters KMP/KM≈96 . The last parameter is expected to remain approximately constant in neuronal cells . The average external glucose concentration and glucose gradient are expected to vary substantially depending on the glycemic environment to which the neuron is exposed . We note that ΔG^ext has a maximum possible value since the minimal glucose concentration cannot drop below 0zero . We proceed to analyze the limiting case where the glucose gradient is as steep as possible for any given value of average external glucose ( ΔG^ext=2 ) . We quantify the amount of mitochondrial accumulation at the high glucose side of the domain by calculating the total mitochondrial density within the distal 10% of the interval compared to a uniform distribution , in analogy to experimental measurements ( Pekkurnaz et al . , 2014 ) . Substantial enrichment in the high glucose region occurs when glucose entry into the cell cannot keep up with consumption ( γ≪1 ) and the intracellular glucose levels drop below the hexokinase saturation concentration KM , as can be seen in the glucose and mitochondrial distributions plotted in Figure 6a–c . The interplay between external glucose levels and the entry/consumption rates is illustrated in Figure 6d . For external glucose concentrations well above KM there is a sharp transition to mitochondrial enrichment at γ<1 . At the lowest levels of intracellular glucose , accumulation is again reduced because a very small fraction of mitochondria are found in the stopped state . In the limit of high ks , mitochondrial accumulation would occur for arbitrarily low values of γ ( Figure 6—figure supplement 1 ) . We note that because glucose entry and turnover are much faster than diffusive spread for biologically relevant parameter regimes , the model results do not depend on the cell length L ( Appendix 3 ) . Experimental measurements of mitochondrial enrichment in cultured neurons subjected to a gradient of 0 to 5mM glucose have indicated an approximately 20% enrichment in mitchondrial counts at the axonal region exposed to high glucose . We note that using published estimates of typical glucose permeability and mitochondrial glucose turnover for mammalian cells ( Table 1 ) yields a ratio of entrance and consumption rates of γ≈1 . 9 for this experimental system . Because this ratio is above 1 , we would not expect to see substantial mitochondrial enrichment . To result in the experimentally observed enrichment at high glucose , the ratio γ would need to be reduced by approximately a factor of 2 , implying the existence of additional regulatory mechanisms . Modulation of γ could be achieved by either decreasing the number of glucose transporters in the cell ( reducing P ) or upregulating total hexokinase levels ( increasing kg ) . Neurons are believed to regulate both the density of glucose transporters and hexokinase activity in response to external glucose concentrations and varying metabolic demand ( Fujii and Beutler , 1985; Robey et al . , 1999; Duelli and Kuschinsky , 2001 ) . In particular , adaptation to glycemic levels well above physiological values , as well as possibly reduced synaptic activity in a cultured environment , may result in downregulation of glucose transporters , lowering the value of γ . The discrepancy between model prediction and observed mitochondrial accumulation highlights the existence of additional regulatory pathways not included in the current model whose role could be explored in further studies that directly quantify glucose entry and consumption rates in cultured neurons . Physiological brain glucose levels have been measured at 0 . 7 mM - 1 . 3 mM ( McNay et al . , 2001 ) , with hypoglycemic levels dipping as low as 0 . 1 mM and hyperglycemic levels rising up to 4mM ( Silver and Erecińska , 1994 ) . Axons that stretch across different brain regions with varying glucose levels can thus be subject to a glucose gradient with G¯ext on the order of 1 mM ( white line on Figure 6d ) . We note that the physiological range overlaps substantially with the region of high mitochondrial accumulation , indicating that glucose-dependent halting can modulate mitochondrial distribution under physiologically relevant glycemic levels . By accumulating mitochondria at the cellular region subjected to higher external glucose , the metabolic flux in that region can be substantially enhanced . In ( Figure 6e ) we plot the enhancement in glucose consumption rates ( compared to the case with uniformly distributed mitochondria ) within the 10% of cellular length subjected to the highest glucose concentrations . Metabolic enhancement occurs within a narrow band of the γ parameter . The drop-off in enhancement at low values of the internal glucose concentration ( low γ ) is due to the coupling between glucose levels and mitochondrial localization . Specifically , mitochondrial accumulation at the region subject to high glucose concentration increases the local rate of consumption in that region , driving down local internal glucose levels . Consequently , the difference in internal glucose concentrations between the two ends of the cell is decreased when internal levels fall substantially below M ( Figure 6b ) , reducing the enhancement of metabolic flux . Although mitochondrial accumulation decreases metabolic flux in the low glucose region , the total rate of glucose consumption integrated throughout the cell is enhanced by up to approximately 14% when γ≈1 ( Figure 6f ) . It is interesting to note that the typical physiological range of external glucose levels spans the narrow band of parameter space where metabolic enhancement is expected ( white lines on Figure 6e , f ) . These results implicate glucose-dependent mitochondrial stopping as a quantitatively plausible mechanism of metabolic flexibility , increasing metabolism in regions with high nutrient availability for axonal projections that span between hypoglycemic and euglycemic regions . The magnitude of this effect can be tightly controlled by the cell through modulating overall rates of glucose entry and consumption . Thus , by coupling mitochondrial transport to local glucose levels , whole-cell changes in hexokinase or glucose transporter recruitment can be harnessed to tune the cell’s response to spatially heterogeneous glucose concentrations . The minimal model described here provides a quantitative framework to explore the interdependence of glucose levels and mitochondrial motility and their combined effect on neuronal metabolic flux . Glucose-mediated halting of mitochondrial transport is shown to be a plausible regulatory mechanism for enhancing metabolism in cases with spatially heterogeneous glucose availability in the neuron . We have quantitatively delineated the regions in parameter space where such a mechanism can have a substantial effect on mitochondrial localization and metabolic flux . Specifically , mitochondrial positioning requires both sufficient spatial variation in intracellular glucose and sufficiently low absolute glucose levels compared to the saturation constant of the hexokinase enzyme . In the case of tightly localized glucose entry ( as at the nodes of Ranvier ) , intracellular spatial heterogeneity requires a small value of the dimensionless length scale for glucose decay ( λ^=D⁢KM/kg⁢M¯⁢L2≪1 ) . For physiologically estimated values , mitochondrial localization to the nodes is expected to occur for glucose levels below approximately 2 mM , comparable to physiological brain glucose concentrations ( McNay et al . , 2001; John et al . , 2008 ) . In the case where glucose can enter homogeneously throughout the cell surface ( as with unmyelinated axons ) , heterogeneity can arise from an external glucose gradient . We show that metabolic enhancement through mitochondrial positioning occurs in a narrow range of the key parameter γ= ( 2PKMPG¯ext ) / ( kgM¯ ( KMP+G¯ext ) ) , which describes the ratio of glucose entry to glucose metabolism , and that this narrow range intersects with physiological estimates . The model developed here is intentionally highly simplified , encompassing a minimal set of parameters necessary to describe glucose-dependent mitochondrial localization . Other regulatory pathways that determine mitochondrial positioning are not included in this basal model . In particular , we do not consider here calcium-based transport regulation , which is known to localize mitochondria to regions of synaptic activity ( Zhang et al . , 2010; Wang and Schwarz , 2009; MacAskill and Kittler , 2010; Macaskill et al . , 2009 ) . Upregulating OGT signaling in cultured cells has been shown to decrease the fraction of motile mitochondria by a factor of three , while reducing endogenous OGT nearly doubles the motile fraction , indicating that a substantial number of stationary mitochondria are stopped as a result of OGT activity ( Pekkurnaz et al . , 2014 ) . Our model assumes the extreme case where all stopping events are triggered in a glucose-dependent manner , thereby isolating the effect of glucose heterogeneity . Stopping mechanisms dependent on neuronal firing activity could alter mitochondrial distribution in concert with glucose-dependent halting , increasing the density of mitochondria at presynaptic boutons or near areas of localized calcium influx as at the nodes of Ranvier ( Zhang et al . , 2010 ) . We note that mitochondria have previously been shown to accumulate at spinal nodes of Ranvier in response to neuronal firing activity ( Fabricius et al . , 1993; Zhang et al . , 2010 ) . The mechanism described here provides an additional driving force for mitochondrial localization near the nodes even in quiescent neurons . Additional metabolic feedback loops , not included in our model , may result in a more complex dependence of mitochondrial stopping on glucose concentration . In particular , both the pentose phosphate pathway and glycolysis generate intermediates that feed back into UDP-GlcNAc production by the hexosamine biosynthetic pathway ( Kruger and von Schaewen , 2003; Shirato et al . , 2011 ) . Furthermore , several of the enzymes involved in the metabolic pathways linking glucose levels to Milton O-GlcNacylation may be regulated in a glucose-dependent manner . For example , the activity of the fructose-6-phosphate metabolizing enzyme GFAT is believed to be regulated by intermediates in the hexosamine pathway ( Traxinger and Marshall , 1991 ) and O-GlcNAc transferase ( OGT ) itself is directly regulated by UDP-GlcNAc levels ( Hart et al . , 2007 ) . Other enzymes , such as the de-GlcNAcylating enzyme OGA exhibit long term regulation of expression in response to altered glucose levels ( Zou et al . , 2012 ) . These regulatory mechanisms provide additional potential routes of metabolic control through mitochondrial positioning . Several key parameters that regulate mitochondrial localization in response to glucose heterogeneity can be dynamically regulated in neurons . Specifically , the rate of glucose consumption ( kg⁢M¯ ) can be tuned by modulating the concentration or activity of hexokinase within mitochondria or by altering total mitochondrial size and number . This parameter controls both the glucose decay length λ^ in the case of localized glucose influx and the ratio of glucose entry to consumption γ in the case of spatially distributed entry . We note that our model assumes hexokinase to be localized exclusively to mitochondria . The predominant form of hexokinase in the brain ( HK1 ) is known to bind reversibly to the mitochondrial membrane , with exchange between a mitochondria-bound and a cytoplasmic state believed to contribute to the regulation of its activity ( Golestani et al . , 2007 ) . Release of hexokinase into the cytoplasm would result in more spatially uniform glucose consumption , negating the metabolic enhancement achieved through mitochondrial localization . An additional parameter known to be under regulatory control is the rate of glucose entry into the neuron ( P ) . The glucose transporters GLUT3 ( Simpson et al . , 2008; Duelli and Kuschinsky , 2001; Weisová et al . , 2009 ) and GLUT4 ( Ashrafi et al . , 2017 ) have been shown to be recruited to the plasma membrane in response to neuronal firing activity . Interestingly , transporter densities are themselves spatially heterogeneous , concentrating near regions of synaptic activity ( Ashrafi et al . , 2017; Ashrafi and Ryan , 2017 ) . The model described in this work quantifies the extent to which a locally increased glucose influx can enhance total metabolic flux , given the ability of mitochondria to accumulate at regions of high intracellular glucose . A number of possible feedback pathways linking glucose distribution and mitochondrial positioning are not included in our basic model . For instance , hexokinase release from mitochondria into the cytoplasm ( potentially altering kg ) is known to be triggered at least in part by glucose-6-phosphate , the first byproduct in glucose metabolism ( Crane and Sols , 1954 ) . Chronic hypoglycemia has been linked to an upregulation in GLUT3 in rat neurons ( Uehara et al . , 1997 ) , which would in turn lead to an increased glucose uptake ( P ) . The fraction of glucose funneled into the hexosamine biosynthetic pathway ( incorporated within ks ) can also be modified through feedback inhibition of GFAT by the downstream metabolic product UDP-GlcNAc ( Li et al . , 2007 ) . Such feedback loops imply that several of our model parameters ( P , kg , ks ) are themselves glucose-dependent and may become spatially non-uniform in response to heterogeneous glucose . Incorporating these effects into a spatially resolved model of metabolism would require quantifying the dynamics of both the feedback pathways and mitochondrial positioning , and forms a promising avenue for future study . Control of glucose entry and consumption underlies cellular metabolic flexiblity , and defects in the associated regulatory pathways can have grave consequences for neuronal health . Misregulation of hexokinase has been highlighted as a contributor to several neurological disorders , ranging from depression ( Regenold et al . , 2012 ) to schizophrenia ( Shan et al . , 2014 ) . Neuronal glucose transporter deficiency has been linked to autism spectrum disorders ( Zhao et al . , 2010 ) and Alzheimer’s disease ( Liu et al . , 2008 ) . Furthermore , defects in mitochondrial transport , with the consequent depletion of mitochondria in distal axonal regions , contribute to peripheral neuropathy disorders ( Baloh , 2008 ) . Glucose-dependent mitochondrial localization provides an additional layer of control , beyond conventionally studied regulatory mechanisms , which allows the cell to respond to spatial heterogeneity in glucose concentration . Our analysis paves the way for quantitative understanding of how flexible regulation of metabolism can be achieved by controlling the spatial distribution of glucose entry and consumption . We simulate the internodal space of the axon , between localized nodes of glucose entry , as a one-dimensional domain for a reaction diffusion system with motile reaction sinks . The glucose concentration field is discretized over 100 equidistant points along the domain . Its dynamics are governed by the reaction diffusion equation ( Equation 1 ) , evolved forward over time-steps of δ⁢t using the forward Euler method . Because forward Euler methods have stringent conditions for stability and convergence , we use a time-step that is much smaller than both the glucose decay time-scale and the time-scale associated with diffusion over our spatially discretized grid ( see below ) . The number of mitochondria in the domain is calculated according to N=M¯⁢L⁢π⁢r2≈38 , where the mitochondrial density M¯ , internodal distance L , and axonal radius r are estimated from published data ( Table 1; Appendix 1 ) . The mitochondria are treated as discrete intervals of length Δ = 1 μm , with the position of each mitochondrial center updated at each timestep . Over each time step , every motile mitochondrion moves a distance of ±v⁢δ⁢t , ( with transport velocity v = 1 μm/s ) and switches to a stationary state with probability 1-exp⁡ ( -ks⁢δ⁢t ) , where ks⁢ ( x ) is a function of the center position of that mitochondrion ( Equation 3 ) . Mitochondria that reach within a distance of Δ/2 from the ends of the domain are reflected , reversing their velocity while remaining motile . Analogously , every stationary mitochondrion switches to a motile state on each time-step with probability 1-exp⁡ ( -kw⁢δ⁢t ) . Processive walks are initiated with equal probability in either direction . At any given time , the spatial density of mitochondria is calculated from the location of mitochondrial centers at positions x1 , …⁢xN , according to M⁢ ( x ) =n⁢ ( x ) / ( π⁢r2⁢Δ ) , where n ( x ) =∑i=1N[θ ( x−xi+Δ/2 ) −θ ( x−xi−Δ/2 ) ] , is the number of mitochondria overlapping spatial position x and θ is the Heaviside step function . We integrate the simulation forward in time-steps of δ⁢t=0 . 2⁢Δ⁢x2D , where Δ⁢x is the spatial discretization . This time-scale is much smaller than the relevant decay time for glucose consumption [τg= ( kg⁢M¯KM ) -1] . Using these small time-steps allows for stability and robust convergence with the forward Euler method . The simulation proceeds for 107 steps . Simulations are repeated 100 times to obtain the histogram shown in Figure 2 . Convergence to steady-state is established by comparing to calculations with the continuum model described in the subsequent sections . For an arbitrary spatial distribution of stopping rates ks⁢ ( x ) the corresponding steady-state mitochondrial distribution can be calculated directly by solving the equations for mitochondrial transport ( Equation 4 ) : ( 9 ) S=ks ( x ) ( W−+W+ ) kwvdW+dx=12ks ( x ) ( W−−W+ ) vdW−dx=12ks ( x ) ( W−−W+ ) . Because our model assumes symmetry between anterograde and retrograde mitochondrial transport , as well as equal glucose concentrations at either boundary of the domain , we take W-=W+ , implying that the population of walking mitochondria must be spatially constant . Consequently , the population of stopped mitochondria is proportional to the stopping rate ( S=C⁢ks⁢ ( x ) /kw ) . The constant C can be calculated from the normalization condition , ∫0LM ( x ) dx=∫0L[W− ( x ) +W+ ( x ) +S ( x ) ]dx=M¯L The overall steady-state distribution of mitochondria is then given by , ( 11 ) M⁢ ( x ) =W-⁢ ( x ) +W+⁢ ( x ) +S⁢ ( x ) =M¯1+1L⁢∫0Lks⁢ ( x ) kw⁢d⁢x⁢[ks⁢ ( x ) kw+1] Because the stopping rate is an explicit function of glucose concentrations [ks⁢ ( x ) =ks⁢G⁢ ( x ) KM+G⁢ ( x ) ] , this approach allows us to find the steady-state mitochondrial distribution for any fixed distribution of glucose . We solve for steady-state glucose and mitochondrial distributions using a numerical method that evolves the glucose concentration forward in time while explicitly setting the mitochondrial concentration to its steady-state value at each step . The glucose distribution is initialized according to the steady-state solution for uniform consumption ( Equation 13 ) . Mitochondrial density M⁢ ( x ) is calculated from the glucose distribution according to Equation 11 and Equation 3 . The glucose distribution G⁢ ( x ) , in turn , evolves according to the mitochondrial distribution as given by Equation 1 and Equation 2 . The glucose profile is integrated forward with a timestep δ⁢t=10-5⁢L2/D . The distributions are assumed to be converged once the root mean squared rate of glucose change drops below the minimal cutoff: 10-6⁢kg⁢M¯ . Results of the continuous mitochondrial distribution model are shown to match the discrete mitochondria simulations ( Figure 2b ) . All subsequent analysis is done in the continuum limit . We validate our numerical calculations by comparing to the analytically tractable solution in the limit of low glucose and nearly uniform mitochondrial distribution . In the limit of spatially uniform , linear consumption , the steady-state reaction-diffusion equation for glucose can be expressed as0=D∂2G∂x2−kG ( x ) where k=kg⁢M¯/KM is the constant consumption rate . Assuming fixed glucose concentrations ( c0 ) at the boundaries of the domain , the steady-state glucose distribution is then given byG ( x ) =c0cosh⁡ ( xλ ) cosh⁡ ( L2λ ) with λ=Dk defining the glucose decay length-scale . This quantity is a measure of how far glucose diffusively penetrates into the domain before being consumed by hexokinase . It is scaled by the size of the domain to give the dimensionless decay length scale λ^=D⁢KMkg⁢M¯⁢L2 used as a key parameter in our model with localized glucose entry: For the model with spatially uniform glucose permeability , we solve directly for the steady state distributions of glucose and mitochondria in the limit of slow diffusivity . When diffusion along the domain is slow compared to the timescales of glucose consumption and glucose import , the steady-state equation for glucose concentration is given by a simplified form of Equation 6:−k ( x ) G ( x ) +P ( x ) ( Gext ( x ) −G ( x ) ) =0 Substituting k ( x ) =kgM ( x ) G ( x ) G ( x ) +KM and P ( x ) = ( 2/r ) PKMPKMP+\|Gext ( x ) −G ( x ) \| , we get a quadratic equation in G⁢ ( x ) ; ( 15 ) [1−2PKMPrkgM]G ( x ) 2+[2PKMPGextrkgM−2PKMPKMrkgM−Gext−KMP]G ( x ) +[2PKMPKMGextrkgM]=0 For a given mitochondrial profile , this quadratic equation is solved to find G⁢ ( x ) =G⁢ ( M⁢ ( x ) ) . The mitochondrial distribution , M⁢ ( x ) is then updated according to Equation 11 and Equation 3 . We thus arrive at an iterative solution for G⁢ ( x ) and M⁢ ( x ) .	Cells are equipped with power factories called mitochondria that turn nutrients into chemical energy to fuel processes in the cell . Hundreds of mitochondria move throughout the cell , shifting their positions in response to energy demands . This happens via molecular motors that pick the mitochondria up and carry them to new locations . Such movements enable the mitochondria to accumulate in parts of the cell with the greatest energy needs . Mitochondria of nerve cells or neurons have a particular challenging job , as neurons can be very long and different parts within the cells can have different energy needs . It has been shown that mitochondria stop in regions where nutrients such as sugar are most concentrated . So far , it has been unclear whether this regulated stopping helps control energy balance in neurons . Here , Agrawal et al . used a computational model of rat neurons to find out whether sugar levels are sufficient in guiding mitochondria . The results showed that the mitochondria only accumulated in high-nutrient regions when the sugar concentrations were moderate – not too low and not too high . A specific range of sugar levels was necessary to make this mechanism useful for increasing the efficiency of energy production . Such concentrations match the ones observed in healthy rat brains . When neurons are unable to meet their energy demands , they stop working and sometimes even die . This is the case in many diseases , including diabetes , dementia , and Alzheimer’s disease . Computer models allow us to explore the complex energy regulation in detail . A better understanding of how neurons regulate their energy production and demand may help us discover how they become faulty in these diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems" ]	2018	Spatial control of neuronal metabolism through glucose-mediated mitochondrial transport regulation
Prostate is the most frequent cancer in men . Prostate cancer progression is driven by androgen steroid hormones , and delayed by androgen deprivation therapy ( ADT ) . Androgens control transcription by stimulating androgen receptor ( AR ) activity , yet also control pre-mRNA splicing through less clear mechanisms . Here we find androgens regulate splicing through AR-mediated transcriptional control of the epithelial-specific splicing regulator ESRP2 . Both ESRP2 and its close paralog ESRP1 are highly expressed in primary prostate cancer . Androgen stimulation induces splicing switches in many endogenous ESRP2-controlled mRNA isoforms , including splicing switches correlating with disease progression . ESRP2 expression in clinical prostate cancer is repressed by ADT , which may thus inadvertently dampen epithelial splice programmes . Supporting this , treatment with the AR antagonist bicalutamide ( Casodex ) induced mesenchymal splicing patterns of genes including FLNB and CTNND1 . Our data reveals a new mechanism of splicing control in prostate cancer with important implications for disease progression . Prostate is the most common male sex-specific cancer ( Center et al . , 2012 ) . Prostate cancer progression is controlled by androgen steroid hormones including testosterone and its active metabolite 5-α dihydroxytestosterone . Androgens stimulate androgen receptor ( AR ) signalling in prostate cancer cells to control transcription , including of genes that regulate the cell cycle , central metabolism and biosynthesis , as well as housekeeping functions ( Livermore et al . , 2016 ) . The roles of both androgens and the AR in transcription have been intensively investigated . However , androgens and the AR also regulate alternative pre-mRNA splicing through still largely unknown mechanisms ( Munkley et al . , 2018; Rajan et al . , 2011 ) . This represents a very important knowledge gap: alternative splicing patterns in cancer cells can generate protein isoforms with different biological functions ( Oltean and Bates , 2014 ) , and is a key process in the generation of biological heterogeneity in prostate cancer ( Rajan et al . , 2009 ) . Androgens are also closely linked to prostate cancer treatment , with androgen deprivation therapy ( ADT ) being the principal pharmacological strategy for locally advanced and metastatic disease . ADT utilises drugs to inhibit gonadal and extra-gonadal androgen biosynthesis and competitive AR antagonists to block androgen binding and abrogate AR function ( Livermore et al . , 2016 ) . ADT delays disease progression , but after 2–3 years tumours often grow again , developing castration resistance with a median survival time of 16 months ( Karantanos et al . , 2013 ) . The central role of androgens and the AR in prostate cancer , and the poor clinical outlook of castration-resistance prostate cancer ( CRPCa ) , have made it crucially important to identify androgen-regulated target genes and mechanisms of function –particularly those that relate to metastasis . The process of epithelial-mesenchymal transition ( EMT ) plays a pivotal role in prostate cancer metastasis ( Gravdal et al . , 2007; Matuszak and Kyprianou , 2011; Min et al . , 2010; Saini et al . , 2011; Xie et al . , 2010 ) . While the mechanisms driving EMT in prostate cancer are poorly understood , ADT has recently been shown to directly induce EMT in both mouse and human prostate tissue ( Sun et al . , 2012; Zhifang et al . , 2015 ) . Importantly , changes in alternative splicing patterns can have dramatic effects on EMT and on metastatic disease progression ( Pradella et al . , 2017 ) . While the mechanisms through which androgens regulate splicing control are not well understood , splicing itself takes place in the spliceosome , which is a multi-component structure containing a core of essential proteins and small nuclear RNAs ( Papasaikas and Valcárcel , 2016 ) . Splicing inclusion of alternative exons is often controlled by splicing regulator proteins that bind either to regulated exons or within their adjacent flanking intron sequences ( Gabut et al . , 2008 ) . The estrogen and progesterone steroid nuclear hormone receptors control splicing via recruitment of alternative splicing regulators ( including the RNA helicases Ddx5 and Ddx17 ) ( Auboeuf et al . , 2007; Auboeuf et al . , 2004; Auboeuf et al . , 2002 ) , and by changing RNA polymerase II extension rates and chromatin structure to affect splice site selection ( Kornblihtt et al . , 2009; Naftelberg et al . , 2015 ) . Steroid hormones can also drive selection of alternative promoters to include different upstream exons in mRNAs ( Dutertre et al . , 2010; Munkley et al . , 2018 ) . However , to what extent the above mechanisms may contribute to androgen-mediated splicing is largely unknown . We reasoned that a potential model to unify the role of androgens and the AR in transcription and splicing control could be via transcriptional regulation of genes that encode splicing regulatory proteins . To address this we analysed a recently described set of genes that reciprocally change expression in response to androgen stimulation in culture and ADT in patients ( Munkley et al . , 2016 ) . Here we identify AR-mediated transcriptional control of the key splicing regulator protein Epithelial Splicing Regulator Protein 2 ( ESRP2 ) . Importantly , many ESRP2-regulated exons switch splicing in response to androgen stimulation . ESRP2 and its close relative ESRP1 ( 60% identical to ESRP2 protein ) are important regulators of epithelial alternative splicing patterns ( Bebee et al . , 2015; Kalluri and Weinberg , 2009; Oltean and Bates , 2014; Valastyan and Weinberg , 2011; Warzecha et al . , 2010; Warzecha et al . , 2009a; Warzecha et al . , 2009b ) , reduced expression of which can drive critical aspects of EMT ( Hayakawa et al . , 2017; Pradella et al . , 2017; Warzecha et al . , 2010 ) . Our data identify an AR-ESRP2 axis controlling splicing patterns in prostate cancer cells , and further suggest that reduced ESRP2 levels in response to ADT may inadvertently help prime prostate cancer cells to facilitate longer term disease progression . To first gain insight into how androgens may mediate patterns of splicing control , we analysed a recently generated dataset of genes that exhibit reciprocal expression patterns on acute androgen stimulation in vitro versus clinical ADT ( Munkley et al . , 2016 ) . While a number of genes encoding splicing factors changed expression in response to acute androgen stimulation in vitro , ESRP2 also showed a reciprocal expression switch between acute androgen stimulation in culture and ADT in patients ( Munkley et al . , 2016 ) . ESRP2 expression decreased following ADT in 7/7 prostate cancer patients ( Rajan et al . , 2014 ) ( Figure 1A ) . Furthermore , RNAseq data prepared from different stages of LTL331 patient-derived xenografts ( Akamatsu et al . , 2015 ) showed reduced ESRP2 mRNA levels following castration and relapse neuroendocrine prostate cancer ( NEPC , Figure 1B ) . We similarly analysed expression of ESRP1 . ESRP1 is a close paralog of ESRP2 , but was not identified in our initial screen to identify androgen-regulated genes ( Munkley et al . , 2016 ) . ESRP1 expression levels also reduced following ADT ( Figure 1A ) . However , ESRP1 showed less change in gene expression compared to ESRP2 in patient-derived xenografts following castration or relapse NEPC ( Figure 1C ) ( Akamatsu et al . , 2015 ) . Further analyses supported androgen-mediated control of ESRP2 but not ESRP1 in prostate cancer cell lines . Western blots detected high endogenous levels of both ESRP1 and ESRP2 proteins within the AR positive LNCaP and CWR22 RV1 prostate cancer cell lines , as compared to the AR negative PC3 and DU145 prostate cancer cell lines ( Figure 1D and E ) . However , qPCR analysis showed that while androgens activated ESRP2 gene expression in response to in AR-positive LNCaP cells , this was not observed for ESRP1 gene expression ( Figure 1F ) . Androgen mediated-control of ESRP2 expression was also detected in two additional AR-expressing prostate cell lines ( VCaP and RWPE-1 , Figure 1G ) . ESRP2 protein expression was detected 48 hr after androgen exposure , with ESRP1 protein levels not changing over this same time-period ( Figure 1H ) . The specificity of the ESRP1 and ESRP2 antibodies used in these experiments was confirmed by detection of over-expressed protein and detection of siRNA mediated protein depletion by western blot ( Figure 1—figure supplement 1 ) . Further experimental analyses also supported ESRP2 as an early and so likely direct target for transcriptional control by the AR: ( i ) ESRP2 gene expression in LNCaP cells was rapidly induced in response to 10 nM of the synthetic androgen analogue R1881 ( methytrienolone ) ( Figure 1I ) . ( ii ) Chromatin immunoprecipitation ( ChIP ) from LNCaP cells confirmed direct AR binding to a site within 20 Kb of the ESRP2 gene promoter that had been previously predicted from a genome-wide study ( at position chr16: 68210834–68211293 on human genome assembly hg38 ) ( Massie et al . , 2011 ) ( Figure 1J ) . The AR ChIP signal adjacent to ESRP2 was similar to that detected in parallel for KLK3 ( encoding prostate specific antigen , or PSA ) , which is a known AR-regulated gene . ( iii ) Consistent with ESRP2 regulation at physiological androgen concentrations , ESRP2 transcription in LNCaP cells was induced over a wide range of R1881 concentrations , ranging from 1 nM to 100 nM ( Figure 1K ) . Each of these above data are consistent with AR-mediated regulation of ESRP2 expression levels within prostate cancer cell lines as well as tissue . We next monitored ESRP1 and ESRP2 expression profiles from prostate cancer patients . Meta-analysis of 719 clinical prostate cancer tumours from 11 previously published studies detected significant up-regulation of both ESRP1 and ESRP2 in 9/11 datasets ( Figure 2—source data 1 ) ( Arredouani et al . , 2009; Cancer Genome Atlas Research Network , 2015; Fraser et al . , 2017; Grasso et al . , 2012; Lapointe et al . , 2004; Liu et al . , 2006; Luo et al . , 2002; Taylor et al . , 2010; Tomlins et al . , 2007; Vanaja et al . , 2003; Varambally et al . , 2005; Wallace et al . , 2008 ) . We experimentally validated this meta-analysis using two independent panels of clinical samples . Real-time PCR showed significant up-regulation of both ESRP1 and ESRP2 mRNA in ( 1 ) prostate carcinoma relative to benign prostate hyperplasia ( BPH ) ( Figure 2A ) ; and ( 2 ) in nine prostate tumour samples relative to matched normal tissue from the same patient ( Figure 2B ) . A recent study by Walker et al . ( 2017 ) identified a molecular subgroup of prostate cancers with metastatic potential at presentation . Within this dataset ESRP1 was 2 . 76 fold up-regulated in the ‘metastatic-subgroup’ compared to the ‘non-metastatic subgroup’ . Using RNA from a subset of samples from the Walker et al . study , we confirmed significant ( p<0 . 05 ) upregulation of the ESRP1 gene in primary prostate cancer patients presenting with a metastatic biology ( Figure 2C ) . ESRP2 gene expression did not significantly increase in the 20 samples studied . We also used these same samples to assess if the observed up-regulation of ESRP1 and ESRP2 could result from prostate tumours consisting of a more pure population of epithelial-derived cells compared to matched tissue . Arguing against this possibility , levels of E-Cadherin were not significantly increased between BPH compared to prostate carcinoma , or between matched tumour and normal prostate tissue from patients ( Figure 2—figure supplement 1 ) . Each of the above data showed that ESRP1 and ESRP2 expression levels are relatively high in primary prostate cancer compared to normal prostate tissue . High ESRP2 expression was not prognostic of disease progression in the TCGA ( PRostate ADenocarcinoma ) PRAD cohort analysed using KM-express ( Chen et al . , 2018 ) , but high expression of ESRP1 associated with a significantly reduced time to first biochemical recurrence ( p=0 . 022 ) ( Figure 2D ) . We tested our antibodies against ESRP1 and ESRP2 proteins on prostate cancer FFPE tissue and cell blocks , but they did not pass our stringent quality control tests ( Figure 1—figure supplement 1C ) . While this manuscript was in preparation , another group used an alternative ESRP1 antibody to show up-regulation of ESRP1 in 12 , 000 prostate cancer tissue microarray tumours ( Gerhauser et al . , 2018 ) . We next investigated the effects of ESRP1/2 expression on the biology of prostate cancer cells in vivo . Because of their low normal endogenous expression profiles ( Figure 1C and D ) , we selected PC3 and DU145 cells to study the effects on prostate cancer cells of ESRP1/ESRP2 protein up-regulation . Ectopic expression of ESRP1 and ESRP2 protein expression in AR negative PC3 and DU145 cell line models reduced prostate cancer cell growth in vitro ( Figure 2—figure supplement 2 ) . Over-expression of both ESRP1 and ESRP2 ( either alone or together ) in PC3 cells also significantly slowed growth of prostate cancer xenografts in vivo ( Figure 2E–G ) . Taken together , the above data show that ectopic expression of ESRP1 and ESRP2 proteins slow the growth of PC3 and DU145 prostate cancer cell lines and are strongly suggestive that high levels of ESRP2 protein inhibit growth of prostate cancer cells . To enable us to test whether androgens may control splicing indirectly via transcriptional regulation of ESRP2 , we next set out to identify a panel of endogenous ESRP2-responsive exons within prostate cancer cells . We first used siRNAs to jointly deplete both ESRP1 and ESRP2 proteins from LNCaP cells ( since ESRP1 and ESRP2 can regulate overlapping targets ) ; and in parallel treated LNCaP cells with a control siRNA . We then used RNAseq to monitor the effects of these treatments on the LNCaP transcriptome . Bioinformatic analysis ( Trincado et al . , 2018 ) of these RNAseq data ( GSE129540 ) predicted 446 ESRP1/ESRP2 regulated alternative splicing events across 319 genes ( ΔPSI > 10% , p<0 . 05 ) ( Figure 3—source data 1 ) . We experimentally validated splicing switches for 44 predicted ESRP1/ESRP2-controlled exons by RT-PCR analysis , after LNCaP cells were treated with either of two independent sets of siRNAs directed against ESRP1 and ESRP2 or control siRNAs ( Figure 3 and Figure 3—source data 2 ) . We detected similar splicing switches for 35/44 of these skipped exons after jointly depleting ESRP1 and 2 from the AR-positive CWR22 RV1 prostate cancer cell line . 28/44 of these splicing switches were also observed after jointly depleting ESRP1 and ESRP2 from the AR positive PNT2 cells that model the normal prostate epithelium ( Figure 3 and Figure 3—source data 2 ) . Given this set of endogenous target exons , we carried out further analyses to next identify target exons that respond to increasing levels of either ESRP2 or ESRP1 expression in PC3 cells ( which normally express low levels of endogenous ESRP1/ESRP2 ) ( Figure 1D and E ) . Ectopic expression of either ESRP1 or ESRP2 in PC3 cells induced splicing switches for 31/42 exons analysed . Importantly , the splicing switches induced by ectopic expression of either ESRP2 or ESRP1 were reciprocal to the splicing switches detected after siRNA depletion of ESRP1/ESRP2 ( Figure 3 ) . Experimentally validated ESRP-regulated exons fell into two groups . Splicing of one group of exons were repressed by ectopic expression of ESRP1 or ESRP2 in PC3 cells , and reciprocally activated by endogenous ESRP1/ESRP2 depletion in LNCaP cells ( these exons are in the top of the heatmap in Figure 3 , from OSBL3 to FN1 ) . Splicing of the second group of exons were activated by ectopic expression of ESRP1 or ESRP2 , and reciprocally repressed by ESRP1/ESRP2 depletion ( from TRIP10 to ITGA6 in Figure 3 ) . The above data identified a robust panel of alternative exons within prostate cancer cells that responded to ESRP1/ESRP2 expression levels . We next tested if this panel of ESRP2-regulated exons are additionally regulated by ambient androgen concentrations . LNCaP cells were harvested after growth in steroid deplete media and after 48 hr of androgen stimulation ( this timing was designed to enable full levels of androgen-mediated ESRP2 protein induction , Figure 1H ) . Our prediction was that androgen stimulation of LNCaP cells would activate ESRP2 expression to regulate our panel of endogenous test exons . If this was the case , splicing switches in response to androgen stimulation should occur in a reciprocal direction to splicing changes induced by ESRP1/ESRP2 protein depletion in LNCaP cells . Consistent with these expectations , more than 70% ( 37/44 ) exons in our test panel demonstrated androgen regulated splicing ( Figure 3—source data 2 ) . Importantly , plotting the percent spliced-in ( PSI ) for each exon after 48 hr androgen stimulation ( Y axis ) versus the PSI after ESRP1/ESRP2 depletion ( X axis ) showed a significant negative correlation ( slope = −0 . 66 , R2 = 0 . 64 , p<0 . 0001 ) ( Figure 4A ) . Thus , exons that showed more exon skipping in response to ESRP1/ESRP2 depletion had higher splicing inclusion after androgen stimulation ( which would induce ESRP2 expression ) ( examples shown in Figure 4A and B ) . Reciprocally , exons that showed higher splicing inclusion in response to ESRP1/2 depletion also had less splicing inclusion after androgen stimulation ( examples shown in Figure 4A and C ) . These results experimentally support an androgen-ESRP2 axis that controls splicing patterns in prostate cancer cells . The genes containing ESRP-activated exons that were also activated by androgen exposure ( Figure 4B ) included: MINK1 ( exon 18 ) which encodes a pro-migratory serine/threonine kinase; MAP3K7 ( exon 12 ) which encodes a serine/threonine kinase that regulates signalling and apoptosis , activates NFKappaB , and is lost in aggressive prostate cancer ( Kluth et al . , 2013; Rodrigues et al . , 2015 ) ; GRLH1 ( exon 5 ) that encodes a transcription factor involved in epithelial cell functions ( Jacobs et al . , 2018 ) ; and FLNB ( exon 30 ) , alternative splicing of which has been identified as a key switch contributing to breast cancer metastasis ( Li et al . , 2018; Ravipaty et al . , 2017 ) . Amongst the genes containing ESRP2-repressed exons that were also skipped in response to androgen stimulation ( Figure 4C ) were DOCK7 ( exon 23 ) , which encodes a guanine nucleotide exchange factor involved in cell migration ( Gadea and Blangy , 2014 ) ; and RPS24 ( exon 5 ) , a gene that is highly expressed in prostate cancer ( Arthurs et al . , 2017 ) . To visualise the amplitude of ESRP2-mediated splicing control , we plotted PSIs measured in vitro after ectopic expression of ESRP1/ESRP2 versus PSI values after siRNA mediated depletion of ESRP1/ESRP2 ( Figure 5A , using data from Figure 3 and Figure 3—source data 2 , slope = −0 . 74 , R2 = 0 . 6221 , p<0 . 0001 ) . Consistent with the heat map ( Figure 3 ) , ESRP2-regulated exons fell into two groups . Splicing of one group of exons were ESRP2-activated , and splicing of these were conversely repressed by ESRP1/ESRP2 depletion , while the second group of ESRP2-repressed exons had the reverse properties . To assess how important ESRP2-regulated mRNAs might be in prostate cancer we monitored associated data for time taken to first biochemical tumour recurrence available in the TCGA PRAD cohort , in which information for 38/44 ESRP-regulated exons was available . This analysis revealed 3 groups of ESRP-regulated exons with different clinical associations . The group of ESRP1/ESRP2-promoted splice isoforms that correlated with decreased time to biochemical recurrence are shown in black on Figure 5A ( individual plots are shown in Figure 5—figure supplement 1 , and the functions of these genes and their associated splice isoforms in Figure 5—source data 1 ) . Skipping of RPS24 exon five correlates with a worse prognosis , and is the splice isoform promoted by ESRP2 . Splicing inclusion of RPS24 exon five is needed to maintain the RPS24 open reading frame ( Wang et al . , 2015 ) . Splicing inclusion of NUMB exon three also correlated with a worse prognosis , and is activated by ESRP2 . NUMB exon three encodes peptide information enabling protein interactions between NUMB and MDM2 , a protein that influence p53 stability ( Colaluca et al . , 2018 ) . Expression of the second group of ESRP1/ESRP2-promoted mRNA isoforms correlated with an increased time to biochemical occurrence . These exons are shown in green in Figure 5A , and include exons in the FLNB , SLK and ITGA6 genes ( functions of these genes and exons are summarised in Figure 5—source data 2 ) . For example , inclusion of ITGA6 exon 25 is activated by ESRP2 , and predicted to alter signalling pathways activated by the encoded protein ( Groulx et al . , 2014 ) . Splicing inclusion of the third set of exons did not correlate with time to biochemical recurrence ( identified as grey dots in Figure 5A , and summarised in Figure 5—source data 3 ) . These exons included GRHL1 exon 5 , splicing of which is needed to maintain the GRHL1 reading frame . GRHL1 encodes a transcription factor important for the operation of epithelial enhancer sequences ( Cieply et al . , 2016; Jacobs et al . , 2018 ) . To provide some measurement of the enrichment for clinically-relevant events , we compared the significance of optimal biochemical reoccurrence ( BCR ) survival difference between ESRP-regulated and all other exons whose PSI variance across TCGA primary tumours was ≥0 . 005 ( approximately the minimum for regulated events , to avoid biasing the potential relevance towards these ) . As illustrated in the violin plot in Figure 5D , there was a significant trend for a stronger prognostic value amongst the ESRP-regulated exons . Further analysis of the PRAD cohort revealed that 19/38 ESRP-regulated exons also have different patterns of splicing inclusion between tumour and normal tissue ( Figure 5E and Figure 3—source data 2 ) . These differentially spliced exons include the AR-ESRP2-controlled alternative exons in the DOCK7 and RPS24 genes ( both of which were excluded in prostate tumours compared to normal prostate tissue ) ; and the alternative exons in the MINK1 and MAP3K7 genes ( each of which had increased levels of splicing inclusion in prostate tumours compared to normal tissue ) . Further qRT-PCR analysis of an independent cohort confirmed more frequent skipping of DOCK7 ( exon 23 ) and RPS24 ( exon 5 ) in prostate tumour tissue compared to normal prostate ( Figure 5F and G ) . Some exons had more subtle changes than would be apparent from just comparing overall exon skipping and exon inclusion in prostate cancer . NUMB exon three and ITGA6 exon 25 ( both activated by ESRP2 ) are predominantly skipped in prostate tumours compared to normal tumour tissue , yet their PSI levels increase in larger , more advanced tumours to produce their respective mRNA isoforms that are associated with a decreased time to biochemical recurrence ( Figure 5—figure supplement 2A and B ) . RAC1 exon 3A ( activated by ESRP2 ) falls into the ‘grey’ area when comparing inclusion in normal versus prostate cancer , but more detailed analysis show that this exon is highly included in higher Gleason grades of prostate cancer , again to produce the RAC1 splicing isoform associated with a decreased time to biochemical recurrence ( Figure 5—figure supplement 2C ) . RPS24 exon 5 ( repressed by ESRP2 , and overall more skipped in tumours ) is skipped more in larger more advanced tumours , making the mRNA isoform associated with a decreased time to biochemical recurrence ( Figure 5—figure supplement 2D ) . Similarly , MYO1B exon 23 ( skipped in response to ESRP2 ) is both overall more skipped in prostate tumour versus normal , and more skipped in higher Gleason grade cancers ( Figure 5—figure supplement 2E ) . FLNB exon 31 ( activated by ESRP2 ) actually shows slightly reduced splicing inclusion in larger , more aggressive tumours ( Figure 5—figure supplement 2F ) . The above data identified a subset of ESRP2-regulated splicing switches that associated with biochemical recurrence of prostate cancer after treatment . Since ESRP2 expression was repressed by ADT in patient prostate cancer tissue , we next investigated whether AR inactivation may influence mRNA splice isoforms that correlate with cancer progression . To test this , androgen induction of ESRP2 mRNA expression was blocked using the androgen antagonist bicalutamide ( Casodex ) ( Figure 6A ) . Consistent with Casodex preventing expression of some potentially harmful isoforms in prostate cancer cells , the splicing inclusion of NUMB exon three and TUFT1 exon two were reduced by Casodex ( both these exons are normally activated by androgen exposure and ESRP2 ) . Likewise , exon skipping events in the RPS24 , FN1 and MYH10 genes that correlated with a poorer prognosis were also reduced by Casodex ( these exons are normally skipped in response to ESRP2 ) . Not all the splicing switches induced by Casodex correlate with increased time to biochemical recurrence . Skipping of CTNND1 exon 2 and 3 correlates with a decreased time to biochemical recurrence within the TGCA dataset ( Figure 5—figure supplement 1 ) , and this is the mRNA isoform promoted by Casodex treatment ( Figure 6C ) . Splicing inclusion of MAGI1 exon 7 ( normally repressed by ESRP2 ) and RALGPS2 exon 15 were also increased by Casodex treatment ( Figure 6B and C ) . ESRP2 and ESRP1 are important to maintain epithelial splicing programmes . We thus considered whether by repressing ESRP2 expression , ADT might also inadvertently switch splicing towards mesenchymal patterns that could facilitate metastasis . Consistent with this prediction , treatment of LNCaP cells with Casodex reduced splicing inclusion levels of the FLNB gene exon 30 by almost 20% ( Figure 6B ) . Although it is not differentially spliced between normal prostate and prostate cancer ( Figure 5E ) , increased skipping of FLNB exon 30 has been recently reported as a key driver of EMT in breast cancer development ( Li et al . , 2018 ) . Similarly , Casodex treatment also increased splicing inclusion of what are normally mesenchymal-expressed exons in the CTNND1 gene ( Warzecha et al . , 2009b ) ( Figure 6C ) . We used siRNA as a further strategy to reduce AR expression ( Figure 6—figure supplement 1A ) . As predicted , ESRP2 protein expression was reduced by siRNA depletion of the AR ( Figure 6—figure supplement 1A ) . Furthermore , siRNA-mediated depletion of AR reduced levels of FLNB splicing inclusion from 84% to 69% , and levels of TUFT1 exon 2 splicing from 23% to 9% ( Figure 6—figure supplement 1B ) . Both these data support a scenario where splicing inclusion of ESRP2-dependent exons are controlled by expression levels of the AR . The above data suggested a model where decreases in ESRP2 expression in response to inhibition of AR activity are sufficient to induce splicing changes , even though ESRP1 was still expressed . To further investigate whether loss of ESRP2 alone would be sufficient to induce splicing changes we carried out individual siRNA-mediated depletion of ESRP2 both within both LNCaP and CWR22RV1 cells . Consistent with our model , single ESRP2 depletion was able to switch splicing patterns of exons within the MAP3K7 , ARFGAP2 and CTNND1 genes ( Figure 6—figure supplement 2; Figure 3—source data 2 ) . As examples , individual depletion of ESRP2 reduced splicing inclusion of MAP3K7 exon 12 , and activated splicing inclusion of CTNND1 exons 2 and 3 ( Figure 6 – Figure 3—source data 2 ) . Furthermore , splicing patterns of ESRP1/ESRP2 target exons were also responsive to single up-regulation of either ESRP1 or ESRP2 ( Figure 3—source data 2 ) . In this study we report a novel molecular mechanism that explains how androgen steroid hormones control splicing patterns in prostate cancer cells , and unifies the functions of the AR both as a transcription factor and being able to control splicing . In this model , the AR controls expression of the master splicing regulator protein ESRP2 , which then regulates the splicing patterns of key genes important for prostate cancer biology ( Figure 7 ) . Amongst the key data supporting this proposed model , we find that ESRP2 is a direct and early target for transcriptional activation by the AR in prostate cancer cells . Furthermore , endogenous splice isoform patterns controlled by ESRP1 and ESRP2 also respond to androgen stimulation , siRNA-mediated depletion of the AR and/or the AR inhibitor bicalutamide ( Casodex ) . While intuitively straightforward , this model is conceptually different from the mechanisms through which estrogen and progesterone have been shown to regulate splicing ( via recruitment of splicing regulators as transcriptional cofactors , and by modulation of transcription speeds and chromatin structure ) . Androgens are already known to substantially modify transcriptional levels in prostate cancer , with important implications for cell behaviour and cancer progression ( Munkley et al . , 2016 ) . The data presented here imply that androgens also have an important role in controlling splicing patterns , particularly those that relate to epithelial/mesenchymal functions . Previous studies identified just a small number of alternative exons that are controlled by androgens in prostate cancer cells , none of which overlapped with the current study ( Munkley et al . , 2018; Rajan et al . , 2011 ) . We suggest that an important reason for this discrepancy is because previously splicing patterns were monitored after 24 hr of androgen exposure . Since we now show that splicing regulation by androgens operates indirectly through transcriptional control of ESRP2 , 24 hr androgen exposure would not be sufficient to upregulate ESRP2 levels . In the current study we analysed androgen-dependent splicing switches after 48 hr , to allow sufficient time for ESRP2 induction at the protein level and re-equilibration of downstream splice isoform ratios . ESRP1 expression levels also decreased in prostate tumours following ADT so might also be under androgen-control in tissue , although did not reciprocally increase following androgen stimulation of cultured cells . ESRP1 has recently been shown to be amplified in an aggressive subgroup of early onset prostate cancer , but how this contributes to disease progression has been not well understood ( Gerhauser et al . , 2018 ) . Our data here show that ESRP1 and ESRP2 control a number of individual mRNA splice isoforms that correlate with time to biochemical recurrence ( Figure 5—figure supplement 2 and Figure 5—figure supplement 1 ) , including of MAP3K7 exon 12 inclusion which is associated with a shorter time to biochemical reoccurrence in the TGCA database . Deletion of the MAP3K7 gene occurs in 30–40% of prostate tumours , and is associated with a poor clinical prognosis ( Goodall et al . , 2016; Kluth et al . , 2013; Liu et al . , 2007; Wu et al . , 2012 ) . MAP3K7 is a key gene in prostate cancer , and MAP3K7 exon 12 splicing is associated with epithelial properties of prostate cancer cells ( Dittmar et al . , 2012 ) . More generally , epithelial splicing patterns may play an important role early in prostate cancer development in establishing primary tumours ( Figure 7 ) . The expression of ESRPs appears to be plastic during cancer progression ( Hayakawa et al . , 2017; Ishii et al . , 2014; Ueda et al . , 2014 ) . ESRPs have previously been shown to have a dual role in carcinogenesis with both gain and loss associated with poor patient prognosis ( Hayakawa et al . , 2017 ) . ESRP1 expression is linked to poor survival and metastasis in lung cancer ( Yae et al . , 2012 ) , and both ESRP1 and ESRP2 are upregulated in oral squamous cell carcinoma relative to normal epithelium ( Ishii et al . , 2014 ) . Since ESRP2 is a critical component of epithelial-specific splicing programmes , we suggest that down-regulation of ESRP2 levels in response to ADT could dampen epithelial splicing patterns , helping to prime prostate cancer cells for future mesenchymal development and possibly contribute to development of metastasis . Supporting this , mesenchymal splicing patterns were induced by bicalutamide ( Casodex ) treatment of LNCaP cells , including in the FLNB , CTNND1 and MAP3K7 genes . FLNB encodes an actin binding protein which is linked to cancer cell motility and invasion ( Del Valle-Pérez et al . , 2010; Iguchi et al . , 2015 ) . Skipping of FLNB exon 30 is sufficient to initiate metastatic progression in breast cancer ( Li et al . , 2018 ) . In experiments reported here androgens promote the FLNB isoform that is not associated with metastasis . Expression of the metastatic FLNB variant is promoted by bicalutamide ( Casodex ) treatment . In breast cancer , the metastatic effects of FLNB alternative splicing are mediated via the FOXC1 transcription factor . The role FOXC1 plays in prostate cancer progression is unknown , but FOXC1 expression may be linked to androgen receptor levels ( van der Heul-Nieuwenhuijsen et al . , 2009 ) . ESRP2 also promotes skipping of epithelial-expressed exons in the CTNND1 gene ( catenin delta 2 , encoding a protein involved in cell adhesion and signalling ) , while Casodex treatment induces expression of a normally mesenchyme-specific splice isoform ( Warzecha et al . , 2009b ) . The Map3k12Δexon12 splice isoform is produced in response to ESRP2 depletion , and is usually expressed in highly metastatic cancer cell lines ( Tripathi et al . , 2019 ) . The clinical prognosis of metastatic prostate cancer is poor ( Livermore et al . , 2016 ) . This makes the mechanisms that control metastasis of prostate cancer cells , and any links with ADT of prime importance . In prostate cancer EMT has been linked to a common mechanism underlying therapeutic resistance and is associated with poor prognosis ( Gravdal et al . , 2007 ) . Sun et al . showed that although ADT can effectively control prostate tumour size initially , it simultaneously promotes EMT , an unintended consequence that could ultimately lead to CRPCa ( Sun et al . , 2012 ) . Such direct links between ADT and EMT uncover an important yet overlooked consequence of the standard care treatment for advanced prostate cancer ( Byrne et al . , 2016 ) . Although the causes of EMT in prostate cancer progression to CRPCa are likely to be complex , the down-regulation of ESRP proteins has been shown to be essential for EMT progression ( Horiguchi et al . , 2012 ) . Thus , loss of ESRP expression may provide a molecular explanation why AR positive prostate cancer cells show increased susceptibility to EMT in response to ADT , and so is relevant to consider with regard to therapy . Our findings have important implications for second line treatment strategies in a clinical setting , and suggest an alternative approach may be to inhibit EMT in combination with ADT to prevent disease progression . Cell culture and androgen treatment of cells was as described previously ( Munkley et al . , 2015a; Munkley et al . , 2015b; Munkley et al . , 2015c; Munkley et al . , 2014; Rajan et al . , 2011 ) . All cells were grown at 37°C in 5% CO2 . LNCaP cells ( CRL-1740 , ATCC ) were maintained in RPMI-1640 with L-Glutamine ( PAA Laboratories , R15-802 ) supplemented with 10% Fetal Bovine Serum ( FBS ) ( PAA Laboratories , A15-101 ) . For androgen treatment of LNCaP cells , medium was supplemented with 10% dextran charcoal stripped FBS ( PAA Laboratories , A15-119 ) to produce a steroid-deplete medium . Following culture for 72 hr , 10 nM synthetic androgen analogue methyltrienolone ( R1881 ) ( Perkin–Elmer , NLP005005MG ) was added ( Androgen + ) or absent ( Steroid deplete ) for the times indicated . Similarly , LNCaP cells were pre-treated with with 10 μM bicalutamide ( Casodex ) or ethanol ( vehicle ) for 2 hr prior to addition of 10nM R1881 for 48 hr . Cell line validation was carried out using STR profiling was according to the ATCC guidelines . All cell lines underwent regular mycoplasma testing . The following antibodies were used for western blotting: Anti-ESRP2 rabbit antibody ( Genetex , GTX123665 ) , anti-rabbit ESRP1 ( Sigma , HPA023719 ) , anti-AR mouse antibody ( BD Bioscience , 554226 ) , anti-actin rabbit antibody ( Sigma , A2668 ) , anti-FLAG mouse monoclonal antibody ( Sigma , F3165 ) , normal rabbit IgG ( 711-035-152 Jackson labs ) and normal mouse IgG ( 715-036-150 Jackson labs ) . For immunohistochemistry the following ESRP antibodies were tested: anti-rabbit ESRP1 ( Sigma , HPA023719 ) and anti-rabbit ESRP2 ( Abcam ab113486 ) but were found not to be specific for FFPE cell pellets . Cells were harvested and total RNA extracted using TRI-reagent ( Invitrogen , 15596–026 ) , according to the manufacturer’s instructions . RNA was treated with DNase 1 ( Ambion ) and cDNA was generated by reverse transcription of 500 ng of total RNA using the Superscript VILO cDNA synthesis kit ( Invitrogen , 11754–050 ) . Quantitative PCR ( qPCR ) was performed in triplicate on cDNA using SYBR Green PCR Master Mix ( Invitrogen , 4309155 ) using the QuantStudio 7 Flex Real-Time PCR System ( Life Technologies ) . ESRP1 was detected using ( ESRP1 for AGCACTACAGAGGCACAAACA; ESRP1 Rev TGGAGAGAAACTGGGCTACC ) . ESRP2 was detected using the primer combination ( ESRP2 For CCT GAA CTA CAC AGC CTA CTA CCC; ESRP2 Rev TCC TGA CTG GGA CAA CAC TG ) . Samples were normalised using the average of three reference genes: GAPDH ( GAPDH For AAC AGC GAC ACC CAT CCT C; GAPDH Rev TAGCACAGCCTGGATAGCAAC ) ; β–tubulin ( TUBB For CTTCGGCCAGATCTTCAGAC; TUBB Rev AGAGAGTGGGTCAGCTGGAA ) ; and actin ( ACTIN For CATCGAGCACGGCATCGTCA; ACTIN Rev TAGCACAGCCTGGATAGCAAC ) . siRNA mediated protein depletion of ESRP1/2 was carried out using Lipofectamine RNAiMAX Transfection Reagent ( Thermo Fisher , 13778075 ) as per the manufacturer’s instructions and for the times indicated . The siRNA sequences used were ESRP1 siRNA1 ( hs . Ri . ESRP1 . 13 . 1 ) ; ESRP1 siRNA2 ( hs . Ri . ESRP1 . 13 . 2 ) ; ESRP2 siRNA 1 ( hs . Ri . ESRP2 . 13 . 1 ) ; ESRP2 siRNA 2 ( hs . Ri . ESRP2 . 13 . 2 ) ; and a negative control siRNA ( IDT ( 51-01-14-04 ) ) . AR esiRNA was as described previously ( Munkley et al . , 2016 ) . Freshly cut tissue sections were analysed for immunoexpression using Ventana Discovery Ultra autostainer ( Ventana Medical Systems , Tucson , Arizona ) . In brief , tissue sections were incubated in Cell conditioning solution 1 ( CC1 , Ventana ) at 95°C to retrieve antigenicity , followed by incubation with respective primary antibodies described above . Bound primary antibodies were visualized using UltraMap DAB anti-Rb Detection Kit . LNCaP cells were stimulated with 10 nM R1881 overnight . The ChIP assay was performed using the one step ChIP kit ( Abcam ab117138 ) as per manufacturer’s instruction . Briefly , cells were fixed and crosslinked in 1% formaldehyde for 10 min at 37°C and incubated with protease inhibitors . Chromatin was isolated from cell lysates and enzymatically fragmented using an EZ-Zyme Chromatin Prep Kit ( Merck 17 375 ) . 10 ug of anti - AR antibody ( Abcam ab74272 ) or IgG control antibody was used to precipitate DNA crosslinked with the androgen receptor . Enriched DNA was then probed by qPCR using primers targeting the ESRP2 regulatory region to assess AR binding intensity . Primer sequences used to detect PSA were ( PSA ChIP for GCC TGG ATC TGA GAG AGA TAT CAT C; PSA Chip rev ACA CCT TTT TTT TTC TGG ATT GTT G ) . Primers used to detect AR binding near to ESRP2 were ( ESRP2 Chip for TCCCGAGTAGCTGGGACTAC; ESRP2 Chip rev CAGTGGCTTACACCTGGGAG ) . The ESRP1 plasmid ( PIBX-C-FF-B-ESRP1 ) was a gift from Prof Russ Carstens ( University of Philadelphia . USA ) and the ESRP2 plasmid ( pBIGi hESRP2-FLAG ) from Dr Keith Brown ( University of Bristol , UK ) . PC3 cells were transfected using FuGene HD Transfection Reagent as per manufacturer’s instructions . Stable transfectants with ESRP1 was selected using 10 µg/ml Blasticidin and ESRP2 plasmid was selected using 150 ug/ml Hygromycin . ESRP2 Plasmid was inducible by 2 . 5 ug/ml doxycycline for 48 hr . PC3 ESRP1 overexpressed cells were transfected with pBIGi hESRP2-FLAG plasmid using the same protocol . For cell growth curves ( carried out for in vitro analysis of PC3 stable cell lines ) , PC3 cells were seeded 100 , 000 cells per well in 12-well plate in eight plates . Cells were counted every 24 hr after seeding in the plate . All the treatments had 12 repeats . WST assays were carried out over 7 days as per manufacturer’s instructions ( Cayman , CAY10008883 ) . For DU145 cells 10 , 000 cells were seeded per well in a 96 well plate . All data was tested by two-way ANOVA . LNCaP cells ( passage 19 ) were treated with either control siRNAs or siRNAs targeting ESRP1 and ESRP2 for 72 hr ( samples prepared in triplicate ) . RNA was extracted 72 hr after siRNA treatment using the Qiagen RNAeasy kit ( Cat No . 74104 ) as per the manufacturer's instructions . RNAseq was carried out using TruSeq Stranded mRNA Sequencing NextSeq High-Output to obtain 2 × 75 bp reads . Quality control of reads was performed using FastQC . Reads were mapped to the hg38 transcriptome using Salmon . Differential gene expression analysis was performed using DESeq2 . Percent spliced-in ( PSI ) estimates for splicing events were calculated using SUPPA2 ( Trincado et al . , 2018 ) based on isoform transcripts per million ( TPM ) estimates from Salmon ( Patro et al . , 2017 ) . Quantification utilised Gencode gene models ( release 28 ) . Differential PSI was calculated using DiffSplice using the empirical method ( Hu et al . , 2013 ) . Events with a delta PSI > 10% and FDR < 0 . 05 were considered as significant . Clinical expression patterns of ESRP2-regulated exons were monitored using psichomics ( Saraiva-Agostinho and Barbosa-Morais , 2019 ) . Differential splicing analysis between primary solid tumour and solid tissue normal samples were subsequently performed to evaluate relative higher inclusion levels in either tumour or normal tissue samples using Δ median and t-test p-value ( Benjamini-Hochberg adjusted ) values . Survival analysis based on TCGA clinical data derived from prostate cancer patient samples was performed with time to first PSA biochemical recurrence being the event of interest . Additional statistical analyses and generation of plots were performed in R ( R Development Core Team , 2019 ) . Violin plots were created with R package vioplot ( Adler and Kelly , 2018 ) . Stable overexpression of ESRP1 and stable doxycycline-inducible overexpression of either ESRP2 alone or ESRP1 and 2 were obtained using PC3 cells ( that have the low endogenous levels of both proteins ) . One million PC3 overexpressing ESRP1 or control cells were injected subcutaneously in the flank of male nude mice and tumour volumes were monitored . Two million PC3 cells overexpressing ESRP2 , overexpressing ESRP1 and 2 , or control cells were injected subcutaneously in the flank of male nude mice and tumour volumes were monitored . PC3 ESRP2 and PC3 ESRP1/2 cells were cultured in medium supplemented with 2 . 5 ug/ml doxycycline for 48 hr prior to injecting into nude mice to induce ESRP2 expression and mice were administered Doxycycline repeatedly . Tumour diameters were measured using calipers . Our study made use of RNA from 32 benign samples from patients with benign prostatic hyperplasia ( BPH ) and 17 malignant samples from transurethral resection of the prostate ( TURP ) samples . Malignant status and Gleason score were obtained for these patients by histological analysis . We also analysed normal and matched PCa tissue from nine patients obtained by radical prostectomy . The samples were obtained with ethical approval through the Exeter NIHR Clinical Research Facility tissue bank ( Ref: STB20 ) . Written informed consent for the use of surgically obtained tissue was provided by all patients . The RNA samples analysed in Figure 2C were previously published ( Walker et al . , 2017 ) . All statistical analyses were performed using GraphPad Prism 6 ( GraphPad Software , Inc ) . Statistical analyses were conducted using the GraphPad Prism software ( version 5 . 04/d ) . PCR quantification of mRNA isoforms was assessed using the unpaired student’s t-test . Data are presented as the mean of three independent samples ± standard error of the mean ( SEM ) . Statistical significance is denoted as p<0 . 05 , p<0 . 01 , p<0 . 001 and **p<0 . 0001 .	Cancers often begin as cells that grow in connected sheets or clumps known as epithelial cells . To spread , the cancer cells need to change into cells that can break away from the group and move through the tissues . In prostate cancer , this process can happen years after successful treatment , but researchers are not sure why . Prostate cancer grows in response to testosterone . This hormone circulates around the body , and when it goes into a cell it helps select which genes are switched on or off . Testosterone-blocking drugs can help slow prostate cancer growth by changing this switching on and off of genes . But , over time , some cancers become resistant to the effects of these drugs and start to spread . This may be down to complexities in how testosterone controls gene activity . To produce a protein , a human cell first makes a copy of the corresponding gene . This copy is then modified , cutting and pasting different parts of the sequence ( a process called ‘splicing’ ) before the protein is produced . The patterns of splicing a cell exhibits depend on splicing regulator proteins . Testosterone can change splicing patterns in prostate cancer cells , but researchers did not know how . To find out , Munkley et al . examined a set of genes that turn off in response to testosterone-blocking drugs in people with prostate cancer . This revealed that testosterone controls a master splicing regulator called ESRP2 , which is normally present in epithelial cells . In prostate cancer cells in mice , extra ESRP2 slowed tumour growth . But , although ESRP2 levels are high in human prostate cancer cells to begin with , they drop in response to testosterone-blocking drugs . In the laboratory grown cells , the result was a switch away from 'epithelial-like' gene splicing patterns . Some of the new splicing patterns correlated with better patient prognosis , but other splicing patterns might help cancer cells to spread around the body . These results raise the possibility that blocking testosterone may impair prostate cancer growth , but also inadvertently prepare cancer cells to break away from tumours . A more complete understanding of how testosterone controls splicing could help explain why some tumours initially shrink when testosterone is blocked , but then later spread . Identifying the genes controlled by ESRP2 may reveal new drug targets to improve prostate cancer treatment .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2019	Androgen-regulated transcription of ESRP2 drives alternative splicing patterns in prostate cancer
Cell differentiation is controlled by individual transcription factors ( TFs ) that together activate a selection of enhancers in specific cell types . How these combinations of TFs identify and activate their target sequences remains poorly understood . Here , we identify the cis-regulatory transcriptional code that controls the differentiation of serotonergic HSN neurons in Caenorhabditis elegans . Activation of the HSN transcriptome is directly orchestrated by a collective of six TFs . Binding site clusters for this TF collective form a regulatory signature that is sufficient for de novo identification of HSN neuron functional enhancers . Among C . elegans neurons , the HSN transcriptome most closely resembles that of mouse serotonergic neurons . Mouse orthologs of the HSN TF collective also regulate serotonergic differentiation and can functionally substitute for their worm counterparts which suggests deep homology . Our results identify rules governing the regulatory landscape of a critically important neuronal type in two species separated by over 700 million years . Cell identities are characterized by the expression of specific transcriptomes that are activated through cell-type-specific regulatory landscapes . Large efforts have been made to identify functional enhancers in different tissues and developmental stages . The approaches include the occupancy of combinations of transcription factors ( TFs ) , identifying DNA regions displaying open chromatin states , analyzing specific histone marks and assessing enhancer function by transgenesis in vivo ( Junion et al . , 2012; Mo et al . , 2015; Nord et al . , 2013; Pattabiraman et al . , 2014; Visel et al . , 2013; Zinzen et al . , 2009 ) . These studies have revealed a highly dynamic organization of active enhancers that change depending on the cell type and developmental stage . However , to date , it is unclear what features of the DNA sequences distinguish enhancer regions from the rest of the genome . The identification of such features is critical both for understanding fundamental biological processes such as cell fate specification , as well as for biomedicine , given that most disease-associated mutations are thought to be located within regulatory sequences ( Mathelier et al . , 2015; Nishizaki and Boyle , 2017 ) . TFs are the main regulators of enhancer function . Each enhancer is bound by specific combinations of TFs that will either activate or repress transcription ( Reiter et al . , 2017 ) . The distribution of TF-binding sites ( TFBS ) has been studied in detail in only a few enhancers . For example , a study of sparkling , a specific enhancer of the Drosophila Pax2 gene , revealed it to be densely packed with TFBS that required specific arrangements for its functionality ( Swanson et al . , 2010 ) . However , these one-by-one approaches are not able to reveal any general molecular logic underlying cell-type-specific regulatory landscapes . Chromatin immunoprecipitation combined with deep sequencing ( ChIP-seq ) has been used to generate genome-scale binding profiles of specific TFs . It is now clear that TF-binding profiles are dynamic during cell differentiation and vary in related species ( Garber et al . , 2012; Heinz et al . , 2010; Khoueiry et al . , 2017; Nord et al . , 2013; Stefflova et al . , 2013; Wilczyński and Furlong , 2010; Zinzen et al . , 2009 ) . However , it is unclear what distinguishes TFBS actually bound by the TF from those that are unoccupied . Moreover , despite the fact that only a small fraction of bound TFBS are located in enhancers ( Kwasnieski et al . , 2014; Pattabiraman et al . , 2014; Whitfield et al . , 2012 ) , the molecular organization that distinguishes functional enhancers from the rest of non-coding regions is still unknown . Collective binding of several TFs is emerging as an important feature that distinguishes TFBS at functional enhancers from other genomic regions bound by individual TFs ( Junion et al . , 2012; Khoueiry et al . , 2017; Mazzoni et al . , 2013; Zinzen et al . , 2009 ) . However , it is still unclear how these combinations of TFs are collectively recruited to and activate cell-type-specific regulatory landscapes . The study of the transcriptional regulatory mechanisms underlying neuronal subtype specification in vivo in complex model organisms , such as rodents , is a challenging task . Here , we take advantage of the simple model organism C . elegans to study neuron type specification in vivo . C . elegans is especially suitable for transcriptional regulatory studies because its cell lineage is fully described , it is easy to genetically manipulate and its genome is very compact ( despite containing a similar number of genes to the human genome ) ( Gerstein et al . , 2010 ) . In this work , we focus on the study of the transcriptional regulatory logic of serotonergic neurons . Serotonergic neurons are present in all eumetazoan groups and are universally defined by their ability to synthesize and release serotonin ( 5HT ) ( Flames and Hobert , 2011 ) . They regulate multiple processes and their dysfunction has been linked to bipolar disorder , depression , anxiety , anorexia and schizophrenia ( Deneris and Wyler , 2012; Mathelier et al . , 2015 ) . Several TFs are known to be involved in mammalian serotonergic differentiation ( Deneris and Wyler , 2012 ) . However , little is known about their function in the regulation of specific serotonergic neuron enhancers . Here , we focus on this clinically relevant and highly conserved neuronal subtype , and exploit the amenability of C . elegans to unravel the rules governing the activation of the serotonergic transcriptome . This work reveals the phylogenetically conserved action of a collection of TFs on the selection of a specific regulatory landscape from the genome and allows for the identification of neuron subtype specific functional enhancers merely based on the presence of the TF collective regulatory signature . Serotonergic neurons are characterized by the coordinated expression of a battery of phylogenetically conserved enzymes and transporters known as the 5HT pathway genes ( Figure 1A ) . C . elegans adult hermaphrodites contain three functionally distinct serotonergic neuron subclasses: the NSM neurosecretory neuron , the ADF chemosensory neuron and the HSN motor neuron ( Figure 1B ) , which arise from different progenitors and , with the exception of the 5HT pathway genes , express different effector genes ( Figure 1—figure supplement 1 ) . The HSN neuron is , by far , the best characterized , thus we focused on dissecting the transcriptional rules governing the differentiation of this subclass . The HSN motor neuron controls vulval muscle contraction and its dysfunction leads to an egg laying defective ( egl ) phenotype . To identify the TF combination controlling HSN terminal differentiation , we selected among previously described egl mutants , those that both code for TFs and had reduced or no staining of 5HT in HSN . At least four genes matched this criteria: the POU domain TF unc-86 , the Spalt-type Zn finger TF sem-4 , the bHLH domain TF hlh-3 and the Insm-type Zn finger TF egl-46 ( Basson and Horvitz , 1996; Doonan et al . , 2008; Sze et al . , 2002; Wu et al . , 2001 ) . In addition to previous reports , we found that the GATA factor egl-18 , a regulator of HSN migration ( Desai et al . , 1988 ) and the ETS TF ast-1 , a regulator of dopaminergic fate ( Flames and Hobert , 2009 ) , also exhibit an HSN 5HT staining phenotype not previously published . Thus , although additional TFs with subtler egl phenotypes or with pleiotropic lethal effects and no available hypomorphic alleles are likely to be required for correct HSN differentiation , we initially focused our study in this set of six TFs that we refer as the HSN TF combination . To confirm previous observations , we analyzed null alleles for each member of the HSN TF combination except for ast-1 , where we used a hypomorphic allele as the null allele is lethal prior to HSN differentiation ( Schmid et al . , 2006 ) . All TF mutants indeed displayed a defective egg laying phenotype and 5HT staining and 5HT pathway gene expression defects , further supporting their roles in HSN differentiation ( Figure 1C , E and Source data 1 ) . We observed similar 5HT staining and 5HT pathway gene expression defects by RNAi knock down and in the analysis of additional mutant alleles for each candidate TF confirming that the phenotype is due to mutations in each corresponding TF and not to background strain effects ( Figure 1—figure supplement 2 and Supplementary file 1 ) . Importantly , 5HT pathway gene expression defects were specific for the HSN serotonergic subclass , while ADF and NSM neurons were unaffected ( Source data 1 ) , the exception being unc-86 ( n846 ) which showed a previously reported NSM differentiation phenotype ( Sze et al . , 2002; Zhang et al . , 2014 ) . We next assessed whether the HSN TF combination is also required for the expression of non-5HT related genes by analyzing nine additional reporters . We observed expression defects in all mutant strains ( Figure 1D , E and Source data 1 ) . Although most terminal features were affected , the expression of some genes remained normal indicating that , in each single mutant , HSN neuron is present and shows broad but partial differentiation defects ( Figure 1D , E ) . HSN neuron is born embryonically and remains in a quiescent undifferentiated state until fourth larval stage ( L4 ) , when it activates the expression of most effector genes , including the 5HT pathway genes . We did not observe precocious expression of any of the analyzed terminal features in any of the single mutant backgrounds , suggesting that the HSN TF combination acts mainly as activator of transcription . Most notably , the phenotypic profile of each mutant was slightly different from the others , which suggests that these TFs do not function in a cascade-like linear pathway ( Figure 1E ) . The expression of the HSN TF combination has only been partially studied ( Basson and Horvitz , 1996; Doonan et al . , 2008; Finney et al . , 1988; Wu et al . , 2001 ) . We used fosmid reporter strains ( for unc-86 , sem-4 and egl-18 ) , endogenous locus tagging ( for ast-1 and hlh-3 , Figure 2—figure supplement 1 ) and a transcriptional reporter strain ( for egl-46 ) to analyze their expression pattern in HSN throughout development . We find that all six TFs are expressed in the HSN at L4 coinciding with the onset of differentiation ( Figure 2A ) . The HSN TF combination is also expressed in other neurons , including expression of UNC-86 , EGL-18 and EGL-46 in the NSM serotonergic neuron , while none of them are expressed in the ADF serotonergic neuron . A deeper analysis of the developmental expression of each TF shows a very diverse array of expression dynamics ( Figure 2B ) . Some HSN TFs are expressed embryonically ( such as UNC-86 , HLH-3 or EGL18 ) . SEM-4 is widely expressed in the embryo in the area were HSN is located , although it is likely to be expressed at this early stage in HSN , we could not unequivocally identify it . In contrast , AST-1 and EGL-46 initiate their expression at different postnatal stages ( Figure 2B ) . Interestingly , HLH-3 shows two waves of expression: it is present in the mother cell of HSN ( around 280 min of embryonic development , see lineage in Figure 2—figure supplement 1 ) and its expression becomes fainter in postmitotic HSN and PHB neurons ( data not shown ) . At first larval stage ( L1 ) , HLH-3 expression is undetectable in HSN and expression reappears at third larval stage ( L3 ) , preceding AST-1 onset of expression and HSN maturation and is quickly downregulated at the end of L4 [Figure 2B , Figure 2—figure supplement 1 and ( Doonan et al . , 2008 ) ] . Thus , to further study the temporal requirements of HLH-3 activity in HSN differentiation , we induced HLH-3 expression at early L4 state in hlh-3 mutants . This late expression is sufficient to rescue tph-1 reporter expression defects indicating that embryonic HLH-3 expression is not required for correct HSN terminal differentiation ( Figure 2C ) . Little is known about the temporal control of HSN differentiation . Heterochronic genes have been described to regulate the onset of HSN axon extension ( Olsson-Carter and Slack , 2010 ) , although the molecular mechanisms underlying this process are unknown . AST-1 and HLH-3 expression correlates with HSN maturation suggesting they might have a role in determining the onset of this process . We used an early active HSN promoter to induce ast-1 and hlh-3 expression precociously from first larval stage ( L1 ) . Our results show that AST-1 significantly advances tph-1 expression to L2-L3 larval stages ( Figure 2D ) . On the contrary , early hlh-3 induced expression either alone or in combination with ast-1 leads to both a delay in onset and expression defects of tph-1 reporter gene ( Figure 2D ) . Despite tph-1 expression defects , HSN was still present as we could identify it by differential interference contrast ( DIC ) ( data not shown ) . These results suggest that AST-1 activity is an important determinant of HSN maturation onset . We also found that lin-41 heterochronic mutants show ast-1 expression defects in the HSN ( data not shown ) further supporting the role of AST-1 as a downstream effector controlling HSN maturation timing . Additionally , these experiments underscored the importance of the dynamic regulation of hlh-3 expression . HLH-3 is a proneural TF of the asc family , ortholog of mouse Ascl1 and Drosophila Scute ( Figure 2—figure supplement 1 ) . Proneural factors regulate both neural progenitor specification and neuronal differentiation and their functions are conserved through evolution from cnidarians to mammals ( Guillemot and Hassan , 2017 ) . Ascl1 is required for correct mouse serotonergic specification ( Pattyn et al . , 2004 ) and its activity is required to induce serotonergic fate from human fibroblasts ( Vadodaria et al . , 2016; Xu et al . , 2016 ) . HLH-3 shows several features common to ASCL1: ( 1 ) HLH-3 is transiently expressed in all neuronal progenitors and differentiating neuroblasts and it is required for correct differentiation of several neuronal types , including HSN ( Doonan et al . , 2008; Gruner et al . , 2016; Krause et al . , 1997; Luo and Horvitz , 2017; Murgan et al . , 2015 ) . ( 2 ) Both HLH-3 and ASCL1 are required to induce correct neurotransmitter identity ( Pattyn et al . , 2004; Sommer et al . , 1995 ) . ( 3 ) As will be explained in a later section , Ascl1 can rescue HSN differentiation defects of hlh-3 mutants , supporting its functional conservation . ( 4 ) As would be expected for a proneural gene , HLH-3 is also required for correct expression of panneuronal features in the HSN ( Figure 1 , rab-3 expression defects ) . ( 5 ) We find that HLH-3 expression needs to be tightly temporally regulated to correctly induce HSN fate . Temporal regulation of ASCL1 and SCUTE activities is also required for correct neuronal specification ( Andersen et al . , 2014; Imayoshi et al . , 2013; Quan et al . , 2016; Urbán et al . , 2016 ) . ( 6 ) HLH-3 regulates egl-46 expression ( discussed in next section , Figure 3A ) and similarly , Insm1 ( ortholog of egl-46 ) is a direct target of Ascl1 ( Castro et al . , 2011 ) . Taken together , our data suggests that HLH-3 acts as a proneural factor in HSN specification . Future experiments will determine if , similar to ASCL1 ( Wapinski et al . , 2013 ) , HLH-3 acts as a pioneer factor to facilitate binding of other TFs and promote neural differentiation . We next examined whether these TFs could exhibit cross regulation by analyzing their expression in each of the six different TF mutant backgrounds . In most cases , expression of each TF was largely independent of the integrity of the rest of the HSN TFs ( Figure 3A and Source data 1 ) . However , UNC-86 is a notable exception as it is required for the expression of most factors ( Figure 3A ) . Noteworthy , SEM-4 , that is downstream UNC-86 , is also required for AST-1 and partially for HLH-3 expression . Thus , UNC-86 effects could be , at least in part due to SEM-4 regulation . Finally , additional more modest effects are also observed between other TF pairs , such as the regulation of AST-1 and EGL-46 by HLH-3 ( Figure 3A and summarized in Figure 3B ) . Since TFs required for neuronal terminal differentiation are often also required to maintain the correct differentiated state ( Deneris and Hobert , 2014 ) , we explored whether this was also the case for the HSN TFs . We find that UNC-86 , SEM-4 , AST-1 , EGL-46 and EGL-18 expression is maintained in HSN after differentiation while HLH-3 expression is not observed after larval L4 stage ( Figure 2A and B ) . RNAi experiments to knock down the expression of the adult expressed TFs after HSN maturation produce defects in the maintenance of tph-1 and cat-1 expression ( Figure 3C ) . Additionally , the use of temperature-sensitive alleles for ast-1 , unc-86 and sem-4 leads to similar maintenance defects ( Figure 3—figure supplement 1 ) . Our results revealed that these five TFs are continuously required to maintain the correct HSN differentiated state . We next performed a comprehensive , in vivo analysis of the cis-regulatory modules ( CRMs ) for 5HT pathway genes to analyze how the HSN TFs regulate HSN terminal differentiation . First , we dissected the regulatory regions of the 5HT pathway genes by in vivo reporter analysis and isolated the minimal CRMs able to direct expression in each serotonergic neuron subclass ( Figure 4 and Figure 4—figure supplement 1 ) . We found that for each gene , different CRMs were active in specific subclasses of serotonergic neurons ( HSN , NSM or ADF ) . This suggests that expression of the same 5HT pathway gene is independently regulated in each of the three serotonergic neuron subclasses . These results , together with previous reports of different TF mutants affecting specific subclasses of serotonergic neurons ( Desai et al . , 1988 , Olsson-Carter and Slack , 2010 , Zhang et al . , 2014 , Zheng et al . , 2005 ) support the presence of subclass-specific serotonergic differentiation programs . Of note , 5HT pathway gene CRMs in some cases partially overlap ( Figure 4 ) . This overlap might be due to the presence of shared TFBS among serotonergic neuron subclasses and indeed , UNC-86 regulates both NSM and HSN differentiation ( Sze et al . , 2002 ) . We found that disruption of POU TFBS in tph-1 and bas-1 HSN CRMs but not in cat-1 CRM ( discussed in the following section ) affects both HSN and NSM expression . TFs are pleiotropic and it is known that the same TF can act with different combinations of TFs in different neuronal types to control neuron-type-specific genetic programs ( Hobert , 2016 ) . The observed serotonergic subclass independent regulation of 5HT pathway genes is in sharp contrast with our previous study of the dopaminergic regulatory logic in which all four subclasses of dopaminergic neurons are regulated by the same combination of TFs and through unique CRMs ( Doitsidou et al . , 2013; Flames and Hobert , 2009 ) . Dopaminergic neuron subclasses are functionally similar ( mechanosensory neurons ) , which may explain why they can share a unique TF combination to select a similar transcriptome . Conversely , the functional and molecular diversity of serotonergic neuron subclasses would require independent TF programs to select diverse terminal transcriptomes . Next , to assess whether the action of the HSN TF combination was direct on the serotonergic regulatory regions , we focused our analyses on the HSN minimal CRMs from the three 5HT pathway genes that showed the strongest phenotypes in our previous mutant analysis: tph-1 ( TPH ) , cat-1 ( VMAT ) and bas-1 ( AADC ) . We performed site-directed mutagenesis on predicted TFBS in these CRMs and analyzed in vivo the effect of the mutations . Our analysis , explained in detail below , revealed that all members of the HSN TF combination act directly upon 5HT pathway gene CRMs . Each CRM has a different disposition of TFBS arrangements supporting the flexible function of the HSN TFs . Additionally , we found examples of redundancy between TFBS that provide robustness of expression to the system and whose functionality can only be revealed in the context of smaller CRMs or mutant backgrounds . Notably , redundancy is specific to the CRM architecture as two TFs can act redundantly in one CRM but not in others . Finally , we also found that short HSN CRMs that lack TFBS for some HSN TF members can drive partially penetrant HSN expression , while longer CRMs with functional binding sites for additional members of HSN TFs drive more robust expression . This direct but flexible action of a combination of TFs to directly regulate cell type specification has been previously termed ‘TF collective’ mode of regulation ( Junion et al . , 2012; Spitz and Furlong , 2012 ) , accordingly , we termed this set of TFs the ‘HSN TF collective’ . The HSN minimal CRM for tph-1 ( TPH ) ( tph-1prom2 , Figure 4A ) contained predicted binding sites for all six HSN TF members ( Figure 5A ) . In vivo mutation reporter analyses revealed that all except the SPALT- and GATA-binding sites were required for proper tph-1 expression in HSN ( Figure 5A and Figure 5—figure supplement 1 ) . SEM-4 ( SPALT ) is required for ast-1 expression thus its effect on tph-1 expression could be indirect . Paradoxically , egl-18 ( GATA ) mutants showed defects in tph-1prom2 expression , similar to what was observed for the full-length reporter ( Figure 5B ) . Taking into account that EGL-18 does not regulate the expression of any member of the HSN TF collective , it may act upstream of another unidentified TF to regulate tph-1prom2 expression . Alternatively , EGL-18 may be recruited to the tph-1 promoter even in the absence of functional GATA -binding sites , perhaps through interactions with other members of the HSN TF collective . Similar binding site-independent recruitment of TFs , when combinatorially binding in a TF collective , has been reported for other combinations of TFs ( Junion et al . , 2012; Uhl et al . , 2016 ) . The HSN minimal CRM for cat-1 ( VMAT ) ( cat-1prom14 , Figure 4B ) also contained predicted binding sites for the HSN TF collective ( Figure 5C ) . Point mutation analyses revealed functionality of all but INSM-binding sites ( Figure 5C and Figure 5—figure supplement 1 ) . In agreement with this observation , we found that cat-1prom14 expression does not require EGL-46 ( INSM ) factor ( Figure 5D ) . However , the penetrance of HSN expression for the minimal cat-1prom14 was much lower than for the full-length reporter ( 55% versus 100% expression , respectively , Figure 5D ) . This indicates that additional TFBS outside of the minimal CRM are required to promote robust HSN expression . Indeed , full-length reporter ( cat-1prom1 ) expression was affected in egl-46 mutants ( Figure 5D ) . These results suggest that , although partial expression from cat-1 can be achieved without EGL-46 , this TF is required for robust expression in the context of the full cat-1 promoter . The requirement for GATA sites in the cat-1 minimal CRM contrasted with the lack of an expression defect of a full-length cat-1 reporter in egl-18 ( GATA ) mutants ( Figure 5D ) . However , when we analyzed the minimal cat-1 CRM ( cat-1prom14 ) in egl-18 mutants we found that its activity was affected in this mutant background ( Figure 5D ) . Thus , EGL-18 directly regulates cat-1 expression but its loss can be compensated in the context of a large regulatory region by other unknown factors . We confirmed EGL-18 direct binding to the cat-1 promoter in vitro using electrophoretic mobility shift assays ( EMSA ) ( Figure 5—figure supplement 2 ) . The HSN minimal CRM for bas-1 ( AADC ) ( bas-1prom18 , Figure 4C ) contained predicted binding sites for four TFs from the HSN TF collective: ETS , POU , GATA and SPALT TFs , but lacked any predicted INSM- or HLH-binding sites . Reporter analyses of the minimal CRM revealed that ETS- , POU- and SPALT- but not GATA-binding sites were required for expression in HSN ( Figure 5E and Figure 5—figure supplement 1 ) . Similar to cat-1 , a bas-1 functional binding site for several TFs was detectable only in the context of the minimal small CRMs while there was no defect in expression of the full-length reporter in the corresponding TF mutant backgrounds . For example , we found functional ETS- ( ast-1 ) binding sites in bas-1prom18 while expression of the full-length bas-1 reporter was unaffected in ast-1 ( ot417 ) ( Figure 5G ) . As ast-1 ( ot417 ) is a hypomorphic allele , we confirmed that ast-1 is not required for bas-1 full-length reporter expression by mosaic analyses with a rescuing array in a null ast-1 allele ( hd92 ) ( 87 out of 87 ast-1 null HSN neurons expressed bas-1 ) . We analyzed minimal CRM bas-1prom18 activity in ast-1 ( ot417 ) mutants and found a small but significant reduction in the percentage of GFP-positive HSNs ( Figure 5G ) . We also confirmed AST-1 binding to the bas-1 promoter in vitro using EMSA ( Figure 5—figure supplement 2 ) . Altogether , these results suggest that AST-1 can bind and activate transcription from the bas-1 minimal CRM as can EGL-18 from cat-1 minimal CRM . In both cases , however , other factors can compensate for their loss by activating transcription from regulatory sequences outside the minimal CRMs . This genetic redundancy for some members of the HSN TF collective at specific 5HT pathway genes possibly acts as a mechanism to ensure that differentiation is robust . Although HLH-3 ( bHLH ) and EGL-46 ( INSM ) were required for full-length bas-1 expression ( Figure 1E ) , no functional HLH- or INSM-binding sites were found in the minimal bas-1 CRM ( bas-1prom18 ) ( Figure 5E ) . Similar to the minimal cat-1 CRM ( cat-1prom14 ) , GFP expression of bas-1prom18 was partially penetrant ( ranging from 38% to 83% depending on the transgenic line , Figure 5—figure supplement 1 ) , while a longer construct ( bas-1prom13 ) was more robustly expressed ( 90% expression in all lines , Figure 5—figure supplement 1 ) . bas-1prom13 contains bHLH- and INSM-binding sites and INMS-binding site mutation , but not bHLH mutation , leads to expression defects which suggest a direct role for EGL-46 in robust bas-1 expression ( Figure 5F ) . We did not find functional GATA-binding sites in bas-1 CRMs , and egl-18 ( GATA ) mutants did not show bas-1 expression defects either . This would suggest that GATA factors are dispensable for the regulation of this gene . However , as we had already observed genetic redundancy in other CRMs , we considered that this could also be the case for bas-1 regulation . First , we analyzed bas-1 minimal CRM expression ( bas-1prom18 ) in egl-18 ( ok290 ) mutants and found that EGL-18 was required for its normal expression ( Figure 5G ) . Next , to determine whether the role for GATA factors in bas-1 expression was direct , we analyzed the expression of a bas-1 minimal CRM carrying GATA-binding site mutations ( bas-1prom78 ) in the ast-1 ( ot417 ) genetic background . Interestingly , while GATA-binding site mutations had no significant effects in wild type worms , we found a complete loss of expression of this construct in ast-1 ( ot417 ) mutants ( Figure 5H ) . These results revealed both a direct role for GATA factors in bas-1 expression and redundancy/compensatory effects between egl-18 and ast-1 . Interestingly , these two factors do not act redundantly in other CRMs such as tph-1 . Of note , despite the fact that HLH-3 expression is not maintained during adulthood ( Figure 2B ) we find functional bHLH-binding sites both in tph-1 and in cat-1 CRMs . These results suggest that HLH-3 is directly required to initiate expression of some HSN effector genes . Similar direct action on effector genes has been described for mouse ortholog ASCL1 in the regulation of neuronal differentiation ( Raposo et al . , 2015 ) . Our cis-regulatory analysis revealed compensatory effects among the HSN TF collective , thus , to increase our understanding of the TF collective action , we performed double mutant analysis . We analyzed tph-1 , cat-1 and bas-1 reporters because their HSN CRMs contain functionally verified binding sites for all six factors ( Figure 5 ) . unc-86 and sem-4 null mutants show complete loss of expression of tph-1 , cat-1 and bas-1 , thus we used hypomorphic alleles with partial phenotypes for double mutant analysis . Synergism was the most common effect in our double mutant analysis , although we also found epistatic effects , additivity and suppression ( Figure 6 , Figure 6—figure supplement 1 and Figure 6—source data 1 ) . We found synergism among different members of the HSN TF collective in their action upon tph-1 , cat-1 and bas-1 reporters ( Figure 6A–H ) . Interestingly , the same pair of TFs acting synergistically in the regulation of one reporter can show a different genetic relationship in the regulation of a different gene ( Figure 6E–H ) . For example , while unc-86 acts synergistically with sem-4 and hlh-3 in the regulation of cat-1 expression , it shows additive effects with both TFs in the regulation of tph-1 reporter ( Figure 6E , F ) . Similarly , hlh-3 shows synergy with egl-18 and egl-46 in the regulation of cat-1 and tph-1 respectively , while it is epistatic to egl-18 in the regulation of bas-1 ( Figure 6G , H ) . Reporter specific synergistic effects have been previously described ( Doitsidou et al . , 2013; Zhang et al . , 2014 ) and are likely a direct consequence of the flexibility of the TF collective mode of action that shows different disposition of functional binding sites in each enhancer . Unexpectedly , hlh-3 mutation suppresses egl-46 phenotype in the regulation of cat-1 expression ( Figure 6H ) . Genetic suppression is an intriguing phenotype that could reflect complex effects of competition for protein-protein interactions . We found additional examples of suppression in our double mutant analysis ( Figure 6—figure supplement 1 ) The HSN TF collective has pleiotropic functions . To try to avoid pleiotropic effects , we took advantage of our cis-regulatory data to perform combinations of TFBS mutations . To check for TFBS interactions , our analysis was limited to those sites which mutations produced only partial defects ( Figure 5 ) . We tested three out of the four possible combined TFBS mutations; however , none of these constructs showed synergistic effects ( Figure 6—figure supplement 1 ) . Interestingly , double cis SPALT- and GATA- BS mutations do not show synergy in the context of the minimal tph-1 CRM despite the synergistic effect observed for sem-4 , egl-18 double mutants in the regulation of tph-1 full length reporter . As these two factors do not regulate each other's expression ( Figure 3B ) , it is possible that TFBS mutations are more easily compensated than mutations in the corresponding trans-activating factors . As we were limited by the lack of additional BS mutations with partial effects , we next combined cis mutations of the TFBS with trans effects of TF single mutants . Combined cis/trans mutant analysis revealed synergistic relationships among two additional pairs of the HSN TF collective ( Figure 6I ) . Altogether , we found synergistic relationships among 9 out of the 15 possible HSN TF collective pair combinations ( Figure 6—figure supplement 1 and Figure 6—source data 1 ) . In addition to synergism , additivity and epistasis , we found several examples of genetic suppression both in the double mutant and the cis/trans mutant analysis ( Figure 6I and Figure 6—figure supplement 1 ) . Similar to the other genetic interactions , suppression is also TF pair and enhancer context specific . Altogether , the emerging picture is that of a joint action of the HSN TF collective upon their direct target genes . This regulation is flexible and often partially redundant showing synergistic relationships among different members of the HSN TF collective . Importantly , specific relationships and dependencies are determined by the CRM context and by the specific TF pairs tested . Our results suggest that the HSN TF collective is required for broad activation of HSN effector genes ( and not only for 5HT pathway gene expression ) ( Figure 1 ) and it acts directly on the regulatory regions of their target genes ( Figure 5 ) . Since the members of the HSN TF collective belong to six different TF families that recognize very different binding sites ( Figure 7A ) , we wondered whether the clustering of binding sites for the HSN TF collective in regulatory regions of HSN effector genes might confer sufficient specificity to impose a defining regulatory signature . There are 96 genes known to be expressed in the HSN neuron ( Supplementary file 2 ) ( Hobert et al . , 2016 ) , excluding pan-neuronal features which are regulated in a very redundant manner ( Stefanakis et al . , 2015 ) . We analyzed upstream and intronic sequences of HSN expressed genes in search of DNA windows ( up to 700 bp length ) containing at least one position weight matrix match for all six members of the HSN TF collective ( termed the ‘HSN regulatory signature’ ) ( Figure 7A ) . We found that known HSN expressed genes contained large upstream and intronic sequences , thus , for comparison purposes , we built ten thousand sets of 96 random genes with similar upstream and intronic length distribution . A significantly higher percentage of HSN expressed genes contain the HSN regulatory signature compared to the random sets of genes ( p<0 . 05 ) ( Figure 7B , Figure 7—source data 1 ) . Studies in Drosophila and vertebrates have shown that functional enhancers that are bound by combinations of TFs show higher interspecific conservation compared to enhancers bound by single TFs ( Ballester et al . , 2014; Khoueiry et al . , 2017; Stefflova et al . , 2013 ) . Thus , we performed a similar motif search in C . brenneri , C . remanei , C . briggsae and C . japonica genomes and calculated , for each C . elegans gene , the proportion of its orthologs that had , in its upstream or intronic sequence , at least one 700 bp window with binding sites for all the six TFs . We considered the HSN regulatory signature as phylogenetically conserved when orthologous genes in all species displayed the signature within their upstream or intronic regions . We found that the inclusion of the conservation criteria in this analysis slightly increased the difference between HSN and the random sets of genes ( p<0 . 01 ) ( Figure 7B ) . The higher prevalence of conserved signature in HSN expressed genes supports the idea that the HSN TF collective broadly selects the HSN transcriptome . We tested HSN regulatory signature windows from four of the known HSN expressed genes by in vivo reporter assays and confirmed that they correspond to active HSN enhancers ( three out of four tested contructs show HSN expression , Figure 7—figure supplement 1 and Figure 7—source data 2 ) . Of note , C . elegans functional HSN regulatory signature windows do not show a high level of sequence conservation ( Figure 7—figure supplement 1 ) , which is in agreement with rapid evolution of regulatory sequences ( Villar et al . , 2015 ) . Next , we examined the distribution of the HSN regulatory signature windows across the entire C . elegans genome . Remarkably , we found that it was preferentially found in the putative regulatory sequences of genes known to be expressed in neurons or that have a neuronal function compared to the rest of the genome , as would be expected for genes controlled by the HSN TF collective ( Figure 7C ) . Filtering of conserved regulatory signatures further increased the difference between ‘neuronal’ and ‘non-neuronal’ genomes , which adds support to its functionality ( Figure 7C ) . Gene ontology analysis of all genes in the C . elegans genome with HSN regulatory signature revealed enrichment of processes controlling transcription , axon guidance , synaptic transmission and oviposition , all characteristic of HSN differentiation and function ( Figure 7D ) . Our experimental data ( Figure 5 ) , in agreement to the TF collective model ( Spitz and Furlong , 2012 ) , shows that the presence of TFBS for all TF collective members is not required in specific enhancer contexts . Thus , we aimed to analyze if HSN regulatory windows lacking TF-binding sites for one or two TF classes show also an enriched distribution in HSN expressed genes and in the neuronal genome . We find that , in contrast to the six-motif HSN regulatory signature , windows containing only five or four types of HSN TF motifs are not preferentially found in HSN expressed genes compared to the 10 . 000 random sets of genes with or without filtering for conservation ( p>0 . 05 in all conditions ) ( Figure 7—figure supplement 2 ) . Additionally , genomic distribution of the HSN regulatory signature is less enriched in neuronal genes compared to non-neuronal genes when including windows lacking one or two HSN TF collective motifs ( Figure 7—figure supplement 2 ) . Moreover , while only 25% of the genes ( 4 , 968 ) contain at least one assigned six-motif regulatory window , regulatory windows with five or more motifs are found in 52% of the genes ( 10 , 415 ) and 72% of the genes ( 14 , 325 ) contain windows with four or more motifs . Finally , GO comparative analysis shows that genes with assigned 6-motif HSN regulatory windows show the highest enrichment in terms related to HSN function and that the additional GO terms obtained when including windows lacking either one or two HSN TF motifs are not related to neuronal functions ( Figure 7—figure supplement 2 ) . Altogether , our data shows that the most prevalent mark of HSN expressed genes is the regulatory signature with all six TFBS . Even if some HSN enhancers can still be functional with a partial complement of HSN TF collective binding sites , at the genomic level , including enhancers with missing TFBS abolishes cell type specificity . Next , we aimed to identify new genes expressed in HSN based solely on the presence of the HSN regulatory signature . To this end , we randomly selected 35 neuronal genes with a conserved HSN regulatory signature and generated transgenic reporter lines . We found that 13 out of the 35 constructs ( 37% ) showed GFP expression in HSN ( Figure 7E and Figure 7—source data 2 ) , while none of 10 randomly picked similar-sized intergenic regions of neuronal genes lacking the HSN regulatory signature led to reporter expression in HSN ( Figure 7—source data 2 ) . Importantly , all reporter constructs , including the negative controls , did drive GFP expression in a variable set of additional neurons , which might be due to the compact nature of the C . elegans genome . Finally , to analyze if the activity of the identified HSN regulatory windows was under the control of the HSN TF collective we crossed them into the unc-86 ( n846 ) mutant . The expression of 12 out of 15 reporter constructs ( 80% ) was significantly reduced in unc-86 mutants ( Figure 7—figure supplement 3 ) . Of note , onset of expression of the HSN regulatory window reporters can be used to predict the effect of unc-86 mutation: while all reporters with L4 onset of expression are strongly dependent on unc-86 , HSN regulatory windows that initiate expression at earlier stages show more modest dependency on unc-86 function ( Figure 7—figure supplement 3 ) . Our results reveal that the presence of a conserved HSN regulatory signature can be successfully used to de novo identify HSN expressed genes . However , our high level of false positives ( 63% ) indicates that the signature itself is not sufficient to induce HSN expression . Additional TFs might be part of the HSN TF collective and thus active HSN regulatory signature windows would contain additional TFBS . Repressive elements or chromatin accessibility could also block HSN expression of non-functional HSN regulatory signature windows , indeed members of the SWI/SNF chromatin remodeling complex are required for correct HSN terminal differentiation ( Weinberg et al . , 2013 ) . It is also possible that specific syntactic rules ( TFBS order , distance and disposition ) discriminate functional from non-functional HSN regulatory signature windows . Future studies will help identify additional players and rules for HSN terminal differentiation . Mouse orthologs for four out of the six TFs of the HSN TF collective are involved in mammalian serotonergic specification: ASCL1 ( bHLH TF ortholog of HLH-3 ) ( Pattyn et al . , 2004 ) , GATA2/3 ( GATA TF ortholog of EGL-18 ) ( Haugas et al . , 2016 ) , INSM1 ( Zn Finger Insm TF ortholog of EGL-46 ) ( Jacob et al . , 2009 ) and PET1 ( ETS TF ortholog of AST-1 ) ( Hendricks et al . , 2003 ) . Additionally , BRN2 ( also known as POU3F2 , a POU TF from the same family that UNC-86 ) has been associated with serotonergic specification , although its expression in serotonergic neurons has not been studied ( Nasu et al . , 2014 ) . We analyzed BRN2 expression in serotonergic differentiating neurons and found it expressed in serotonergic progenitors and serotonergic newborn neurons at embryonic stage E11 . 5 , when serotonergic neurons are differentiating ( Figure 8A and B ) . Finally , SALL2 is the closest mouse ortholog for C . elegans SEM-4 , but there is no known role for any SALL TFs in serotonergic specification . We found that SALL2 is also expressed in serotonergic progenitors and serotonergic newborn neurons at embryonic stage E11 . 5 ( Figure 8A and C ) . In evolutionary biology , the term deep homology refers to the relationship between two structures that share the genetic mechanisms governing their differentiation ( Shubin et al . , 1997 ) . As C . elegans HSN and mouse raphe serotonergic neurons share many of the TFs required for their differentiation , we hypothesized that they might be homologous structures . If this were the case , then HSN neurons and mouse serotonergic raphe should not merely share the expression of 5HT pathway genes , which are also present in the other C . elegans serotonergic neuron classes , but also be more broadly similar in molecular terms . To address this , we used available gene expression data from the WormBase to generate partial expression profiles for the 118 neuronal classes of the C . elegans hermaphrodite . This partial expression profile can be successfully used to reproduce the anatomical classification of C . elegans neuron subtypes ( Hobert et al . , 2016 ) . We assigned mouse orthologs to C . elegans neuronal genes and merged the resulting table with another one featuring the available mouse raphe serotonergic neuron transcriptome ( Okaty et al . , 2015 ) . Hierarchical clustering of this data set shows that HSN is , molecularly , the closest neuron to the mouse raphe neurons ( Figure 8D ) . Importantly , hierarchical clustering generated from mouse orthologs of C . elegans genes resembles the neuron class clustering generated directly from C . elegans genes ( Hobert et al . , 2016 ) . HSN and mouse raphe close relationship is not merely due to 5HT pathway genes expression because NSM and ADF serotonergic neurons are molecularly more distant to the mouse raphe serotonergic neurons than HSN ( Figure 8D and Source Data 5 ) . Moreover , HSN remained the most similar neuron to mouse serotonergic raphe neurons even after removing the 5HT pathway genes from the HSN expression profile ( Figure 8—figure supplement 1 ) . Shared orthologous genes between HSN and mouse raphe serotonergic neurons belong to different functional categories including axon guidance and migration , neurotransmission , or transcriptional regulation ( Table 1 ) . Importantly , HSN proximity to mouse serotonergic neurons was not maintained with other mouse neuronal populations ( Figure 8—figure supplement 1 ) . Finally , to test if there is deep homology between HSN and mouse raphe serotonergic neurons , we tested if mouse orthologs of the HSN TF collective can functionally substitute for their worm counterparts . We performed cell-specific rescue experiments of C . elegans mutants and found that mouse Pet1 , Ascl1 , Insm1 , Gata2 and Sall2 could respectively substitute ast-1 , hlh-3 , egl-46 , egl-18 and sem-4 , which suggest that this regulatory program could be phylogenetically conserved ( Figure 8E and F ) . Of note , our rescue experiments , both with C . elegans or mouse genes , restore tph-1 expression but do not rescue egg-laying defects . The HSN TF collective has pleiotropic actions in other tissues that also contribute to the egg laying phenotype ( Basson and Horvitz , 1996; Doonan et al . , 2008; Eisenmann and Kim , 2000; Koh et al . , 2002 ) what could explain the persistence of egg-laying defects in the HSN specific rescue experiments . In sum , these results revealed an unexpected level of regulatory and molecular proximity between C . elegans HSN and mouse serotonergic raphe neurons suggesting that deep homology might exist between these two neuronal types . Neuronal terminal differentiation programs have been best characterized in C . elegans . So far , relatively simple TF combinations , composed of two or three members , were shown to be required , and in some contexts sufficient , to select specific neuronal types . These TFs have been termed Terminal Selectors ( Doitsidou et al . , 2013; Serrano-Saiz et al . , 2013; Van Buskirk and Sternberg , 2010; Zhang et al . , 2002 ) . In some cases , additional TFs act together with Terminal Selectors to partially modulate the transcriptomes of specific neuronal subclasses ( Kerk et al . , 2017; Kratsios et al . , 2017 ) . Accordingly , it has been suggested that , in C . elegans , a rather simple organization of CRMs control the expression of neuronal terminal features ( Holmberg and Perlmann , 2012 ) . Our results , however , demonstrate a more complex scenario in the regulation of the HSN transcriptome . We have identified six TFs required for HSN terminal differentiation acting directly upon the regulatory regions of HSN expressed genes . Nonetheless , additional unidentified factors are likely to compose the HSN TF collective . We found that the HSN TF collective includes a proneural TF ( hlh-3 ) that is required to initiate HSN differentiation but whose expression , like all proneural factors ( Guillemot and Hassan , 2017 ) , is not maintained in the mature neuron . Future experiments should determine if , as has been proven for its mouse ortholog Ascl1 ( Wapinski et al . , 2013 ) , hlh-3 acts as a pioneer factor for HSN terminal differentiation . In light of our findings , nematode neuronal terminal differentiation programs are not necessarily simpler than those found in vertebrates , as previously proposed ( Holmberg and Perlmann , 2012 ) . Considering the technical advantages of C . elegans as a simple model system , our work is an example on how its study may help to identify rules of terminal differentiation in eumetazoa . The combination of our extensive cis-regulatory analysis and the double mutant characterization allowed us to describe the flexible action of the HSN TF collective that can activate enhancers with very different dispositions of TFBS . This flexibility is also made evident by the specific synergistic relationships in the regulation of some enhancers and not others . We propose that these redundant actions , globally considered , confer robustness of expression to the system . Co-binding of specific combinations of TFs to the same genomic region , assessed by ChIP-seq , has been successfully used to identify , de novo , cell-type-specific enhancers in Drosophila embryos ( Busser et al . , 2015; Junion et al . , 2012; Zinzen et al . , 2009 ) . However , this approach fails to address why specific genomic regions work as enhancers . Recently , massively parallel reporter assays ( MPRA ) have been used to identify generic rules of enhancer function . The analysis of synthetic enhancers revealed that highest levels of expression are achieved with clusters of binding sites for different TFs ( Smith et al . , 2013 ) . Another MPRA study has analyzed enhancer activity of regions bound by the adipocyte terminal selector PPARγ and has determined that the best predictor for enhancer functionality is the presence of nearby TFBS for more than 30 different TFs expressed in adipocytes ( Grossman et al . , 2017 ) . In accordance to this complex scenario of combinatorial action of multiple TFs in the global selection of cell type regulatory landscapes , we found that clusters of bioinformatically predicted TFBS for the HSN TF collective can be used for the de novo identification of HSN enhancers . Of note , our analysis still shows a high rate of false positives , which suggests that additional features are present in HSN functional enhancers . Future analyses based on more complex paradigms should facilitate the identification of such features that could include additional TFBS ( or the absence of repressor sites ) or specific syntactic rules . The diversity of C . elegans serotonergic neuronal classes ( NSM , ADF and HSN ) contrasts with that of tetrapod vertebrates , in which serotonergic neurons are limited to the raphe system ( Flames and Hobert , 2011 ) . Other chordates contain additional serotonergic populations ( Flames and Hobert , 2011 ) and serotonergic subclass diversity is also prevalent in other phyla such as arthropoda and mollusca ( Flames and Hobert , 2011 ) , which suggests a loss of serotonergic diversity in the tetrapod branch . As in nematodes , serotonergic subclass specification in other organisms is likely to be independently regulated . For instance , in Drosophila , the TFs islet , hunchback and engrailed are required for serotonergic specification of the ventral ganglion , while are dispensable for fly brain serotonergic specification ( Lundell et al . , 1996; Thor and Thomas , 1997 ) . Similarly , in zebrafish , Pet1 regulates raphe serotonergic specification but is dispensable for the specification of other serotonergic subclasses ( Lillesaar et al . , 2007 ) . Considering the homologous regulatory network between HSN and mouse raphe and their molecular proximity , our results suggest that the C . elegans HSN serotonergic neuron , but not the NSM or ADF , could share deep homology with mouse raphe neurons . It would be interesting to explore if NSM or ADF regulatory programs show homology to any of the programs controlling the non-raphe serotonergic populations present in other organisms . Noteworthy , despite the homology in TFs regulating HSN and raphe specification , both systems also show discrepancies . For example , while the LIM TF Lmx1b is known to be a key player in mouse serotonergic differentiation ( Ding et al . , 2003 ) , we failed to identify a similar role for any C . elegans LIM TF ( A . JM and N . F unpublished ) . Similarly , while C . elegans GATA factor egl-18 has very redundant effects on HSN differentiation , GATA2/3 factors are fundamental in mouse serotonergic differentiation ( Haugas et al . , 2016 ) . Considering the evolutionary distance between mammals and nematodes , the complexity of the regulatory network ( composed in C . elegans at least by six and most likely more factors ) and the fast evolutionary rate of regulatory regions , it is conceivable that the ancestral common serotonergic regulatory network has significantly diverged between these two animal groups . We propose that this deep homology might be the result of a common ancestor cell type , although , as we do not have enough information about the serotonergic regulatory programs in other animal groups , an alternative scenario is that they might have arisen independently in nematodes and vertebrates and thus although some components would have been convergently employed others could be species specific . If HSN and mouse raphe serotonergic neurons were homologous cell types , we would predict that they are also functionally homologous . Serotonergic systems in all animal groups function as facilitators of motor output , with 5HT promoting a switch between states ( Gillette , 2006 ) . Interestingly , C . elegans 5HT signaling in HSN neurons also facilitates motor output . Egg-laying behavior transitions from inactive to active states of egg laying , and 5HT signaling in HSN mediates the onset of the active phase ( Waggoner et al . , 1998 ) . Thus , HSN and mouse serotonergic neurons would share deep homology , as well as molecular and functional homology . Deep homology of specific nervous system structures has been previously proposed . Conserved TF expression patterns in annelid antero-posterior nervous system axis , including the serotonergic progenitor region , was used to propose the existence of a common Bilaterian ancestor with centralized nervous system ( Denes et al . , 2007; Tomer et al . , 2010 ) . Additionally , homologous TF expression patterns have also been used to propose the presence of a visceral nervous system in the common Bilaterian ancestor ( Nomaksteinsky et al . , 2013 ) . Altogether , these results suggest that , despite considerable divergence in neuronal architecture and connectivity , deep homology could underlie the specification of diverse neuron subtypes . The identification of homologous regulatory programs could help identify homologous neuronal types in distant species . C . elegans culture and genetics were performed as described ( Brenner , 1974 ) . Strains used in this study are listed in Supplementary file 3 . Animals of C57Bl/6JRccHsd genetic background were housed in an animal care facility with a 12 hr dark/light cycle and had free access to food and water . All experiments were performed according to the animal care guidelines of the European Community Council ( 86 ⁄ 609 ⁄ EEC ) and to Spanish regulations ( RD1201 ⁄ 2005 ) , following protocols approved by the ethics committees of the Consejo Superior Investigaciones Científicas ( CSIC ) . Gene constructs for cis-regulatory analyses were generated by cloning into the pPD95 . 75 vector . For the identification of the putative binding sites the following consensus sequences were used: ETS: CGGAWR ( Wyler et al . , 2016 ) , GATA: GATA ( Merika and Orkin , 1993 ) ; HLH: CAGAA/ACGTG MatInspector Software ( Cartharius et al . , 2005 ) ; INSM: KNNWGSGG ( Breslin et al . , 2002 ) ; SPALT: TTGTST ( Toker AS 2003 ) and MatInspector Software ( Cartharius et al . , 2005 ) ; POU: WTKCAT ( Weirauch et al . , 2014 ) and ( Sze et al . , 2002 ) . Mutagenesis was performed by Quickchange II XL site-directed mutagenesis kit ( Stratagene , Santa Clara , CA ) . Reporters for HSN regulatory signature analysis were generated by fusion PCR ( Hobert , 2002 ) . Generated strains and primers are listed in the Supplementary file 3 and 4 . For hlh-3 mutant rescue experiments , the entire coding sequence of hlh-3 was cloned in front of the heat shock inducible promoter ( hsp16-2 ) . The transgenic DNA mix was composed by hlh-3 cDNA ( 50 ng/µl ) , together with the co-injection markers rol-6 ( su1006 ) ( 50 ng/µl ) and ttx-3::mCherry ( 50 ng/µl ) . For HSN precocious maturation experiments , cDNAs of ast-1 and hlh-3 were amplified by PCR and cloned in front of an HSN-specific promoter that drives early expression in the HSN ( see promoter sequence below ) and the transgenic DNA mix concentrations were the same as above . When both cDNAs were co-injected , we used 25 ng/µl for co-injection markers . For rescue experiments , cDNAs corresponding to the entire coding sequence of ast-1 , hlh-3 , egl-46 , Pet1 , Ascl1 , Insm1 , Sall2 and Gata2 were amplified by PCR and cloned in front of cell-specific promoters: bas-1prom , cat-4prom and kal-1prom ( primers in Supplementary file 4 ) . The transgenic DNA mix was composed by the DNA of interest [ast-1 ( 50 ng/µl in HSN early maturation experiments and 5 ng/µl in HSN rescue experiments ) , hlh-3 ( 50 ng/µl ) , egl-46 ( 50 ng/µl ) , Pet1 ( 10 ng/µl ) , Ascl1 ( 50 ng/µl ) , Insm1 ( 50 ng/µl ) , Gata2 ( 50 ng/ul ) and Sall2 ( 20 ng/µl ) ] , the co-injection markers rol-6 ( su1006 ) ( 50 ng/µl ) and ttx-3::mCherry ( 50 ng/µl ) and , when necessary , pBlueScript as carrier DNA . DNA was injected into N2 animals and then crossed with their respective mutant strains . ast-1 and hlh-3 reporter strains were generated using CRISPR/Cas9-mediated fluorescent protein knock-in , as described in ( Dickinson et al . , 2013; Dickinson et al . , 2015 ) . For homology arm recombination , we used plasmids containing a self-excising selection cassette: the GFP-containing pJJR82 plasmid ( Addgene , Cambridge , MA ) in the case of ast-1 , and the mNeonGreen-containing pDD268 plasmid ( Dickinson et al . , 2015 ) in the case of hlh-3 . To target Cas9 to the specific genomic locus , we used the single guide RNA sequence GGGGTGACTATCGATAAAGA for ast-1 , and GCTATGATGATCACCAGAAG for hlh-3 , cloned in the pDD162 ( Addgene ) and the pJW1219 ( Addgene ) plasmids respectively . Injection mixes consisted on the Cas9–sgRNA plasmid ( 50 ng/µl for ast-1 and 100 ng/µl for hlh-3 ) , the repair template ( 10 ng/µl for ast-1 and 20 ng/µl for hlh-3 ) , and a pharyngeal co-injection marker [2 . 5 ng/μl pCFJ90 ( Pmyo-2::mCherry ) ; Addgene] . Scoring and images were performed using 60X objective in a Zeiss Axioplan2 microscope . Lack of GFP signaling was considered OFF phenotype . As we observed no appreciable bias in reporter expression between left and right HSN neurons , percentages were calculated regardless of side . cis-regulatory reporter and mutant scoring was performed using young adult worms maintained at 25°C , unless indicated . For cis-regulatory analysis a minimum of 30 animals ( 60 HSN cells ) per line were scored . For mutant analysis at least 100 HSN cells , roughly corresponding to 50 animals , were scored for each genotype . For double mutant analysis , we scored young adult worms and we included an extra phenotype category termed ‘dim’ whenever fluorescence was obviously reduced but still detectable . For HSN regulatory signature analysis , the three lines showing strongest GFP under the dissecting scope were selected for scoring under the microscope . To prepare figures for publication , images were cropped and rotated , brightness and contrast were adjusted , and maximum intensity projections ( where applicable ) were performed using FIJI . No other image manipulations were performed . For wild type TF expression analysis at HSN birth , an unc-86 fosmid reporter was crossed with the desired TF reporter in order to construct double reporter strains , when possible . UNC-86 is expressed in the HSN after cell cycle exit , approximately 400 min after fertilization and coinciding with embryonic comma stage , which was chosen as analytical time point ( Desai et al . , 1988; Finney and Ruvkun , 1990 ) . In the particular case of hlh-3 , 1 to 2 cell-stage embryos with the endogenous gene tagged were selected and mounted [0 hr post-fertilization ( hpf ) to 0 . 8 hpf , respectively] , incubated at 25°C and analyzed at different time points . We determined that HLH-3 is initially expressed in the HSN/PHB precursor cell ( approximately five hpf ) and maintained in the postmitotic HSN . HSN cells were identified relative to nearby landmark cell deaths ( Sulston et al . , 1983 ) . The rest of developmental stages of the worm were identified by standard anatomical features . For hlh-3 mutant time-specific rescue experiments using the hsp16-2 promoter , synchronized worms were grown until early L4 larva stage , when they received three heat shock pulses ( 30 min at 37°C ) with 2 hr resting intervals . Animals were analyzed the next morning at young adult stage . Data was categorically classified as ‘on’ or ‘off’ and the significance of the association was examined using the two tailed Fisher’s exact test . For double mutant analysis , ‘phenotype’ vs . ‘no phenotype’ was compared and thus , ‘dim’ and ‘off’ were considered under the category ‘phenotype’ . The null hypothesis was that the level of expression in the double mutant would be equal to the product of the levels of expression in single mutants ( Mani et al . , 2008 ) . Whatever statistically deviated from the expected , was considered genetic interaction; Pearson’s chi-squared test was used . C . elegans serotonin antibody staining was performed using the tube fixation protocol ( McIntire et al . , 1992 ) . Briefly , synchronized young adult hermaphrodites were fixed in 4% paraformaldehyde ( PFA ) for 18 hr , with β-mercapto-ethanol for another 18 hr , with 1 mg/ml collagenase ( Sigma Aldrich , Merk , Darmstadt , Germany ) for 90 min and incubated for 24 hr with rabbit anti-5HT antibody ( 1:5000; Sigma Aldrich ) . Alexa 555-conjugated donkey anti-rabbit ( 1:500; Molecular probes ) was used as secondary antibody . For mouse immunohistochemistry , freshly isolated E11 . 5 embryos from C57Bl/6JRccHsd were fixed by immersion in 4% PFA . Rabbit anti-Sall2 ( 1:100; Sigma Aldrich ) , goat anti-Brn2 ( 1:100; Santa Cruz Biotechnology , Santa Cruz , CA ) , rabbit anti-5HT ( 1:5000; Sigma Aldrich ) and goat anti-5HT ( 1:200; Abcam , Cambridge , UK ) antibodies were used . As secondary antibodies Alexa 555-conjugated donkey anti-rabbit and anti-goat , and Alexa 488-conjugated donkey anti-rabbit and anti-goat were used ( 1:600; Molecular probes , Invitrogen , Eugene , OR ) . Immunofluorescence samples were analyzed and photographed using a confocal TCS-SP8 Leica microscope . HSN early promoter was generated from tph-1prom2 in which we incidentally found a point mutation that caused L1 expression: GTAGTAAGCTCCGATGCGTTCCCGTTCATTATTCTTCTTCAATAAATTCGAA ATCTGACATCATTCTCATCTTTTCCCATCATCACAAGCCGTGGGCTCATTTA TTCTCCCACGGAAACCATGACAGCAAAAATAAATAGAGTGGCGCCTTATTC GACTCATTTCGTTTTTTTTTCTCCGGATATTAGATTGTGTGGCAGGCGGCTC CATTGTATATTcCGaaCCGAATTtttGAAGCACCACGCCATCGGATATCTAAAA GAGGAGGTGTCTTTGTTTGCGCATAATAAAACAATCAATCAACACAGCAAA GACCCCTCTCAACCTCATTTCATGATTTTCTTTGGTTTTTAGGTAGCATTGC TCTCTTCAATCAT * Mutated nucleotides with respect to tph-1prom2 are indicated in lowercase letters . RNAi experiments were performed by the standard feeding protocol ( Kamath et al . , 2003 ) . rrf-3 ( pk1426 ) background was used to sensitize worms to the RNAi effects . For maintenance experiments , animals were grown under normal food ( OP50 ) until young adult stage . At this stage we scored tph-1::yfp and cat-1::MDM2::gfp expression in the HSN to confirm that all animals expressed the fluorescent protein and then we transferred animals to RNAi plates with HT115 bacteria ( Novagen ) transfected with RNAi clones . Worms were incubated at 15°C for 72 hr and then HSN fluorescent expression was scored . For F1 RNAi scoring , we bleached gravid adults in OP50 plates , eggs were allowed to hatch and worms grew in RNAi treated food . We scored their progeny , which had developed under the embryonic effects of RNAi knock down ( F1 scoring ) . The experiment was performed in two independent replicates with similar results . As a negative control the L4440 empty vector was used ( pPD129 . 36 , Addgene ) . Full-length unc-86 and ast-1 cDNA into the pET-21b His tag expression vector ( EMD Millipore , Merk ) were kindly provided be Oliver Hobert . They were transformed into E . coli Rosetta2 ( DE3 ) ( Novagen ) strain . Overexpression was done by first growing the cells at 37°C in LB and Power Broth medium ( Molecular Dimensions ) respectively , supplemented with 100 μg/ml ampicillin , 100 μg/ml chloramphenicol to OD600 = 0 . 5–0 . 6 and then inducing expression with 0 . 5 mM iso-propyl-b-D thiogalactopyranoside ( IPTG , Acros Organics , Thermo Fisher , Waltham , MA ) at 37°C for 3 hr or 20°C for 16 hr , respectively . UNC-86 protein was obtained as previously explained ( Zhang et al . , 2014 ) with minor changes . Briefly , cells were collected by centrifugation and resuspended in buffer A ( 100 mM NaH2PO4 , 10 mM Tris [pH 7 . 5] , 10% glycerol ) supplemented with 1 mM phenylmethanesulfonyl fluoride ( PMSF ) . Cells were lysed by sonication . Soluble and insoluble fractions were separated by centrifugation and analyzed by SDS/PAGE . Protein was substracted from insoluble fraction as follow: insoluble fraction was resuspended in solubilization buffer ( buffer A supplemented with 8 M urea ) and loaded on a pre-equilibrated His Trap HP column ( GE Healthcare , Chicago , IL ) . The resin was washed with solubilization buffer supplemented with 10 mM imidazole , and protein was eluted with the same buffer supplemented with 500 mM imidazole . Elution buffer was exchanged by progressive dialysis to 20 mM HEPES [pH 7 . 5] , 100 mM NaCl 10% glicerol , 2 mM MgCl2 , and the protein was concentrated by centrifugation up to 1 . 3 µg/µl and stored at −80°C . For AST-1 protein , cells were collected by centrifugation and resuspended in buffer B ( 200 mM MES [pH 6 . 0] , 500 mM NaCl , 2 mM MgCl2 , 10% glycerol ) supplemented with 1 mM PMSF . Cells were lysed by sonication and soluble proteins were loaded on a His Trap HP column ( GE Healthcare ) pre-equilibrated with buffer B . The resin was washed with buffer B supplemented with 10 mM imidazole , and protein was eluted with buffer B supplemented with 300 mM imidazole . Eluted fraction was analyzed by SDS/PAGE . Imidazole was removed and protein concentrated by centrifugation up to 0 . 3 µg/ul , and stored at −80°C . egl-18 cDNA was cloned into pcDNA . 3 vector followed by His tag sequence and transfected with Lipofectamine-2000 ( Invitrogen ) in HEK293T cells . HEK293T cells were grown in DMEM 10% FBS . After 24 hr , cells were lysed with the following buffer: 1 mM EDTA , 0 . 5% Triton , 20 mM β-glicerolP , 0 . 2 mM PMSF , 100 µM Na3VO4 and protease inhibitor . EMSAs were performed incubating UNC-86 and AST-1 proteins in a buffer containing 10 mM Tris [pH 7 . 5] , 50 mM NaCl , 1 mM MgCl2 , 4% glycerol , 0 . 5 mM DTT , 0 . 5 mM EDTA , 1 µg of poly ( dIdC ) , 6 µg of bovine serum albumin and labeled probes for 20 min at room temperature . For EGL-18 , protein extracts were incubated in 20 mM Hepes , 50 mM NaCl , 5 mM MgCl2 , 5% glycerol , 1 mM DTT , 0 . 1 mM EDTA , 1 µg of poly ( dIdC ) , 6 µg of bovine serum albumin and 1 µg anti-6xhistag antibody ( Abcam ) at 4°C for 30 min . As negative control , anti-GFP antibody ( Roche , Basel , Switzerland ) was used . Then , labeled probes were added and incubated for 20 min at room temperature . Finally , samples were loaded onto a 6% ( 37 . 5:1 acrylamide: bisacrylamide ) gel and run at 150 V for 4 hr . Gels were then dried and visualized using Fujifilm FLA-500 . Probe sequences are listed in Supplementary file 4 . Primers were annealed and end-labeled with ATP [γ−32P] ( Perkin Elmer , Waltham , MA ) using T4 PNK ( Thermo Scientific ) according to the manufacturer’s specifications . Unless otherwise indicated , all the analyses were performed using R and Bioconductor ( Huber et al . , 2015 ) . For C . elegans regulatory signature analysis , we built PWMs from the functional motifs found in the 5HT pathway genes CRMs ( Figure 5 ) . Next , we downloaded upstream and intronic gene regions from WormBase version 262 and classified genes in three groups: genes known to be expressed in HSN , genes expressed in neurons and non-neuronal genes , according to WormBase annotations on gene expression and/or belonging to the published neuronal genome ( Hobert , 2013 ) . PWMs were aligned to genomic sequences and we retrieved matches with a minimum score of 70% . To increase specificity , we removed all matches that did not bear an exact consensus sequence for the corresponding TF family ( ETS: YWTCCG , GATA: DGATAD , HLH: SCAGAA , INMS: CCSCWNNM , SPALT: TTGTST , POU: WTKCAT ) . Then , we performed a sliding window search to find regions that included at least one match for four or more of the 6 TF types . Windows were separated according to the number of different motifs that they bore ( 4 , 5 or 6 ) , and then overlapping regions were merged . Embryonic stem cell enhancers median size has been reported to be around 800 bp ( Parker et al . , 2013 ) ; therefore , the initial search was performed with a maximum length restriction of either 600 , 700 or 800 bp . Differences between HSN-expressed genes and other gene groups was greater when the maximum length was set to 700 bp , thus we kept this maximum window length for the rest of the analyses . To assess enrichment in signature in HSN-expressed genes we sampled 10 , 000 groups of 96 genes that ( 1 ) had not previously been reported to be expressed in the HSN , ( 2 ) at least one ortholog had been described in other Caenorhabditis species ( C . briggsae , C . japonica , C . remanei or C . brenneri ) , and ( 3 ) such that their upstream and intronic regions were similar in length , on average , to those of the HSN-expressed genes ( Mann-Whitney U test , p-value>0 . 05 ) . We compared the distribution of the proportion of genes with signature ( 4 , 5 or 6 different motifs ) in these groups to the HSN-expressed gene group . We consider the enrichment in signature to be significant when the percentile of the HSN-expressed group is above 95 . In order to assess signature conservation , we performed a similar motif search using other nematode genomes also available from WormBase ( C . briggsae , C . japonica , C . remanei , C . brenneri ) and we considered the signature to be conserved if HSN regulatory windows were found in all orthologous genes , at least 4 , 5 or 6 motifs for 4 , 5 and 6-motif C . elegans windows . Gene ontology analysis was performed using GOrilla software , using C . elegans coding genome ( 19 . 276 genes ) as control list ( Eden et al . , 2009 ) . For hierarchical clustering , we used curated data from WormBase ( Hobert et al . , 2016 ) to generate a matrix with gene expression profiles for the 118 C . elegans hermaphrodite anatomical neuronal classes . Pan-neuronal genes and neurons in which less than 30 genes had been reported to be expressed were excluded . We built a similar matrix with mouse gene expression data from RNA-seq experiments , either from adult raphe nuclei divided into different rhombomeres ( R1Dorsal , R1 Medial , R2 , R3 , R5 , R6 ) ( Guillemot and Hassan , 2017 ) or from cortical neurons that served as negative control ( Molyneaux et al . , 2015 ) ) . To transform the quantitative RNA-seq data into a presence-absence binary matrix . We considered values above 19 CPM as present and values below that threshold as absent because this cut-off produces a list of approximately 7000 expressed genes in each Raphe sample ( roughly a third of the genome that is what is being estimated as expressed in a given cell type ) . Nevertheless , results were consistent in all conditions when considering cutoffs ranging from 9 to 140 CPM , after which HSN-raphe cluster robustness started to decline ( low AU and BP values , not shown ) . To assign mouse ortholgs to C . elegans genes , we combined orthology relationships between mouse and worm genes annotated in the ENSEMBL database and worm-human orthology relationships reported in Shaye and Greenwald ( 2011 ) . In the last case , we used ENSEMBL database again to assign mouse orthologs to human genes . In ( Molyneaux et al . , 2015 ) ) , ENSEMBL , OrthoMCL , InParanoid and Homologene methods are combined to identify orthologs . Thus , we combined both sources to have a wider coverage of orthology relationships than using ENSEMBL or ( Molyneaux et al . , 2015 ) data alone . Worm genes without any mouse ortholog and genes that were not expressed in any worm neuron were removed . Whenever a worm gene had more than one mouse ortholog , it was duplicated in the worm data set . For hierarchical clustering , this binary matrix containing mouse and worm expression data was fed to the pvclust function in the pvclust R package ( Shimodaira , 2002 ) , which uses a bootstrapping technique to calculate p-values for each cluster , the AU and BP values ( Shimodaira , 2002 ) . Parameters were set as follows: method . hclust = ‘average’ , method . dist= ‘binary’ , nboot = 10 , 000 , r = seq ( 0 . 5 , 1 . 4 , by = 0 . 1 ) . The standard error of the PV and AU values was approximately 0 . 1% for most clusters , including the HSN-raphe cluster . Also , as a control , 100 random sets of 96 expressed genes ( the same number of genes that are expressed in the HSN ) were generated from the worm gene pool . Each random set contained the four 5HT pathway genes ( tph-1 , cat-1 , cat-4 and bas-1 ) plus 92 randomly picked genes from the genes expressed in C . elegans neurons . This data set was merged with mouse raphe nuclei expression profile and pvclust was run as before .	All cells in the body essentially share the same DNA , despite looking very different and playing a range of roles . The reason that cell types are so different from one another is because of the way they interpret the DNA . Each different type of cell uses a specific subset of the genes within the genome . The part of the DNA that controls which cell will use which genes and when is called the regulatory genome; this DNA is not translated into proteins . The regulatory genome is much less well understood than the protein-coding genome . At present , when a new species is discovered , it is often possible to sequence its DNA and deduce where the protein-coding genes are and what roles they might play . However , it is not yet possible to do the same for the regulatory genome . Finding a way to do this is an important step towards understanding when and where each of the organism’s genes is active . Lloret-Fernández , Maicas , Mora-Martínez et al . focused on the regulatory genome of nerve cells that use a chemical messenger called serotonin in the nematode worm Caenorhabditis elegans . First , they studied mutations in six genes that code for transcription factors that are active in this cell type . Transcription factors are proteins that identify and bind to specific regions of the genome to control the activity of nearby genes . These six mutants failed to correctly activate the regulatory genome of this nerve cell , which was measured using a genetic approach that caused the nerves to glow green under a microscope when the regulatory genome was active . Further experiments then confirmed that all six transcription factors must act together to identify and activate the regulatory genome in this particular nerve cell . The fact that the DNA sites that these transcription factors bind are clustered close to each other means they can be used as a marker to help decode the active regulatory genome of this class of nerve cell . This is a small step towards understanding how the regulatory genome works . Comparisons with similar nerve cells from mammals found that the equivalent transcription factors have the same role , suggesting that they may be broadly conserved across species . Understanding the regulatory genome better could eventually lead to new treatments for certain genetic conditions , as many mutations associated with diseases appear outside the protein-coding genome .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2018	A transcription factor collective defines the HSN serotonergic neuron regulatory landscape
The degradation and recycling of cellular components is essential for cell growth and survival . Here we show how selective and non-selective lysosomal protein degradation pathways cooperate to ensure cell survival upon nutrient limitation . A quantitative analysis of starvation-induced proteome remodeling in yeast reveals comprehensive changes already in the first three hours . In this period , many different integral plasma membrane proteins undergo endocytosis and degradation in vacuoles via the multivesicular body ( MVB ) pathway . Their degradation becomes essential to maintain critical amino acids levels that uphold protein synthesis early during starvation . This promotes cellular adaptation , including the de novo synthesis of vacuolar hydrolases to boost the vacuolar catabolic activity . This order of events primes vacuoles for the efficient degradation of bulk cytoplasm via autophagy . Hence , a catabolic cascade including the coordinated action of the MVB pathway and autophagy is essential to enter quiescence to survive extended periods of nutrient limitation . Evolutionary conserved selective and non-selective protein degradation pathways are essential for cell growth and survival . The ubiquitin-proteasome system ( UPS ) mediates selective poly-ubiquitination of cytoplasmic proteins and their degradation at 26S proteasomes for regulatory and quality control functions . Mis-folded proteins in the endoplasmic reticulum ( ER ) are also ubiquitinated , extracted by the ER-associated protein degradation system ( ERAD ) and degraded at 26S proteasomes in the cytoplasm ( Vembar and Brodsky , 2008 ) . Macro-autophagy ( hereafter autophagy ) non-selectively transports bulk cytoplasm into lysosomes . Therefore the induction of autophagy is tightly controlled: under normal growth conditions autophagy operates on a basal level because it is suppressed by signaling from the target of rapamycin complex 1 ( TORC1 ) ( Kamada et al . , 2000; Loewith and Hall , 2011; Zoncu et al . , 2011b ) . In response to cellular stress , such as nutrient depletion ( e . g . , of amino acids ) , TORC1 is inactivated ( Loewith and Hall , 2011 ) and autophagy is strongly induced . Deregulation of autophagic processes is implicated in metabolic and infectious diseases as well as in cancer or neurodegeneration ( Rubinsztein et al . , 2012 ) . Once induced , the autophagic machinery begins to sequester cytoplasmic components , ribosomes and organelles within a large double-membrane compartment termed the autophagosome ( Yang and Klionsky , 2010; Kraft and Martens , 2012; Mizushima et al . , 2011 ) . In addition , some core components of the autophagic machinery such as LC3/Atg8 are transcriptionally induced ( Kirisako et al . , 1999 ) . Direct fusion of autophagosomes with lysosomes delivers autophagic bodies and the sequestered cargo into the lysosomal lumen . Alternatively , autophagosomes can first fuse with multivesicular bodies ( MVBs ) to form so-called amphisomes , before they fuse with lysosomes ( Seglen et al . , 1991 ) . Finally , the breakdown of autophagic bodies and the efficient degradation of autophagic cargo inside lysosomes is required to recycle amino acids , nucleotides , carbohydrates and lipids back to the cytoplasm . The recycling of these key metabolic building blocks protects cells from their fatal depletion and thus maintains cellular homeostasis to survive nutrient limitation ( Onodera and Ohsumi , 2005; Vabulas and Hartl , 2005; Jones et al . , 2012; Suraweera et al . , 2012 ) . Therefore evolutionary conserved starvation programs in mammalian cells and yeast expand and strengthen this intracellular recycling system by enhancing the de novo synthesis of vacuolar/lysosomal hydrolases ( Gasch et al . , 2000; Sardiello et al . , 2009; Settembre et al . , 2011; Shen and Mizushima , 2014 ) . In addition to autophagy , TORC1 also regulates ubiquitin-mediated endocytosis of integral plasma membrane proteins . On the one hand , TORC1 signaling was required to promote the endocytosis of certain plasma membrane proteins ( MacGurn et al . , 2011 ) . On the other hand , inactivation of TORC1 either by rapamycin or starvation , triggered the endocytosis of other plasma membrane proteins that were subsequently degraded in an ESCRT ( endosomal sorting complex required for transport ) -dependent manner via the MVB pathway ( Schmidt et al . , 1998; Jones et al . , 2012; Lang et al . , 2014 ) . The extent to which starvation induces plasma membrane remodeling has yet to be determined . Furthermore , how the subsequent ubiquitin-dependent degradation of membrane proteins via the MVB pathway helps to meet the specific metabolic and energetic demands of cells during nutrient limitation is not fully understood . Therefore it is also not clear how selective ( MVB ) and non-selective ( autophagy ) lysosomal proteolysis pathways cooperate to mediate cell survival during nutrient limitation . To comprehensively address these questions we have used quantitative proteomics . Our results demonstrate that within the first 3 hr of amino acid starvation many integral plasma membrane proteins , including high-affinity amino acid permeases , glucose transporters and G-protein coupled receptors , were selectively removed from the cell surface by endocytosis and subsequently targeted into vacuoles via the ESCRT-dependent MVB pathway and degraded , while others remained stable or were up-regulated ( e . g . , the general amino acid permease , Gap1 ) . This comprehensive and selective remodeling of the plasma membrane appeared to be completed within 3–4 hr of starvation . Autophagy was also immediately activated upon starvation and remained active throughout starvation . Surprisingly , early during starvation the selective degradation of membrane proteins via the MVB pathway was mainly responsible to maintain critical levels of free intracellular amino acids that were sufficient to uphold protein synthesis and promote the corresponding adaptation of the proteome . Most notably this included the de novo synthesis of vacuolar hydrolases , which boosted the proteolytic activity of vacuoles to support the efficient degradation of autophagic cargo . The continuous delivery and degradation of autophagic cargo further enhanced intracellular amino acid recycling and was ultimately essential to restore intracellular amino acid pools of cells during extended starvation . These findings reveal an unexpected role for the MVB pathway in maintaining intracellular amino acid homeostasis and thereby promoting the up-regulation of vacuolar hydrolases early during starvation , which is tightly coordinated with autophagy . This catabolic cascade is ultimately required to allow starving cells to complete their cell division cycle and enter a quiescent state for survival . To understand how the MVB pathway , autophagy and proteasomal degradation cooperate during nutrient limitation , we first analyzed the starvation-induced degradation of model proteins in yeast . To assess selective membrane protein degradation via the MVB pathway , we followed the ubiquitin-dependent endocytosis of the plasma membrane methionine permease , Mup1-GFP and its transport into the vacuole in response to starvation ( for amino acids and nitrogen sources ) ( Beck et al . , 1999; Menant et al . , 2006; Jones et al . , 2012 ) . Under rich growth conditions Mup1-GFP is mainly found at the plasma membrane and very little is degraded ( Figure 1A , B ) . Yet , within 3 hr after the onset of starvation the majority of Mup1-GFP was removed from the cell surface , delivered into vacuoles and degraded ( Figure 1A , B ) . The proteolytic degradation of Mup1-GFP inside vacuoles released free GFP , which remained stable and was monitored by western blotting ( Figure 1A ) . The starvation-induced delivery of Mup1-GFP into the vacuole was dependent on the ESCRT machinery but was not affected in an autophagy ( atg8∆ ) mutant ( Figure 1B ) . In an ESCRT ( vps4∆ ) mutant , the MVB pathway was blocked and Mup1-GFP was not delivered into the vacuole but instead accumulated on the class E compartment and at the plasma membrane ( Figure 1B ) . 10 . 7554/eLife . 07736 . 003Figure 1 . Starvation induces selective and non-selective protein degradation pathways . ( A ) WT cells expressing Mup1-GFP , Rpl25-GFP , Rps2-GFP or GFP-Atg8 were grown in rich medium ( 0 hr ) or starved as indicated . Cell lysates were analyzed by SDS-PAGE and western blot ( WB ) using the indicated antibodies . residual anti-GFP signal after re-probing the membrane with anti-Pgk1 antibody . ( B ) Fluorescence microscopy of Mup1-GFP in WT cells , vps4∆ mutants and atg8∆ mutants growing under rich or starvation conditions . ( V ) acuoles , ( P ) lasma ( M ) embrane and class ( E ) compartments . Scale bar = 5 µm . ( C , D ) Whole cell lysates of WT cells or the indicated mutants grown under rich conditions or starved for the indicated times were separated by SDS-PAGE and analyzed by western blot using the indicated antibodies . ( C ) pdr5∆ cells were treated with the proteasome inhibitor MG132 ( 50 µM ) or vehicle ( DMSO ) during starvation . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 00310 . 7554/eLife . 07736 . 004Figure 1—figure supplement 1 . Induction of autophagy . ( A ) SDS-PAGE and western blot analysis of WT cells grown under rich conditions or starved using the indicated antibodies . ( B ) Vacuolar hydrolase-deficient cells ( pep4∆ , prb1∆ , prc1∆ ) analyzed as in ( A ) . ( C ) Pho8∆60-specific alkaline phosphatase activity was measured in WT , and atg8∆ cells under rich conditions and after starvation ( n = 6 , mean ± SD ) . WT Pho8∆60 activity after 20 hr of starvation was set to 100% . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 004 To define the timing of starvation-induced degradation of Mup1-GFP in the context of eukaryotic starvation programs , we compared it to the delivery of bulk cytoplasm via autophagy . Therefore we determined the degradation of highly abundant selective ( ribosomes ) and non-selective ( Fba1 ) autophagic cargoes . Growing yeast cells contain about 200 , 000 ribosomes that occupy up to 30–40% of the cytoplasmic volume ( Warner , 1999 ) . Upon starvation , otherwise stable ribosomes are among the first autophagic cargoes and rapidly degraded by selective ( ribophagy ) and non-selective autophagy ( Kraft et al . , 2008; Ossareh-Nazari et al . , 2014 ) . We monitored the release of free GFP from two different ribosomal proteins by western blotting: Rpl25-GFP ( large subunit ) and Rps2-GFP ( small subunit ) . Both are fully functional GFP fusion proteins that incorporate into ribosomes ( Kraft et al . , 2008 ) . When equal amounts of cell lysates were subjected to western-blot analysis , the protein levels of full length Mup1-GFP and the GFP-tagged ribosomal subunits were comparable ( Figure 1A , lanes 6 , 16 ) . After 3 hr , at a time when the majority of full length Mup1-GFP was already degraded , free GFP from Rpl25 was first detected , showing that autophagy was also delivering cytoplasmic contents into the vacuole ( Figure 1A , lane 8 ) . During subsequent starvation the protein levels of free GFP from both ribosomal subunits increased . Monitoring the autophagy-dependent degradation of Fba1-GFP , one of the most abundant cytoplasmic proteins with approximately 1 . 000 . 000 molecules/cell ( Ghaemmaghami et al . , 2003 ) , yielded similar results . Free GFP was first detected after 3 hr of starvation and the protein levels free GFP strongly increased during subsequent starvation ( Figure 1—figure supplement 1A ) . To determine the earliest possible starvation-induced autophagic activity , we monitored the transport and degradation of fully functional GFP-Atg8 . Atg8 is a core component of the autophagic machinery that remains conjugated to the inner membrane of all selective and non-selective autophagosomes , including cytoplasm to vacuole targeting ( cvt ) -vesicles . Therefore Atg8 is degraded together with autophagic cargo inside vacuoles . To be able to compare the degradation of GFP-Atg8 to Mup1-GFP , 10 times more lysate of cells expressing GFP-Atg8 was subjected to western blot analysis ( Figure 1A ) . Small amounts of free GFP released from GFP-Atg8 inside vacuoles could be readily detected by western blot analysis 1 hr after the onset of starvation and the levels of free GFP strongly increased at 3 hr of starvation ( Figure 1A , lane 27–30 ) . These findings are consistent with the strong increase of endogenous Atg8 levels during starvation ( Figure 1—figure supplement 1B ) as observed earlier ( Kirisako et al . , 1999 ) . Previous work also demonstrated that Atg8 protein levels control the size of autophagosomes but not the frequency ( about 9 autophagosomes/hour ) by which they are formed ( Abeliovich et al . , 2000; Xie et al . , 2008 ) . Hence , the increase in Atg8 protein levels during the first 4 hr of starvation would result in the formation of bigger ( but not more ) autophagosomes that could capture larger volumes of cytoplasm later during starvation . Our results for the early degradation of GFP-Atg8 and the continuous increase in autophagic degradation of highly abundant selective as well as non-selective cargoes throughout starvation are fully consistent with this model . This idea was further supported using the Pho8∆60 assay , a sensitive method to measure bulk autophagy ( Noda et al . , 1995 ) . Pho8∆60 activity was low under rich conditions , began to increase during the first 3 hr of starvation and continuously increased during extended periods of starvation ( Figure 1—figure supplement 1C ) . These results show that autophagy is immediately activated upon starvation and delivers increasing volumes of cytoplasmic material into the vacuole with ongoing starvation ( Figure 1 ) . Additionally , we investigated how cytoplasmic proteins were degraded at proteasomes upon starvation . Therefore , we employed a ubiquitin-GFP ( Ub-GFP ) fusion protein , which is an established reporter for proteasomal activity ( Johnson et al . , 1992; Vabulas and Hartl , 2005 ) . It is detected at low levels in proliferating cells reflecting the equilibrium between its rapid degradation and its synthesis . Upon starvation , Ub-GFP was rapidly degraded at 26S proteasomes . The degradation of the reporter was exclusively dependent on proteasomal degradation but did not require autophagy or the MVB pathway ( Figure 1C , lanes 5–7; Figure 1D ) . Overall , these findings indicate that starvation triggered protein degradation by different selective and non-selective degradation pathways: the constitutive protein degradation via the proteasome was active from the onset of starvation and was previously suggested to play a key role upon acute nutrient restriction ( Vabulas and Hartl , 2005 ) . Our results further suggest that both autophagy and starvation induced-endocytosis were simultaneously activated early during starvation . Autophagy continuously delivered ever-increasing volumes of cytoplasm into vacuoles , whereas the starvation-induced degradation of membrane proteins was completed within 3 hr . These findings suggest an important role for the MVB pathway early during starvation . While protein degradation at 26S proteasomes provides an immediate amino acid pool for protein synthesis already within minutes of acute starvation ( Vabulas and Hartl , 2005 ) and autophagy is required to supply amino acids during extended periods of starvation ( Onodera and Ohsumi , 2005 ) , the relative contribution of the MVB pathway to overall amino acid homeostasis was not clear . Therefore we next measured the intracellular levels of 18 different amino acids in isogeneic WT cells , MVB ( vps4∆ ) or autophagy ( atg8∆ ) mutants as well as double mutants ( vps4∆ , atg8∆ ) by liquid chromatography ( Altmann , 1992 ) . These strains were auxotrophic for the amino acids lysine and leucine . When grown in synthetic medium supplemented with amino acids ( rich ) , the intracellular free amino acid levels were comparable in WT cells and autophagy mutants ( atg8∆ ) ( Figure 2A ) , but slightly lower in vps4∆ mutants , which was mainly due to reduced lysine and arginine levels ( Figure 2A , B ) . 10 . 7554/eLife . 07736 . 005Figure 2 . Changes in free amino acid levels and protein synthesis during starvation . ( A ) Cells were grown to mid-log phase ( rich ) and starved as indicated . Free amino acids were extracted and analyzed by liquid chromatography . Data are represented as the sum of free amino acids ( mg ) per gram of dry yeast . Mean ± SD , n ≥ 3 . ( B ) Changes in individual amino acids from ( A ) normalized to maximal values . Mean ± SD , n ≥ 3 . ( C , D ) Cells grown under the indicated conditions were incubated for 5 min with 35S-labeled Met and Cys . ( C ) 35S-incorporation was analyzed by SDS-PAGE and digital autoradiography . Coomassie staining shows equal protein loading . ( D ) Quantification of 35S-incorporation under rich conditions and after 1 , 2 and 4 hr of starvation by liquid scintillation counting . Incorporation under rich conditions was set to 100% . Mean ± SEM , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 00510 . 7554/eLife . 07736 . 006Figure 2—figure supplement 1 . Changes in free amino acids levels during starvation in prototrophic yeast . ( A ) Prototrophic cells were grown to mid-log phase ( rich ) and starved as indicated . Free amino acids were extracted and analyzed by liquid chromatography . Data are represented as the sum of free amino acids ( mg ) per gram of dry yeast . Mean ± SD , n = 3 . ( B ) Changes in individual amino acids from ( A ) were normalized to maximal values . Mean ± SD , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 006 1 hr after starvation in synthetic medium without amino acids and ammonium salts , the total free amino acid pool decreased to similar levels in WT cells and all mutant strains ( Figure 2A , B ) . In WT cells the levels of most amino acids continued to decrease for another hour . Interestingly , at around 4 hr of starvation the overall levels of amino acids almost fully recovered , suggesting strong amino acid recycling . However , the levels of arginine and lysine , which were among the most abundant free amino acids , decreased further . The levels of glutamine , threonine and glycine did not recover very well , while the levels of other amino acids ( particularly of glutamate ) increased . The recovery of amino acid levels after 4 hr of starvation was strongly dependent on autophagy . These results are at large consistent with previous findings , where an approximately threefold reduction in intracellular amino acid levels was detected during the first 2 hr of starvation and autophagy was required for the partial recovery of amino acid levels from 3 to 6 hr of starvation ( Onodera and Ohsumi , 2005 ) . In our strain background the levels of amino acids were generally lower under rich growth and we observed an approximately twofold reduction in intracellular amino acids during the first 2 hr of starvation . At 4 hr of starvation amino acid levels fully recovered in an autophagy dependent manner ( Figure 2A , B ) . In addition , our findings showed that the MVB pathway essentially contributed to maintain the overall levels of free intracellular amino acids . 1 hr after starvation , the amino acids levels decreased similar in vps4∆ mutants , autophagy mutants ( atg8∆ ) and WT cells . However , after 2 hr the overall amino acid levels were lower in vps4∆ mutants compared to WT cells and autophagy mutants . The levels of 14 individual amino acids were lower in vps4∆ mutants when compared to WT cells or autophagy mutants . Moreover the amino acid levels failed to recover during extended starvation ( Figure 2A ) . From 2 hr onwards , the amino acid levels were always lowest in the double mutants ( vps4∆ , atg8∆ ) ( Figure 2A , B ) . To exclude effects contributed by amino acid auxotrophies , the same analysis was performed in a different genetic background with fully prototrophic WT cells and the respective vps4∆ and atg8∆ single mutants ( Mülleder et al . , 2012 ) . During the first 2 hr of starvation , the amino acid levels initially declined in the prototrophic WT cells , but not as strongly as in auxotrophic strains , and recovered at around 4 hr of starvation , which was dependent on autophagy ( Figure 2—figure supplement 1A , B ) . In prototrophic vps4∆ mutants , the levels of most amino acids ( 12 ) were lower after 2 hr of starvation when compared to WT cells or autophagy mutants ( atg8∆ ) , as observed in the auxotrophic strains ( Figure 2A , B; Figure 2—figure supplement 1A , B ) . These results showed that the MVB pathway was essential to maintain the levels of most free intracellular amino acids within the first 2 hr of starvation , while autophagy was essential to restore intracellular amino acids later during starvation . Based on these results we next tested how the MVB pathway would contribute to uphold protein synthesis during starvation . Therefore we measured 35S-methionine/cysteine incorporation into newly synthesized proteins . Under rich conditions , 35S-label incorporation was comparable in WT cells , atg8∆ and vps4∆ single or double mutants ( Figure 2C , D ) , although the methionine permease Mup1 was more abundant in ESCRT mutants . Already 1 hr after starvation , 35S-label incorporation was reduced to 40% in WT cells . During the next 3 hr of starvation , WT cells managed to maintain protein synthesis at this level ( Figure 2C , lanes 1 , 5 , 9 , 13 , Figure 2D ) . In the autophagy-deficient atg8∆ mutant , 35S-label incorporation was initially similar to WT cells for up to 2 hr , but began to decline after 4 hr of starvation ( Figure 2C , lanes 3 , 7 , 11 , 15 , Figure 2D ) , which is consistent with the key role of autophagy in amino acid recycling . In ESCRT mutants ( vps4∆ ) , protein synthesis declined faster when compared to WT cells or autophagy mutants , which seems consistent with the more rapid decline of intracellular amino acids ( Figure 2C , lanes 2 , 6 , 10 , 14 , Figure 2D ) . vps4∆ , atg8∆ double mutants showed an additive effect , since even less 35S-label was incorporated upon starvation compared to the single deletion mutants ( Figure 2C , D ) . Taken together these findings suggest that ( i ) the MVB pathway is essential to maintain a critical pool of free amino acids for protein synthesis early during starvation . ( ii ) In the absence of the MVB pathway autophagy can only partially uphold amino acids levels and protein synthesis ( iii ) The MVB pathway and autophagy cooperate to maintain intracellular amino acids during starvation , potentially in a consecutive manner . Recent reports have shown an important role for the ESCRT machinery in higher eukaryotic cells in regulating autophagy at the stage of amphisomes fusing with lysosomes ( Nara et al . , 2002; Filimonenko et al . , 2007; Lee et al . , 2007; Rusten et al . , 2007; Metcalf and Isaacs , 2010; Spitzer et al . , 2015 ) . Therefore we next carefully examined the role of the ESCRT machinery in distinct steps of autophagy in yeast . The induction of autophagy is tightly controlled by TORC1 . Under nutrient rich growth conditions , TORC1 was active and its direct targets Sch9 and Atg13 were phosphorylated in WT cells , vps4∆ and atg8∆ mutants ( Figure 3A , B , lane 1 , 3 , 5 ) ( Kamada et al . , 2000; Urban et al . , 2007 ) . When autophagy and the MVB pathway were simultaneously disrupted ( vps4∆ , atg8∆ ) , TORC1 signaling appeared to be reduced under rich growth conditions , but not completely switched off ( Figure 3A , lane 7 ) . Upon starvation TORC1 signaling was efficiently turned off and the autophagy core component Atg13 was dephosphorylated in all strains , which is a prerequisite for the induction of autophagy ( Figure 3B ) ( Kamada et al . , 2000 ) . 10 . 7554/eLife . 07736 . 007Figure 3 . Autophagy in ESCRT mutants . ( A , B ) SDS-PAGE and western blot analysis of total cell lysates from WT cells and vps4∆ , atg8∆ single and double mutants grown in rich medium ( 0 hr ) or during starvation using the indicated antibodies . ( C ) Live-cell fluorescence microscopy of WT cells and vps4∆ mutants expressing GFP-Atg8 ( green ) and mCherry-CPS ( red ) under rich conditions or 4 hr after starvation . ( D ) Pho8∆60-specific alkaline phosphatase activity was measured in WT , vps4∆ and atg8∆ cells under rich conditions and after 4 hr of starvation ( n = 8 , ±SD ) . WT Pho8∆60 activity under rich conditions was normalized to 100% . ( E ) Fluorescence microscopy of pHluorin-Atg8 ( green ) and mCherry-CPS ( red ) in WT cells and indicated mutants under rich conditions or after 4 hr of starvation . ( F ) Quantification of quenching of vacuolar pHluorin-Atg8 from E . ( C , E ) ( V ) acuoles and class ( E ) compartments . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 007 To assess the formation and the delivery of autophagosomes into vacuoles , we followed the transport of GFP-tagged Atg8 using live cell fluorescence microscopy . Upon starvation , GFP-Atg8 was efficiently transported into the lumen of vacuoles in WT cells and vps4∆ mutants ( Figure 3C ) , indicating that the autophagic machinery was fully operational and independent of the ESCRT machinery . This conclusion was further strengthened using the Pho8∆60 autophagy-reporter assay ( Noda et al . , 1995 ) . Pho8∆60 activity increased after 4 hr of starvation in WT and vps4∆ mutants , but not in atg8∆ cells ( Figure 3D ) . Collectively , these results show that in yeast autophagosomes together with their cargo were delivered into the vacuoles of vps4∆ mutants in response to starvation , which is consistent with earlier reports ( Reggiori et al . , 2004 ) . Next , we analyzed autophagic processes further downstream and examined the lysis of autophagic bodies . This is a prerequisite for the subsequent proteolytic breakdown of autophagic cargo ( Takeshige et al . , 1992; Yang et al . , 2006 ) and depends on vacuolar acidification , the catabolic activity of Pep4 and Prb1 and the lipase Atg15 ( Teter et al . , 2000; Epple et al . , 2001 ) . To determine the breakdown of autophagic bodies in living cells , we generated a functional pHluorin-Atg8 chimera . The fluorescence of pHluorin-Atg8 is detectable at cytosolic pH but not at the lower pH within the vacuole ( Prosser et al . , 2010 ) . In WT cells the fluorescence of pHluorin-Atg8 was efficiently quenched in the lumen of vacuoles upon starvation ( >90% of cells , n = 157 ) . In contrast , in mutants that are either deficient in vacuolar peptidases ( prb1∆ , prc1∆ , pep4∆ ) or vacuolar acidification ( vma4∆ ) pHluorin-Atg8 was not quenched ( <10% of cells , n = 198 and n = 40 , respectively ) and pHluorin-Atg8 positive vesicular structures were detected inside their vacuoles , suggesting that autophagic bodies were not efficiently lysed ( Figure 3E , F ) . In the vast majority of vps4∆ mutants ( >90% of cells , n = 121 ) , the fluorescence of pHluorin-Atg8 was quenched in the vacuoles similar to WT cells , but occasionally few perivacuolar pHluorin-Atg8 positive structures were observed ( <17% of cells , n = 121 ) . Overall , it seemed that autophagic bodies were efficiently lysed in acidified vacuoles of vps4∆ mutants ( Figure 3E , F ) . These findings emphasize that in yeast the autophagic machinery , the fusion of autophagosomes with the vacuole per se and the lysis of autophagic bodies is not impaired in ESCRT mutants . Next we determined in detail how the MVB pathway would contribute to the starvation program of yeast . Therefore we measured how the proteome of WT ( vps4∆ complemented with VPS4 ) cells changed within the first 3 hr of starvation using stable isotope labeling with amino acids in cell culture ( SILAC ) ( de Godoy et al . , 2008 ) . WT cells were grown under rich conditions with heavy 13C615N2-lysine or light 12C614N2-lysine , and light cells were subsequently starved for 3 hr . Equal cell numbers were mixed prior to lysis and mass spectrometry ( MS ) analysis ( Figure 4—figure supplement 1A , upper panel ) . In total 2941 proteins were quantified ( peptide count ≥ 2 ) , comprising 58% of the characterized yeast ORFs ( Figure 4A , Figure 4—figure supplement 1A , upper panel , Supplementary file 1 ) . In this early phase of starvation , the yeast proteome already underwent extensive remodeling and 264 proteins significantly changed in abundance ( MaxQuant significance B ) ( Cox and Mann , 2008 ) . 101 proteins were significantly down- and 163 proteins significantly up-regulated . 10 . 7554/eLife . 07736 . 008Figure 4 . The MVB is required for starvation induced proteome remodeling . ( A ) Schematic presentation of proteome remodeling in WT cells during starvation . Starvation induced changes in protein levels were measured using SILAC based quantitative proteomics ( see also Figure 4—figure supplement 1A , B , Supplementary files 1 , 2 ) . The major changes in WT cells under starvation as indicated by Gene Ontology ( GO ) analysis of significantly changed proteins are shown . Green: up-regulated; red: down-regulated under starvation . ( B ) Correlation of changes in protein abundance in WT cells and vps4∆ mutants during starvation ( see also Figure 4—figure supplement 1A , Supplementary file 3 ) . WT and vps4∆ mutant protein ratios ( log2 [starved/rich] ) are plotted against each other . Green: significantly regulated in both datasets; blue: significantly regulated only in vps4∆; purple: significantly regulated only in WT . Grey: not significantly regulated . ( C ) Density plot showing log2-transformed protein ratio distributions in the three quantitative proteome datasets . The significant protein changes are excluded . Blue: WT ( starved ) /WT ( rich ) ; green: vps4∆ ( starved ) /vps4∆ ( rich ) ; red: vps4∆ ( rich ) /WT ( rich ) ( see also Figure 4—figure supplement 1 , Supplementary file 4 ) . p-value according to Kolmogorov–Smirnov <6 × 10−16 ( D ) The 135 significantly differentially regulated proteins during starvation between WT cells ( black bars ) and vps4∆ mutants ( white bars ) ( see also Supplementary file 5 ) . ( E ) Enrichment of GO terms in 135 significantly differentially regulated proteins ( from D ) . Data are represented as fold-enrichment over whole genome frequency ( see also Supplementary file 6 ) . significantly represented GO terms . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 00810 . 7554/eLife . 07736 . 009Figure 4—figure supplement 1 . Quantitative proteomics and GO analysis . ( A , C ) Representation of SILAC based quantitative proteomic changes . Protein ratio is plotted against signal intensity . Proteins that were significantly ( significance B ) down-regulated ( red dots ) or up-regulated ( green dots ) during 3 hr of starvation are shown . Non-regulated proteins are shown as grey bars . ( A ) Upper panel: protein ratios of WT cells starved/rich ( labeled with heavy lysine ) ; lower panel: protein ratios of vps4∆ mutants starved/rich ( labeled with heavy lysine ) . ( C ) Protein ratios of vps4∆ ( rich ) /WT ( rich , labeled with heavy lysine ) . ( B ) GO terms ( cellular processes and components ) that were significantly up- or down-regulated in the proteomic analysis of WT cells under starvation . Data are represented as enrichment over whole genome frequency for each GO term . ( Bold: more than 1 . 5-fold enriched ) . ( D ) SDS-PAGE and western blot analysis of cells grown under rich conditions using the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 00910 . 7554/eLife . 07736 . 010Figure 4—figure supplement 2 . Starvation induced endocytosis . ( A ) Fluorescence microscopy of Ste2-GFP , Can1-GFP , Fur4-GFP , and Pma1-GFP in WT and vps4∆ cells growing under rich conditions or after starvation . Vacuole ( V ) , class E compartment ( E ) and plasma membrane ( PM ) are labeled . Scale bar = 5 µm . ( B ) Fluorescence microscopy of Gap1-GFP in WT cells growing under rich conditions or after starvation . Plasma membrane ( PM ) is labeled . Scale bar = 5 µm . ( C ) WT and vps4∆ cells expressing Gap1-GFP were grown in rich medium or starved as indicated . Cell lysates were analyzed by SDS-PAGE and western blot ( WB ) using the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 010 To identify cellular components and biological processes that were regulated early during starvation , we determined the enrichment of specific Gene Ontology ( GO ) terms ( Harris et al . , 2004 ) ( Figure 4A , Figure 4—figure supplement 1B , Supplementary file 2 ) . Amino acid- and carbohydrate-metabolic pathways were up-regulated to meet the demands of starving cells . Most vacuolar hydrolases were also up-regulated in response to nutrient limitation , indicating that starving cells enhanced the catabolic activity and thus the recycling capacity of their vacuoles . Components of the autophagic machinery , including Atg1 and Atg8 , were also up-regulated ( Figure 4A ) . In contrast , cell wall components , proteins required for ribosome biogenesis and cell cycle/division were down-regulated ( Figure 4A , Figure 4—figure supplement 1B ) . This reflects known cellular responses to starvation , such as reduced cell growth and protein synthesis , exit from mitosis and entry into quiescence . Strikingly , the GO analysis also revealed that plasma membrane proteins were among the most frequently down-regulated cellular components during starvation . Not only Mup1 , but many other plasma membrane proteins , including diverse high affinity nutrient permeases required for the transport of sugars ( Itr1 , Hxt2 , 3 ) , nucleobases ( Fur4 ) , amino acids ( Bap3 , Gnp1 , Tat1 , Can1 ) and ammonium ( Ato3 ) but also the G-protein coupled receptor Ste2 were down-regulated ( Figure 4A ) and degraded via the MVB pathway ( Figure 4—figure supplement 2A ) . This comprehensive starvation-induced remodeling of the plasma membrane was highly selective . The protein levels of the plasma membrane H+-ATPase , Pma1 , were not dramatically altered . The majority of Pma1 remained at the cell surface and only a small portion was delivered into the vacuole ( Figure 4—figure supplement 2A ) . Moreover , the general amino acid permease Gap1 and the ammonium permease Mep2 were strongly up-regulated and the majority of Gap1 was retained at the plasma membrane ( Figure 4A , Figure 4—figure supplement 2B , C ) . To define how the MVB pathway contributed to proteome remodeling , the same proteomic experiment was performed with an isogenic ESCRT mutant ( vps4∆ ) ( Figure 4—figure supplement 1A lower panel , Supplementary file 3 ) . To compare starvation-induced proteome remodeling in WT cells and vps4∆ mutants , we restricted our analysis to 2694 proteins that were reliably quantified in both strains under rich and starvation conditions ( Figure 4B , Supplementary file 3 ) . Correlation analysis ( R = 0 . 82 , p < 1e-16 ) revealed that WT cells and vps4∆ mutants up- or down-regulated similar proteins in response to starvation , suggesting that ESCRT mutants are not generally deficient in inducing a starvation response ( Figure 4B ) . However , data correlation analysis indicated that a majority of proteins in vps4∆ mutants showed less pronounced changes during starvation ( Figure 4B ) . To address this observation over the entire datasets , we calculated the frequency by which changes in protein abundance occurred in WT cells or vps4∆ mutants . This analysis showed that the distribution of protein ratios was broader in WT cells ( blue curve ) than in the vps4∆ mutants ( green curve ) ( Figure 4C , p < 6e-16 ) . Hence in WT cells more proteins were stronger up- or down-regulated under starvation compared to ESCRT mutants , where changes in protein levels were less pronounced . An additional direct quantitative analysis of the proteomes of WT cells ( labeled with heavy 13C615N2-lysine ) and vps4∆ mutants growing under rich conditions ( Figure 4—figure supplement 1C , D , Supplementary file 4 ) showed the narrowest ratio distribution ( red curve ) ( Figure 4C ) . These results indicate that the MVB pathway has a small and selective effect on the proteome of cells growing under rich conditions , but becomes critical to support proteome remodeling once extra-cellular amino acids become limiting . To pinpoint processes that were particularly dependent on the MVB pathway during starvation , we identified 135 proteins that showed the most significant differences in changes of protein abundance between WT cells and vps4∆ mutants ( Figure 4D , Supplementary file 5 ) . From this analysis it became additionally evident that most proteins were ( with few exceptions ) stronger up- or down-regulated in WT cells ( black bars ) than in vps4∆ mutants ( white bars ) . GO analysis of these 135 proteins identified three processes that were primarily differentially regulated between vps4∆ mutants and WT cells upon starvation ( Figure 4E , Supplementary file 6 ) . Based on this analysis we conclude that in the first 3 hr of starvation the MVB pathway is particularly required ( i ) for the degradation of plasma-membrane proteins , as expected , which probably helps to maintain intracellular amino acid levels; ( ii ) to increase the protein levels of vacuolar hydrolases and thereby enhance the catabolic processes in vacuoles and also ( iii ) to down-regulate proteins that control the cell division cycle . Our quantitative proteomic analysis indicated that most vacuolar proteases , in particular Prb1 , but also Ape1 , Cps1 and Pep4 , were up-regulated during the first 3 hr of starvation in WT cells ( Figures 4A , 5A ) . Also other types of vacuolar hydrolytic enzymes were induced during that time , like the alpha-mannosidase Ams1 or the ribonuclease Rny1 . To directly assess how the catabolic activity of vacuoles changed during starvation , we measured the enzymatic activity of vacuolar alkaline phosphatase , Pho8 . The transmembrane protein Pho8 is delivered to the vacuole via the AP-3 pathway , which functions independently of the ESCRT machinery ( Cowles et al . , 1997 ) . Pho8 activity requires proteolytic maturation of Pho8 by Pep4 on the C-terminus . An additional uncharacterized endoproteolytic activity further cleaves mPho8 to yield a soluble sPho8 inside the vacuole that can be specifically measured ( Figure 5B , C ) ( Song , 2006 ) . In a yeast mutant that was deficient for three major vacuolar peptidases ( pep4∆ , prb1∆ , prc1∆ ) , Pho8 was not matured ( Figure 5B , lane 1 ) and Pho8 activity was not detected ( Figure 5C ) . In vps4∆ mutants the Pep4-dependent maturation to mPho8 was not impaired under rich growth conditions or starvation ( Figure 5B , lane 3–5 ) . In all cells sPho8 levels were low under rich conditions , which corresponded to low Pho8 activity ( Figure 5B , C ) . Within the first 3 hr of starvation , sPho8 activity increased at least fivefold in WT cells ( Figure 5C ) , consistent with the de novo synthesis of vacuolar proteases resulting in increased endoproteolytic activity generating sPho8 ( Figure 5B lane 3 ) . In the following 15 hr of starvation Pho8 activity only doubled ( Figure 5C ) . Hence the major boost for the catabolic activity of vacuoles occurs during the first 3 hr of starvation . In two different mutants that block autophagy ( atg8∆ and atg5∆ mutants ) , sPho8 activity still increased 3–5 fold during the first 3 hr of starvation similar to WT cells , but failed to increase further upon extended starvation ( Figure 5C ) , suggesting that autophagy was not required to boost the catabolic activity of vacuoles early during starvation . 10 . 7554/eLife . 07736 . 011Figure 5 . Boosting the catabolic activity of vacuoles during starvation requires the MVB pathway . ( A ) Starvation-induced changes in the protein levels of various vacuolar hydrolases based on SILAC data in WT ( black ) and vps4∆ mutants ( white ) . ( B ) Indicated yeast strains were grown in rich medium ( 0 hr ) or starved for 3 hr . Cell lysates were subjected to SDS-PAGE and western blot analysis with the indicated antibodies . p , precursor form; m , mature form; s , soluble form . ( C , D ) Soluble Pho8 ( sPho8 ) activity in rich medium and upon starvation of the indicated strains ( mean ± SEM , n = 4 ) . ( E ) Fluorescence microscopy of Pep4-GFP ( green ) in WT cells and vps4∆ mutants growing under rich or starvation conditions . ( V ) acuoles ( FM4-64 , red ) and class ( E ) compartments . Scale bar = 5 µm . vps10∆ + VPS10 ( 2 µ ) cells were used as the isogenic WT control in ( D ) and ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 01110 . 7554/eLife . 07736 . 012Figure 5—figure supplement 1 . Boosting the catabolic activity of vacuoles requires membrane protein degradation via the MVB pathway . ( A ) Soluble vacuolar alkaline phosphatase ( sPho8 ) activity in WT and ESCRT mutants ( vps36∆; vps23∆ ) . Data are represented as fold increase in activity after 3 hr starvation over rich growth condition ( n = 3; mean ± SEM ) . ( B ) Fluorescence microscopy of CPY ( 1–50 ) -mRFP and Mup1-GFP in WT cells , vps4∆ mutants and vps4∆ mutants over-expressing Vps10 growing under rich or starvation conditions . ( V ) acuoles and class ( E ) compartments . ( C ) Indicated yeast strains were grown in rich medium ( 0 hr ) or starved for 3 hr . Cell lysates were subjected to SDS-PAGE and western blot analysis with the indicated antibodies . p , precursor form; m , mature form; s , soluble form; unspecific background band . ( D ) WT cells , vps4∆ mutants and vps4∆ mutants over-expressing Vps10 were grown to mid-log phase ( rich ) and starved as indicated . Free amino acids were extracted and analyzed by liquid chromatography . Data are represented as the sum of free amino acids ( mg ) per gram of dry yeast . Mean ± SD , n = 3 . ( E ) 35S-Met/Cys incorporation into proteins of WT ( vps10∆ + VPS10 , 2 µ ) , vps4∆ mutants and vps4∆ mutants over-expressing Vps10 under rich growth conditions and during starvation , was analyzed by SDS-PAGE and digital autoradiography . Coomassie staining shows equal protein loading . ( F ) Quantification of 35S-incorporation under rich conditions and after 1 , 2 and 4 hr of starvation by liquid scintillation counting . Incorporation under rich conditions was set to 100% . Mean ± SEM , n = 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 012 In vps4∆ mutants growing under rich conditions , sPho8 protein levels were slightly lower ( Figure 5B , lane 4 ) , consistent with lower Pho8 activity ( Figure 5C ) . Upon starvation sPho8 activity merely doubled in vps4∆ mutants during first hours of starvation and never reached levels comparable to WT cells or autophagy mutants ( Figure 5C ) . ESCRT-I ( vps23∆ ) or ESCRT-II ( vps36∆ ) mutants were also severely impaired in increasing the catabolic activity of their vacuoles during starvation ( Figure 5—figure supplement 1A ) . SILAC based quantification of vacuolar hydrolases in starving vps4∆ mutants revealed that the protein levels of most vacuolar hydrolases including Prb1 , Ape1 , Cps1 and Pep4 were only marginally induced ( Figure 5A ) . Therefore mPho8 was not efficiently cleaved to sPho8 in vps4∆ mutants ( Figure 5B lane 5 ) and the catabolic activity of vacuoles in ESCRT mutants failed to increase early during starvation ( Figure 5C ) . The inability of ESCRT mutants to increase the protein levels of vacuolar proteases appears to culminate in a failure to boost the catabolic activity of vacuoles early during starvation . This is best explained by the central role of the MVB pathway in maintaining intracellular amino acid levels for protein synthesis early during starvation . ESCRT mutants also have a minor sorting defect for soluble vacuolar hydrolases , which are aberrantly secreted , mainly because Vps10 , the sorting receptor for multiple vacuolar hydrolases ( e . g . : Pep4 and Prc1 ) recycles less efficiently between endosomes and the golgi ( Bankaitis et al . , 1986; Rothman and Stevens , 1986 ) . Hence , severe mis-sorting and strong secretion of vacuolar hydrolases might alternatively explain the failure of ESCRT mutants to recycle amino acids , maintain protein synthesis and boost the catabolic activity of vacuoles . Therefore we next analyzed the extent to which the mis-sorting of vacuolar hydrolases would contribute to the here describe phenotypes of ESCRT mutant . First , we determined the subcellular localization of the master protease Pep4-GFP and Prc1/CPY-RFP in ESCRT mutants under rich growth or starvation using live cell fluorescence microscopy . Pep4-GFP localized to the class E compartment in ESCRT mutants , but a large fraction of Pep4-GFP was also delivered into the lumen of the vacuole ( Figure 5E ) . Similar results were obtained for a construct containing the vacuolar sorting signal of Prc1/CPY fused to RFP ( Figure 5—figure supplement 1B ) . Hence , a considerable fraction of vacuolar hydrolases still arrived in the lumen of the vacuole where they fully matured and were active ( Figures 3E , 5 ) . Notably , ESCRT mutants , unlike other endo-lysosomal trafficking complexes including HOPS , CORVET or retromer , were never identified as pep mutants ( Jones , 1977 ) because they displayed relatively minor mis-sorting of vacuolar hydrolases ( Bankaitis et al . , 1986; Rothman and Stevens , 1986 ) , which kept their vacuoles catabolically active . Earlier reports suggested that the overexpression of Vps10 can selectively rescue the partial mis-sorting of vacuolar hydrolases ( CPY/Prc1 , Pep4 ) in vps4∆ mutants but not the degradation of membrane proteins ( Babst et al . , 1998 ) . As expected , Mup1-GFP still localized to class E compartments and was not delivered into vacuoles in vps4∆ cells overexpressing Vps10 ( Figure 5—figure supplement 1B ) . Yet , Vps10 overexpression alleviated mis-sorting of vacuolar enzymes ( Figure 5—figure supplement 1C ) and thus increased the catabolic activity of vacuoles in vps4∆ mutants to 80% of WT levels under rich growth conditions ( Figure 5D ) . Despite this restoration of vacuolar catabolic activity prior to starvation , overexpression of Vps10 failed to rescue intracellular amino acid levels ( Figure 5—figure supplement 1D ) and protein synthesis ( Figure 5—figure supplement 1E , F ) in vps4∆ mutants throughout starvation . It seems that the cellular defects of ESCRT mutants during starvation can be mostly attributed to their inability to degrade membrane proteins . vps4∆ mutants but also vps4∆ mutants overexpressing Vps10 could not efficiently up-regulate the de novo synthesis of vacuolar hydrolases in time . Therefore the catabolic activity of vacuoles remained low during starvation in vps4∆ mutants overexpressing Vps10 ( Figure 5D ) . This is further emphasized by the impairment of ESCRT mutants to efficiently increase the protein levels of two other hydrolases , Ape1 and Ams1 , during starvation ( Figure 5A ) . Both Ape1 and Ams1 are delivered to vacuoles via the cvt-pathway and hence will not be secreted or mis-sorted in ESCRT mutants . All of these findings are consistent with the idea that selective membrane protein degradation via the MVB pathway , rather than the mere sorting of vacuolar hydrolases , is essential to maintain sufficient free intracellular amino acids to uphold protein synthesis for proteome remodeling early during starvation . It thereby contributes essentially to the de novo synthesis of vacuolar hydrolases to concomitantly boost of the catabolic activity of vacuoles . Next we tested if the MVB-dependent de novo synthesis of vacuolar hydrolases and the subsequent boost in hydrolytic activity was also required to break down and recycle autophagic cargo . Our proteomic studies comparing WT cells and vps4∆ mutants growing under rich conditions indicated that Atg8 protein levels were increased in vps4∆ mutants when compared to WT cells ( Figure 4—figure supplement 1C , Supplementary file 4 ) . This was also confirmed by western blot analysis ( Figure 6A , lane 5 , Figure 4—figure supplement 1D ) . Despite the efficient transport of GFP-Atg8 into vacuoles and lysis of autophagic bodies ( Figure 3C , E ) , the release of free GFP from Atg8 , which depends on efficient vacuolar proteolysis , was delayed in vps4∆ mutants , but not completely blocked ( Figure 6A ) . 10 . 7554/eLife . 07736 . 013Figure 6 . Boosting the catabolic activity of vacuoles is essential for the efficient degradation of autophagic cargo . ( A , B , C ) SDS-PAGE and western blot analysis of total cell lysates from WT cells and vps4∆ mutants grown in rich medium or during starvation using the indicated antibodies . p ( ro ) Ape1 , m ( ature ) Ape1 . residual anti-GFP signal after re-probing the membrane with anti-Pgk1 antibody . ( D ) Fluorescence microscopy of Rpl25-GFP ( green ) and mCherry-CPS ( red ) in WT cells and vps4∆ mutants under rich conditions or after starvation . Dashed lines indicate the vacuolar membrane . ( V ) acuoles and class ( E ) compartments . Scale bar = 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 01310 . 7554/eLife . 07736 . 014Figure 6—figure supplement 1 . Proteolytic processing of autophagic cargo . ( A ) SDS-PAGE and western blot analysis of total cell lysates from WT cells and the indicated mutants starved for 4 hr using the indicated antibodies . p , precursor form; m , mature form; Ψ-m , pseudo-mature form generated in prb1∆; *residual anti-GFP signal after re-probing the membrane with anti-Pgk1 antibody . ( B ) SDS-PAGE and western blot analysis of total cell lysates from vps4∆ mutants and vps4∆ mutants overexpressing Vps10 grown in rich medium or during starvation using the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 014 Similarly , already under rich growth conditions vps4∆ mutants had higher protein levels of mature mApe1 and immature pApe1 ( Figure 6B , lane 1 , 2 ) , which is delivered to vacuoles via the cvt-pathway ( Klionsky et al . , 1992 ) . Starving WT cells strongly induced the expression of Ape1 and increased its autophagy-dependent transport to the vacuole ( Baba et al . , 1997 ) , indicated by its efficient proteolytic processing ( Figure 6B , lanes 3 , 5 , Figure 5A ) . In starving vps4∆ mutants , mApe1 also increased , but not as strongly as in WT cells , which is consistent with our quantitative proteomic data ( Figure 5A , Supplementary file 3 ) . In vps4∆ mutants we always detected more pApe1 , indicating a delay in proteolytic maturation ( Figure 6B , lane 6 ) . In contrast when PEP4 or PRB1 were deleted , the release of free GFP from GFP-Atg8 and Ape1 maturation were fully blocked , further confirming that the catabolic activity of ESCRT mutants vacuoles was by no means completely defective ( Figure 6—figure supplement 1A ) . Additionally , the autophagy-dependent proteolytic processing of the ribosomal subunit Rpl25-GFP was delayed in vps4∆ mutants ( Figure 6C , lane 7–10 ) , although it was transported into the vacuoles of vps4∆ mutants ( Figure 6D ) . The degradation of autophagy cargo during starvation was also not restored in vps4∆ mutants by Vps10 overexpression ( Figure 6—figure supplement 1B ) . These results provide further evidence that amino acid recycling through the selective degradation of membrane proteins via the MVB pathway is required to boost the vacuolar catabolic activity early during starvation , mainly by maintaining protein synthesis and therefore promoting the up-regulation of vacuolar hydrolases . This order of events primes vacuoles for the efficient degradation of autophagic cargo . Starving cells have to exit proliferation and enter quiescence to survive this stress condition . In yeast , starvation results in a stable G1/G0 arrest , which can be scored by the increase of unbudded cells . Starving WT cells grew slowly , but still doubled their optical density ( Figure 7A ) and the majority ( >80% ) no longer displayed budding daughter cells ( Figure 7B ) . Hence WT cells managed to complete a final cell division cycle and efficiently arrested in G1/G0 during starvation . Autophagy mutants ( atg8∆ or atg5∆ ) grew slower when compared to WT cells upon starvation and failed to complete cell division and thus could not enter G1/G0 arrest ( Figure 7A , B ) , consistent with recent reports showing that autophagy is essential to overcome a Swe1-dependent checkpoint mechanism ( Matsui et al . , 2013; An et al . , 2014 ) . Our proteomic analysis indicated that cell cycle regulatory proteins ( including Swe1 ) were less efficiently down-regulated during starvation in vps4∆ mutants ( Figure 4E , Supplementary file 3 ) . Strikingly , when vps4∆ mutants were subjected to starvation , their growth slowed down prior to the growth of autophagy mutants ( Figure 7A ) . ESCRT-I ( vps23∆ ) , ESCRT-II ( vps36∆ ) , ESCRT-III ( snf7∆ ) and vps4∆ mutants failed to complete their final round of cell division and hence could not enter a G1/G0 arrest ( Figure 7B ) . When the partial sorting defect of vacuolar hydrolases of ESCRT mutants was rescued by Vps10 overexpression , vps4∆ mutants still stopped to grow and failed to enter quiescence during starvation ( Figure 7—figure supplement 1A , B ) . A mutant deficient for three vacuolar peptidases ( pep4∆ , prb1∆ , prc1∆ ) ( Figure 7B ) also failed to complete cell division and could not enter G1/G0 arrest ( Figure 7B ) . In this mutant the sequestration of cargo into MVBs or autophagosomes and their transport into vacuoles is not affected ( Figure 3E ) . Hence these results implicate that proteolytic degradation of MVB cargo and autophagic cargo inside the vacuoles in general , rather than removal/sequestration of a specific factor , is essential to complete a final cell division cycle and enter quiescence during starvation . 10 . 7554/eLife . 07736 . 015Figure 7 . The coordinated function of the MVB pathway and autophagy is required to enter quiescence upon starvation . ( A ) Growth of WT cells and the indicated mutants after shift from logarithmic growth in rich medium ( 0 hr ) to starvation measured with OD600nm . Mean ± SEM , n = 4 . ( B ) Quantification of unbudded cells ( G1/G0 arrested ) under rich conditions ( 0 hr ) or after indicated time of starvation . ( C ) Cells were starved for the indicated times and equal amounts of cells in serial dilutions were placed on rich medium ( YPD ) . ( D ) Model representing the coordinated action of lysosomal protein degradation pathways for amino acid maintainance and recycling as well as protein synthesis during starvation . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 01510 . 7554/eLife . 07736 . 016Figure 7—figure supplement 1 . Cell growth and entry into quiescence upon starvation . ( A ) Growth of WT cells and the indicated mutants overexpressing Vps10 after shift from logarithmic growth in rich medium to starvation measured with OD600nm . Mean ± SEM , n = 4 . ( B ) Quantification of unbudded cells ( G1/G0 arrested ) under rich growth or after 4 hr or 24 hr of starvation . n > 210 . DOI: http://dx . doi . org/10 . 7554/eLife . 07736 . 016 Finally , survival of long-term starvation was assessed by placing equal amounts of cells from starving cultures onto rich medium plates . WT cells barely lost viability even when they were starved for 13 days . In contrast , both vps4∆ and atg8∆ mutants gradually lost viability over time , and the vps4∆ , atg8∆ double mutants showed severe synthetic survival defects ( Figure 7C ) . In summary these results show that starvation-induced endocytosis and the subsequent selective degradation of these membrane proteins via the MVB pathway maintains intracellular amino acids pools that are required for the synthesis of new proteins early during starvation . This includes the de novo synthesis of vacuolar hydrolases , which is essential to boost the catabolic activity of vacuoles ( Figure 7D ) . This order of events primes vacuoles for the efficient degradation of bulk cytoplasm by autophagy , enables the continuous recycling of nutrients to maintain cellular homeostasis during extended periods of starvation and thereby promotes a stable cell cycle arrest that ensures cell survival . Here we show that eukaryotic cells utilize a catabolic cascade of selective and non-selective degradation pathways to ensure cell survival upon starvation ( Figure 7D ) . Immediately upon nutrient limitation , the degradation of cytoplasmic proteins by the UPS acutely supplies amino acids to maintain protein synthesis ( Vabulas and Hartl , 2005 ) . Starvation also inactivates TORC1 signaling , which probably simultaneously stimulates autophagy as well as starvation-induced endocytosis . Once TORC1 signaling is turned off , Atg13 is quickly dephosphorylated , which allows the formation of the Atg1/Ulk1 kinase complex to induce autophagy ( Kamada et al . , 2000 , 2010 ) . The TORC1/Npr1 signaling axis also controls at least in part starvation-induced endocytosis , by regulating different arrestin-like adaptors ( ARTs ) for Rsp5 , the major ubiquitin ligase required for endocytosis in yeast . Thereby TORC1 signaling orchestrates the selective remodeling of the plasma membrane proteome under different growth conditions and additionally could increase protein turnover via the MVB pathway ( Schmidt et al . , 1998; Beck et al . , 1999; Léon et al . , 2008; MacGurn et al . , 2011; Jones et al . , 2012; Boeckstaens et al . , 2014; Crapeau et al . , 2014 ) . Our results demonstrate for the first time that starvation induces a massive but still selective remodeling of the plasma membrane proteome . At least 18 different integral plasma membrane proteins undergo starvation-induced endocytosis including different amino acid permeases , sugar transporters and the G-protein coupled receptor , Ste2 . Within a few hours these membrane proteins are transported via the MVB pathway into vacuoles for degradation . How TORC1 signaling could control Rsp5-dependent cargo specificity and the selective ubiquitinylation of many different cargoes at the same time is currently not clear . Our results further demonstrate that the ESCRT-dependent degradation of these membrane proteins in vacuoles is essential to maintain a critical pool of free amino acids for protein synthesis . This enables the de novo synthesis of vacuolar hydrolases that are required to boost the catabolic activity of vacuoles ( Figure 7D ) . Thereby , the MVB pathway ensures that the catabolic activity of the vacuole is up-regulated in time to allow subsequent , efficient degradation of autophagic cargo . Only when these conditions are met , autophagy considerably contributes to amino acid recycling during extended starvation to restore intracellular amino acid levels ( Komatsu et al . , 2005; Onodera and Ohsumi , 2005 ) . Since the MVB pathway is also in part required for the proper targeting of hydrolases into the vacuole , it might even fulfill a dual function . Yet , our results emphasize that a significant fraction of vacuolar hydrolases are sorted into the lumen of vacuoles in ESCRT mutants , where they mature and become active . Furthermore , if the observed phenotypes in ESCRT mutants would be caused solely by mis-sorting of vacuolar hydrolases , this would lead to a delay in autophagy-mediated amino acid recycling . In this case their defects during starvation should be similar or even weaker when compared to autophagy mutants . The partial sorting defects for vacuolar hydrolases are thus not consistent with the stronger defects of ESCRT mutants in maintaining amino acid levels , protein synthesis and cell growth early during nutrient limitation . Moreover it was possible to restore the catabolic activity of vacuoles in ESCRT mutants under rich growth conditions , while this was no longer possible during starvation . Hence it seems that mainly the selective degradation of membrane proteins as a source for amino acids , rather than just the sorting of vacuolar hydrolases , contributes to the key role of the MVB pathway during starvation . At the moment it is not clear why the MVB pathway appears to be more critical than autophagy to maintain intracellular amino acids levels early during starvation . We speculate that membrane proteins that have been selected for degradation via the MVB are for the most part not re-synthesized , while non-selective autophagy will inevitably also capture and degrade proteins that are still needed and thus have to be replaced . Hence , selective protein degradation may at least initially provide a bigger added value for the free intracellular amino acid pool as compared to the non-selective degradation of bulk cytoplasm by autophagy . The ESCRT machinery has recently been described to contribute to selective micro-autophagy on late endosomes in mammalian cells ( Sahu et al . , 2011 ) . Hence , ESCRT-dependent catabolic pathways may not be limited to membrane proteins , although our study did not experimentally address this possibility . Ultimately our results suggest that only the coordinated action of the MVB pathway and autophagy provides sufficient intracellular recycling capacity during starvation to allow efficient mitotic exit and entry into a stable G1/G0 quiescent state and thereby ensures cell survival ( Figure 7D ) . The underlying mechanisms are not fully understood but probably require amino acid recycling from the vacuole to the cytoplasm via vacuolar amino acid permeases . Alternatively , the evolutionary conserved EGO/LAMTOR/Ragulator complex in conjugation with a lysosomal amino acid sensor ( Dubouloz et al . , 2005; Sancak et al . , 2010; Zoncu et al . , 2011a; Rebsamen et al . , 2015; Wang et al . , 2015 ) could somehow measure free amino acids in the lumen of lysosomes and transiently re-activate TORC1 to complete the final cell division cycle prior to entry into quiescence ( Matsui et al . , 2013; An et al . , 2014 ) . In Drosophila and human cells , loss of the ESCRT machinery interferes with a late step in autophagy , namely the fusion of amphisomes with lysosomes ( Nara et al . , 2002; Filimonenko et al . , 2007; Rusten et al . , 2007; Lee and Gao , 2009 ) . Amphisomes are acidic pre-lysosomal hybrid organelles of MVBs/late endosomes and autophagosomes ( Stromhaug and Seglen , 1993 ) , Thus , our findings that a functional MVB pathway is key to boost the catabolic activity may not be restricted to lysosomes but may also include MVB-derived amphisomes and thereby ensure the efficient degradation of autophagic cargo in higher eukaryotes . In yeast autophagosomes appear to fuse directly with vacuoles and amphisomes have not been described . Consistently , we and others demonstrated that loss of ESCRT function does not significantly impair the autophagic machinery itself or the delivery of autophagic cargo into lysosomes in yeast or Caenorhabditis elegans ( Reggiori et al . , 2004; Djeddi et al . , 2012; Jones et al . , 2012 ) . Our results show that the MVB pathway takes a central role in cellular homeostasis to preserve and redistribute biomass that can be used to maintain cell growth under nutrient limitation . It is tempting to speculate that during transient nutrient fluctuations or changes in metabolism a more stepwise activation of this catabolic cascade with initial selective protein degradation via UPS and the MVB pathway might help to delay massive non-selective breakdown of cytoplasm via autophagy , at least for some time . This might provide a safety mechanism to protect cells from the immediate need for non-selective protein degradation . While the ESCRT machinery has acquired additional roles in diverse biological processes in higher eukaryotes , we propose that the central role of the MVB pathway in the catabolic cascade of eukaryotic cells during starvation is evolutionary conserved . All experiments were performed with SEY6210 yeast strains , except for Pho8∆60-expressing strains , strains used in Figure 1C and Figure 6—figure supplement 1A and prototrophic strains ( Figure 2—figure supplement 1A , B ) , which were derived from BY4741 or BY4742 . For growth under rich conditions , cells were incubated in YNB synthetic medium supplemented with amino acids/nucleobases ( Ade , Arg , Lys , Thr , Tyr plus Ura , Trp , Leu or His when required for auxotrophic strains ) and 2% glucose , at 26°C , except for Figure 3B ( YPD ) . For starvation experiments , cells were kept at mid-log phase for 24 hr before they were twice washed with and resuspended in YNB with 2% glucose but w/o amino acids and ( NH4 ) 2SO4 . For growth on agar plates , yeast cells were diluted to OD600nm = 0 . 05 and spotted in 10× dilutions on YPD or YNB plates . Protein synthesis was inhibited by treatment with cycloheximide ( Sigma Aldrich , Austria , 50 µg/ml ) and proteasomal activity was blocked by MG132 ( Sigma Aldrich , 50 µM ) . Genetic modifications were done by PCR and/or homologous recombination using standard techniques . Where applicable , tags were introduced at the C-terminus to preserve the endogenous promoter sequences . Plasmid-expressed genes including their endogenous promoters were amplified from yeast genomic DNA into centromeric vectors ( pRS series ) . All constructs were analyzed by DNA-sequencing and transformed into yeast cells using standard techniques . Yeast strains and plasmids used in this study are listed in Supplementary file 7 and primer in Supplementary file 8 . For quantitative proteomics yeast cells were grown in complex synthetic medium ( CSM -His , -Arg , -Lys , complemented with Arginine and Lysine , SunriseScience Products , San Diego , CA ) and kept at mid-log phase for 24 hr . Subsequently , cells were washed with their corresponding labeling medium and then used to inoculate 500 ml of labeling medium . Cells were kept in log phase for 10 generations with either , heavy 13C615N2-L-Lysine or unlabeled 12C614N2-L-Lysine . For starvation experiments , logarithmically growing cells from unlabeled medium were washed twice with starvation medium and incubated for 3 hr in starvation medium . Cells were harvested by centrifugation . Labeled and unlabeled cells were mixed in a 1:1 ratio according to their OD600nm . Cell were mechanically disrupted with glass beads at 4°C in PBS containing protease inhibitors ( Aprotinin 10 µg/ml; Pepstatin 1 µg/ml , Leupeptin 10 µg/ml , Pefablock SC 100 µg/ml ) . Cell lysates were cleared from intact cells by centrifugation ( 5 min 1500 rpm , 4°C ) . Cleared cell lysates were TCA-precipitated and washed twice with acetone . Proteins were dissolved in water , lyophilized ( in order to remove traces of organic solvents ) and solubilized in 100 mM NH4HCO3 ( pH 8 . 0 ) . Solubilization was attained by sonication for 3 × 40 s . Resolubilized proteins were reduced with dithiothreitol , alkylated with iodoacetamide , and in-solution digested with LysC ( 1:75 wt/wt ) in 100 mM NH4HCO3 ( pH 8 . 0 ) . The resulting peptides were fractionated by reverse phase chromatography using an EC 250/4 . 6 Nucleosil 120-3 µm C18 column ( Macherey–Nagel , Germany ) and resulting fractions were analyzed by capillary electrophoresis-mass spectrometry ( Sarg et al . , 2013 ) . Peptide separation was performed applying ultra low flow conditions ( 10 nl/min ) using a neutral capillary installed into a PA800plus capillary electrophoresis system ( Beckman Coulter , Germany ) , which was coupled via sheathless porous sprayer interface ( Faserl et al . , 2011 ) to an LTQ Orbitrap XL mass spectrometer ( Thermo Scientific , Austria ) . Alternatively , cell lysates were fractionated in cytosolic and membrane associated proteins by ultracentrifugation ( 100 . 000×g ) and proteins precipitated in 10% trichloroacetic acid ( TCA ) . Precipitates were washed twice with acetone and resolubilized in SDS/urea sample buffer . SDS-PAGE in Tris-HCl gradient gels ( 4–15% , ReadyGel BioRad , Austria ) was used to reduce sample complexity . The SDS gel was stained with SimplyBlue Safe Stain ( Invitrogen , Austria ) , cut into 21 slices per lane and proteins in each slice were in-gel digested by Trypsin ( 50 ng/µl; biological sample 1 ) or LysC ( 50 ng/µl; biological sample 2 ) . Resulting peptides were analyzed by liquid chromatography-mass spectrometry ( LC -MS ) using an UltiMate 3000 nano-HPLC system ( Dionex/Thermo Scientific ) coupled to an LTQ Orbitrap XL ( Thermo Scientific ) . MS settings were as described ( de Godoy et al . , 2008 ) . MS data were analyzed using MaxQuant ( Version 1 . 2 . 2 . 5 ) ( Cox and Mann , 2008 ) . The yeast ORF sequences from the Saccharomyces Genome Database ( Dwight et al . , 2004 ) were used for protein identification ( last modified January 2010 ) . The parameters for the enzymes , labels , maximum charge and variable modifications were chosen according to the experimental setup . All other settings were default . Quality control of the experiments was performed by comparing their label incorporation , peptide length distribution , calibrated and uncalibrated mass error distribution of retention time , fraction of matched MS/MS scans , and correlation of protein ratios between different replicates . To compare WT and vps4∆ mutants under rich conditions , we only used proteins with at least three heavy and light peptide counts . Each protein had at least one unique peptide and MS scans in at least two biological replicates . No additional filtering criteria were applied for the WT and vps4∆ data under starvation conditions . Differential regulation was estimated using the significance B ( Perseus v1 . 0 . 2 . 13 ) . The GO enrichment analysis was performed using the differentially regulated proteins . They were mapped against the GOSlim Generic biological processes and cellular components ( Harris et al . , 2004 ) . GO term fusion was performed based on the GO tree ( http://www . geneontology . org/ ) . The enriched GO term at the highest level in the GO hierarchy was selected and its child terms were excluded . In cases when the parent process has a higher p-value , the child term was chosen . A hypergeometric test was used to estimate if the mapped GO term is significantly enriched with the selected proteins . The null hypothesis is that the selected proteins are randomly sampled from all yeast proteins . The resulting p-values were corrected with the Benjamini–Hochberg method . All adjusted p-values below 0 . 05 were reported . We also calculate the ratio of the observed ( dataset frequency ) vs the expected number of proteins ( genome frequency ) associated with the GO term , referred to as enrichment over genome ( McClellan et al . , 2007 ) . The density plots show the computed density estimates of the protein ratios quantified in all the three experiments , using Gaussian kernel ( R software environment ) . The protein ratio distributions were compared pairwise using the Kolmogorov–Smirnov test , under the null hypothesis that the two tested groups are samples from the same distribution , that is , have the same median , variability and distribution shape . The significant protein changes were excluded . Additionally , paired Student's t test was used to test for mean difference in the log2-transformed protein ratios . All p-values were adjusted using the Benjamini–Hochberg test . To individually compare the ratios of each protein in WT and ESCRT mutant cells during starvation , we analyzed the ratio of ratios . For each protein quantified under the two starvation conditions , we took the ratio ( fold change ) of WT and vps4∆ ratios and then transformed it on a log scale with base 2 . To extract the most differentially regulated proteins , we calculated the z-scores from the normal distribution of the ratio of ratios and selected the critical values ( i . e . those z-scores that are less likely to occur ) . The significance level was 0 . 05 . The two-sided test resulted with 135 significant differences between WT and vps4∆ protein ratios . To explore and visualize the data we used the R language for statistical computing and graphics . To calculate the linear correlation between protein ratios we used Pearson correlation . All p-values below or equal to 0 . 05 were reported . A Zeiss Axio Imager M1 equipped with standard fluorescent filters and a SPOT Xplorer CCD camera was used . VisiView software was used for image-acquisition . Brightness and contrast were linearly adjusted . For vacuole staining ( Vida and Emr , 1995 ) growing or starving cells were labeled for 10 min with 10 µg/ml FM4-64 ( stock solution 1 mg/ml in DMSO ) , washed twice with and subsequently chased for 1 hr in the respective medium before microscopy was performed . To prepare whole cell lysates , yeast cells were pelleted , resuspended in ice-cold water with 10% trichloroacetic acid ( TCA ) , incubated on ice for at least 30 min and washed twice with acetone . The precipitate was resolubilized in boiling buffer ( 50 mM Tris-HCl [pH 7 , 5]; 1 mM EDTA , 1% SDS ) , solubilized with glass beads and boiled at 95°C . Urea sample buffer ( 150 mM Tris-HCl [pH 6 , 8] , 6 M Urea , 6% SDS , bromphenol blue , 10% β-mercaptoethanol ) was added and the cleared cell lysate was separated by SDS-PAGE . Alternatively , proteins were extracted by alkaline extraction ( Kushnirov , 2000 ) . For analysis of protein phosphorylation ( Figure 3A , B ) , TCA extraction was performed as described ( Papinski et al . , 2014 ) . Whole cell protein extracts were prepared by TCA extraction or alkaline lysis , separated by SDS-PAGE and transferred to PVDF membranes . Antibodies used in this study include: α-Flag ( Sigma , Austria ) , α-GFP ( IgG1K , Roche , Austria ) , α-Pgk ( 22C5D8 , Life technologies , Austria ) , α-ALP ( 1D3A10 , Life technologies ) , α-HA ( 12CA5 , Abcam , UK ) , α-Pep12 ( 2C3G4 , Abcam ) , α-CPY/Prc1 ( clone 10A5 , Invitrogen ) , α-Ape1 and α-Atg8 ( Papinski et al . , 2014 ) . The α-Atg13 antibody was kindly provided by Daniel Klionsky , University of Michigan . α-Vps4 ( Babst et al . , 1998 ) , α-Pep4 ( Klionsky et al . , 1988 ) and α-Vps21 ( Horazdovsky et al . , 1994 ) antibodies were kindly provided by Scott Emr , Cornell University . The α-Sch9 and α-Sch9pT737 antibodies were kindly provided by Robbie Loewith , University of Geneva . The Pho8∆60 assay was performed as described ( Noda et al . , 1995; Klionsky , 2007 ) . Soluble endogenous vacuolar Pho8 activity ( sPho8 ) was measured using a fluorigenic method described for Pho8∆60 ( Noda and Klionsky , 2008 ) . Mid-log cells grown in rich YNB medium ( not containing methionine and cysteine ) or starved cells were labeled with 15–30 µCi [35S]-Met/Cys labeling mix ( Hartmann analytics IS-103 ) for 5 min at 30°C and stopped with excess L-Met ( 5 mM ) and 75 µg/ml cycloheximide . Cell extracts were analyzed by autoradiography of SDS-PAGE ( Amersham Biosciences , Austria , STORM 840 ) or quantified by liquid scintillation counting ( Beckmann Coulter LS6500 ) . Isogenic strains were grown to mid-log phase in YNB medium , washed twice with and inoculated in starvation medium at 0 . 6 OD600nm/ml . Equal cell numbers ( at least 30 OD600nm ) were harvested for each yeast strain and time point . Cultures from rich medium were harvested by vacuum filtration , washed twice with ice cold YNB medium ( with 2% glucose and ammonium sulfate but without amino acids ) and twice with ice cold 60% methanol . Starving cultures were washed twice with ice cold 60% methanol . All cell pellets were air-dried over night and weighed . Amino acids were ethanol-extracted and analyzed as described ( Altmann , 1992 ) . Norvalin ( 8 nmol/mg dry weight ) was the internal standard . For the representation of each individual amino acid , the measured values were normalized by the maximum measured amino acid content across all conditions and replicates . Cells were briefly sonicated and visually scored for emerging buds by bright-field microscopy .	Yeast and other organisms have evolved to survive extended periods of starvation by digesting their own proteins and other cell materials and thereby recycle them into new proteins and structures . One way in which these cell materials can be destroyed is by a process called autophagy . A membrane forms around the cell material to isolate it from the rest of the cell . In yeast , the resulting structure fuses with a cell compartment called the vacuole , which contains enzymes that break down the cargo into smaller molecules that can be re-used by the cell . When cells experience starvation , autophagy is not very selective in what it destroys and so it is tightly controlled to avoid damaging important structures in healthy cells . Alongside autophagy , specific proteins in the membrane surrounding a yeast cell can be targeted for destruction by another process called the MVB pathway . Certain membrane proteins are tagged with a small protein called ubiquitin , which leads them to being selectively incorporated into cell compartments called MVBs that then go on to fuse with the vacuole . However , it is not clear how the MVB pathway and autophagy may cooperate to enable the cell to survive periods of starvation . Here , Müller et al . monitored the changes in the proteins present in yeast cells during a period of starvation . The experiments show that many different membrane proteins in the yeast cells were destroyed via the MVB pathway within three hours of the removal of their food source . This was essential to allow the cells to carry on producing new proteins at this early stage in starvation . These new proteins included the enzymes found in vacuoles , which increased the ability of the cells to break down the proteins and other cell materials that were transported there via autophagy . These findings show how the MVB pathway and autophagy are co-ordinated to allow cells to survive periods of starvation . The next challenge is to work out how the MVB pathway is regulated at the molecular level in response to fluctuations in nutrient availability .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2015	The coordinated action of the MVB pathway and autophagy ensures cell survival during starvation
Two ER membrane-resident transmembrane kinases , IRE1 and PERK , function as stress sensors in the unfolded protein response . IRE1 also has an endoribonuclease activity , which initiates a non-conventional mRNA splicing reaction , while PERK phosphorylates eIF2α . We engineered a potent small molecule , IPA , that binds to IRE1's ATP-binding pocket and predisposes the kinase domain to oligomerization , activating its RNase . IPA also inhibits PERK but , paradoxically , activates it at low concentrations , resulting in a bell-shaped activation profile . We reconstituted IPA-activation of PERK-mediated eIF2α phosphorylation from purified components . We estimate that under conditions of maximal activation less than 15% of PERK molecules in the reaction are occupied by IPA . We propose that IPA binding biases the PERK kinase towards its active conformation , which trans-activates apo-PERK molecules . The mechanism by which partial occupancy with an inhibitor can activate kinases may be wide-spread and carries major implications for design and therapeutic application of kinase inhibitors . Roughly 30% of all proteins encoded in eukaryotes pass through the endoplasmic reticulum ( ER ) , where they are folded , modified , and assembled , before they are delivered to the plasma membrane , the outside of the cell , or to various way-stations in the secretory and endocytic pathways . To ascertain fidelity during protein maturation , ER-resident unfolded protein sensors continuously monitor the folding status in the ER lumen . The unfolded protein response ( UPR ) is induced when the protein folding capacity of the ER is surpassed , triggering the activation of three transmembrane sensors/signal transducers , IRE1 ( inositol-requiring enzyme 1 ) , PERK ( protein kinase RNA ( PKR ) -like ER kinase ) , and ATF6 ( activating transcription factor-6 ) . Two of the sensors , IRE1 and PERK , are protein kinases that are amenable to modulation by small molecule ATP-mimetics . IRE1is the most conserved of these proteins . It contains an ER-lumenal sensor domain that is activated by binding directly to unfolded polypeptides ( Credle et al . , 2005; Gardner and Walter , 2011 ) . As a result , IRE1 oligomerizes , activating its cytosolic kinase and endoribonuclease domains ( Cox et al . , 1993; Sidrauski et al . , 1996; Calfon et al . , 2002; Korennykh et al . , 2009; Li et al . , 2010 ) . IRE1's RNase domain initiates a non-conventional splicing reaction that results in the excision of an intron from the mRNA encoding the transcription factor XBP1 . XBP1 produced from the spliced mRNA drives transcription of UPR target genes to remedy ER stress . The luminal domain of PERK is homologous to that of IRE1 and thus its activation is presumably also driven by direct binding to unfolded polypetides ( Gardner and Walter , 2011 ) . Active PERK phosphorylates the α-subunit of eukaryotic translation initiation factor 2 ( eIF2α ) ( Harding et al . , 1999 ) , leading to trapping eIF2α in its GDP-bound inactive state , blocks eIF2α recycling . As a result , global protein synthesis is attenuated , while a few mRNAs , including that encoding the transcription factor ATF4 , are preferentially translated ( Harding et al . , 2000; Wek et al . , 2006 ) . Recent chemical genetic work in our laboratories revealed that phospho-transfer by Ire1's kinase domain can be bypassed using an ATP mimetic ( 1NM-PP1 ) ( Papa et al . , 2003; Rubio et al . , 2011 ) . Starting with studies in Saccharomyces cerevisiae , we showed that Ire1's RNase modality can be activated using a small molecule ATP mimetic ( 1NM-PP1 ) , when used in conjunction with a mutant form of Ire1 ( ‘IRE1-as’ for analog sensitized ) that allows 1NM-PP1 to bind to the ATP-binding site of IRE1's kinase domain . Subsequent work showed that the RNase activity of wild-type Ire1 can too be activated pharmacologically with the broad-acting kinase inhibitors APY29 and Sunitinib in vitro ( Korennykh et al . , 2009 ) . The crystal structure of the Ire1 kinase/RNase domains bound to APY29 , combined with biophysical and enzyme kinetic analyses , showed that binding of ATP-mimetic ligands to Ire1's active kinase site predisposes the enzyme to oligomerization , which activates its RNase activity . Ligand binding to the ATP binding pocket in IRE1's kinase domain , however , does not always result in oligomerization and RNase activation . Rather , activation requires that IRE1's kinase domain is in its active conformational state , characterized by the inward positioning of the αC helix and the DFG-loop in the kinase active site ( DFG/αC-in conformation ) ( Korennykh et al . , 2011; Korennykh and Walter , 2012; Wang et al . , 2012; Sanches et al . , 2014 ) . Thus , ATP-mimetic ligands that trap IRE1's kinase domain in the inactive , DFG/αC-out conformation act as inhibitors , rather than activators , of IRE1 oligomerization and signaling via its RNase domain . Because RNase activation can occur in the absence of a phospho-transfer reaction , IRE1 is unique in that it is possible to monitor the functional consequences of conformational changes in the kinase domain induced by ligand occupancy of the ATP-binding site without concerns of losing the kinase activity . The model depicting IRE1's kinase domain as a switch that becomes trapped in two states ( DFG/αC-in and DFG/αC-out ) depending on the ligand bound to its active site is an over-simplification . Different ligands yield different plateaus of maximal oligomerization and RNase activation , even when saturating the active site . This seemingly perplexing property is reconciled by the model in which different ligands predispose IRE1's kinase domain to populate the DFG/αC-in and DFG/αC-out states to different degrees; a strong IRE1 RNase activator would stabilize the DFG/αC-in state , whereas a weaker one would bias the IRE1 molecules in the population towards the DGF/αC-in state , without completely trapping them in this state . The reverse would be true for IRE1 RNase inhibitors , which would bias IRE1's kinase domain towards the DFG/αC-out state . To date , models of IRE1 activation have largely been derived from in vitro characterizations that lack in vivo confirmation , as the available tools were non-selective ( and hence overtly toxic ) to test in living cells ( Wang et al . , 2012 ) . Moreover , while 1NM-PP1 predisposes IRE1-as towards activation , it proved insufficient to activate IRE1 in cells in the absence of ER stress ( which vastly concentrates IRE1 by virtue of oligomerization of the lumenal domain ) or over-expression . Here , we describe the development of a novel small molecule , IPA , as the lead compound of a series of second-generation IRE1 activators . Surprisingly , IPA activates not only IRE1's RNase , but also PERK signaling but , by contrast to its ability to activate IRE1 , only at low concentrations . We propose that PERK activation results from ligand-induced conformational changes in a small percentage of the molecules in the population that then interact with and activate PERK molecules that contain an empty active site . Recent work identified an ATP mimetic that activates mammalian IRE1α's RNase activity in vitro ( Wang et al . , 2012; Sanches et al . , 2014 ) . These results , along with the co-crystal structure of S . cerevisiae Ire1 with the aminopyrazole-based inhibitor APY29 ( PDB ID: 3FBV ) ( Korennykh et al . , 2009 ) , provided a starting point to develop more selective and more potent IRE1 activators . We reasoned that ( 1 ) the cyclopropyl substituent on the pyrazole ring , which binds to the gatekeeper pocket in the S . cerevisiae Ire1 structure , would be a key determinant of human IRE1α binding , ( 2 ) interactions of the hinge-binding element of the APY29 scaffold would be essential to stabilizing IRE1α's kinase domain in a conformation leading to RNase activation , and ( 3 ) the pyrimidine ring , which occupies the adenine pocket in the S . cerevisiae structure , would provide an appropriate space filling moiety that further enhances affinity to the ATP binding pocket ( Figure 1A ) . We therefore kept these three elements constant in further optimizations and explored varying substituents attached to the pyrimidine ring for their ability to improve properties of the compounds . 10 . 7554/eLife . 05434 . 003Figure 1 . Design and characterization of IRE1α activators . ( A ) The core scaffold of APY29 ( aminopyrazole pyrimidine-base indicated in beige ) . ( B ) Structure-activity analysis of activating compounds . Compounds were assayed at 1 μM in a RNA cleavage assay containing IRE1α-KR43 ( 200 nM ) and 5′ [32P]-labeled RNA substrate HP21 ( see ‘Materials and methods’ ) . IPAx is methylated at the N[1] position in the pyrazole ring indicated as shown in R1 . ( C ) The effects of activating compounds ( 20 μM ) as a function of IRE1α-KR43 concentration . 1/2kmax , obs and Hill coefficients ( ‘n’ ) were calculated ( IPA: 1/2kmax , obs = 0 . 43 μM , n = 4 . 2; cmp1: 1/2kmax , obs = 3 μM , n = 3 . 8; IPAx 1/2kmax , obs = 8 μM , n = 3 . 3 , DMSO control: 1/2kmax , obs = 8 μM , n = 3 . 3 ) . ( D ) Compounds and FDA-approved drugs ( all at 1 µM ) were screened against a panel of 266 recombinant human kinases ( ‘S/T’: Ser/Thr protein kinases; ‘Y’: Tyr protein kinases , ‘PI’: phosphatidylinositide lipid kinases ) . APY24 contains the aminopyrozole pyrimidine-base , which is similar to APY29 ( Figure 1—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 00310 . 7554/eLife . 05434 . 004Figure 1—figure supplement 1 . Concentration dependence of IPA , cmp1 , and APY29 on IRE1α-KR43 RNase activity . A fluorescence RNA substrate ( HP17 ) containing a 5′ fluorescein and 3′ black hole quencher was used to determine RNase activity at a constant concentration of IRE1α ( [IRE1α-KR43] = 100 nM; IPA EC50 = 0 . 36 μM; APY29 EC50 = 5 . 2 μM; cmp1 EC50 = 0 . 95 μM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 00410 . 7554/eLife . 05434 . 005Figure 1—figure supplement 2 . Structure of APY based compound . The structure of APY24 contains a aminopyrozole pyrimidine core similar to APY29 . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 005 To this end , we modified the molecule by substituting urea-linked and variably m- and p-substituted phenyl groups for the benzimidazole of APY29 . Diphenyl urea linkers have previously been used successfully in the design of kinase inhibitors ( Dar et al . , 2008; Dar and Shokat , 2011; Korennykh et al . , 2012 ) . In exploring the addition of this chemotype on the APY29 scaffold , we generated a panel of 10 variants ( Figure 1B ) . Each member of the panel was tested in an RNase activity assay using recombinant IRE1α-KR43 and a labeled short hairpin RNA substrate derived from human XBP1 mRNA . IRE1α-KR43 comprises the soluble cytoplasmic portion of IRE1α , composed of the kinase and RNase domains plus 43 amino acids of the linker that bridges the kinase to its transmembrane domain . As expected from the chemical design strategy , we produced a series of IRE1α RNase activators . We observed that compounds bearing m-trifluoromethyl , p-methyl , p-ethyl , m-metoxy , and p-methoxy groups on the terminal phenyl ring ( cmp1 , cmp2 , cmp3 , cmp4 , and cmp5 ) activated IRE1α-KR43 activity 10–100-fold under assay conditions , as compared to fivefold by the parent compound APY29 ( Figure 1B ) . These results were surprising , because the presence of an m-trifluoromethyl group was previously shown to be important to retain activity of related kinase inhibitors but , as shown here , is not essential for activation of IRE1α-KR43 . We were also surprised that the activity was improved further by the presence of a p-substituted thioether group and , similarly albeit weaker , by the presence of a p-iodo group ( cmp6 and cmp7 , 900-fold RNase activation under assay conditions ) . We conclude that IRE1α prefers a substituent with polarizable character at the para-position of the terminal phenyl ring , perhaps indicating that the ring occupies a hydrophobic region in IRE1α's active site . Increasing bulkiness at the para-position resulted in decreased activities ( cmp8 , cmp9 , and cmp10 ) , suggesting that the size of this pocket must be limited . We decided to further characterize compound cmp6 , based on its robust activation properties . Henceforth , we refer to cmp6 as IPA ( for IRE1/PERK Activator ) for reasons to be discussed below . To validate that binding of IPA to IRE1α's kinase domain is critical for IRE1α RNase activation , we generated a control compound , IPAx . In IPAx , the pyrazole ring bears an additional N1-methyl group that is predicted to sterically interfere with binding to the gatekeeper pocket in IRE1α ( Figure 1A ) . Indeed , when assayed for activation of IRE1α RNase activity , IPAx was inactive showing an indistinguishable effect from the DMSO control ( Figure 1B ) . ATP mimetics , such as APY29 , induce IRE1α-KR43 oligomerization , which in turn activates the RNase activity ( Korennykh et al . , 2011; Wang et al . , 2012 ) . To assess whether IPA acts through a corresponding mechanism , we measured RNase activity as a function of enzyme concentration in the presence of saturating compound concentrations ( 20 μM; Figure 1C and Figure 1—figure supplement 1 ) . When no activator was added , apo-IRE1α–KR43 RNase activity , measured as kobs under single-turnover conditions , increased sharply with increasing enzyme concentration , reaching ½ kmax , obs at 8 μM ( Figure 1C , purple circles ) . The measured Hill coefficient of the activation was n = 3 . 3 , indicating that , consistent with previous work ( Li et al . , 2010 ) , enzymatic activation parallels oligomerization . By contrast , in the presence of cmp1 or IPA , IRE1α–KR43 RNase activity reached ½ kmax , obs at 3- and 20-fold lower enzyme concentrations , respectively ( ½ kmax , obs = 3 μM and 0 . 43 μM; and n = 3 . 8 and 4 . 2 , Figure 1C , red and blue circles ) , indicating that cmp1 and IPA binding significantly predispose IRE1α–KR43 to oligomerization and activation . By contrast , addition of IPAx did not enhance IRE1α activation , showing activation kinetics that were in all respects indistinguishable from the no-compound control ( Figure 1C , pink circles ) , consistent with its inability to bind to IRE1α-KR43 ( Figure 1C and Figure 1—figure supplement 1 ) . Kinome-wide screening of cmp1 and IPA demonstrated a dramatic improvement of selectivity for both molecules when compared to APY24 ( Figure 1D and Figure 1—figure supplement 1 ) , a close analog of APY29 , containing the same core-scaffold . In this assay , compounds were screened at a fixed concentration of 1 μM against a panel of 266 kinases , and the number of those inhibited by ≥ 80% was scored ( Figure 1D ) . We also included in the analysis several bench-mark inhibitors , such as the promiscuous natural product staurosporine ( STS ) and several clinically approved kinase inhibitors . We note that both cmp1 and IPA show better selectivity profiles when compared to the clinically approved compounds Sunitinib and Dasatinib . To examine the effects of IPA in living cells , we monitored the UPR in HEK293T cells using RT-PCR to measure splicing of XBP1 mRNA . We found that IPA-induced XBP1 mRNA splicing in a time- and dose-dependent manner ( Figure 2A ) . At a concentration of 2 μM IPA , splicing was induced within 2 hr ( lane 5 ) . As expected , IPAx did not induce XBP1 mRNA splicing ( Figure 2B , lane 3 ) . We used tunicamycin ( Tm ) , which induces ER stress by blocking N-linked glycosylation , as a positive control to induce the UPR ( Figure 2A , lanes 7 , and Figure 2B , lane 4 ) . 10 . 7554/eLife . 05434 . 006Figure 2 . IPA activates the IRE1 branch of the unfolded protein response ( UPR ) in HEK293T cells . ( A ) HEK293T cells were treated with increasing concentrations of IPA as a function of time . Tunicamycin ( Tm , 2 µg/ml ) was used as positive control to induce endoplasmic reticulum stress . The resulting XBP1 mRNA spliced products detected by RT-PCR ( ‘u’: unspliced and ‘s’: spliced ) are indicated . Control cells were treated with DMSO only . The asterisk identifies a hybrid amplicon resulting from spliced and unspliced XBP1 mRNA . ( B ) The effects of IPAx on the splicing of XBP1 mRNA in HEK293T cells were detected by RT-PCR after 4-hr incubation ( [IPA] and [IPAx] = 2 µM ) . ( C ) Inhibition of IPA-mediated XBP1 mRNA splicing in HEK293T cells by AD60 ( incubation time = 4 hr; [AD60] = 1 μM ) . ( D ) IRE1-GFP foci formation in T-REx293 cells . IRE1-GFP was visualized by confocal microscopy . ( E ) Effects of IPA on HEK293T cell viability ( LD50 = 0 . 83 μM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 00610 . 7554/eLife . 05434 . 007Figure 2—figure supplement 1 . AD60 inhibition of XBP1-luciferase-splicing reporter activation in HEK293T cells . A dose response of AD60 in the presences of Tm ( 2 μg/m ) is shown . Luciferase activity of cells treated with AD60 was normalized to cells treated with Tm only . AD60 IC50 = 0 . 75 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 007 To further rule out off-target effects through which IPA might induce ER stress indirectly , we use a recently discovered IRE1α inhibitor , AD60 , that binds to the ATP binding pocket of IRE1α's kinase domain , locking it into its inactive DFG/αC-out conformation and inhibiting its RNase activity ( Dar et al . , 2008; Korennykh et al . , 2012 ) . Using an XBP1 mRNA-splicing reporter that produces an XBP1-luciferase fusion protein only after IRE1-mediated splicing of its mRNA , we determined an IC50 of 0 . 75 μM for AD60 in HEK293T cells ( Figure 2C and Figure 2—figure supplement 1 ) . As expected , AD60 inhibited Tm-induced XBP1 mRNA splicing ( Figure 2C , lane 3 ) . Likewise , AD60 fully ablated IPA-induced XBP1 mRNA splicing ( Figure 2C , lanes 9–12 ) , even at the highest IPA concentration tested . Taken together , these results confirm that IPA binds to the ATP-pocket of IRE1α to induce XBP1 mRNA splicing in cells . We next used a T-REx293 cell line , containing a genome-integrated doxycycline-inducible IRE1α-GFP fusion gene , to monitor IRE1α oligomerization in vivo ( Li et al . , 2010 ) . In the presence of doxycycline , IRE1α-GFP was expressed and localized to the ER ( Figure 2D ) . As previously described , Tm rapidly induced the relocalization of IRE1α-GFP into discrete foci , indicative of IRE1α oligomerization and activation ( Li et al . , 2010 ) . Furthermore , the foci dissolved by 8 hr of treatment as cells attenuated IRE1α signaling even in the presence of unmitigated ER stress ( Figure 2D , upper panels ) ( Li et al . , 2010 ) . As expected , IPA caused relocalization of IRE1α-GFP into foci , consistent with sustained IRE1 activation ( Figure 2D , lower panels ) . While Tm-induced foci became larger and fewer over time as previously observed , IPA-induced foci remained small and numerous . Interestingly , IPA-induced foci remained stable even at late time points , consistent with the results obtained for XBP1 mRNA splicing ( Figure 2A ) . These results suggest that IPA-binding to IRE1α stabilizes IRE1α's oligomeric state , interfering with its attenuation . We next tested the effect of IPA on cell viability . As shown in Figure 2E , IPA killed cells with an LD50 of 0 . 82 μM . This strong toxicity was surprising considering that previous studies have shown that IRE1α has a cytoprotective role and that IRE1α knock-out cells are viable . The result therefore is best explained by a possible off-target effect connected to ER stress , likely to be mediated through another kinase in the cell . Moreover , as the interplay between the different UPR branches controls the switch over from cytoprotection to apoptosis ( Lin et al . , 2007 ) , we next probed for possible effect of IPA on PERK and ATF6 signaling . To assess the effects of IPA on the activity of the two other UPR branches , we first examined activation of ATF6 . To this end , we monitored mRNA levels by RT-PCR for two ATF6-driven transcriptional target genes , HERP1 and DERL3 , in HEK293T cells . As expected , treatment with Tm induced the expression of these two genes ( Figure 3A , B , lanes 6–9 ) . By contrast , IPA caused no detectable induction ( Figure 3A , B , lanes 2–5 ) , indicating that IPA does not induce general ER stress . 10 . 7554/eLife . 05434 . 008Figure 3 . Effects of IPA on the ATF6 and PERK branches of the UPR . ( A ) The levels of ATF6 transcriptional target mRNAs HERPUD1 ( HERP1 ) and DERLIN3 ( DERL3 ) were measured by RT-PCR in HEK293T cells treated with 1 μM IPA . GAPDH was used as a loading control . ( B ) Same as in ( A ) but cells were treated with 4 μM IPA . ( C ) Phosphorylation of PERK and eIF2α in HEK293T cells treated with IPA were detected by immunoblotting ( incubation time = 4 hr ) ; PERK phosphorylation is apparent from its shift in gel mobility; eIF2α phosphorylation was detected using a phospho-specific antibody . β-actin was used as a loading control . ( D ) HEK293T cells were treated for 4 hr with IPA ( 1 µM ) , IPAx ( 1 µM ) , or Tm ( 2 µg/ml ) . PERK shift and eIF2α phosphorylation were detected as in C . ( E ) Wild-type ( WT ) , Ire1−/− , and Perk−/− mouse embryonic fibroblast ( MEFs ) cells were treated with IPA ( 1 µM ) or Tm ( 5 µg/ml ) for 4 hr . Induction of XBP1 mRNA splicing by IPA was measured by RT-PCR . ( F ) Cells were treated as above . PERK gel mobility shift and eIF2α phosphorylation were detected by immunoblotting . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 008 To evaluate effects of IPA on the PERK branch , we measured the phosphorylation state of PERK and eIF2α by Western blotting . Upon activation , PERK becomes hyperphosphorylated and displays a characteristic mobility shift in SDS-PAGE gels , as observed in cells treated with Tm ( Figure 3C , lane 7 ) . Unexpectedly , we observed that IPA also induced a shift in the mobility of PERK ( Figure 3C , lanes 2–4 ) . This shift , however , was only observed at low-to-intermediate concentrations ( 0 . 1 , 0 . 5 , and 1 µM ) , whereas at higher concentrations ( 2 and 4 μM , Figure 3C , lanes 5 and 6 ) , PERK phosphorylation was severely diminished or not detectable . We also monitored PERK activation by phosphorylation of eIF2α , detected with a phospho-eIF2α-specific antibody ( Figure 3C ) . These results paralleled those obtained for PERK phosphorylation . By contrast , equivalent concentrations of IPAx did not stimulate PERK or eIF2α phosphorylation ( Figure 3D , lane 3 ) . PERK activation by IPA could be due to direct binding of IPA to PERK , or it could be caused by indirect activation of PERK by IRE1 . If the latter were true , IPA should not trigger PERK activation in the absence of IRE1 . To test this notion , we treated mouse embryonic fibroblasts ( MEFs ) derived from Ire1α−/− and Perk−/− knock-out mice with IPA and monitored XBP1 mRNA splicing , and PERK and eIF2α phosphorylation . As in HEK293T cells , treatment of wild-type MEFs with IPA induced both the PERK and IRE1 branches of the UPR . As expected , we observed no IPA-induced XBP1 mRNA splicing in Ire1α−/− MEFs ( Figure 3E , lane 4 ) and only a trace amount of IPA-induced eIF2α phosphorylation in Perk−/− MEFs ( Figure 3F , lane 6 ) . By contrast , we observed no IPA-induced XBP1 mRNA splicing , but pronounced PERK phosphorylation ( indicated by its mobility shift ) and eIF2α phosphorylation in the Ire1α−/− knock-out MEFs . These results indicate that IPA activates the IRE1 and PERK branches of the UPR independently . To test whether IPA binds PERK at its ATP binding site , we purified the cytosolic domain of PERK as a GST-PERK fusion protein . This preparation purified as a stable GST-PERK dimer , which in the presence of radiolabelled [γ-32P]ATP efficiently phosphorylated eIF2α in vitro . We next tested the effects of IPA and IPAx using this kinase assay . IPA inhibited the PERK kinase reaction with an IC50 = 2 . 8 μM ( Figure 4A , solid red circles and Figure 4—figure supplement 1 ) . These data suggest that IPA binds to PERK directly , where it would act as a competitive inhibitor for ATP and thus block its phosphorylation activity . 10 . 7554/eLife . 05434 . 009Figure 4 . In vitro effects of IPA on GST-PERK kinase activity . ( A ) The effects of IPA , cmp1 , and APY29 on GST-PERK kinase activity were monitored measuring phosphorylation of purified S . cerevisiae eIF2α with γ-[32P]-labeled ATP ( IPA IC5o = 2 . 8 μM; cmp1 IC5o = 4 . 5 μM , APY29 IC5o = 0 . 69 μM ) . ( B ) HEK293T cells were pre-treated with IPA for 30 min . Cells were then either co-incubated with IPA and Tm ( 2 µg/ml ) for an additional 4 hr ( lanes 4 , 5 , and 6 ) or washed ( IPA wash-out ) and treated with Tm alone ( lanes 7 , 8 , and 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 00910 . 7554/eLife . 05434 . 010Figure 4—figure supplement 1 . Audioradiographs of IPA , Cmp1 and APY29 tested against PERK-GST fusion protein . Dose responses were based on a 12-point inhibitor titration , using a twofold dilution series starting from 30 μM to determine the IC50 values of PERK inhibition . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 010 The observation of IPA-induced PERK inhibition ( shown in Figure 3C and Figure 4A ) poses the conundrum of why we observed IPA-induced activation in the low micromolar IPA range , producing an unusual , bell-shaped dose response ( Figure 3C ) . This behavior could be explained by a model posing that at lower IPA concentrations , IPA occupies the ATP-binding sites of a subset of PERK molecules , triggering kinase activation of neighboring , unoccupied PERK molecules , as previously shown for other kinases ( Hall-Jackson et al . , 1999; Hatzivassiliou et al . , 2010; Poulikakos et al . , 2010 ) . At higher IPA concentrations , IPA would saturate all PERK molecules , thereby inhibiting the pathway . We treated HEK293T with both Tm and IPA , expecting that Tm would not be able to activate PERK signaling in the presence of sufficiently high IPA concentration . Indeed , at 4 μM IPA , we observed that Tm addition did not activate PERK ( Figure 4B , lane 6 ) . Moreover , at 1 µM IPA , which induces PERK activation ( Figure 3C and Figure 3D ) , we found that Tm did not enhance the PERK mobility shift or eIF2α phosphorylation further than what was achieved by IPA alone ( Figure 4B , lane 5 ) . IPA inhibition of PERK was reversible , since washing IPA out allowed for activation by Tm ( Figure 4B , lane 9 ) . We next demonstrated that it is possible to keep the PERK pathway off in the presence of IPA . To this end , we added the selective PERK inhibitor GSK2606414 ( Axten et al . , 2012 ) ( henceforth abbreviated as ‘GSK’ ) to HEK293T cells treated with IPA , at concentrations that exert the greatest effect on PERK pathway activation . As expected , GSK abolished IPA-induced PERK and eIF2α phosphorylation ( Figure 5A; lanes 5 and 6 and Figure 5—figure supplement 4 ) . 10 . 7554/eLife . 05434 . 011Figure 5 . Inhibition of IPA-mediated PERK activation by GSK2606414 . ( A ) Reversal of IPA-mediated PERK activation ( [IPA] = 0 . 5 and 1 μM; lanes 5 and 6; note peak in Figure 3C at these concentrations ) in combination with GSK2606414 ( GSK , 1 μM ) was observed using immunoblotting as described in ( Figure 3C ) . As a control , GSK ( 1 μM ) was used to also inhibit PERK branch activation by thapsigargin ( Tg; 100 nM; lane 3 ) . ( B ) HEK293T cells were incubated with [35S] methionine to monitor protein translation upon addition of IPA ( 1 μM ) or a combination of IPA ( 1 μM ) and GSK ( 1 μM ) . The UPR was induced with Tg ( 100 nM ) or DMSO as indicated . ( C ) Cotreatment of HEK293T cells with IPA + GSK . HEK293T cell viability was measured as a dose-response of IPA in combination with 1 μM of GSK ( pink circles ) . The presence of 1 μM GSK shifted the IPA LD50 from 0 . 82 μM to 6 . 21 μM . The change in cell viability ± GSK inhibitor is overlaid on both dose responses ( green bars ) . Cells were treated with compounds for 24 hr and cell viability was normalized to DMSO controls . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 01110 . 7554/eLife . 05434 . 012Figure 5—figure supplement 1 . Quantification of [35S] incorporation . Quantification of [35S] incorporation was performed a total of 3-time , and mean errors were derived . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 01210 . 7554/eLife . 05434 . 013Figure 5—figure supplement 2 . The effects of GSK on HEK293T cell viability . A dose response of GSK was administered in HEK293T cells for 24 hr . No effect on viability was observed at 1 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 01310 . 7554/eLife . 05434 . 014Figure 5—figure supplement 3 . The effect of staurosporine ( STS ) on HEK293T cell viability in the presence of GSK PERK inhibitor . A dose response of STS in the presence of 1 μM GSK PERK inhibitor . No change in cell viability was observed in the presence of GSK when STS was present . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 01410 . 7554/eLife . 05434 . 015Figure 5—figure supplement 4 . Inhibition of IPA-mediated PERK activation using GSK2606414 . Reversal of IPA-mediated PERK activation ( [IPA] = 0 . 5 and 1 μM; lanes 7 and 8; note lanes 5 and 6 which show robust activation of the PERK pathway ) in combination with GSK2606414 ( GSK , 1 μM ) was observed using immunoblotting as described in ( Figure 3C ) . As a control , GSK ( 1 μM ) was used to also inhibit PERK branch activation by thapsigargin ( Tg; 100 nM; lane 3 ) . No effect on the pathway was observed with GSK alone ( lane 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 015 The phosphorylation of eIF2α resulting from PERK activation inhibits translation initiation . Indeed , when we incubated HEK293T cells with IPA in the low concentration range where PERK becomes activated , incorporation of [35S]-methionine into newly synthesized proteins was impaired ( Figure 5B , lanes 3 and 5 ) , consistent with translation attenuation . Interestingly , treatment with GSK-restored translation to baseline levels ( Figure 5B , lanes 4 and 6 and Figure 5—figure supplement 1 ) . This pharmacological manipulation therefore allowed us to experimentally uncouple IPA-induced IRE1α and PERK activation . In particular , we were interested in testing whether the unexpected toxicity of IPA ( Figure 2E ) could , at last in part , be explained by PERK activation , which is known to induce apoptosis ( Lin et al . , 2007 ) . To explore this notion , we tested cell viability at increasing IPA concentrations in the presence GSK . GSK inhibits PERK in cells with an EC50 of 200 nM ( Axten et al . , 2012; Moreno et al . , 2013 ) and shows no cell toxicity at 1 μM ( Figure 5C and Figure 5—figure supplement 2 ) . Indeed , GSK added at 1 μM protected cells from IPA-induced cell death , shifting the LD50 from 0 . 8 μM to 6 . 2 μM ( Figure 5C ) . As a further control , we tested whether GSK would also rescue cells treated with STS , a rather pleiotropic kinase inhibitor . GSK had no effect under these conditions ( Figure 5C and Figure 5—figure supplement 3 ) . To address directly whether PERK activation resulted from IPA binding to PERK itself , we recombinantly expressed the cytosolic portion of human PERK and eIF2α , the only known PERK substrate , and reconstituted the phosphorylation reaction in vitro . To this end , the cytosolic portion of PERK was dephosphorylated and then proteolytically severed and chromatographically separated from its glutathione S-transferase tag , known to promote-constitutive multimerization . The resulting PERK cytosolic fragment was responsive to activation , by contrast to the constitutively active , phosphorylated PERK-GST fusion protein assayed in Figure 4—figure supplement 1 . As shown in Figure 6A , we observed a large increase in the rate of PERK phosphorylation of eIF2α , which peaked at 0 . 5 μM IPA ( 3 . 2-fold increase over DMSO control ) and then declined at higher IPA concentrations , whereas IPAx had no effect ( Figure 6A , Figure 6—figure supplements 1 , 2 ) . The in vitro bell-shaped activity profile mimicked closely the effect observed in intact cells ( Figure 3C ) . Furthermore using chemical crosslinking , we observed that IPA—but not its inactive analog IPAx—significantly increased PERK oligomerization , apparent as an enhanced formation of cross-linked species migrating at the size expected for PERK dimers , trimers , and tetramers ( Figure 6B ) . The formation of higher-ordered PERK oligomers during ER stress was previously reported ( Bertolotti et al . , 2000; Marciniak et al . , 2006 ) . 10 . 7554/eLife . 05434 . 016Figure 6 . Reconstitution of activation of cytosolic PERK protein in vitro . ( A ) Recombinant PERK cytoplasmic domain was incubated at a set concentration of IPA . The fold-change in the rate of eIF2α was normalized to the DMSO control and plotted for all concentration . The greatest effects were observed at 500 nM ( 3 . 3-fold change ) and 3 μM ( 0 . 31-fold change ) in activity . IPAx showed no effect on the rate of PERK activity at a concentration of 20 μM . ( B ) Recombinant PERK cytoplasmic domain ( 2 μM ) was preincubated with varying concentrations of IPA ( or IPAx ) and subjected to chemical cross-linking . An IPA-dependent increase in the dimer , trimer , and tetramer complexes was observed , whereas IPAx ( 50 μM ) showed no effect when compared to the DMSO control . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 01610 . 7554/eLife . 05434 . 017Figure 6—figure supplement 1 . Biochemical reconstitution of PERK activation . SDS-polyacrylamide gel showing a dose response of IPA . eIF2α phosphorylation was used as a read out of activity . A dramatic activation and inhibition profile is observed . At maximal activity IPA elicits a 3 . 3-fold change in activation and 0 . 31-fold change at inhibitory concentrations . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 01710 . 7554/eLife . 05434 . 018Figure 6—figure supplement 2 . The effect of IPAx on PERK activation in vitro . To test effect of IPAx on PERK activation two concentrations ( 0 . 3 and 0 . 5 μM ) were tested in vitro . The SDS-polyacrylamide gel shows the effect of IPAx on PERK activation at 0 . 3 and 0 . 5 μM . No activation was observed when normalized to the DMSO control . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 018 The kinase domain of IRE1 can be targeted by small molecules to modulate its function . Based on the co-crystal structure of S . cerevisiae Ire1 with APY29 , we developed a series of new small molecule activators , including IPA . We showed that IPA is a strong activator of IRE1α signaling in vitro , by trapping IRE1α's kinase domain in its active ( DFG/αC-in ) conformation , which promotes IRE1 oligomerization . In the oligomer , IRE1's RNase becomes activated ( Korennykh et al . , 2009 ) , offering the unique opportunity to read-out the conformational status of a kinase domain in the absence of phospho-transfer . Conversely , a compound , AD60 , that traps IRE1's kinase domain in its inactive ( DFG/αC-out ) conformation acts as an inhibitor of IRE1 signaling . Our studies confirm in mammalian cells that IRE1's kinase domain acts as a conformational switch , in which ligand binding to the ATP binding pocket—rather than enzymatic phospho-transfer—controls activity and down-stream signal transduction events ( Korennykh et al . , 2011 ) . To rule out that IPA exerts its IRE1 activating activity by causing general ER stress , we explored its effects on the two other branches of the UPR that signal through ATF6 and PERK . Unexpectedly , we discovered that IPA also activated PERK . By contrast to IRE1 however , for which IPA drives activation in vitro and in vivo as IPA concentrations were increased , PERK activation displayed a bell-shaped dose response: PERK was activated at low IPA concentrations while being inhibited at higher ones . The paradoxical activation of kinase signaling by kinase inhibitors was first noted for Raf kinase inhibitors almost 15 years ago ( Hall-Jackson et al . , 1999 ) . The observation of kinase activation in patients undergoing Raf inhibitor ( vemurafenib ) clinical trials then led to an intense investigation of the cellular basis for the phenomenon ( Hatzivassiliou et al . , 2010; Poulikakos et al . , 2010 ) . Current models suggest that drug-induced dimerization of Raf causes an increase in Raf kinase activity . This contention is supported by apo- and drug-kinase complex crystal structures , which show Raf dimers in the asymmetric crystal unit ( Rajakulendran et al . , 2009 ) . Other models have been proposed that Raf inhibitor-induced Raf activation ( Holderfield et al . , 2013 ) , and , to date , the critical prediction of the model that the Raf kinase inhibitor vemurafenib should induce Raf kinase activity in a purified system has resisted all attempts at biochemical reconstitution . This paucity in direct experimental access has forced mechanistic studies of inhibitor-induced activation to be carried out in cells , where the effects of the many components of the Ras-Raf-Mek-Erk pathway are necessarily confounding . Our work with the reconstitution of PERK's paradoxical activation by IPA shows for the first time that no other components of the kinase pathway are necessary and that inhibitor binding is sufficient for activation through the formation of homo-dimers or higher-ordered homo-oligomers . It was previously suggested that PERK activation is depended upon dimerization and the formation of higher-ordered oligomers ( Bertolotti et al . , 2000 ) . This behavior can be explained in a model in which ligand binding to a few kinase molecules biases them towards the DFG/αC-in conformation that nucleates assembly with apo-kinases , enabling their trans-activation ( Figure 7 ) ( Korennykh and Walter , 2012 ) . Assuming a Kd of 2 . 8 μM for IPA-binding to PERK as determined by PERK inhibition ( Figure 4A ) , we estimate that approximately 15% of the PERK molecules are occupied by IPA under assay conditions that yield maximal activation . Since PERK•IPA triggered activation of unoccupied apo-PERK molecules would bias their conformation towards the active state and hence is likely to enhance their affinity for the inhibitor , the estimate of IPA-occupancy in the population defines an upper limit . This back-of-the-envelope calculation therefore suggests that PERK may form oligomers larger than dimers , in which one PERK•IPA may suffice to activate more than one apo-PERK . At low concentrations of IPA , heterodimers of PERK•IPA/PERKapo , or perhaps larger oligomers of PERK•IPA/[PERKapo]n , where all PERK molecules are in the active DFG/αC-in conformation , are active in phospho-transfer . At higher , saturating IPA doses , all PERK molecules would be occupied by IPA and thus inhibited through competition with ATP binding . 10 . 7554/eLife . 05434 . 019Figure 7 . Proposed model for PERK activation . The mechanism of PERK activation suggests that at low concentrations of IPA PERK protein is hyper activated presumably through movements of the ( DFG/αC-in ) . At higher doses of IPA all active sites are filled blocking PERK's activity . DOI: http://dx . doi . org/10 . 7554/eLife . 05434 . 019 Thus , the PERK/IPA enzyme/ligand combination mimics the RAF/vemurafenib combination and broadens the number of examples where an ATP-competitive binder can activate rather than inhibit a kinase . This mode of kinase regulation , now documented for two different protein kinases , may be more widespread than currently appreciated . Indeed , we observed a small amount of activation of eIF2α phosphorylation in PERK−/− cells ( Figure 3F , lane 6 ) , indicating that at least one of the other known eIF2α kinases ( GCN2 , HRI , and PKR ) may be similarly regulated . Kinase activation by partial occupancy of the active site , therefore , may be an important consideration in dosing kinase inhibitors for therapeutic applications , imposing a ‘minimal tolerable dose’ below which potential drugs may exert detrimental effects that oppose the desired therapy . This dangerous scenario may result from many kinase inhibitors that bind and stabilize kinases in the DFG/αC-in conformation . Based on the observation that ATF6 was not activated in cells treated with IPA , we conclude that IPA-mediated activation of IRE1 and PERK occurred in the absence of ER stress . This property contrasts with that of the previously described activator 1NM-PP1 , which binds in the ATP binding pocket to IRE1-as , the cognate , analog-sensitized allele . Activation of IRE1-as by 1NM-PP1 requires the additional induction of ER stress ( Wang et al . , 2012 ) , or overproduction of IRE1-as driving IRE1-as oligomerization by mass action ( Ghosh et al . , 2014 ) . IPA therefore presents a unique pharmacological tool with which activation of wild-type IRE1 can be studied in living cells directly in the absence of ER stress . The select application of combinations of UPR modulators has allowed us to begin dissecting the individual contributions of the signaling branches of the UPR . In particular , we have shown that the lethality that IPA displays at high doses can be partially overcome if the PERK pathway is inactivated using a selective PERK inhibitor , GSK , increasing the EC50 with which IPA drives cells into apoptosis by about an order of magnitude . This observation is consistent with the proposed roles of both the IRE1 and PERK branches , providing cytoprotective and pro-apoptotic outputs , respectively ( Lin et al . , 2007 ) . Moreover , AD60 , as a DFG-out kinase inhibitor , reversed the effects of IPA on IRE1 . IPA , AD60 , and other compounds developed to date therefore provide a stepping-stone towards developing novel methodologies for the selective pharmacological tuning of the UPR . Diseases , such as multiple myeloma , a cancer of highly secretory plasma cells where the UPR is thought to play a major cytoprotective role ( Carrasco et al . , 2007; Leung-Hagesteijn et al . , 2013 ) , or triple-negative breast cancer , in which high XBP1 activity has been correlated with poor patient prognosis ( Chen et al . , 2014 ) , have exposed the potential significance of targeting the UPR in cancers . Additionally , mutations found in cancer cells that weaken IRE1α RNase activity may be amenable targets for allosteric modulation . The chemical biology tools developed in this work provide an important step forward towards exploring the utility of UPR-based therapies , as well as offer fundamental mechanistic insights into a key mechanism that keeps the healthy balance of protein folding in the ER . HEK293T , T-REx293 and wild type , Ire1−/− and Perk−/− cells MEF cells were maintained at 37°C , 5% CO2 in Dulbecco's Modified Eagle Medium ( DMEM ) supplemented with 10% Fetal Bovine Serum ( FBS ) , 10 units/ml penicillin and 10 μg/ml streptomycin ( Life Technologies , San Francisco , CA ) . Tm was obtained from Sigma ( Milwaukee , WI ) . Transient and stable transfections were performed using the Lipofectamine 2000 ( Invitrogen , Carlsbad CA ) and FuGene6 reagent ( Roche ) . Stable cell lines expressing IRE1-3F6HGFP T-REx293 were described previously ( Li et al . , 2010 ) . T-REx293 cells were split 2 days before imaging onto glass-bottom micro-well dishes ( MatTek ) at 5 × 104 cells/dish . Doxycyclin-containing medium ( 10 nM doxycycline ) was added for 24 hr , withdrawn before imaging and replaced with imaging media ( Hank's Balanced Salt Solution [Gibco] , 2% FBS , and 5 mM HEPES pH 7 . 0 ) . Images were acquired on a spinning-disk confocal microscope as described ( Li et al . , 2010 ) . HEK293T cells ( 500 , 000 cells/ml ) were plated in 12-well flat bottom culture cluster dishes ( Costar , Fisher Scientific ) 24 hr before the experiment . IPA ( 1 μM ) , or a combination of IPA with GSK2606414 ( TRC inc . Toronto Canada ) , were added to the cells culture for no longer than 2 hr . 15 min prior to lysing cells , 4 μCi of Express [35S] protein-labeling mix ( Perkin–Elmer ) was added . Media were removed , and cells were immediately lysed in a buffer containing 25 mM Tris-HCl pH 8 , 8 mM MgCl2 , 1 mM dithiothreitol ( DTT ) , 1% Triton X-100 , 15% glycerol , 10 mM leupeptin , 153 μM aprotenin , and 1 mM phenylmethanesulfonyl fluoride . Cells lysates were kept on ice for 30 min before high-speed centrifugation followed by denaturation at 95°C . The samples were resolved on an SDS-polyacrylamide gel , which were then stained with Coomassie blue to ascertain protein loading . Dry gels were exposed to a blank phosphorscreen and scanned using a Typhoon variable mode imager ( GE Healthcare ) . The resulting autoradiograms were quantified using ImageQuant software ( Molecular Dynamics ) . HEK293T cells were grown in DMEM ( Sigma ) complete media containing 10% FBS , 10 units/ml penicillin ( Invitrogen ) , and 10 μg/ml streptomycin ( Invitrogen ) . Cells were plated at a density of 30 , 000 cells per well in 96-well black plates with clear flat bottoms ( Corning , Sigma-Aldrich ) 24 hr before the experiment . A 12-point semi-logarithmic dilution series of compounds was made starting at 30 μM not to exceed 0 . 1% DMSO after final dilution into the growth media . Viability assays were conducted over the course of 24 hr . Cell viability was determined using CellTiter-Blue ( Promega ) following the manufacturer's instructions . A Spectramax5 Microplate reader equipped with SoftMax Pro 5 software ( Molecular Devices ) was used to read out viability , and data were plotted using SigmaPlot software package ( Systat Software ) . Experiments were done in triplicate , and mean values were computed for each data point . Cells were lysed and total RNA was collected ( Total RNA Kit 1 , EZ-RNA , EZNA [USA] ) . PolyA+ mRNA was reverse transcribed with M-MLV ( Invitrogen ) , and the resulting cDNA was used as a template for PCR amplification across the fragment of XBP1 cDNA containing the intron . Primers used to amplify human XBP1: 5′-TTACGAGAGAAAACTCATGGC-3′ and 5′-GGGTCCAAGTTGTCCAGAATGC-3′ . To amplify murine XBP1: 5′-GAACCAGGAGTTAAGAACACG-3′ and 5′-AGGCAACAGTGTCGAGTCC-3′ . PCR conditions were as follows: 95°C for 5 min , 95°C for 1 min , 58°C for 30 s , 72°C for 30 s , and 72°C for 5 min , with 20 cycles of amplification . PCR products were resolved on a 2 . 5% agarose tris-acetate EDTA ( TAE ) gel . As previously reported , a hybrid amplicon species consisting of unspliced Xbp-1 annealed to spliced Xbp-1 can also be produced through this PCR reaction and was visible as a slower migrating band above the unspliced amplicon . The same cDNA was used as template for PCR of DERLIN-3 and HERPUD1 , both ATF6 targets , and GAPDH , which was used as loading control . Primers used included the following: DERLIN-3 , 5′-AGTTCCACTCTTTGATGGAGGGCA-3′ and 5′-AGCCAGCTGTGAGGAATATGGGAA-3′; HERPUD1-1 , 5′-ACAAGGTGGCCCTATTGTGGAAGA-3′ and 5′-AGTCCATTCCTGTCAAAGCCTCCA-3′; GAPDH , 5′-CCATGTTCGTCATGGGTGTGA-3′ and 5′-CATGGACTGTGGTCATGAGT-3′ . PCR conditions were: 95°C × 5 min , 95°C × 1 min , 56°C × 30 s , 72°C for 30 s for 25 cycles , 72°C for 5 min at 4°C . PCR products were resolved on a 2% agarose TAE gel . Human IRE1α-KR43 with a hexa-histidine ( 6xHis ) tag in its N-terminus was expressed and purified in SF21 cells as described ( Li et al . , 2010 ) . Cells were lysed in buffer containing 20 mM Tris-HCl , pH 7 . 5 , 600 mM NaCl , 2 mM MgCl2 , 3 mM imidazole , 10% glycerol , 1% Triton X-100 , 3 mM β-mercaptoethanol , COMPlete protease inhibitors ( Roche ) , and PhosSTOP phosphatase inhibitor cocktail ( Roche ) and passed through an AVESTIN emulsiflex-C3 3× times for complete lysis . The lysate was cleared using centrifugation at 100 , 000×g . Clear lysate was allowed to incubate on Ni-NTA agarose ( Qiagen ) beads for 2 hr before being washed 3× times with buffer containing 20 mM Tris-HCl , pH7 . 5 , 600 mM NaCl , 2 mM MgCl2 , 30 mM imidazole , 10% glycerol , and 3 mM β-mercaptoethanol . Protein was eluted from the column by raising the imidazole concentration to 250 mM . The eluate was then passed through a HiTrap desalting column ( GE Healthcare ) . Lastly , the 6xHis was cleaved as described ( Li et al . , 2010 ) and the resulting protein was then loaded onto a HisTrap HP , 5 × 5 ml column to remove the uncleaved protein . The cleaved protein was then loaded on a Mono-S 5/50 GL column ( GE Healthcare ) , and the eluate was then concentrated to 5 mg/ml and loaded on a Superdex 200 HR 10/300 ( GE Healthare ) column in buffer containing 20 mM Tris-HCl pH 7 . 5 , 250 mM NaCl , 5% Glycerol and 5 mM tris ( 2-carboxyethyl ) phosphine ( TCEP ) . The pure protein solution was aliquoted and stored at −80°C . To express the cytosolic kinase domain of PERK , the murine PERK kinase domain ( 580–1077aa ) was cloned into a pGEX4T1 vector to create a fusion protein containing N-terminal GST . The plasmid was transformed into Escherichia coli strain BL21DE3 RIPL ( Agilent Technologies ) . Cells were grown in LB medium containing ampicillin and chloramphenicol until OD600 = 0 . 6 . Expression was induced with 0 . 2 mM IPTG at 18°C for 16 hr . Cells were harvested by centrifugation , resuspended in buffer A ( 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 3 mM β-mercaptoethanol , 2 . 5% Glycerol ) and lysed by sonication . After centrifugation , the supernatant was applied to a GST-Sepharose column and washed with buffer A and buffer A containing 500 mM KCl and 1 mM ATP . Sample was eluted with buffer A containing 20 mM glutathione . GST-PERK kinase domain was then concentrated and further purified on a Superdex 200 10/300 gel filtration column equilibrated in buffer B ( 50 mM Tris pH 7 . 5 , 50 mM NaCl , 5 mM MgCl2 , 3 mM DTT , 1% Glycerol ) . Fractions containing GST-PERK kinase domain were concentrated and flash-frozen in liquid nitrogen and stored at −80°C . Full-length cytosolic human PERK was codon-optimized for E . coli expression by Genewiz Inc . A construct was then cloned into a Pgex-6p-2 vector for expression using two rounds of In-Fusion cloning ( Clontech ) ( 535–1093 Δ660–868 ) . The cytosolic portion of PERK , lacking the unstructured loop region ( amino acids 535–1093 Δ660–868 ) was then co-expressed with a tagless lambda phosphatase to produce a fully dephosphorylated PERK protein in BL21 star ( DE3 ) ( Life Technologies ) ( Dar et al . , 2008 ) . Cells were grown to an OD600 of 0 . 5 before induction with 0 . 1 mM IPTG at 15°C for 25 hr . Cells were harvested and lysed using AVESTIN Emulsiflex-C3 in a buffer containing 50 mM Tris-HCl , pH 8 . 0 , 500 mM NaCl , 10% glycerol , 5 mM TCEP ( buffer A ) , and EDTA-free COMPlete protease inhibitor cocktail ( Roche ) . Lysate was cleared by centrifugation at 100 , 000×g before batch binding to a GST-Sepharose resin . The resin was washed 5× times with buffer A , and on-column tag cleavage was preformed using PreScission protease ( GE Healthcare ) . The protein was loaded onto a HiTrap Q HP column to remove remaining protease . The PERK ( 535–1093 Δ660–868 ) protein was then concentrated and fractionated on a Superdex 200 GL ( GE Healthcare ) to remove uncleaved GST-PERK protein . Cells were lysed in 50 mM Tris-HCl , 2% SDS , 10% glycerol , 0 . 01% bromophenol blue , pH 6 . 8 supplemented with phosphatase inhibitors ( Sigma ) and protease inhibitors ( Roche ) . 30 μg to 50 μg of total protein was loaded on each lane of 8% or 10% SDS-PAGE gels . The separated proteins were then transferred onto nitrocellulose membranes for immunoblotting . The following antibodies and dilutions were used: anti-β-actin 1:10 , 000 ( A5441 , Sigma ) , anti-total PERK at 1:1000 ( C33E10 , Cell Signaling Technology ) , anti-phospho eIF2α at 1:1000 ( 3597S , Cell Signaling Technology ) , anti-total eIF2α at 1:1000 ( L57A5 , Cell Signaling Technology ) . After overnight incubation with primary antibody , membranes were washed in PBS with 0 . 05% Tween and incubated in HRP-coupled secondary antibody anti-rabbit ( 611–1302 , Rockland ) or anti-mouse ( 610–1302 , Rockland ) diluted at 1:5000 in wash buffer . Immunoreactive bands were detected by chemiluminescence . IRE1 endoribonuclease activity was detected employing two in vitro assays . kobs and Hill coefficients were determined from the cleavage kinetics of [32P]-labeled RNA substrates as previously described ( Korennykh et al . , 2009 ) . The assay was started by adding 1 μl of [32P]-labeled RNA to 9 μl of premixture containing 20 mM HEPES pH 7 . 4 , 70 mM NaCl , 2 mM MgCl2 , 4 mM DTT , 5% glycerol , 1 µl of 10 μM compound in DMSO . Reactions were performed at 30°C . Reactions were conducted under single turnover conditions . They contained ≤1 pM [32P]-labeled RNA and increasing concentrations of purified IRE1α-KR43 . Reactions were quenched at time intervals with 6 μl stop solution ( 10 M urea , 0 . 1% SDS , 0 . 1 mM EDTA , 0 . 05% xylene cyanol , and 0 . 05% bromophenol blue ) . Samples were analyzed in 10–20% Urea-PAGE gels . Gels were scanned using a Typhoon variable mode imager ( GE Healthcare ) and quantified using the ImageQuant and GelQuant software packages ( Molecular Devices ) . The data were plotted and fit to exponential curves using SigmaPlot software package ( Systat Software ) to determine observed rate constants as previously described ( Korennykh et al . , 2009 ) . EC50 values were determined using a FRET-based cleavage assay . A FRET probe ( FRET_IDT_17 56- FAM/rCrArCrCrUrCrUrGrCrArGrCrArGrGrUrG/IABlk_FQ ) was purchased from IDT ( RNase free HPLC purification ) and dissolved in RNase-free H2O . Excitation: 485 ( 480–490 ) and emission: 517 ( 512–520 ) . The reactions were started by addition of RNA FRET probe ( to a final concentration of 100 nM ) to 9 . 4 μl of premixture containing 20 mM HEPES , pH 7 . 4 , 70 mM NaCl , 2 mM MgCl2 , 4 mM DTT , 5% glycerol , and compound in DMSO not to exceed 1% . Reactions were performed at 30°C and contained 150 nM enzyme under single turnover conditions . Reactions were quenched at time intervals with equal volumes of formamide . Samples were analyzed using a SpectraMax M5 plate reader equipped with SoftMax Pro 5 and data acquisition software ( Molecular Devices ) . The data were plotted using SigmaPlot software package ( Systat Software ) . Experiments were repeated 3–4 times and mean values were computed . A plasmid encoding a modified ER stress reporter construct consisting of a C-terminally FLAG-tagged firefly luciferase coding sequence fused to an N-terminal hemagglutinin-tagged tandem repeat of a partial sequence of XBP1 of human origin lacking its DNA-binding domain and containing the IRE1-cognate intron ( pCAX-HA-2xXBP1ΔDBD ( anATG ) -LUC-F , kind gift of Takao Iwawaki [Hosoda et al . , 2010] ) , was used as a template to generate a retroviral expression construct . The coding sequence of the reporter was amplified by PCR using primers with engineered BamHI and EcoRI sites and was subsequently cloned into the cognate sites of the retroviral expression vector pBABE . puro ( Addgene ) to generate construct DAA-A171 . DAA-A171 was used to generate recombinant retroviral particles using standard methods and the resulting retroviral supernatant was used to transduce HEK293T cells , which were then subsequently selected with puromycin to create a stable reporter cell line . IPA and cmp1 were assayed by Invitrogen to derive percent inhibition of kinase activity . All compounds were screened at 1 μM and raw values are shown in supplementary tables . Detailed procedures for kinase reactions , ATP concentrations used and Z′-LYTE or Adapta assay formats are described in SelectScreen Customer Protocol ( http://www . invitrogen . com/kinaseprofiling ) . Kinase-inhibition data for 1 μM inhibitor of each of the clinical and tool compounds STS , Sunitinib , Dasatinib , Imatinib , SB202190 , Erlotinib and Gefitinib were provided by Invitrogen based on stock results and not an independent comparison conducted or commissioned by the authors . Raw data of percent inhibition of all 266 kinases can be located in Supplementary file 1 . Screening analysis for APY24 was previously reported ( Statsuk et al . , 2008 ) . A solution of PERK cytosolic domain ( 535–1093 Δ660–868 ) ( 2 μM ) was incubated with IPA or IPAx for 15 min before the addition of 250 μM of the amino group-reactive cross-linking agent bis[sulfosuccinimidyl] suberate ( BS2g-d0 , Thermo Scientific ) . After 30-min incubation at room temperature Tris-HCl pH 7 . 5 was added to a final concentration of 55 mM . Samples were then boiled in Laemmli sample buffer and loaded into an Any kD Mini-protean TGX gel ( Bio Rad ) . Cross-linked protein was visualized using Colloidal Coomassie G-250 stain . GST-PERK kinase domain ( 580–1077 aa ) was pre-incubated with compounds for 30 min . The kinase reaction was initiated by addition of ATP including 0 . 2 mCi γ-[32P]-ATP . The final concentrations of the reactants were 50 nM GST-PERK , 50 μM eIF2α , 100 μM ATP and varying concentrations of compound . At 30 min , 2 μl of each reaction was spotted onto a P81 phospho-cellulose membrane . The membrane was washed in 1% phosphoric acid and the sheets were washed five times in buffer , dried , and transferred radioactivity was measured using a Typhoon variable mode imager ( GE Healthcare ) and quantified using the ImageQuant ( Molecular Devices ) . Titration data were fit to a sigmoidal dose response to derive IC50 values using the SigmaPlot software package ( Systat software ) . Dose responses were based on a 12-point inhibitor titration , using a semi-logarithmic dilution series starting from 30 μM . Experiments were completed 2–4 times and mean values were computed . PERK activation was measured using PERK ( 535–1093 Δ660–868 ) in a buffer containing 20 mM Tris-HCl , 150 mM NaCl , 4 mM MgCl2 , 5 mM TCEP , 1% glycerol , pH 8 . 0 . PERK concentration was 1 μM , and activity was measured at 30°C . PERK was pre-incubated with 50 μM of cold ATP in the presence or absence of IPA or IPAx . The reaction was then initiated by adding S . cerevisiae eIF2α ( 1–180 ) and 0 . 2 mCi γ-[32P]-ATP . The reaction was monitored over a time course of 30 min , during which 1/6 of the reaction volume was removed at 0 . 5 , 1 , 5 , 10 , 20 , and 30 min and stopped in Laemmli buffer supplemented with 50 mM EDTA . Reactions were then boiled and loaded onto a 12 . 5% Criterion precast gel ( Bio-rad ) . The gel was then dried before imaging using a Typhoon variable mode imager ( GE Healthcare ) . Gels were quantified using ImageQuant software and data were processed using SigmaPlot software package . Rates were determined using non-linear regression . Experiments were performed 3 times , and mean errors were determined . The fraction PERK bound by IPA was calculated using the standard Langmuir isotherm equation . Using the calculated IC50 described in Figure 4A . fraction bound=11+IC50[I] . 10 . 7554/eLife . 05434 . 024Figure 8 . Synthesis of intermediate 3 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 024 2 , 4-dichloropyrimidine 2 ( 13 . 30 g , 89 . 31 mmol ) and 3-amino-5-cyclopropylpyrazole 1 ( 11 . 00 g , 89 . 31 mmol ) were dissolved in tetrahydrofuran ( TFA ) ( 100 ml ) followed by the addition of deionized water ( 100 ml ) . The solution was then treated with potassium acetate ( 261 . 9 g , 2 . 5 mol ) . The resulting mixture was kept at 55°C for 48 hr . The mixture was filtered to remove excess potassium acetate . The organic layer was separated and dried using magnesium sulfate and concentrated before loading on an AnaLogix silica column . Intermediate 3 was separated as previously reported ( Hosoda et al . , 2010 ) . 1H NMR ( 400 MHz , DMSO ) δ 12 . 14 ( s , 1H ) , 10 . 23 ( s , 1H ) , 1 . 84 ( m , 1H ) , 0 . 88 ( m , 2H ) , 0 . 64 ( m , 2H ) . 13C NMR ( 100 MHz , DMSO ) δ 161 . 4 , 160 . 7 , 160 . 0 , 153 . 9 , 148 . 6 , 147 . 8 , 146 . 8 , 8 . 4 , 8 . 2 . MS calculated for C10H10N5 235 . 06 , found 236 . 15 ( M+ ) . 10 . 7554/eLife . 05434 . 025Figure 9 . Synthesis of intermediate 5 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 025 Pyrimidine monochloride 3 ( 4 g , 16 . 9 mmol ) ( Oakwood chemical ) and p-phenylene-diamine 4 ( 1 . 9 g , 16 . 9 mmol ) ( Sigma ) were dissolved in butanol ( BuOH ) ( 50 ml ) ( Sigma ) followed by the addition of concentrated HCl ( 0 . 1 ml ) ( Fisher Scientific ) . The resulting mixture was kept at 100°C overnight . Purple precipitate was collected by filtration , washed with 30 ml of cold BuOH and dried under vacuum yielding intermediate 5 , which was used in the next step without further purification . MS calculated for C16H17N7 307 . 15 , found 308 . 5 . 10 . 7554/eLife . 05434 . 026Figure 10 . Synthesis of cmp1 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 026 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry dimethylformamide ( DMF ) ( Sigma ) . The flask was placed in an ice bath before addition of 3- ( trifluoromethyl ) phenyl isocyanate ( 93 . 5 mg; 0 . 069 ml; 0 . 5 mmol ) 6 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by the addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and distilled water ( dH2O ) for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yielded cmp1 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 30 ( s , 1H ) , 10 . 00 ( s , 2H ) , 8 . 70 ( s , 3H ) , 7 . 43 ( m , J = 2 . 3 Hz , 6H ) , 6 . 62 ( d , J = 2 . 1 Hz , 5H ) , 5 . 29 ( s , 1H ) 2 . 51 ( m , J = 10 . 2 Hz , 4H ) , 1 . 79 ( t , J = 9 . 6 Hz , 1H ) , 0 . 87 ( s , 4H ) 13C NMR ( 151 MHz , DMSO-d6 ) δ 155 . 7 , 152 . 8 , 141 . 8 , 138 . 4 , 135 . 5 , 133 . 6 , 130 . 2 , 129 . 3 , 128 . 9 , 118 . 8 , 112 . 6 , 97 . 6 , 8 . 9; ESI-MS m/z [M + H] found 495 . 49 calculated 494 . 49 . 10 . 7554/eLife . 05434 . 027Figure 11 . Synthesis of cmp2 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 027 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . Flask was placed in an Ice bath before addition of p-Toyl isocyanate ( 0 . 066 mg; 0 . 063 ml; 0 . 5 mmol ) 7 ( Sigma ) drop wise using a Hamilton syringe under argon . Resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield cmp2 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 21 ( s , 1H ) , 9 . 13 ( s , 1H ) , 8 . 58 ( q , J = 1 . 3 Hz , 9H ) , 8 . 35 ( d , J = 21 . 6 Hz , 3H ) , 6 . 54 ( s , 3H ) 2 . 51 ( m , J = 10 . 2 Hz , 4H ) , 1 . 89 ( s , 1H ) 1 . 29 ( t , J= 1 . 1 Hz , 3H ) , 0 . 87 ( s , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 155 . 4 , 153 . 8 , 146 . 8 , 140 . 3 , 139 . 7 , 136 . 7 , 131 . 2 , 127 . 8 , 117 . 3 , 115 . 4 , 63 . 5 , 33 . 5 , 32 . 1 8 . 1; ESI-MS m/z [M + H] found 440 . 28 calculated 440 . 21 . 10 . 7554/eLife . 05434 . 028Figure 12 . Synthesis of cmp3 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 028 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . The flask was placed in an ice bath before addition of 4-Ethylphenyl isocyante ( 0 . 073 mg; 0 . 072 ml; 0 . 5 mmol ) 8 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield product cmp3 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 30 ( s , 1H ) , 10 . 00 ( s , 2H ) , 7 . 50 ( d , J = 2 . 1 Hz , 3H ) , 7 . 33 ( s , 6H ) , 7 . 08 ( t , J = 7 . 2 Hz , 3H ) , 1 . 79 ( s , 2H ) , 1 . 13 ( m , J = 13 . 4 Hz , 5H ) , 0 . 87 ( s , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 162 . 4 , 153 . 0 , 141 . 16 , 130 . 4 , 130 . 0 , 129 . 9 , 123 . 8 , 122 . 21 , 118 . 4 , 114 . 5 , 113 . 9 , 35 . 6 , 8 . 4; ESI-MS m/z [M + H] found 455 . 5 calculated 454 . 31 . 10 . 7554/eLife . 05434 . 029Figure 13 . Synthesis of cmp4 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 029 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . The flask was placed in an ice bath before addition of 3-methoxyphenyl isocyanate ( 0 . 075 mg; 0 . 066 ml; 0 . 5 mmol ) 9 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by the addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield cmp4 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 43 ( s , 1H ) , 11 . 1 ( s , 1H ) , 10 . 17 ( s , 1H ) , 8 . 78 ( d , J = 19 Hz , 2H ) , 8 . 68 , 7 . 82 ( s , 1H ) , 7 . 57 ( q , j = 6 . 1 Hz , 2H ) , 7 . 41 ( t , J = 8 . 9 Hz , 2H ) , 7 . 18 ( q , J = 7 . 2 Hz , 3H ) , 6 . 84 ( d , J = 5 . 8 Hz , 1H ) , 3 . 89 ( s , 3H ) , 1 . 79 ( s , 1H ) , 0 . 87 ( s , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 160 . 1 , 152 . 95 , 141 . 4 , 130 . 0 , 119 . 9 , 110 . 9 , 107 . 6 , 104 . 4 , 55 . 3 , 8 . 37; ESI-MS m/z [M + H] found 457 . 6 calculated 456 . 45 . 10 . 7554/eLife . 05434 . 030Figure 14 . Synthesis of cmp5 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 030 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . The flask was placed in an ice bath before addition of 4-methoxyphenyl isocyanate ( 0 . 075 mg; 0 . 065; 0 . 5 mmol ) 10 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by the addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield cmp5 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 40 ( s , 1H ) , 11 . 18 ( s , 1H ) , 10 . 20 ( s , 1H ) , 9 . 73 ( q , J = 8 . 6 Hz , 2H ) , 8 . 78 ( d , J = 8 . 9 Hz , 3H ) , 7 . 62 ( t , J = 2 . 5 Hz , 2H ) , 7 . 44 ( t , j = 6 . 1 Hz , 3H ) , 6 . 98 ( t , J = 7 . 2 Hz , 2H ) 3 . 78 ( m , J = 12 . 2 Hz , 3H ) , 1 . 79 ( s , 1H ) , 0 . 87 ( s , 4H ) , 0 . 54 ( s , 2H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 159 . 3 , 153 . 1 , 141 . 3 , 125 . 3 , 121 . 4 , 110 . 4 , 108 . 6 , 104 . 3 , 58 . 3 , 8 . 4; ESI-MS m/z [M + H] found 457 . 5 calculated 456 . 51 . 10 . 7554/eLife . 05434 . 031Figure 15 . Synthesis of cmp6 ( IPA ) . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 031 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . The flask was placed in an ice bath before addition of 4- ( methylthio ) phenyl isocyanate ( 0 . 082 mg; 0 . 07 ml; 0 . 5 mmol ) 11 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield IPA . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 10 ( s , 1H ) , 9 . 95 ( s , 2H ) , 8 . 61 ( d , J = 3 . 7 Hz , 2H ) , 7 . 48 ( d , J = 2 . 3 Hz , 8H ) , 7 . 22 ( d , J = 21 . 3 Hz , 3H ) , 2 . 21 ( m , J = 10 . 2 Hz , 3H ) , 1 . 79 ( s , 1H ) , 0 . 87 ( s , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 161 . 3 , 155 . 3 , 142 . 3 , 140 . 6 , 137 . 4 , 135 . 8 , 133 . 3 , 131 . 2 , 128 . 3 , 119 . 5 , 114 . 3 , 16 . 5 , 8 . 2; ESI-MS m/z [M + H] found 472 . 18 calculated 472 . 20 . 10 . 7554/eLife . 05434 . 032Figure 16 . Synthesis of cmp7 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 032 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . The flask was placed in an ice bath before addition of 4-iodophenyl isocyanate ( 0 . 122 mg , 0 . 5 mmol ) 12 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield cmp7 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 21 ( s , 1H ) , 9 . 10 ( s , 1H ) , 8 . 81 ( s , 1H ) , 8 . 79 ( s , 1H ) , 8 . 52 ( t , J = 8 . 5 Hz , 1H ) , 8 . 08 ( S , 1H ) , 7 . 68 ( m , J = 11 . 8 Hz , 4H ) , 7 . 38 ( m , J = 10 . 2 , 4H ) , 6 . 14 ( d , J = 15 . 2 Hz , 2H ) , 1 . 79 ( s , 1H ) , 0 . 87 ( s , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 156 . 4 , 152 . 9 , 140 . 3 , 137 . 7 , 136 . 1 , 133 . 5 , 120 . 7 , 120 . 2 , 119 . 3 , 116 . 9 , 98 . 1 , 84 . 7 , 8 . 1; ESI-MS m/z [M + H] found 553 . 3 calculated 552 . 08 . 10 . 7554/eLife . 05434 . 033Figure 17 . Synthesis of cmp8 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 033 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . The flask was placed in an ice bath before addition of 4-bromophenyl isocyanate ( 0 . 099 mg , 0 . 5 mmol ) 13 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield cmp8 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 21 ( s , 1H ) , 9 . 51 ( s , 1H ) , 8 . 81 ( m , J = 21 . 8 HZ , 2H ) , 8 . 68 ( d , J = 6 . 3 Hz , 1H ) , 7 . 82 ( s , 1H ) , 7 . 48 ( m , J = 22 . 5 Hz , 8H ) , 7 . 33 ( d , J = 23 . 2 Hz , 2H ) , 7 . 08 ( t , J = 7 . 2 , 3H ) , 1 . 79 ( s , 1H ) , 0 . 87 ( s , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 155 . 3 , 152 . 3 , 140 . 9 , 138 . 3 , 137 . 6 , 132 . 6 , 122 . 4 , 121 . 8 , 120 . 8 , 118 . 4 , 95 . 4 , 82 . 7 , 8 . 2; ESI-MS m/z [M + H] found 505 . 5 calculated 504 . 10 . 10 . 7554/eLife . 05434 . 034Figure 18 . Synthesis of cmp9 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 034 Intermediate 5 ( 153 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . The flask was placed in an ice bath before addition of 4-isopropylphenyl isocyanate ( 0 . 080 mg; 0 . 08; 0 . 5 mmol ) 14 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield cmp9 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 11 ( s , 1H ) , 8 . 51 ( t , J = 13 . 6 Hz , 4H ) , 7 . 58 ( m , J = 8 . 3 Hz , 2H ) , 7 . 22 ( q , J = 12 . 6 Hz , 8H ) , 6 . 48 ( d , J = 5 . 1 Hz , 1H ) , 1 . 13 ( m , J = 13 . 4 , 8H ) , 0 . 87 ( m , J = 19 . 4 Hz , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 158 . 3 , 151 . 6 , 144 . 7 , 139 . 3 , 138 . 6 , 136 . 9 , 134 . 5 , 124 . 8 , 119 . 1 , 118 . 5 , 36 . 5 , 32 . 5 , 8 . 1; ESI-MS m/z [M + H] found 469 . 5 calculated 468 . 24 . 10 . 7554/eLife . 05434 . 035Figure 19 . Synthesis of cmp10 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 035 Intermediate 5 ( 200 mg , 0 . 5 mmol ) was dissolved in 10 ml of dry DMF . The flask was placed in an ice bath before addition of 4-tert-butylphenyl isocyanate ( 0 . 087 mg; 0 . 089 ml; 0 . 5 mmol ) 15 ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield cmp10 . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 12 . 11 ( s , 1H ) , 9 . 43 ( s , 1H ) , 8 . 81 ( s , 1H ) , 8 . 68 ( m , J = 22 . 3 Hz , 2H ) , 7 . 92 ( s , 1H ) , 7 . 43 ( m , j = 28 . 1 Hz , 8H ) , 6 . 28 ( t , J = 7 . 2 Hz , 2H ) 1 . 79 ( s , 1H ) , 1 . 13 ( s , 9H ) , 0 . 87 ( s , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 153 . 1 , 144 . 3 , 137 . 7 , 135 . 8 , 133 . 9 , 125 . 8 , 119 . 1 , 118 . 3 , 119 . 07 , 118 . 3 , 34 . 3 , 31 . 7 , 30 . 9 , 8 . 1; ESI-MS m/z [M + H] found 483 . 5 calculated 482 . 25 . 10 . 7554/eLife . 05434 . 036Figure 20 . Synthesis of IPAx intermediate 1 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 036 Pyrimidine monochloride 3 ( 0 . 4 g , 1 . 67 mmol ) was dissolved in dry round bottom flask containing 30 ml of dichloromethane ( anhydrous ) ( Sigma ) . The flask was then chilled to 4°C in an ice bath under argon . Potassium carbonate ( K2CO3 ) ( 0 . 92 g , 6 . 68 mmol ) was added , and the reaction was stirred for 10 min . Iodomethane ( 0 . 425 g , 2 . 5 mmol ) was then added drop-wise and kept at 4°C until determined complete by thin layer chromatography . The reaction was then filtered to remove excess K2CO3 and an equal volume of dH2O was added to reaction mixture . The mixture was washed three times with brine solution and then dried using magnesium sulfate . The solution was then filtered and concentrated before being loaded on to an AnaLogix silica column . Monomethylated intermediate 16 was isolated using a solution of 5% methanol in chloroform over 30 min . ESI-MS m/z [M + H] found 250 . 3 calculated 249 . 09 . 10 . 7554/eLife . 05434 . 037Figure 21 . Synthesis of IPAx intermediate 2 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 037 Intermediate 16 ( 0 . 157 g , 0 . 633 mmol ) and p-phenylene-diamine ( 0 . 068 g , 0 . 633 mmol ) were dissolved in BuOH ( 3 ml ) followed by the addition of concentrated HCl ( 0 . 01 ml ) resulting mixture was kept at 100°C for 8 hr . Isolation and purification of 17 is as described for intermediate 3 . ESI-MS m/z [M + H]+ found 322 . 5 calculated 321 . 19 . 10 . 7554/eLife . 05434 . 038Figure 22 . Synthesis of IPAx intermediate 3 . DOI:http://dx . doi . org/10 . 7554/eLife . 05434 . 038 Intermediate 17 ( 0 . 038 mg , 0 . 123 mmol ) was dissolved in 5 ml of dry DMF . The flask was placed in a ice bath before addition of 4- ( methylthio ) phenyl isocyonate ( 0 . 023 mg , 0 . 123 mmol ) ( Sigma ) drop wise using a Hamilton syringe under argon . The resulting mixture was allowed to warm to room temperature overnight . Crude compound was isolated by the addition of cold H2O ( 30 ml ) and isolated by filtration . Solid was washed with 100% CH3CN and then suspended in 1:1 CH3CN and dH2O for reverse-phase purification using HPLC . A gradient from 2–80% CH3CN:H2O ( 0 . 1% TFA ) was used to further during purification on a C18 column ( Agilent Technologies ) followed by lyophilization to yield IPAX . 1H NMR ( 400 MHz , DMSO-d6 ) : δ 10 . 00 ( s , 2H ) , 8 . 81 ( s , 1H ) , 8 . 68 ( d , J = 2 . 3 Hz , 1H ) , 7 . 82 ( s , 1H ) , 7 . 50 ( d , j = 2 . 1 Hz , 3H ) , 7 . 33 ( s , 6H ) , 7 . 08 ( t , J = 7 . 2 , 3H ) , 6 . 08 ( s , 2H ) , 1 . 79 ( s , 1H ) , 0 . 87 ( s , 4H ) ; 13C NMR ( 151 MHz DMSO-d6 ) δ 162 . 4 , 153 . 0 , 141 . 1 , 131 . 3 , 130 . 1 , 129 . 3 , 123 . 8 , 122 . 1 , 118 . 2 , 115 . 2 , 114 . 2 , 8 . 4; ESI-MS m/z [M + H]+ found 487 . 5 calculated 486 . 21 .	Cells contain thousands of proteins that carry out the essential tasks needed for survival . Before they can work , proteins must first fold into specific three-dimensional shapes . The endoplasmic reticulum , a cellular compartment that specializes in properly folding newly made proteins into their native states , is critical for this protein maturation process . If folding-enzymes in the endoplasmic reticulum are not properly balanced with the load of proteins they must fold , the endoplasmic reticulum can be overwhelmed with unfolded proteins that accumulate , leading to ‘endoplasmic reticulum stress’ . The cell copes with endoplasmic reticulum stress by triggering the ‘unfolded protein response’ ( UPR ) . This response helps to clear the unfolded proteins by increasing the size of the endoplasmic reticulum and the concentration of folding enzymes within it , and by decreasing the influx of newly made protein into the endoplasmic reticulum . The UPR engages signaling molecules in the endoplasmic reticulum membrane , among them two signaling enzymes called IRE1 and PERK . Drugs that activate these signaling enzymes could help the cell to deal with unfolded proteins , prevent toxicity resulting from endoplasmic reticulum stress , and ward off the diseases that result from it . Mendez , Alfaro , Morales-Soto et al . developed a small molecule , called IPA ( short for IRE1/PERK Activator ) , that was designed to bind to and activate IRE1 . Serendipitously , IPA not only activated IRE1 but also activated PERK . Surprisingly , PERK activation was only observed at low IPA concentrations in which IPA occupied the active sites in only a few PERK molecules , whereas at higher concentrations and full occupancy IPA completely inhibited PERK . Mendez , Alfaro , Morales-Soto et al . proposed that , under conditions of partial IPA occupancy , a minority of IPA-bound PERK molecules assume an activated state that propagates to adjacent PERK molecules that have no IPA bound to them , and activates them . Similar dose-dependent activation was previously observed for a clinically used drug designed to inhibit a similar signaling enzyme that is important in cancer progression . Together with the observations of Mendez , Alfaro , Morales-Soto et al . , these results suggest that research into similar treatments must consider that a ‘minimal dose’ can exist , below which drugs may have the opposite effect to what is desired . Further work is still needed to fully understand the mechanisms that produce such behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2015	Endoplasmic reticulum stress-independent activation of unfolded protein response kinases by a small molecule ATP-mimic
V1 and V2b interneurons ( INs ) are essential for the production of an alternating flexor–extensor motor output . Using a tripartite genetic system to selectively ablate either V1 or V2b INs in the caudal spinal cord and assess their specific functions in awake behaving animals , we find that V1 and V2b INs function in an opposing manner to control flexor–extensor-driven movements . Ablation of V1 INs results in limb hyperflexion , suggesting that V1 IN-derived inhibition is needed for proper extension movements of the limb . The loss of V2b INs results in hindlimb hyperextension and a delay in the transition from stance phase to swing phase , demonstrating V2b INs are required for the timely initiation and execution of limb flexion movements . Our findings also reveal a bias in the innervation of flexor- and extensor-related motor neurons by V1 and V2b INs that likely contributes to their differential actions on flexion–extension movements . Terrestrial animals use their limbs to generate a broad array of motor behaviors , from stereotypical movements that include protective reflexes and locomotion to complex volitional tasks that are exemplified by reaching and grasping movements ( Grillner , 1975; Alstermark and Isa , 2012 ) . Charles Sherrington ( 1906 ) first demonstrated that all such motor behaviors rely on the reciprocal actions of flexor and extensor muscles around each limb joint and that terrestrial animals require reciprocal inhibition to move and articulate their limbs . It is known that reciprocal flexor–extensor motor activity is produced by inhibitory neurons in the spinal cord , many of which appear to be core components of the locomotor central pattern generator ( CPG ) ( reviewed in Kiehn , 2006; Goulding , 2009; Grillner and Jessell , 2009; Arber , 2012 ) . However , the functional organization of the inhibitory circuits that control flexor–extensor activity and the contribution that different inhibitory neuron cell types make to flexor–extensor motor control is still poorly understood . Studies in the cat have identified a number of physiologically defined IN cell types that are candidates for exercising flexor–extensor control . The most prominent of these are reciprocal Ia inhibitory interneurons ( IaINs ) , which are activated by muscle spindle afferents and inhibit antagonist motor neurons . IaINs are rhythmically active during locomotion ( Feldman and Orlovsky , 1975; Pratt and Jordan , 1987 ) and scratching ( Deliagina and Orlovsky , 1980 ) . Peak IaIN activity coincides with the phase in which antagonist motor neurons are hyperpolarized ( Pratt and Jordon , 1987; Geertsen et al . , 2011 ) . This activity profile is strong evidence of a central role in reciprocal inhibition . The contribution that non-reciprocal inhibitory Ib interneurons ( IbINs ) make to locomotion is less clear . While the IbINs that are innervated by Golgi tendon organs ( GTOs ) generally inhibit homonymous motor neurons under non-locomotor conditions ( Jankowska , 1992; Pearson and Collins , 1993 ) , during locomotion the Ib pathway exerts an excitatory effect on extensor motor activity ( Conway et al . , 1987; Pearson and Collins , 1993; Gossard et al . , 1994; Angel et al . , 2005 ) . Nonetheless , inhibition mediated by IbINs has been observed during the extension phase of walking in humans ( Shoji et al . , 2005 ) . This inhibition is associated with the unloading of the limb , and it suggests that IbINs may contribute to the phase transition from stance ( extension ) to swing ( flexion ) . The role that Renshaw cells play in shaping flexor–extensor locomotor activity appears to be more limited ( Pratt and Jordan , 1987 ) . While these cells are rhythmically active during locomotor activity , reducing Renshaw cell transmission by pharmacological or genetic blockade of cholinergic transmission changes the periodicity of the locomotor rhythm with little discernible effect on flexor–extensor alternation ( Noga et al . , 1987; Myers et al . , 2005 ) . Genetic analyses in mice have identified a number of molecularly defined interneuron populations that provide inhibition to motor neurons ( reviewed in Goulding , 2009; Arber , 2012 ) . Among these are three prominent populations of ipsilaterally projecting inhibitory interneurons: dorsal Lbx1-derived inhibitory INs ( Gross et al . , 2002; Tripodi et al . , 2011 ) and ventral V1 and V2b INs ( Saueressig et al . , 1999; Sapir et al . , 2004; Zhang et al . , 2014 ) . The V1 and V2b IN populations are both heterogeneous , being made up of multiple physiological cell types including Renshaw cells ( V1 INs ) , IaINs ( V1 and V2b INs ) and putative IbINs ( V2b INs ) ( Sapir et al . , 2004; Alvarez et al . , 2005; Zhang et al . , 2014 ) , as well as other as yet unidentified inhibitory cell types . Recently , we have found that the composite activities of the V1 and V2b INs are required to secure flexor–extensor alternation in the in vitro neonatal spinal cord and in newborn mice ( Zhang et al . , 2014 ) . This finding is consistent with our demonstration that cells with the features of IaINs are derived from both V1 and V2b INs and that disynaptic reciprocal inhibition is only abolished when both of these populations are functionally inactivated ( Wang et al . , 2008; Zhang et al . , 2014 ) . The limited repertoire of motor behaviors that can be assayed using the in vitro neonatal spinal cord preparation , together with our inability to discern any marked differences in the function of V1 and V2b INs with respect to generating an alternating flexor–extensor motor rhythm in vitro , prompted us to examine the contribution that V1 and V2b INs make to motor control in awake behaving mice . In particular , we were interested in determining whether these two inhibitory interneuron classes control discrete aspects of limb movement with regard to flexor–extensor-driven motor behaviors . Our results demonstrate a striking functional bias in the actions of V1 and V2b INs on flexor–extensor motor activity , whereby the selective ablation of V1 vs V2b INs results in hindlimb hyperflexion and hyperextension , respectively . This functional bias was observed both in air-stepping juvenile mice and in adult animals performing motor tasks such as over-ground walking . In analyzing the inhibitory inputs to motor neurons from these two interneuron populations , we find a genetically defined bias in V1 and V2b connectivity that likely underpins the differential effects on flexor–extensor motor activity . Specifically , a higher proportion of the inhibitory contacts on flexor motor neurons come from V1 INs as compared to extensor motor neurons , whereas V2b INs preferentially contact motor neurons that innervate extensor muscles . To determine the efficiency and specificity of this technique , we monitored V1 IN ablation in P1 En1Cre; Cdx2-FlpO; Maptds-DTR mice treated with a single dose of DTX and analyzed 6 days later . P1 control mice ( En1Cre; Maptds-DTR ) were also treated with DTX and analyzed . Because En1 is no longer expressed postnatally , an Ai6 Cre-dependent reporter allele ( Madisen et al . , 2010 ) was used to independently mark the V1 INs and verify their loss following DTX treatment . Whereas V1 INs were largely spared at upper cervical levels ( Figure 2A , B ) , they were depleted by >90% in the thoracic , lumbar , and sacral spinal cord of En1Cre; Cdx2-FlpO; Maptds-DTR; Ai6 mice ( Figure 2C , D ) . Calbindin+ Renshaw cells , which are derived from En1+ progenitors ( Sapir et al . , 2004 ) , were also largely missing from the caudal spinal cord ( Figure 2E , F ) . By contrast , dorsal calbindin+ neurons that are not derived from En1+ progenitors were still present in normal numbers ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 04718 . 004Figure 2 . Restricted ablation of V1 and V2b INs following diphtheria toxin treatment . Immunohistochemical and histological analysis of P7 control , En1Cre; Cdx2-FlpO; Maptds-DTR and Gata3Cre; Cdx2-FlpO; Maptds-DTR animals 6 days after administering DTX . All sections are from the mid-lumbar cord except for those in panels A , B , M , and N , which are from the cervical cord . ( A–D ) En1-derived V1 INs ( green ) are selectively ablated in the lumbar spinal cord ( c . f . C , D ) , whereas ChAT+ motor neurons ( red ) are spared at lumbar ( D ) and cervical levels ( B ) . ( E , H ) Calbindin+ Renshaw cells ( blue ) are present in control cords ( E , arrow ) but not in V1 IN-ablated cords ( F ) . ( G , H ) Chx10+ V2a INs are present in normal numbers following V1 IN-ablation . ( I , J ) Hematoxylin-eosin staining reveals no evidence of widespread neuronal cell loss or gliosis . ( K , L ) CD86+ microglia were occasionally observed in close proximity to V1 cell debris ( arrowhead ) . ( M–P ) Motor neurons are spared at both cervical ( M , N ) and lumbar ( O , P ) levels following ablation of the V2b INs . ( Q , R ) V2b INs ( green ) are specifically deleted in V2b IN-ablated mice ( R ) while V0c neurons ( red ) are spared . ( S , T ) Chx10+ V2a INs are present in normal numbers following V2b IN-ablation . ( U , V ) Hematoxylin-eosin staining . ( W , X ) Localized CD86 expression ( red ) in lamina VII is associated with dying V2b INs ( green ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 00410 . 7554/eLife . 04718 . 005Figure 2—figure supplement 1 . Selective ablation of ventral calbindin-expressing neurons . ( A , B ) Expression of calbindin in the dorsal spinal cord of P7 control ( A ) and V1 IN-ablated mice ( B ) . ( C ) V1 IN cell numbers per hemisection at thoracic and lumbar levels ( n = 15 sections [30 μm] from 3 spinal cords each for control and V1 IN-ablated mice ) . ( D , E ) Counts of neurons in the lumbar spinal cord of P7 DTX-treated mice ( 6 days p . i . ) . Cell counts at the indicated levels were expressed as a percentage of comparable counts for control sections ( n = 12 sections [30 μm] for control ) , V1 IN-ablated and V2b IN-ablated mice ( n = 3 cords for each genotype ) . Data are expressed as mean ± s . d . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 005 To confirm the specificity of V1 IN cell killing , sections from control and V1 IN-ablated cords were stained with an antibody to choline acetyltransferase ( ChAT ) to visualize motor neurons and cholinergic V0c INs . Both populations were unaffected by the DTR-dependent ablation of V1 INs ( Figure 2A–D , data not shown ) . Moreover , there was no reduction in the number of Chx10+ V2a INs , a prominent population of excitatory neurons that are intermingled with V1 INs in lamina VII ( Figure 2G , H ) . Further histological analysis revealed no change in the integrity of the spinal cord following V1 IN ablation ( Figure 2I , J ) nor was there any reduction in neuronal cell numbers ( Figure 2—figure supplement 1D ) . We did observe a small transient increase in CD86 expression as is expected with the targeted killing of V1 INs ( Figure 2K , L ) . This CD86 expression was localized to lamina VII where V1 INs reside , and in most instances co-localized with GFP-labeled V1 cell debris ( Figure 2L , arrowhead ) . Most importantly , CD86 expression was not widespread nor was there any evidence of infiltration by CD45R-positive B cells or CD3-positive T cells ( data not shown ) . These results agree with other studies showing DTR-mediated cell killing is highly selective and cell autonomous ( Buch et al . , 2005; Hatori et al . , 2008 ) . A similar analysis was performed on Gata3Cre; Cdx2-FlpO; Maptds-DTR mice post DTX-treatment ( Figure 2M–X , Figure 2—figure supplement 1E ) . These mice also displayed a selective loss of V2b INs in the thoracic and lumbosacral spinal cord ( Figure 2R ) . Other spinal neurons including motor neurons , V0c INs and Chx10+ V2a INs were completely spared in these cords ( Figure 2M–T ) . Once again , we observed limited and localized expression of CD86 in lamina VII ( Figure 2X ) . These data demonstrate that the intersectional ablation approach selectively and efficiently deletes V1 and V2b INs in caudal regions of the spinal cord . When DTX was administered to postnatal En1Cre; Cdx2-FlpO; Maptds-DTR pups , it produced a strong hyperflexion phenotype within 3–4 days ( Figure 3 ) . Whereas control P7 pups flexed and extended their hindlimbs when suspended by their tails ( Figure 3A ) , the hindlimbs of P7 DTX-treated En1Cre; Cdx2-FlpO; Maptds-DTR V1 IN-ablated pups remained flexed ( Figure 3B ) , even though their forelimbs were able to extend and display a full range of stepping movements ( Figure 3B , asterisk ) . In contrast to control mice and V1 IN-ablated mice , P7 Gata3Cre; Cdx2-FlpO; Maptds-DTR V2b IN-ablated mice maintained their hindlimbs in an extended state when suspended by their tail ( Figure 3C ) , even though they were still able to flex and extend their forelimbs . 10 . 7554/eLife . 04718 . 006Figure 3 . Mice lacking V1 and V2b INs show abnormal hindlimb movements . Time-lapse sequence images showing the hindlimb movements of P7 mice suspended by their tails . Mice were photographed 4 days after DTX treatment . The genotypes of the mice are indicated . Control mice were littermates that lacked the Cdx2-FlpO allele . ( A ) Control mice are able to flex ( arrowheads ) and extend ( arrow ) their hindlimbs when suspended by their tail . ( B ) Following the ablation of V1 INs , P7 mice lose their ability to extend their hindlimbs , which remain clasped to the body in a flexed position ( arrowheads ) . The forelimbs of these mice are able to undergo extension movements ( asterisk ) . ( C ) Ablation of V2b INs leads to pronounced extension of the hindlimbs ( arrows ) and impairment of hindlimb flexion movements . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 006 We then analyzed the hindlimb locomotor movements of DTX-treated mice that were induced to airstep by subcutaneous injection with L-DOPA . Kinematic analysis revealed strong rhythmic stepping in both the forelimbs and hindlimbs of control animals ( Figure 4 ) . By contrast , V1 IN-ablated animals displayed very little hindlimb movement , in contrast to the forelimbs that showed a normal range of motion . The hindlimbs of V1 IN-ablated animals , while occasionally displaying small rhythmic twitches ( Figure 4B , asterisks ) , remained flexed during stepping ( see Figure 4A , middle panel , arrowheads ) . Moreover , the maximal opening of the ankle joint in these animals was less than 80° compared to 145° for control animals ( Figure 4B , C ) . In those instances where ankle joint did partially open , the overall change in angle was reduced from 90° to less than 30° ( Figure 4C ) . 10 . 7554/eLife . 04718 . 007Figure 4 . Juvenile mice lacking V1 and V2b INs display abnormal hindlimb movements . ( A ) Images of airstepping P7 animals following DTX injection . Flexed limbs are depicted with black arrowheads and extended limbs with white arrows . Control animals ( left ) flex and extend their hindlimbs . The hindlimbs of V1 IN-ablated animals ( middle ) remain flexed and those of V2b IN-ablated mice ( right ) are fully extended . ( B ) Traces showing the change in ankle ( top ) and elbow ( bottom ) joint angle during 5 s of airstepping . Deletion of V1 and V2b INs impairs movement of the ankle joint , whereas rhythmic movements are maintained for the elbow joint . Small rhythmic variations of the ankle joint angle occur in phase with movements of the elbow joint angle ( asterisks ) . ( C ) Measured changes in joint movement showing the amplitude change ( left ) , the maximal joint angle ( filled symbol ) and the minimal joint angle ( open symbol ) joint angle ( right ) . The mean and s . d . is shown for 24 consecutive steps ( n = 3 animals for each genotype ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 007 When V2b IN-ablated mice were induced to airwalk and analyzed in a similar manner , they displayed robust rhythmic forelimb stepping movements ( Figure 4 , right panels ) . By contrast , the hindlimbs of these mice remained extended and displayed very little in the way of flexion movements . Interestingly , L-DOPA also induced small rhythmic deflections of the hindlimb in these mice ( Figure 4B , right panel , asterisk ) , indicating that the hindlimb locomotor CPG still produces an underlying locomotor rhythm . However , as the angle of the ankle joint was not reduced below 130° ( Figure 4C ) , it appears that the hindlimb locomotor CPG is unable to elicit strong flexion movements in the absence of V2b inhibition . To delve more deeply into the nature of the motor deficits that underlie the opposing motor phenotypes that arise when V1 and V2b INs are ablated , fine EMG recording electrodes were implanted unilaterally in the tibialis anterior ( TA ) and gastrocnemius ( GS ) muscles to measure ankle flexor and extensor motor activity during L-DOPA-induced air-stepping . Control mice displayed a regular alternating pattern of EMG activity in both muscles ( Figure 5A ) in which there was little or no overlap in TA and GS burst activity . This profile of TA-flexor and GS-extensor activity in control P7 mice is very similar to that seen in adult mice ( see also Figure 6 ) , where longer duration GS bursts are interspersed with short TA bursts . The GS extensor phase was also seen to increase proportionately with longer step periods , while the TA flexor burst period was constant across all stepping speeds ( Figure 5A , D ) . This is in strong agreement with previous studies performed in rodents and cats ( Grillner , 1975; Halbertsma , 1983; Juvin et al . , 2007; Frigon and Gossard , 2009 ) showing that changes in duration of the extensor/stance phase are preferentially responsible for lengthening or shortening the step cycle . 10 . 7554/eLife . 04718 . 008Figure 5 . Altered rhythmic muscle activities during airstepping in the absence of V1 or V2b cells . ( A ) EMG recordings from the tibialis anterior ( TA , flexor ) and gastrocnemius ( GS , extensor ) muscles during L-DOPA-induced airstepping . Single deleted bursts are indicated by arrowheads . The asterisk marks a prolonged deletion that encompasses two step cycles . ( B ) Scatter plots show the relationship between ankle flexor ( TA ) burst duration and the step cycle period ( black dots ) and between ankle extensor ( GS ) burst duration and the step cycle period ( gray circles ) . Each point represents the ratio of burst to step cycle duration for a single step . Control and V2b IN-ablated animals display an extensor dominant pattern , whereas V1 IN-ablated animals show a flexor-phase dominant pattern . Mean slope of regression calculations for each experimental group shows that the step cycle period in control mice ( aGS = 0 . 72 ) is strongly correlated with GS burst duration , while TA bursts remain relatively constant ( aTA = 0 . 12 ) . The increase in step cycle duration in V1 IN-ablated mice is moderately correlated with flexor phase duration ( aTA = 0 . 46 ) . V2b IN-ablated animals display a pronounced increase in step cycle duration that is very highly correlated with the length of the extensor phase ( aGS = 0 . 96 ) . ( C ) Quantification of step cycle period , flexor and extensor burst duration for control , V1 IN-ablated and V2b IN-ablated mice . ( D ) Percentage of skipped bursts/deletions as measured during twelve 10-s periods of locomotor activity ( n = 3 animals each ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 00810 . 7554/eLife . 04718 . 009Figure 6 . Mice lacking V1 and V2b INs have opposing changes to their gait . Hindlimb kinematics from control ( left ) , V1 IN-ablated ( middle ) , and V2b IN-ablated ( right ) mice at 3 weeks post DTX injection , by which time functional recovery was maximal . ( A ) Representative stick figure diagrams showing one complete step cycle ( swing and stance ) for the hindlimb . The arrow ( middle ) indicates hyperflexion during early swing phase in V1 IN-ablated mice . The arrowhead shows hyperextension of the ankle joint during late stance in V2b IN-ablated mice . ( B ) Comparison of limb positions at the transition from stance to swing ( left ) , at mid-swing ( middle ) , and at the swing to stance transition ( right ) . ( C ) Representative angular changes to the ankle joint . The arrows indicate when the foot is lifted ( stance to swing ) and when the foot is planted ( swing to stance ) . ( D , E ) Simultaneous EMG recordings of the GS and TA muscles in one leg during walking ( D ) and swimming ( E ) . The bar in D indicates the expansion in TA activity . The asterisks indicate co-activation of the TA and GS muscles . Note the synchronous activity of both muscles during swimming in V1 IN-ablated mice . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 00910 . 7554/eLife . 04718 . 010Figure 6—figure supplement 1 . Phase relationship between TA and GS EMG activity . The average period that the TA and GS muscles are active during the step cycle was calculated . Measurements were normalized for each step cycle , with the onset of GS EMG activity ( 0 ) serving as the reference for the beginning of the step cycle and 1 indicating the onset of the following stance phase . Note the expansion of TA activity in mice lacking V1 INs . Mice lacking V2b INs show prolonged GS activity during swing when the TA muscle is active . During swimming V1 IN-ablated mice show extensive co-activation of the GS and TA muscles . The mean period of muscle activity ( gray solid bars ) for each experimental sample is shown ( n = 18 steps ) . Black error bars indicate the standard deviation for each sample . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 01010 . 7554/eLife . 04718 . 011Figure 6—figure supplement 2 . Progressive changes in EMG activity following V1 IN ablation . Representative EMG recordings and kinematics from control ( A ) , and V1 IN-ablated mice 5 ( B ) and 7 ( C ) days after commencing DTX treatment . The EMG recordings ( upper panels ) and joint angle plots ( lower panel ) are synchronized . In control animals , the TA and IP muscles are inactive during stance , and TA and GS activity is strictly alternating ( A ) . The arrow in B indicates a small burst of ectopic TA activity which is associated with hyperflexion of the limb in V1 IN-ablated mice . By day 7 p . i . , TA muscle activity has expanded and is co-active with the GS muscle during stance ( C , bar ) . IP burst activity during swing ( B , bar ) and stance ( arrowhead ) is also increased . The matching lower traces show the accompanying changes in the angle of the hip , knee , and ankle joints during walking . Note the reduced opening of the hip and knee joints that accompanies the loss of V1 INs . At day 5 p . i . , the ankle joint opens up more slowly ( black arrows ) , but is still able to fully open . By day 7 p . i . the ankle joint is flexed throughout the step cycle , both at early stance ( black arrow ) and at the transition from stance to swing ( asterisk ) . The small increase in hip motion seen at day 7 p . i . compared to day 5 p . i . is due to increased swaying of the pelvis during walking . The knee angle is also reduced during the step cycle . The shaded boxes indicate the extensor phase . Following V1 IN cell loss , ectopic TA muscle EMG activity is observed during stance ( arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 01110 . 7554/eLife . 04718 . 012Figure 6—figure supplement 3 . Altered limb and body movements in mice lacking V1 INs . ( A ) Foot placement analysis in control and V1 IN-ablated mice 3 weeks post-DTX treatment . High-speed video was used to capture the foot strike during walking on an elevated walkway . ( Left ) The position of the forepaws during stance is shown in blue , the position of the hindpaws are shown in red . The arrowhead indicates the first forelimb step , with the animals moving from right to left . ( Right ) The stride lengths for the forelimb and hindlimb were calculated for 36 steps ( n = 3 animals each for control and V1 IN-ablated ) . ( B ) V1 IN-ablated mice show an increase in the rotation of the pelvis during walking . ( Left ) The change in the angle of the pelvis was calculated with 90° representing the median position of the pelvis . ( Right ) Quantification of the angular change in pelvis during a single step . The angular change was calculated for 24 steps ( n = 3 animals each for control and V1 IN-ablated ) . Data is expressed as mean ± s . d . ** indicates p < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 012 V1 IN-ablated mice consistently displayed a marked increase in the duration of the TA burst , with the TA muscle remaining active for a proportionately longer period of each step cycle as compared to the GS muscle ( Figure 5B ) . This finding is consistent with the idea that depleting V1 IN inhibition causes a preferential degradation of inhibition to TA flexor motor neurons . The increase in TA burst duration also caused a moderate lengthening of the step cycle period in V1 IN-ablated mice ( Figure 5D ) . V2b IN-depleted mice displayed a strikingly different pattern of EMG activity during airstepping . In addition to an aggregate slowing of the motor rhythm , the duration of GS extensor muscle activity was markedly elongated ( Figure 5C–E ) when compared to control and V1 IN-ablated animals . This slowing of the motor rhythm following the loss of V2b INs can largely be attributed to the increased duration of GS extensor activity , as the TA flexor bursts were similar in length to those seen in control mice ( Figure 5C , D ) . Taken together , these findings demonstrate that a flexor-dominant motor rhythm emerges when V1 INs are deleted , while the loss of V2b INs gives rise to an extensor-dominant motor rhythm . Further analysis of the EMG profiles of V1- and V2b-IN-ablated mice revealed a significant increase in the number and frequency of skipped or deleted GS and TA bursts in air-stepping juvenile mice ( Figure 5A , arrowheads ) . Moreover , quantification of these deletions revealed a strong bias in their valency ( Figure 5D ) . Whereas TA deletions were prevalent in V2b IN-ablated pups ( Figure 5A , right panel , arrowheads ) , mice lacking V1 INs displayed a strong bias toward deletions in GS EMG activity ( Figure 5 , middle panel , arrowhead ) . The increased frequency of deletions in the V1- and V2b-IN-ablated mice ( Figure 5D ) coupled with the changes in phase duration ( Figure 5C ) , strongly suggest that inhibition from V1 and V2b INs facilitates the transition from swing to stance and from stance to swing , respectively . Our observation that the deletions in both V1- and V2b-IN-ablated mice fall into the non-resetting category strongly suggests that both of these genetically defined cell populations are involved in pattern formation rather than rhythm generation ( Lafreniere-Roula and McCrea , 2005 ) . In order to examine the nature of any persistent changes to locomotor behavior that arise from the loss of V1 and V2b INs , we used combined kinematics and EMG recordings to quantitatively analyze changes in the gait of adult animals 3–5 weeks post DTX treatment . By this time , V1- and V2b-IN-ablated mice had regained partial control of their hindlimbs ( see ‘Discussion’ ) and were able to use them to bear weight . V1 IN-ablated mice still displayed marked deficits in their gait , including a persistent hyperflexion of the hindlimbs during walking , which was particularly prominent in the mid-swing and early stance phases of the step cycle ( Figure 6A , B , middle panel ) . Furthermore , V1 IN-ablated mice flexed their hindlimbs faster during the swing phase indicating that they are no longer able to properly modulate the speed and/or force of their flexion movements ( Figure 6C ) . The EMG analysis of V1 IN-ablated mice revealed marked changes in the duration and co-activity of TA and GS muscle activity . These included ( 1 ) a persistent broadening of TA muscle activity , ( 2 ) an overlap in GS and TA burst activity during the transition from stance to swing , and ( 3 ) an increase in GS activity at the end of the stance phase ( Figure 6D , middle panel , asterisk; Figure 6—figure supplements 1 , 2 ) . The increase in GS EMG activity during stance phase appears to be a late behavioral modification that facilitates the opening of the ankle joint during walking , as it is not seen during the early phase of V1 IN ablation ( see Figure 6—figure supplement 2 ) . The co-activation of the TA and GS muscles was even more pronounced during swimming with EMG analysis revealing synchronous TA and GS muscle activity ( Figure 6E , middle panel , asterisk ) that causes the extension of the hindlimbs to be aborted during the power stroke . One factor that may contribute to the synchronous flexor–extensor activity seen during swimming may be the reduction in sensory feedback from GTOs that has been shown to occur ( Akay et al . , 2014 ) . In V1 IN-ablated mice that lack sensory feedback gated by the V1 INs , the likely loss of Ib-mediated sensory inhibition during swimming could account for the observed co-activation of flexor–extensor muscles and abortive hindlimb movements ( Figure 6 , Figure 6—figure supplement 2; see ‘Discussion’ ) . In addition to the hyperflexion phenotype , V1 IN-ablated mice displayed a prominent rear paw overshoot during walking ( Figure 6—figure supplement 3 ) . Kinematic analysis of walking V1 IN-ablated mice revealed two factors that contribute to rear paw overshoot: ( 1 ) an increase in the speed and degree of ankle flexion movements during swing ( Figure 6C ) and ( 2 ) an increase in the angular rotation of the pelvis during stepping ( Figure 6—figure supplement 3 ) . This rotation of the pelvis appears to be a behavioral adaption that helps extend the forward throw of the leg during late swing phase , so as to negate the reduced hindlimb extension that occurs when V1 INs are depleted from the spinal cord . In contrast to V1 IN-ablated mice , the locomotor phenotype that persists in mice lacking V2b INs was much milder , with the most notable change being a prolongation of the stance phase that results in overextension of the ankle joint during walking ( Figure 6A , right panel , arrowhead ) . EMG analysis revealed the likely cause of this delay , namely a second ectopic burst of GS motor activity as the limb is transitioning from stance to swing ( Figure 6D , right panel , asterisk ) . The delay in initiating flexion was not observed when the mice were swimming nor did we detect any marked differences in TA and GS EMG activity between control and V2b IN-ablated mice during swimming ( Figure 6E , right panel ) . A further set of experiments were performed to probe the nature of the early motor deficits that arise when V1 INs are ablated in adult mice . Unfortunately , we were unable to perform a similar analysis of the early motor deficits that occur in adult V2b IN-ablated animals , as treating these animals with a single high dose of DTX caused bowel blockage and increased morbidity . This is likely to be due to the expression of Gata3 in a subset of enteric neurons . Whereas control mice ( n = 5 animals ) displayed a highly reproducible pattern of EMG activity both before and after DTX treatment , with the iliopsoas ( IP ) and TA muscles only being active during the swing phase of the step cycle ( Figure 6—figure supplement 2 , upper left panel ) , the depletion of V1 INs ( n = 6 animals ) resulted in ectopic IP and TA muscle EMG activity within 5 days of DTX treatment ( Figure 6—figure supplement 2B , upper panel ) . This ectopic TA and IP muscle activity was even more pronounced 7 days after DTX treatment ( Figure 6—figure supplement 2C , upper panel ) , with the TA muscle being co-active during stance with the GS muscle ( see bars ) . A similar expansion of IP activity was observed , although to a less extent than that seen in the TA ( Figure 6—figure supplement 2C , arrowhead ) . We posit that the co-activation of the TA and IP muscles during stance constitutes the underlying mechanism that restricts the hip and ankle joints from opening during walking , and it also contributes to the postural changes that occur in V1 IN-ablated mice following DTX treatment . By contrast , there was no increase in GS activity during the swing/flexor phase indicating the loss of V1 INs has an effect on ankle flexor but not extensor activity during walking . A careful comparison of kinematic movements ( Figure 6—figure supplement 2 , lower panels ) and the time locked EMG traces ( Figure 6—figure supplement 2 , upper panels ) revealed a close correspondence between angular movement of the ankle and EMG activity in the TA and GS muscles . During the early phase of V1 IN cell loss ( 5 days p . i . ) , the ankle joint angle at maximal flexion was reduced from 40° to less than 30° ( lower middle panel ) . There was also a progressive decrease in maximal extension of the ankle joint in early swing phase ( Figure 6—figure supplement 2 , lower panels , asterisks ) . A similar hyperflexion phenotype was noted for the knee and hip joints , as indicated by their reduced angular changes during walking ( Figure 6—figure supplement 2 , lower panels ) . In the case of the hip joint , this is consistent with the observed expansion of IP muscle activity ( Figure 6—figure supplement 2C , upper panel , arrowhead ) . In summary , our findings reveal that V1 INs function to ( 1 ) restrict flexor motor activity and promote active limb extension during stance and ( 2 ) moderate flexion movements during the swing phase of the step cycle . The weakening of ankle extensor inhibition in V2b IN-ablated mice , as indicated by ectopic GS burst activity during the swing phase of the step cycle and hyperextension of the ankle during walking ( Figure 6 ) , led us to ask whether V2b INs preferentially inhibit extensor–motor activity . To test this , we took advantage of the neonatal spinal cord preparation , which can be induced to produce fictive locomotion in vitro ( Figure 7A , Kiehn , 2006; Goulding , 2009 ) and is amenable to optogenetic manipulation ( Hagglund et al . , 2010 ) . Spinal cords from P0-P1 Gata3Cre; R26lsl-ChR2 ( Ai32 ) pups displayed a characteristic alternating pattern of L2 flexor-related and L5 extensor-related locomotor rhythm in the presence of NMDA and 5-HT ( Figure 7B–D , n = 6/6 cords ) . However , upon activating channelrhodopsin in V2b INs , there was a strong and highly selective decrease in extensor-related motor activity as measured by extracellular recordings from the L5 ventral root ( Figure 7B–D , bar ) . By contrast , L2 flexor-related motor activity remained largely unchanged following V2b IN activation , although there was some disturbance in the coherence of the motor rhythm . Activation of V1 INs had a more pronounced effect on the motor rhythm , in that it completely suppressed L2 flexor-related and L5 extensor-related rhythmic bursting ( Figure 7E , bar , n = 5/5 cords ) . This is likely to be due to the activation of Renshaw cells , which are known to strongly inhibit motor neurons ( Windhorst , 1996; Bhumbra et al . , 2014 ) . In summary , the suppression of extensor-related L5 motor activity in the in vitro spinal cord preparation when V2b INs are activated provides strong evidence that V2b INs can functionally inhibit extensor-related motor activity during locomotion . This finding is consistent with the expansion of extensor motor neuron activity that we see following V2b IN ablation ( Figures 5 , 6 ) . 10 . 7554/eLife . 04718 . 015Figure 7 . Effect of optogenetic activation of V2b and V1 INs on in vitro locomotion . ( A ) Schematic showing the in vitro recording setup used for the localized light activation of ChR2 in V1 and V2b INs . ( B–D ) Representative ENG recordings from a P0 Gata3Cre; R26lsl-ChR2 ( Ai32 ) spinal cord showing the suppression of L5 extensor-related activity in the presence of blue light photostimulation ( blue bar ) . The lumbar levels that were photostimulated are indicated as upper ( L1–L2 ) , mid ( L3–L4 ) , and lower ( L5–L6 ) . ( E ) Representative recording from a P0 En1Cre; R26lsl-ChR2 ( Ai32 ) spinal cord showing suppression of L2 and L5 ventral root activity following photostimulation of V1 INs at L3–L4 . Upper traces represent the raw filtered ENG recordings . Lower traces display the matching online rectified ENG signal . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 01510 . 7554/eLife . 04718 . 013Figure 8 . Biased V1 and V2b connections onto flexor- and extensor-related motor neurons . ( A ) Schematic of experimental design used for labeling V1 and V2b inhibitory contacts on specific motor pools . Motor neurons ( MNs ) were retrogradely labeled by injecting Cy5-CTB into individual hindlimb muscles at P11 . Interneuron class-specific contacts were labeled with a conditional Thy1lsl-YFP reporter allele ( green ) and inhibitory contacts were detected with antibodies to vGAT and GlyT2 ( red ) . ( B ) Putative V1- and V2b-derived inhibitory contacts ( vGAT/GlyT2 ( red ) and Thy1-YFP reporter ( green ) ) onto CTB-labeled TA and GS motor neurons ( blue ) . Filled arrowheads indicate examples of inhibitory synaptic terminals that co-localize with YFP . YFP-negative inhibitory contacts are marked with open arrowheads . ( C ) Quantification of V1- or V2b-IN-derived inhibitory contacts on the soma/proximal dendrites of defined hindlimb motor neurons ( n = 12 motor neurons per pool ) . A greater number of putative V1 IN inhibitory synapses contact flexor motor neuron pools as compared to their antagonist extensor-related motor pools . In contrast , V2b-derived contacts represent a greater proportion of the inhibitory synapses onto extensor-related motor neurons as compared to antagonist flexor-related motor neurons . Abbreviations: GS , gastrocnemius; TA , tibialis anterior . Error bars: mean ± s . d . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 01310 . 7554/eLife . 04718 . 014Figure 8—figure supplement 1 . Axonal projections of V1 and V2b INs . P0 En1Cre; R26lsl-HTB and Gata3Cre; R26lsl-HTB mice were used to label and trace the projections of V1 and V2b INs , respectively . Rhodamine-conjugated dextran was applied unilaterally to the cut ventral horn at the L2 . Data represent cumulative cell counts from two cords each for V1 and V2b INs . The distribution shown covers the lumbar spinal cord from the T13/L1 border to L6 . DOI: http://dx . doi . org/10 . 7554/eLife . 04718 . 014 In view of the opposing phenotypes that arise from ablating V1 and V2b INs , we asked if differences in the innervation of hindlimb extensor and flexor motor pools by these two classes of inhibitory neuron might contribute to their opposing actions on limb flexion and extension movements . To address this question , defined hindlimb motor pools were visualized by backfilling them from their respective muscles with Cholera Toxin-B ( CTB ) . A conditional Thy1lsl-YFP transgene reporter ( Buffelli et al . , 2003 ) was then used in combination with En1Cre or Gata3Cre to label the terminal processes of V1 and V2b INs . Presumptive inhibitory synapses on CTB-labeled motor neurons were then identified with antibodies to vGAT and GlyT2 ( Figure 8A ) . The analysis of V1 and V2b inhibitory contacts focused on the soma and proximal dendrites of motor neurons , which is where inputs from inhibitory interneurons such IaINs and Renshaw cells tend to be concentrated ( Jankowska and Roberts , 1972; Brown , 1981 ) . Counts of V1-derived inhibitory contacts on motor neurons revealed a strong positive bias in their innervation of the TA ankle flexor motor neurons ( Figure 8 ) . This preferential innervation is also seen for hip flexor motor neurons , which receive a greater percentage of inputs from V1 INs when compared to their antagonist extensor motor pools ( Zhang et al . , 2014 ) . In comparing V1 inhibitory inputs onto TA and GS motor neurons , we counted an approximate twofold greater number of V1-derived synaptic contacts onto TA motor neurons as compared to their antagonist GS counterparts ( Figure 8B , C ) . Conversely , the proportion of inhibitory contacts onto GS ( ankle ) extensor motor neurons from V2b INs was higher than the proportion of V2b contacts onto their TA counterparts ( Figure 8C ) . We have not seen any significant difference in the proportion of V2b IN contacts onto the knee quadriceps ( Q ) and BF/St motor pools ( Zhang et al . , 2014 ) . These two knee muscle groups are bifunctional ( Yakovenko et al . , 2002 ) , and the near neutral weighting of V2b IN inputs onto these motor pools may reflect their bi-functional nature with regard to knee and hip flexion-extension movements . In summary , our findings reveal an overall asymmetry in V1 and V2b in inputs , with flexor-related motor pools receiving proportionately fewer inputs from V2b INs and more from V1 INs , while V2b INs exhibit a bias toward extensor-related motor pools . Multiple methods have been implemented to inactivate genetically defined neurons within a circuit ( Tan et al . , 2006; Luo et al . , 2008; Kim et al . , 2009 ) . These include blocking neurotransmission with tetanus toxin ( Yu et al . , 2004; Zhang et al . , 2008 , 2014 ) and suppressing neuronal excitability with heterologous chloride channels or G-protein-coupled receptors ( GPCRs ) that activate GIRK channels ( Gosgnach et al . , 2006; Tan et al . , 2006; Armbruster et al . , 2007; Ray et al . , 2011 ) . These systems all have drawbacks . Approaches using tetanus toxin are often non-inducible , while ligand-mediated silencing using Gi-coupled receptors can be highly variable and difficult to quantify . By contrast , DTR-mediated neuronal ablation has the advantage of being highly quantifiable , and the timing of cell killing can also be controlled to minimize potential developmental changes due to the chronic silencing or ablation of cells at early developmental times ( Gosgnach et al . , 2006; Crone et al . , 2008; Zhang et al . , 2008 ) . Most importantly , by using an intersectional approach to restrict DTR expression to the caudal CNS , we were able to spare essential functions such as respiration and chewing . With further refinements , this intersectional system can be used to target and ablate discrete populations of neurons , including genetically defined subsets of V1 and V2b INs . Our results demonstrate that V1 INs facilitate the transition from swing to stance , and V2b INs facilitate the transition from stance to swing . In the Xenopus tadpole , V1 ( aIN ) cells promote the transition between the active and inactive phases of the swimming rhythm by providing early phase inhibition to motor neurons ( Li et al . , 2004 ) . The CiA cells , which are homologous to the V1 INs , are likely to function in a similar fashion in zebrafish ( Higashijima et al . , 2004 ) . Interestingly , we see an increase in the incidence of GS ( extensor ) EMG deletions in airstepping mice that lack V1 INs , suggesting the transition from flexion to extension is compromised ( Figure 5 ) . Interestingly , these deletions , and the flexor deletions that arise from the loss of the V2b INs , resemble non-resetting deletions in the cat and turtle ( Grillner and Zangger , 1979; Lafreniere-Roula and McCrea , 2005; McCrea and Rybak , 2008; Stein , 2008 ) . The absence of resetting deletions suggests both populations belong to the pattern forming , rather than rhythm generating , layer of the locomotor CPG . Moreover , the opposing nature of the deletions that occur when V1 vs V2b INs are inactivated suggests that the differential silencing of these two inhibitory cell populations may contribute to the valence of corrective reflexes such as the stumbling corrective reaction ( Forssberg , 1979 ) and crossed extension reflex ( Sherrington , 1910 ) . Our preliminary anatomical analysis of the organization of V1 and V2b inputs to motor neurons ( Figure 8 ) reveals a genetically encoded bias in the innervation of flexor vs extensor motor pools by V1 and V2b INs ( see also Zhang et al . , 2014 ) . Although we were unable to score inhibitory contacts on the distal dendrites of motor neurons , studies showing IaIN and Renshaw cell inhibitory synapses are preferentially located on the soma , and proximal dendrites of motor neurons ( Brown , 1981; Fyffe , 1991 ) suggests that the counts we obtained in this study are likely to be a reasonable measure of the density of V1 and V2b inhibitory synaptic contacts on motor neurons . This in turn suggests that relative differences in the inhibitory drive to motor neurons from the V1 and V2b IN populations may contribute to the opposing actions of V1 and V2b INs on flexor–extensor movements . Our finding that ChR2 activation of V2b INs in the isolated spinal cord selectively suppresses L5 extensor-related motor neuron activity ( Figure 7 ) is consistent with such a model . In considering the differential innervation of motor neurons by V1 and V2b INs , we would like to suggest that there are multiple advantages in having a system where biased rather than segregated inhibitory inputs from V1 and V2b INs determine the valence of flexor–extensor movements . First , biased inputs from the V1 and V2b INs would facilitate the graded activation of synergist and antagonist motor neurons . This graded recruitment of antagonist flexor–extensor motor neurons would be expected to play an important role in modulating limb stiffness and compliance , which is necessary for smooth movements and postural control . Second , the co-innervation of flexor and extensor motor neurons by V1 and V2b INs would enable motor neurons to actively summate and compare inhibitory inputs that are differentially gated by these two inhibitory interneuron populations . There is growing evidence that motor neurons receive a mixture of tonic and dynamic inhibition during locomotion that together with excitatory inputs regulate motor neuron excitability ( Berg et al . , 2007; Johnson et al . , 2012 ) . In particular , the altered membrane conductances produced by concurrent excitation and inhibition are believed to be a fundamental mechanism for changing the gain and dynamic properties of neurons ( Chance et al . , 2002; Mitchell and Silver , 2003; Abbott and Chance , 2005 ) . Consequently , gain modulation and increases in the variability of motor neuron spiking represent an important mechanism for altering the dynamic range of motor neuron activity , so as to produce smooth gradients of muscle force transduction ( Kristan , 2007; Johnson et al . , 2012 ) . Johnson et al . ( 2012 ) have recently shown that inhibitory IaINs are a major source of the tonic inhibitory drive underlying the push–pull control of motor activity in ankle extensor motor neurons . This tonic inhibition , in concert with tonic excitation , causes a net increase in the force modulation produced by ankle extensor muscles . Our observation that V1 and V2b INs are the sole source of Ia inhibition to motor neurons ( Zhang et al . , 2014 ) suggests that they make a major contribution to inhibitory push–pull conductances . As such , the biased innervation of motor neurons by V1 and V2b INs ( Figure 8 ) could facilitate push–pull in limb motor neurons under a range of behavioral conditions . For example , in situations where V2b INs are being dynamically activated by cutaneous feedback , the V1 INs , or subsets thereof , might be tasked with providing tonic inhibition to motor neurons . Our in vivo behavioral analyses showing an inhibitory network comprised of either V1 or V2b INs still generates an alternating pattern of flexor–extensor activity in awake behaving mice , albeit an abnormal one , concurs with our in vitro analysis showing flexor–extensor alternation is only completely degraded when both the V1 and V2b IN populations are inactivated ( Zhang et al . , 2014 ) . These findings argue that the V1 and V2b INs can function in a redundant manner to produce a grossly alternating flexor–extensor locomotor output , and they are in general agreement with the mixed innervation of flexor and extensor motor neurons by V1 and V2b INs ( Figure 8; Zhang et al . , 2014 ) . The co-innervation of flexor and extensor motor neurons by V1 and V2b INs may contribute to the functional recovery that occurs following V1 or V2b IN ablation . This functional recovery is suggestive of a degree of plasticity in the inhibitory control of flexor–extensor movements by both inhibitory populations . Interestingly , the functional deficits that persist after ablating the V1 INs are more pronounced than those found following V2b IN ablation . This might be attributable to the relative abundance of these two cell types in the lumbar spinal cord , where there are twice as many V1 INs as compared to V2b INs ( Zhang et al . , 2014 ) . Differences in the axonal morphology of V1 and V2b INs may also contribute to the persistent hindlimb hyperflexion phenotype V1 IN-ablated mice display , as the motor pools innervating hip and ankle flexors tend to be skewed rostrally in the lumbar spinal cord when compared to their extensor antagonist motor pools ( McHanwell and Biscoe , 1981; Yakovenko et al . , 2002 ) . In this context , it is worth noting that V2b INs predominantly project their axons caudally ( Figure 8—figure supplement 1 ) and may therefore have a limited capacity to substitute for those V1 INs that have rostrally projecting axons . In addition to this , there are fewer inhibitory contacts from V2b INs onto flexor motor neurons ( Figure 8; Zhang et al . , 2014 ) . By contrast , the V2b IN-ablated mice regain a large measure of normal limb movement in the aftermath of V2b IN ablation . In this instance , V1 INs , many of which project caudally ( Figure 8—figure supplement 1 ) , may compensate for the initial loss of V2b IN-derived inhibitory inputs that are proportionately more abundant on extensor motor neurons . The one locomotor deficit that persists when the V2b INs are removed is the delay in the transition from stance to swing during walking ( Figure 6 ) . This transition is controlled in part by Ib pathways from ankle extensor GTOs ( Duysens and Pearson , 1980; Pearson , 2008 ) . Our characterization of presynaptic inputs to V1 and V2b INs showing the V2b INs receive inputs from neurons in the dorsal horn ( F Stam and MG , unpublished findings ) is a strong indication that inhibitory IbINs are derived from the V2b IN population . As such , the presumed loss of inhibitory IbINs that occurs when V2b INs are ablated could account for the delayed transition from stance to swing that we observe ( Pearson , 2008 ) . The loss of IbINs in the V2b IN-ablated mice might also explain why these mice do not show any major change in GS and TA muscle activity during swimming , as Golgi tendon-derived sensory feedback is normally attenuated during this activity ( Akay et al . , 2014 ) . In summary , this study demonstrates that V1 and V2b INs differentially control flexor–extensor motor output , with V1 INs suppressing flexor motor activity during the stance/extension phase of walking to ensure proper extension of the limbs , while the V2b INs suppress extensor activity to facilitate limb extension and ensure the timely transition from stance to swing . Our results are consistent with the V1 and V2b INs contributing to the dynamic control of limb movements during walking via their differential effects on flexor and extensor motor activity . It should be noted that the V1 and V2b IN populations are made up of multiple cell types and the respective contribution that each of these cell types make to flexor–extensor-driven behaviors needs to be assessed . Future efforts to evaluate in depth how these two populations control flexor–extensor motor behaviors will require a better understanding of the molecular , anatomical , and physiological diversity that exists within these two inhibitory IN populations , coupled with a detailed functional characterization of V1 and V2b IN subtypes .	Although there are many different movements an animal can make with its limbs—from reaching to walking—they all basically involve two sets of muscles that act as opposing levers around each joint . ‘Flexor’ muscles contract to bend the limb , and ‘extensor’ muscles contract to extend the limb . When an animal is walking these two sets of muscles contract repeatedly , one after the other . Inhibitory neurons in the spinal cord coordinate these walking movements by preventing the flexor or extensor muscles from contracting at the same time . In 2014 , researchers discovered that two groups of inhibitory neurons , known as the V1 and V2b interneurons , are essential for this alternating pattern of flexing and extending of the limbs of newborn mice . However , these experiments were not able to assess the particular contribution that the V1 and V2b neurons each make to limb movements . Now , Britz et al . —including several of the researchers involved in the 2014 study—have used a sophisticated genetic technique in mice to investigate the role that each group of neurons plays separately . This involved introducing a gene into either the V1 or V2b neurons that makes them susceptible to being killed with the diphtheria toxin . Injecting the mice with diphtheria toxin selectively removed these cells from the regions of the spinal cord that controls hindlimb movements . Britz et al . found that removing either group of neurons prevented the mice from walking normally . Eliminating the V1 neurons caused extreme flexing of the hindlimbs , revealing that the V1 neurons are needed to extend the limb by inhibiting the motor neurons that contract the flexor muscles . In contrast , the loss of V2b neurons caused exaggerated hindlimb extension , indicating that the V2b neurons inhibit the motor neurons that innervate extensor muscles . Both the V1 and V2b groups of neurons contain a wide range of different cell types . Future studies will therefore need to explore how these different cells are involved in coordinating the motions involved in walking .	[ "Abstract", "Introduction", "Results", "Discussion" ]	[ "neuroscience" ]	2015	A genetically defined asymmetry underlies the inhibitory control of flexor–extensor locomotor movements
The status signalling hypothesis aims to explain within-species variation in ornamentation by suggesting that some ornaments signal dominance status . Here , we use multilevel meta-analytic models to challenge the textbook example of this hypothesis , the black bib of male house sparrows ( Passer domesticus ) . We conducted a systematic review , and obtained primary data from published and unpublished studies to test whether dominance rank is positively associated with bib size across studies . Contrary to previous studies , the overall effect size ( i . e . meta-analytic mean ) was small and uncertain . Furthermore , we found several biases in the literature that further question the support available for the status signalling hypothesis . We discuss several explanations including pleiotropic , population- and context-dependent effects . Our findings call for reconsidering this established textbook example in evolutionary and behavioural ecology , and should stimulate renewed interest in understanding within-species variation in ornamental traits . Plumage ornamentation is a striking example of colour and pattern diversity in the animal kingdom and has attracted considerable research ( Hill , 2002 ) . Most studies have focused on sexual selection as the key mechanism to explain this diversity in ornamentation ( Andersson , 1994; Dale et al . , 2015 ) . The status signalling hypothesis explains within-species variation in ornaments by suggesting that these ornaments signal individual dominance status or fighting ability ( Rohwer , 1975 ) . Aggressive contests are costly in terms of energy use , and risk of injuries and predation ( Jakobsson et al . , 1995; Kelly and Godin , 2001; Neat et al . , 1998; Prenter et al . , 2006; Sneddon et al . , 1998 ) . These costs could be reduced if individuals can predict the outcome of such contests beforehand using so-called ‘badges of status’ – that is , two potential competitors could decide whether to avoid or engage in aggressive interactions based on the message provided by their opponent’s signals ( Rohwer , 1975 ) . Patches of ornamentation have been suggested to function as badges of status in a wide range of taxa , including insects ( Tibbetts and Dale , 2004 ) , reptiles ( Whiting et al . , 2003 ) and birds ( Senar , 2006 ) . The status signalling hypothesis was originally proposed to explain variation in the size of mountain sheep horns ( Beninde , 1937; Geist , 1966 ) , but the hypothesis has become increasingly important in the study of variability in plumage ornamentation in birds ( Rohwer , 1975; Senar , 2006 ) . Among the many bird species studied ( Santos et al . , 2011 ) , the house sparrow ( Passer domesticus ) has become the classic textbook example of status signalling ( Andersson , 1994; Searcy and Nowicki , 2005; Senar , 2006; Davies et al . , 2012 ) . The house sparrow is a sexually dimorphic passerine , in which the main difference between the sexes is a prominent black patch on the male’s throat and chest ( hereafter ‘bib’ ) . Many studies have suggested that bib size serves as a badge of status , but most studies are based on limited sample sizes , and have used inconsistent methodologies for measuring bib and dominance status ( Nakagawa and Cuthill , 2007; Santos et al . , 2011 ) . Meta-analysis is a powerful tool to quantitatively test the overall ( across-study ) effect size ( i . e . the ‘meta-analytic mean’ ) for a specific hypothesis . Meta-analyses are therefore able to provide more robust conclusions than single studies and are increasingly used in evolutionary ecology ( Gurevitch et al . , 2018; Nakagawa and Poulin , 2012a; Nakagawa and Santos , 2012b; Senior et al . , 2016 ) . Traditional meta-analyses combine summary data across different studies , where design and methodology are study-specific ( e . g . effect sizes among studies are typically adjusted for different fixed effects ) . These differences among studies are expected to increase heterogeneity , and therefore , the uncertainty of the meta-analytic mean ( Mengersen et al . , 2013 ) . Meta-analysis of primary or raw data is a specific type of meta-analysis where studies are analysed in a consistent manner ( Mengersen et al . , 2013 ) . This type of meta-analysis allows methodology to be standardized so that comparable effect sizes can be obtained across studies and is , therefore , considered the gold standard in disciplines such as medicine ( Simmonds et al . , 2005 ) . Unfortunately , meta-analysis of primary data is still rarely used in evolutionary ecology ( but see Barrowman et al . , 2003; Richards and Bass , 2005; Krasnov et al . , 2009 ) , perhaps due to the difficulty of obtaining the primary data of previously published studies until recently ( Culina et al . , 2018; Schmid et al . , 2003 ) . An important feature of any meta-analysis is to identify the existence of bias in the literature ( Nakagawa and Santos , 2012b; Jennions et al . , 2013 ) . For example , publication bias occurs whenever particular effect sizes ( e . g . larger ones ) are more likely found in the literature than others ( e . g . smaller ones ) . This tends to be the case when statistical significance and/or direction of effect sizes determines whether results were submitted or accepted for publication ( Jennions et al . , 2013 ) . Thus , publication bias can strongly affect the estimation of the meta-analytic mean , and distort the interpretation of the hypothesis ( Rothstein et al . , 2005 ) . Several methods have been developed to identify this and other biases ( Nakagawa and Santos , 2012b; Jennions et al . , 2013 ) ; however , such methods are imperfect and dependent on the number of effect sizes available , and therefore should be considered as types of sensitivity analysis ( Nakagawa et al . , 2017; Nakagawa and Santos , 2012b ) . Here , we meta-analytically assessed the textbook example of the status signalling hypothesis in the house sparrow . Specifically , we combined summary and primary data from published and unpublished studies to test the prediction that dominance rank is positively associated with bib size across studies . We found that the meta-analytic mean was small , uncertain and overlapped zero . Hence , our results challenge the status signalling function of the male house sparrow’s bib . Also , we identified several biases in the published literature . Finally , we discuss potential biological explanations for our results , and provide advice for future studies testing the status signalling hypothesis . Mean sampling effort was 36 interactions/individual ( SD = 24 ) , which highlights that , overall , dominance hierarchies were inferred reliably across groups ( Sánchez-Tójar et al . , 2018b ) . The mean Elo-rating repeatability was 0 . 92 ( SD = 0 . 07 ) and the mean triangle transitivity was 0 . 63 ( SD = 0 . 28 ) . Thus , the dominance hierarchies observed across groups of house sparrows were medium in both steepness and transitivity . Our meta-analyses revealed a small overall effect size with large 95% credible intervals that overlapped zero ( Table 2; Figure 1 ) . Additionally , the overall heterogeneity ( I2overall ) was moderate ( 53%; Table 2 ) . Thus , our results suggested that generally , bib size is at best a weak and unreliable signal of dominance status in male house sparrows . None of the three biological moderators studied ( season , group composition and type of interactions ) explained differences among studies ( Table 3 ) . Sampling effort ( i . e . the ratio of interactions to individuals recorded ) also was not an important moderator ( Table 3 ) . There was no clear asymmetry in the funnel plots ( Figure 2 ) . Also , Egger’s regression tests did not show evidence of funnel plot asymmetry in any of the meta-analyses ( Table 2 ) . However , published effect sizes were larger than unpublished ones , and the latter were not different from zero ( Table 4; Figure 3 ) . Additionally , we found that the overall effect size decreased over time and approached zero ( Table 4; Figure 4 ) . The male house sparrow’s bib is not the strong across-study predictor of dominance status once believed . In contrast to the medium-to-large effect found in the previous meta-analysis ( Nakagawa et al . , 2007 ) , our updated meta-analytic mean was small , uncertain and overlapped zero . Thus , the male house sparrow’s bib should not be unambiguously considered or called a badge of status . Furthermore , we found evidence for the existence of bias in the published literature that further undermines the validity of the available support for the status signalling hypothesis . First , the meta-analytic mean of unpublished studies was essentially zero , compared to the medium effect size detected in published studies . Second , we found that the effect size estimated in published studies has been decreasing over time , and recently published effects were on average no longer distinguishable from zero . Our findings call for reconsidering this textbook example in evolutionary and behavioural ecology , and should stimulate renewed attention to hypotheses explaining within-species variation in ornamentation . The status signalling hypothesis ( Rohwer , 1975 ) has been extensively tested to try and explain within-species trait variation ( e . g . reptiles: Whiting et al . , 2003; insects: Tibbetts and Dale , 2004; humans: Dixson and Vasey , 2012 ) , particularly plumage variation ( Santos et al . , 2011 ) . Soon after the first empirical tests on birds , the black bib of male house sparrows became a textbook example of the status signalling hypothesis ( Andersson , 1994; Searcy and Nowicki , 2005; Senar , 2006; Davies et al . , 2012 ) , an idea that was later confirmed meta-analytically ( Nakagawa et al . , 2007 ) . However , Nakagawa et al . , 2007 meta-analytic mean was over-estimated because only nine low-powered studies were available ( more in Button et al . , 2013 ) . Here , we updated that meta-analysis with newly published and unpublished data . Our results showed that the overall effect size is much smaller and much more uncertain than previously thought . The status signalling hypothesis is thus no longer a compelling explanation for the evolution of bib size across populations of house sparrows . Similar contradicting conclusions have been reported for other model species . An exhaustive review and meta-analysis on plumage coloration of blue tits ( Cyanistes caeruleus ) revealed that , after dozens of publications studying the function of plumage ornamentation in this species , the only robust conclusion is that females’ plumage differs from that of males ( Parker , 2013 ) . Another example is the long-believed effect of leg bands of particular colours on the perceived attractiveness of male zebra finches ( Taeniopygia guttata ) , which has been also experimentally and meta-analytically refuted ( Seguin and Forstmeier , 2012; Wang et al . , 2018 ) . Finally , the existence of a badge of status in a non-bird model species , the paper wasp ( Polistes dominulus; Tibbetts and Dale , 2004 ) has also been challenged multiple times ( e . g . Cervo et al . , 2008; Green and Field , 2011; Green et al . , 2013 ) , generating doubts about its generality . Our findings corroborate studies showing that abundant replication is needed before any strong or general conclusion can be drawn ( Aarts et al . , 2015 ) , and highlight the existence of important impediments ( e . g . publication bias ) to scientific progress in evolutionary ecology ( Forstmeier et al . , 2017; Fraser et al . , 2018 ) . Indeed , our results showed that the published literature on status signalling in house sparrows is likely a biased subsample . The main evidence for this is that the mean effect size of unpublished studies was essentially zero and clearly different from the mean effect size based on published studies , which was of medium size . Furthermore , this moderator ( i . e . unpublished vs . published ) explained a large percentage of the model’s variance . In some of our own unpublished datasets , the relationship between dominance rank and bib size was never formally tested ( D . F . Westneat and V . Bókony , personal communication , February , 2018 ) , that is , our unpublished datasets are not all examples of the ‘file drawer problem’ ( sensu Rosenthal , 1979 ) . Egger’s regression tests failed to detect any funnel plot asymmetry , even in the meta-analyses based on published effect sizes only ( Appendix 2—table 1 ) . However , because unpublished data indeed existed ( i . e . those obtained for this study ) , the detection failure was likely the consequence of the limited number of effect sizes available ( i . e . low power ) and the moderate level of heterogeneity found in this study ( Moreno et al . , 2009; Sterne and Egger , 2005 ) . An additional type of publication bias is time-lag bias , where early studies report larger effect sizes than later studies ( Trikalinos and Ioannidis , 2005 ) . We detected evidence for such bias because the correlation between dominance rank and bib size in published studies has decreased over time and approached zero . Year of publication explained a large percentage of the model's variance , and accounting for year of publication resulted in a strong reduction of the mean effect size across published studies ( Table 4 vs . Appendix 2—table 1 ) . Time-lag bias has been detected in other ecological studies ( Poulin , 2000 ) ; Jennions and Moller , 2002b ) , including a meta-analysis on status signalling across bird species ( Santos et al . , 2011 ) . In the latter study , a positive overall ( across-species ) effect size persisted regardless of the time-lag bias , and no strong evidence for other types of biases was found ( Santos et al . , 2011 ) . However , Santos et al . , 2011 did not attempt to analyse unpublished data , so additional evidence is needed to determine the effect that unpublished data may have on the overall validity of the status signalling hypothesis across bird species . If effect sizes based on unpublished data for other species were of similar magnitude to those obtained for house sparrows , the validity of the status signalling hypothesis across species would need reconsideration . The existence of publication bias in ecology has long been recognized ( Cassey et al . , 2004; Jennions and Moller , 2002b; Palmer , 2000 ) . Publication bias leads to false conclusions if not accounted for ( Rothstein et al . , 2005 ) , and is , thus , a serious impediment to scientific progress . In addition to estimating the overall effect size for a hypothesis , meta-analyses are also used to assess heterogeneity among estimates ( Higgins and Thompson , 2002; Higgins et al . , 2003 ) . Understanding the sources of heterogeneity is an important step towards the correct interpretation of a meta-analytic mean , and can be done using meta-regressions ( Nakagawa and Santos , 2012b ) . Here , we found that the percentage of variance that was not attributable to sampling error ( i . e . heterogeneity ) was moderate . This value is below the average calculated across ecological and evolutionary meta-analyses ( Senior et al . , 2016 ) , and indicates that we accounted for large differences among estimates . Our meta-regressions based on biological moderators explained 20–23% of the variance ( Table 3 ) . However , none of the biological moderators that we tested strongly influenced the overall effect size , possibly because of limited sample sizes . The badge of status idea is more complex than typically portrayed ( reviewed by Diep and Westneat , 2013 ) . Badges of status are expected to be particularly important in large and unstable groups of individuals where individual recognition would otherwise be difficult ( Rohwer , 1975 ) . While the evolution of badges of status in New and Old World sparrows has been related to sociality ( i . e . flocking ) during the non-breeding season ( Tibbetts and Safran , 2009 ) , additional factors need to be involved if the signal is to function in reducing aggression but retaining honesty ( Diep and Westneat , 2013 ) . Our results , however , did not show any evidence for a season-dependent effect as the moderator ‘season’ ( breeding vs . non-breeding ) was not a strong predictor in our models . Badges of status are expected to function both within and between sexes ( Rohwer , 1975; Senar , 2006 ) . Indeed , we found little evidence that the status signalling function of bib size differed between male-only and mixed-sex flocks . Interestingly , when competing for resources , possessing a badge of status would be beneficial for both males and females . However , male but not female house sparrows have a bib . This sexual dimorphism suggests that the bib’s function is likely more important when competing for resources other than essential , a priori non-sex-specific resources such as food , water , sand baths and roosting sites . Møller , 1988 and Pape Moller , 1989 reported that female house sparrows preferentially choose males with large bibs ( but see Kimball , 1996 ) , and bib size has been positively correlated with sexual behaviour ( Veiga , 1996; Møller , 1990 ) , which suggests that the bib may play a role in mate choice . Furthermore , the original status signalling hypothesis posits that the main benefit of using badges of status would be to avoid fights , which should be particularly important when interacting with unfamiliar individuals ( Rohwer , 1975; Senar , 2006 ) . Although we did not have data to test whether unfamiliarity among contestants is an important pre-requisite for the status signalling hypothesis , we found no change in mean effect size when only obviously aggressive interactions were studied . In practice , testing whether the bib is important in mediating aggression among unfamiliar individuals is difficult because the certainty of the estimates of individual dominance increases over time as more contests are recorded , but so does familiarity among contestants . There are some additional explanations for the small and uncertain effect detected by our meta-analyses . First , different populations might be under different selective pressures regarding status signalling . Indeed , the population-specific heterogeneity ( I2population ID ) estimated in our meta-analyses was 15–16% , suggesting that population-dependent effects might exist . Second , although none of the moderators had a strong influence on the overall effect size , the study-specific heterogeneity estimated in our meta-analyses ( I2study ID = 20–21% ) suggests that the uncertainty observed could still be explained by the status signal being context-dependent . However , context-dependence is often invoked post hoc to explain variation among studies , but strong evidence for it is lacking in most cases . Last , most studies testing the status signalling hypothesis in house sparrows are observational ( Table 1 ) , and the only two experimental studies conducted so far were inconclusive ( Diep , 2012; Gonzalez et al . , 2002 ) . Thus , it cannot be ruled out that the weak correlation observed between dominance status and bib size is driven by a third , unknown variable . In this respect , it has been proposed that the association between melanin-based coloration ( such as the bib; e . g . Galván et al . , 2015; Galván and Alonso-Alvarez , 2017 ) and aggression is due to pleiotropic effects of the genes involved in regulating the synthesis of melanin ( reviewed by Ducrest et al . , 2008 ) . Furthermore , bib size has been shown to correlate with testosterone , a hormone often involved in aggressive behaviour ( Gonzalez et al . , 2001 ) but this relationship has not been consistently observed ( Laucht et al . , 2010 ) . Future studies should shift the focus towards understanding the function of bib size in wild populations and increase considerably the number of birds studied per group . The latter is essential because the statistical power of published tests of the status signalling hypothesis in house sparrows is alarmingly low ( power = 8 . 5% for r = 0 . 20 , Appendix 3 ) and lower than the average in behavioural ecology ( Jennions , 2003 ) . Our analyses have several potential limitations . First , although the number of studies included in this meta-analysis is more than double that of the previous meta-analysis ( Nakagawa et al . , 2007 ) , it is still limited . Also , it is likely ( see above ) that additional unpublished data are stored in ‘file drawers’ ( sensu Rosenthal , 1979 ) . Second , most tests included in this study were still low-powered in terms of group size ( median = 6 individuals/estimate , range = 4–41 ) , and the sample size is inflated because some of the published studies pooled individuals from different groups ( Figure 4 ) . Third , although our results showed little evidence of an effect of sampling effort on the overall effect size , the quality of the data on dominance and bib size may still be a potential factor explaining differences among studies . Fourth , experiments will normally yield larger effect sizes than observational studies because effects of confounding factors can be reduced ( Palmer , 2000 ) . Nonetheless , our systematic review only identified two studies where the status signalling hypothesis was tested experimentally in house sparrows ( Gonzalez et al . , 2002; Diep , 2012 ) , preventing us from estimating the meta-analytic mean for experimental studies . Note , however , that the results of those experiments were inconclusive , and potentially affected by regression to the mean ( Forstmeier et al . , 2017 ) . In conclusion , our results challenge an established textbook example of the status signalling hypothesis , which aims to explain within-species variation in ornament size . In house sparrows , we find no evidence that bib size consistently acts as a badge of status across studies and populations , and thus , bib size can no longer be considered a textbook example of the status signalling hypothesis . Furthermore , our analyses highlight the existence of publication biases in the literature , further undermining the validity of past conclusions . Bias against the publication of small ( ‘non-significant’ ) effects hinders scientific progress . We thus join the call for a change in incentives and scientific culture in ecology and evolution ( Forstmeier et al . , 2017; Ihle et al . , 2017; Nakagawa and Parker , 2015; Parker et al . , 2016 ) . We used several approaches to maximize the identification of relevant studies . First , we included all studies reported in a previous meta-analysis that tested the relationship between dominance rank and bib size in house sparrows ( Nakagawa et al . , 2007 ) . Second , we conducted a keyword search on Web of Science , PubMed and Scopus from 2006 to June 2017 to find studies published after Nakagawa et al . , 2007 , using the combination of keywords [‘bib/badge’ , ‘sparrow’ , ‘dominance/status/fighting’] . Third , we screened all studies on house sparrows used in a meta-analysis that tested the relationship between dominance and plumage ornamentation across species ( Santos et al . , 2011 ) to identify additional studies that we may have missed in our keyword search . We screened titles and abstracts of all articles and removed the irrelevant articles before examining the full texts ( Supplementary file 1 ) . We followed the preferred reporting items for systematic reviews and meta-analyses ( PRISMA: Moher et al . , 2009 ) ; see ‘Reporting Standards Documents’ ) . We only included articles in which dominance was directly inferred from agonistic dyadic interactions over resources such as food , water , sand baths or roosting sites ( Appendix 1—table 1 ) . Some studies had more than one effect size estimate per group of birds studied . When the presence of multiple estimates was due to the use of different statistical analyses on the same data , we chose a single estimate based on the following order of preference: ( 1 ) direct reports of effect size per group of birds studied ( e . g . correlation coefficient ) , ( 2 ) inferential statistics ( e . g . t , F and χ2 statistics ) from analyses where group ID was accounted for and no other fixed effects were included , ( 3 ) direct reports of effect size where individuals from different groups where pooled together , ( 4 ) inferential statistics from models including other fixed effects . When the presence of multiple estimates was due to the use of different methods to estimate bib size and dominance rank on the same data , we chose a single estimate per group of birds or study based on the order of preference shown in Appendix 1—tables 1–3 . In each case , the order of preference was determined prior to conducting any statistical analysis , and thus , method selection was blind to the outcome of the analyses ( more details in Appendix 1 ) . We requested primary data ( i . e . agonistic dyadic interactions and bib size measures ) of all relevant studies identified by our systematic review . Additionally , we asked authors to share , if available , any unpublished data that could be used to test the relationship between dominance rank and bib size in house sparrows . We emailed the corresponding author , but if no reply was received , we tried contacting all the other authors listed . One study ( Møller , 1987 ) provided all primary data in the original publication and , therefore , its author was not contacted . Last , we included our own unpublished data ( Appendix 1—table 5 ) . Most studies recorded data from more than one group of birds ( Table 1 ) . For each primary dataset obtained , we inferred the dominance hierarchy of each group of birds from the observed agonistic dyadic interactions ( wins and losses ) among individuals using the randomized Elo-rating method , which estimates dominance hierarchies more precisely than other methods ( Sánchez-Tójar et al . , 2018b ) . We then used the provided measures of individual bib size ( e . g . area outlined from pictures ) or , if possible , calculated bib area from length and width measures following ( Møller , 1987 ) . Subsequently , we estimated the Spearman’s rho rank correlation ( ρ ) between individual rank and bib size for each group of birds . For one study ( Buchanan et al . , 2010 ) , we received the already inferred dominance hierarchies for each group of birds , which we then correlated with bib size to obtain ρ . Regardless of their source ( primary or summary data ) , we transformed all estimates ( e . g . ρ , F statistics , etc ) into Pearson’s correlation coefficients ( r ) , and then into standardized effect sizes using Fisher’s transformation ( Zr ) for among-study comparison . We used the equations from Nakagawa et al . , 2007 and Lajeunesse , 2013 . Since log ( 0 ) is undefined , r values equal to 1 . 00 and −1 . 00 were transformed to 0 . 975 and −0 . 975 , respectively , before calculating Zr . Zr values of 0 . 100 , 0 . 310 and 0 . 549 were considered small , medium and large effect sizes , respectively ( equivalent benchmarks from Cohen , 1988 ) . When not reported directly , the number of individuals ( n ) was estimated from the degrees of freedom . The variance in Zr was calculated as: VZr = 1/ ( n-3 ) . Estimates ( k ) based on less than four individuals were discarded ( k = 33 estimates discarded ) . We ran two multilevel meta-analyses to test whether dominance rank and bib size were positively correlated across studies . The first meta-analysis , in other words ‘meta 1’ , included published and unpublished ( re- ) analysed effect sizes ( i . e . effect sizes estimated from the studies we obtained primary data from ) , plus the remaining published effect sizes obtained from summary data ( i . e . effect sizes for which primary data were unavailable ) . The second meta-analysis , in other words ‘meta 2’ , tested the robustness of the results of meta 1 to the inclusion of non-reported estimates from studies that reported ‘statistically non-significant’ results without showing either the magnitude or the direction of the estimates ( Table 1 ) . Receipt of primary data allowed us to recover some but not all the originally non-reported estimates . Two ‘non-significant’ estimates were still missing . Thus , meta 2 was like meta 1 but included the two non-significant non-reported estimates , which were assumed to be zero ( see Booksmythe et al . , 2017 for a similar approach ) . Note that non-significant estimates can be either negative or positive , and thus , assuming that they were zero may have either underestimated or overestimated them , something we cannot know from non-reported estimates . Meta-analyses based on published studies only are shown in Appendix 2 . We investigated inconsistency across studies by estimating the heterogeneity ( I2 ) from our meta-analyses following Nakagawa and Santos , 2012b . I2 values around 25 , 50% and 75% are considered as low , moderate and high levels of heterogeneity , respectively ( Higgins et al . , 2003 ) . We tested if season , group composition and/or the type of interactions recorded had an effect on the meta-analytic mean . For that , we ran two multilevel meta-regressions that included the following moderators ( hereafter ‘biological moderators’ ) : ( 1 ) ‘season’ , referring to whether the study was conducted during the non-breeding ( September-February ) or the breeding season ( March-August ) ; ( 2 ) ‘group composition’ , referring to whether birds were kept in male-only or in mixed-sex groups; and , ( 3 ) ‘type of interactions’ , referring to whether the dyadic interactions recorded were only aggressive ( e . g . threats and pecks ) , or also included interactions that were not obviously aggressive ( e . g . displacements ) . Because only three of 19 studies were conducted in the wild ( k = 12 estimates; Table 1 ) , we did not include a moderator testing for captive versus wild environments . The three biological moderators were mean-centred following Schielzeth , 2010 to aid interpretation . The ratio of agonistic dyadic interactions recorded to the total number of interacting individuals observed ( hereafter ‘sampling effort’ ) is a measure of sampling effort that correlates positively and logarithmically with the ability to infer the latent dominance hierarchy ( Sánchez-Tójar et al . , 2018b ) . The higher this ratio , the more precisely the latent hierarchy can be inferred ( Sánchez-Tójar et al . , 2018b ) . For the subset of studies for which the primary data of the agonistic dyadic interactions were available ( 12 out of 19 studies; Table 1 ) , we ran a multilevel meta-regression including sampling effort and its squared term as z-transformed moderators ( Schielzeth , 2010 ) . The squared term was included because of the observed logarithmic relationship between sampling effort and the method’s performance ( Sánchez-Tójar et al . , 2018b ) . This meta-regression tested whether sampling effort had an effect on the meta-analytic mean: ( i ) a positive estimate would indicate that the meta-analytic mean may have been affected by the inclusion of studies with unreliable estimates of dominance rank . In contrast , ( ii ) a negative estimate would indicate that effect sizes were larger when based on unreliable estimates of dominance rank and hence provide evidence for the existence of publication bias . For all meta-regressions , we estimated the percentage of variance explained by the moderators ( R2marginal ) following ( Nakagawa and Schielzeth , 2013 ) . All meta-analyses and meta-regressions included the two random effects ‘population ID’ and ‘study ID’ . Population ID was related to the geographical location of the population of birds studied . We used Google maps to estimate the distance over land ( i . e . avoiding large water bodies ) among populations , and assumed the same population ID when the distance was below 50 km ( 13 populations; Table 1 ) . Study ID encompassed those estimates obtained within each specific study ( 19 studies ) . Two studies tested the prediction twice for the same groups of birds ( Table 1 ) and , within each population , some individuals may have been sampled more than once . However , we could not include ‘group ID’ and/or ‘individual ID’ as additional random effects due to either limited sample size or because the relevant data were not available . For the meta-analyses , we assessed publication bias using two methods that are based on the assumption that funnel plots should be symmetrical . First , we visually inspected asymmetry in funnel plots of meta-analytic residuals against the inverse of their precision ( defined as the square root of the inverse of VZr ) for each meta-analysis . Funnel plots based on meta-analytic residuals ( the sum of effect-size-level effects and sampling-variance effects ) are more appropriate than those based on effect sizes when multilevel models are used ( Nakagawa and Santos , 2012b ) . Second , we ran Egger’s regressions using the meta-analytic residuals as the response variable , and the precision ( see above ) as the moderator ( Nakagawa and Santos , 2012b ) for each meta-analysis . If the intercept of such a regression does not overlap zero , estimates from the opposite direction to the meta-analytic mean might be missing and hence we consider this evidence of publication bias ( Nakagawa and Santos , 2012b ) . Further , we tested whether published estimates differed from unpublished estimates . For that , we ran a multilevel meta-regression that included population ID and study ID as random effects , and ‘unpublished’ ( two levels: yes ( 0 ) , no ( 1 ) ) as a moderator . This meta-regression was based on meta 1 ( i . e . it did not include the two non-reported estimates ) . We did not use the trim-and-fill method ( Duval and Tweedie , 2000a; Duval and Tweedie , 2000b ) because this method has been advised against when significant heterogeneity is present ( Moreno et al . , 2009; Jennions et al . , 2013 ) , as it was the case in our meta-analyses ( see section 'Results’ ) . Finally , we analysed temporal trends in effect sizes that could indicate ‘time-lag bias’ . Time-lag bias is common in the literature ( Jennions and Moller , 2002b; Poulin , 2000 ) , and occurs when the effect sizes of a specific hypothesis are negatively correlated with publication date ( i . e . effect sizes decrease over time; Trikalinos and Ioannidis , 2005 ) . A decrease in effect size over time can have multiple causes . For example , initial effect sizes might be inflated due to low statistical power ( ‘winner’s curse’ ) but published more easily and/or earlier due to positive selection of statistically significant results ( reviewed by Koricheva et al . , 2013 ) . We ran a multilevel meta-regression based on published effect sizes only , where ‘year of publication’ was included as a z-transformed moderator ( Nakagawa and Santos , 2012b ) . All analyses were run in R v . 3 . 4 . 0 ( R Core Team , 2017 ) . We inferred individual dominance ranks from agonistic dyadic interactions using the randomized Elo-rating method from the R package ‘aniDom’ v . 0 . 1 . 3 ( Farine and Sánchez-Tójar , 2017; Sánchez-Tójar et al . , 2018b ) . Additionally , we described the dominance hierarchies observed in the groups of house sparrows for which primary data was available . For that we estimated the uncertainty of the dominance hierarchies using the R package ‘aniDom’ v . 0 . 1 . 3 ( Farine and Sánchez-Tójar , 2017; Sánchez-Tójar et al . , 2018b ) and the triangle transitivity ( McDonald and Shizuka , 2013 ) using the R package ‘compete’ 3 . 1 . 0 ( Curley , 2016 ) . We used the R package ‘MCMCglmm’ v . 2 . 24 ( Hadfield , 2010 ) to run the multilevel meta-analytic ( meta-regression ) models ( Hadfield and Nakagawa , 2010 ) . For each meta-analysis and meta-regression , we ran three independent MCMC chains for 2 million iterations ( thinning = 1 , 800 , burn-in = 200 , 000 ) using inverse-Gamma priors ( V = 1 , nu = 0 . 002 ) . Model chains were checked for convergence and mixing using the Gelman-Rubin statistic . The auto-correlation within the chains was <0 . 1 in all cases . For each meta-analysis and meta-regression , we chose the model with the lowest DIC value to extract the posterior mean and its 95% highest posterior density intervals ( hereafter 95% credible interval ) . We report all data exclusion criteria applied and the results of all analyses conducted in our study . We provide all of the R code and data used for our analyses ( Sánchez-Tójar et al . , 2018a ) .	Many bird species have colourful , intricately patterned plumage . This ornamentation is generally believed to exist to attract partners . In the 1970s , however , scientists proposed an alternative idea , called the ‘status signalling hypothesis’ . This suggests that some birds have plumage ornaments that indicate the fighting abilities or dominance status of their bearers , much like the military badges worn by humans . These badges of status might evolve because fights , which commonly determine who gets valuable resources such as food , are a risky business . Individuals would greatly benefit from being able to predict the fighting abilities of any potential competitor and so avoid fights that they will probably lose . Male house sparrows have a black patch on their throat , known as the bib , that has been considered to be a textbook demonstration of the status signalling hypothesis . However , most of the studies that support this idea studied small numbers of birds and used inconsistent methods . Furthermore , some recent studies have failed to replicate previous findings . Sánchez-Tójar et al . collected data from several house sparrow populations across the world and systematically scrutinized the published literature to find all of the studies that tested the status signalling hypothesis in house sparrows . This revealed only weak evidence that the bib of male house sparrows signals the fighting abilities of its bearer . Instead , the published literature is a biased subsample; failures to replicate the hypothesis likely remain unpublished . Currently , failures to replicate previous findings are generally deemed uninteresting , and so are not often published . By demonstrating the need to replicate findings robustly to avoid biasing conclusions , Sánchez-Tójar et al . thus join the call for a change in incentives and scientific culture .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2018	Meta-analysis challenges a textbook example of status signalling and demonstrates publication bias
Genetic screens are powerful tools for the functional annotation of genomes . In the context of multicellular organisms , interrogation of gene function is greatly facilitated by methods that allow spatial and temporal control of gene abrogation . Here , we describe a large-scale transgenic short guide ( sg ) RNA library for efficient CRISPR-based disruption of specific target genes in a constitutive or conditional manner . The library consists currently of more than 2600 plasmids and 1700 fly lines with a focus on targeting kinases , phosphatases and transcription factors , each expressing two sgRNAs under control of the Gal4/UAS system . We show that conditional CRISPR mutagenesis is robust across many target genes and can be efficiently employed in various somatic tissues , as well as the germline . In order to prevent artefacts commonly associated with excessive amounts of Cas9 protein , we have developed a series of novel UAS-Cas9 transgenes , which allow fine tuning of Cas9 expression to achieve high gene editing activity without detectable toxicity . Functional assays , as well as direct sequencing of genomic sgRNA target sites , indicates that the vast majority of transgenic sgRNA lines mediate efficient gene disruption . Furthermore , we conducted the so far largest fully transgenic CRISPR screen in any metazoan organism , which further supported the high efficiency and accuracy of our library and revealed many so far uncharacterized genes essential for development . The functional annotation of the genome is a prerequisite to gain a deeper understanding of the molecular and cellular mechanisms that underpin development , homeostasis and disease of multicellular organisms . Drosophila melanogaster has provided many fundamental insights into metazoan biology , in particular in the form of systematic gene discovery through genetic screens . Forward genetic screens utilize random mutagenesis to introduce novel genetic variants , but are limited by the large number of individuals required to probe many or all genetic loci and difficulties in identifying causal variants . In contrast , reverse genetic approaches , such as RNA interference ( RNAi ) , are gene-centric designed and allow to probe the function of a large number of genes ( Boutros and Ahringer , 2008; Heigwer et al . , 2018; Horn et al . , 2011; Mohr et al . , 2014 ) . In addition , RNAi reagents can be genetically encoded and used to screen for gene function with spatial and temporal precision ( Dietzl et al . , 2007; Kaya-Çopur and Schnorrer , 2016; Ni et al . , 2009 ) . However , RNAi is often limited by incomplete penetrance due to residual gene expression and can suffer from off-target effects ( Echeverri et al . , 2006; Ma et al . , 2006; Perkins et al . , 2015 ) . While genetic screens have contributed enormously to our understanding of gene function , large parts of eukaryotic genomes remain not or only poorly characterized ( Brown et al . , 2009; Dickinson et al . , 2016; White et al . , 2013 ) . For example , in Drosophila only 20% of genes have associated mutant alleles ( Kaufman , 2017 ) . Therefore , there exists an urgent need to develop innovative approaches to gain a more complete understanding of the functions encoded by the various elements of the genome . Clustered Regularly Interspersed Short Palindromic Repeats ( CRISPR ) - CRISPR-associated ( Cas ) systems are adaptive prokaryotic immune systems that have been adopted for genome engineering applications ( Doudna and Charpentier , 2014; Wang et al . , 2016 ) . Cas9 complexed with a single chimeric guide RNA ( sgRNA ) mediates site-specific DNA double strand breaks and subsequent DNA repair can result in small insertions and deletions ( indels ) at the break point . However , not all Cas9-mediated indel mutations abrogate gene function . To compensate for that , strategies have been developed to introduce simultaneously several mutations in the same gene . The efficiency of such multiplexing strategies has been demonstrated in flies , mice , fish and plants , and several sgRNAs are often required to generate bi-allelic loss-of function mutations in all cells ( Port and Bullock , 2016; Xie et al . , 2015; Yin et al . , 2015 ) . Furthermore , to gain a comprehensive understanding of the often multifaceted functions genetic elements have in multicellular organisms requires methods that enable spatial or temporal control of gene disruption . To restrict CRISPR mutagenesis to defined cells , tissues or developmental stages , specific regulatory regions are commonly employed to drive Cas9 expression . However , Cas9 expression vectors with tissue-specific enhancers often display ‘leaky’ Cas9 expression in other tissues and poor control of CRISPR mutagenesis has been observed in multiple systems , including flies , mice and patient derived xenografts ( Chen et al . , 2017; Dow et al . , 2015; Hulton et al . , 2019; Port and Bullock , 2016 ) . It has recently been demonstrated that expressing both Cas9 and sgRNA from conditional regulatory elements can result in tightly controlled genome editing ( Port and Bullock , 2016 ) , but the robustness of such a strategy across many genomic target sites has so far not been explored . Here , we describe a large-scale resource for spatially restricted mutagenesis in Drosophila . The system mediates robust mutagenesis across target genes , giving rise to a large fraction of cells containing gene knock-outs and displays tight spatial and temporal control . We developed a series of tunable Cas9 lines that allow gene editing with high efficiency and low toxicity independent of enhancer strength . These can be used with a growing library of sgRNA transgenes , which currently comprise over 1700 Drosophila strains , for systematic mutagenesis in any somatic tissue or the germline . Furthermore , we present the first large-scale transgenic CRISPR screen using this resource , which confirms its high efficiency and specificity and reveals multiple uncharacterized genes with essential , but unknown function . We set out to develop a large-scale resource that would allow systematic CRISPR-mediated gene disruption with tight spatial and temporal control ( Figure 1A ) . In Drosophila , tissue-specific expression of transgenes is most commonly performed via the binary Gal4/UAS system ( Brand and Perrimon , 1993 ) and thousands of Gal4 lines with specific temporal and spatial expression patterns are publicly available . To harness this resource for tissue-specific CRISPR mutagenesis we aimed to utilize UAS-Cas9 transgenes and combine them with the sgRNA expression vector pCFD6 , which enables Gal4-dependent expression of sgRNA arrays . We have previously shown that conditional expression of both Cas9 and sgRNAs is necessary to achieve tight control of mutagenesis ( Figure 1B; Port and Bullock , 2016 ) . Since this previous proof-of principle study was restricted to testing pCFD6 with two sgRNAs targeting the Wnt secretion factor Evenness interrupted ( Evi , also known as Wntless or Sprinter; Bänziger et al . , 2006; Bartscherer et al . , 2006; Port and Bullock , 2016 ) , we first tested whether this system is robust across target genes and tissues , a prerequisite to generate large-scale libraries of sgRNA strains targeting many or all Drosophila genes . To this end , we created various transgenic fly lines harbouring a pCFD6 transgene encoding two sgRNAs targeting a single gene at two independent positions . These were crossed to flies containing a UAS-cas9 . P2 transgene and a tissue-specific Gal4 driver . We then analysed if mutations were efficiently induced , restricted to the appropriate cells and caused the expected phenotypes . We observed efficient and specific gene disruption in wing imaginal discs with pCFD6 sgRNA transgenes targeting the Drosophila beta-Catenin homolog armadillo ( arm , Figure 1C ) , as well as the transcription factor senseless ( sens ) or the transmembrane protein smoothened ( smo ) ( Figure 1—figure supplement 1A , B ) . To test tissue-specific CRISPR mutagenesis in a different tissue context , we targeted Notch ( N ) in the Drosophila midgut , which is derived from the endoderm . We observed a strong increase in stem cell proliferation and an accumulation of cells with small nuclei , which matches the described phenotype of N mutant clones in the midgut ( Ohlstein and Spradling , 2006; Figure 1D and Figure 1—figure supplement 2 ) . Interestingly , we observed a qualitative difference between perturbation of N expression by RNAi , which only induces hyperplasia in female flies ( Figure 1—figure supplement 2; Hudry et al . , 2016; Siudeja et al . , 2015 ) , and N mutagenesis by CRISPR , which induces strong overgrowth in both male and female midguts ( Figure 1—figure supplement 2 ) . We also tested conditional mutagenesis of neuralized ( neur ) and yellow ( y ) along the dorsal midline and of sepia ( se ) in the developing eye and observed in each case the described null mutant phenotype in the expected domain ( Figure 1E , F , Figure 1—figure supplement 1C ) . Next , we tested whether pCFD6-sgRNA2x also mediates efficient mutagenesis in the germline , where some UAS vectors are silenced ( DeLuca and Spradling , 2018; Huang et al . , 2018 ) . This is a particularly important application , as it allows to create stable and sequence-verified mutant fly lines , which can be backcrossed to remove potential off-target mutations . We crossed previously described nos-Gal4VP16 UAS-Cas9 . P1 flies ( Port et al . , 2014 ) to sgRNA strains targeting either neur , N , cut ( ct ) , decapentaplegic ( dpp ) or Ras85D . Despite the fact that all five genes are essential for Drosophila development and act in multiple tissues , nos-Gal4VP16 UAS-Cas9 . P1 pCFD6-sgRNA2x flies were viable and morphologically normal , demonstrating tightly restricted mutagenesis . We then tested their offspring for CRISPR induced mutations at the sgRNA target sites . Crosses with pCFD6-sgRNA2x targeting neur , N , ct and Ras85D passed on mutations to most or all analysed offspring ( Figure 1G ) . Mutations were often found on both target sites , were frequently out-of-frame and included large deletions of 8 and 14 kb between the sgRNA target sites ( Figure 1G ) . In contrast , nos-Gal4VP16 UAS-Cas9 . P1 pCFD6-dpp2x flies produced only few viable offspring of which only 1/11 carried a mutation , which was in-frame . Since dpp is known to be haploinsufficient ( St Johnston et al . , 1990 ) , this is consistent with a high number of dpp loss-of function alleles being transmitted to the next generation . Together , these experiments demonstrate that sgRNA expression from pCFD6 mediates efficient and tightly restricted mutagenesis in various somatic cell types as well as the germline and establishes that tissue-specific CRISPR mutagenesis in Drosophila is robust across genes and tissues . We and others have shown that expression of high amounts of Cas9 protein is toxic in various organisms ( Jiang et al . , 2014; Poe et al . , 2019; Port et al . , 2014; Yang et al . , 2018 ) . For example , overexpression of Cas9 in the wing imaginal disc of nub-Gal4 UAS-cas9 . P2 animals results in a strong induction of apoptosis ( Figure 2—figure supplement 1A ) . Since only relatively low levels of Cas9 are sufficient for efficient gene editing ( Figure 2—figure supplement 1B ) , we sought to engineer a system that would allow to tune Cas9 expression to optimally balance activity and toxicity . Such a system would ideally allow to modulate Cas9 levels independent of enhancer strength , in order to be compatible with the wide range of available Gal4 lines . We employed a method that uses upstream open reading frames ( uORF ) of different length to predictably reduce translation of the main , downstream ORF ( Ferreira et al . , 2013; Kozak , 2001; Southall et al . , 2013 ) . We created a series of six UAS-cas9 plasmids containing uORFs of different length , ranging from 33 bp ( referred to as UAS-uXSCas9 ) to 714 bp ( UAS-uXXLCas9 , Figure 2A ) . When combined with nos-cas9 these plasmids resulted in Cas9 protein levels inversely correlated with the length of the uORF ( Figure 2B , Figure 2—figure supplement 1C ) . Reducing the amount of Cas9 protein resulted in a strong decrease in the number of apoptotic cells ( Figure 2C ) . Importantly , three UAS-uCas9 transgenes with moderate levels of Cas9 expression and apoptosis levels similar to controls did mediate full on-target gene editing activity at the evi locus in wing imaginal discs ( Figure 2D , Figure 2—figure supplement 1C ) . Together , these experiments demonstrate that the UAS-uCas9 vector series enables titration of Cas9 expression to avoid toxicity without sacrificing gene editing activity . Next , we generated a toolbox of various fly strains harbouring a UAS-uMCas9 transgene and a Gal4 driver on the same chromosome ( Figure 2—figure supplement 2A , B ) . Such stocks can be crossed to transgenic sgRNA lines to induce conditional CRISPR mutagenesis in Gal4-expressing cells . We tested the spatial mutagenesis pattern for a number of novel Gal4 UAS-uMCas9 lines in the wing imaginal disc of third instar larva by either visualizing the loss of protein encoded by the target gene with a specific antibody , or by using the transgenic CIGAR reporter ( Brunner et al . , 2019 ) . CIGAR encodes an ubiquitously expressed fluorescent protein that is only efficiently translated once an upstream sequence has been mutated by CRISPR gene editing ( Brunner et al . , 2019 ) . While not all CRISPR-mediated mutations lead to induction of the fluorophore encoded by CIGAR , this strategy has the advantage that it readily reveals CRISPR activity throughout the entire organism . We found that while some Gal4 UAS-uMCas9 lines resulted in mutagenesis exclusively in cells positive for Cas9 at that stage ( Figure 2—figure supplement 2D , E ) , others had much broader mutagenesis patterns ( Figure 2E , Figure 2—figure supplement 2F , G ) . For example , in third instar wing discs ptc-Gal4 is expressed in a narrow band of cells along the anterior-posterior boundary ( Figure 2E ) . However , CRISPR mutagenesis with ptc-Gal4 frequently leads to mutations throughout the entire anterior compartment ( Figure 2E’ ) , likely reflecting broader expression of ptc-Gal4 in early development or expression at low level in this domain . Similar effects were observed with dpp-Gal4 ( Figure 2—figure supplement 2G ) . Therefore , additional regulatory mechanisms to temporally control Cas9 expression are highly desirable when using Gal4 lines with dynamic expression patterns during development . We first employed the temperature-sensitive Gal80 repressor to suppress Gal4 activity . While Gal80ts mediated strong inhibition of mutagenesis in ptc-Gal4 UAS-uMCas9 tub-Gal80ts flies at the restrictive temperature of 18°C , we still observed mutagenesis in Gal4-expressing cells in 11/24 wing discs , indicating residual Gal4 activity ( Figure 2F ) . We therefore tested an alternative strategy to induce CRISPR mutagenesis at a given time point . We created a transgene that harbors a FRT-flanked GFP Stop-cassette between the UAS promoter and the uMCas9 expression cassette ( UAS-FRT-GFP-FRT-uMCas9 , Figure 2G ) . A brief pulse of Flp recombinase ( from a hs-Flp transgene ) can be used to excise the GFP cassette at the desired time and induce Cas9 expression . We validated this approach by mutagenizing ct in a negatively marked subset of cells in the wing disc and observed loss of Ct protein exclusively in cells that had lost GFP expression ( Figure 2G ) . These experiments highlight the need to critically evaluate spatial mutagenesis patterns in conditional CRISPR experiments and suggest strategies for additional control of gene editing in cases where the Gal4 expression pattern alone does not suffice . We envision that in the future other systems for conditional transgene expression , such as the chemical-dependent GeneSwitch system ( Osterwalder et al . , 2001; Roman et al . , 2001 ) , split-Gal4 ( Luan et al . , 2006 ) or conditional transgene degradation ( Sethi and Wang , 2017 ) will also be combined with CRISPR to further refine mutagenesis patterns . Having established the robustness of our method and developed an optimised Cas9 toolkit , we next focused our efforts on the generation of a large-scale sgRNA resource . First , we generated and validated three sgRNA lines targeting genes with highly restricted expression patterns , which can be used as controls for effects of Cas9/sgRNA expression and induction of DNA damage in the majority of tissues where their target gene is not expressed ( Figure 3—figure supplement 1; Graveley et al . , 2011 ) . To allow systematic screening of functional gene groups we then designed sgRNAs against all Drosophila genes encoding transcription factors , kinases and phosphatases , as well as a large number of other genes encoding fly orthologs of genes implicated in human pathologies ( Figure 3A , see Materials and methods ) . We used CRISPR library designer ( Heigwer et al . , 2016 ) to compile a list of all sgRNAs that do not have predicted off-target sites elsewhere in the genome . We then selected sgRNAs depending on the position of their target site within the target gene . We chose sgRNAs targeting coding exons shared by all mRNA isoforms and target sites that were located in the 5’ half of the open reading frame , where indel mutations often have the largest functional impact . We then grouped sgRNAs in pairs , with each pair targeting sites typically separated by approximately 500 bp of protein coding DNA ( see Materials and methods ) . Next , we devised an efficient cloning protocol to insert defined sgRNA pairs into pCFD6 . This utilized synthesized oligonucleotide pools , which allow cloning of hundreds to thousands of sgRNA plasmids in parallel in a single tube , followed by clonal selection of individual pCFD6-sgRNA2x plasmids and sequence validation ( Figure 3B , see Materials and methods ) . We also generated a derivative of pCFD6 , pCFD6 . FRT , which harbors incompatible FRT2 and FRT5 sites before and after the sgRNA cassette , respectively . These recombination sites can be used to exchange sequences either side of the sgRNA cassette , for example the promoter , or to add additional sgRNAs to the array ( Figure 3C ) . We validated that both FRT sites mediate highly efficient chromosome exchange in vivo ( Figure 3D ) . We then generated a large-scale transgenic sgRNA library , which we collectively refer to as the ‘Heidelberg CRISPR Fly Design Library’ ( short HD_CFD library ) . This growing resource currently contains 2622 plasmids and 1739 fly stocks targeting 1513 unique genes ( Supplementary file 1 ) . Fly lines are so far available for 545/754 ( 72% ) transcription factors , 199/230 ( 87% ) protein kinases and 141/207 ( 68% ) phosphatases ( Figure 2D ) . To test on-target activity of HD_CFD sgRNA strains , we crossed a random selection of 28 HD_CFD lines to an act-cas9;;tub-Gal4/TM3 strain , which is expected to mediate ubiquitous mutagenesis in combination with active sgRNAs . We then sequenced PCR amplicons encompassing the sgRNA target sites ( see Materials and methods ) and analysed editing efficiency by ICE analysis ( Hsiau et al . , 2019 ) . We found that the vast majority ( 26/28 ) of HD_CFD sgRNA lines resulted in gene editing on both target sites ( Figure 4A ) . For 12/28 of lines editing on both sites was inferred to be at least 50% and 23/28 reached this threshold on at least one target site . In contrast , only a single line ( HD_CFD00032 ) resulted in no detectable gene editing at either sgRNA target site . This suggests that HD_CFD sgRNA lines mediate robust and efficient mutagenesis of target genes across the genome . Next , we performed a large-scale transgenic CRISPR screen . We crossed HD_CFD animals to act-cas9;;tub-Gal4/TM3 to induce mutations ubiquitously in the offspring and determined viability at five to seven days after eclosion . 290/639 ( 45% ) of all crosses did not yield any viable offspring , while 269 ( 42% ) lines produced viable adults and 53 ( 8% ) of the lines resulted in lethality with incomplete penetrance ( Figure 4B and Supplementary file 2 ) . In order to benchmark the performance of the screen , we manually curated viability information based on genetic alleles stored in the Flybase database to determine which HD_CFD lines target genes known to be essential or non-essential during Drosophila development . This resulted in a list of 210 lines that target known essential genes . Of those , 167 ( 79% ) resulted in lethality , 20 ( 10% ) were scored as semi-lethal , and 23 ( 11% ) gave rise to viable adult offspring . Interestingly , among the targets of sgRNA lines that produced false-negative results there was a strong enrichment of genes known to play important roles , and to be highly expressed , during early embryonic development . Furthermore , sequencing the sgRNA target sites in randomly selected false-negative lines revealed efficient gene editing on one or both sites in 3/3 lines ( Figure 4—figure supplement 1 ) , suggesting that false-negative results often arise due to mRNA perdurance , not inactive sgRNAs . Next , we analysed our data set for the occurrence of false-positives , that is lines that target non-essential genes , but result in lethality . Among the 639 lines present in our screen , 54 target genes are currently annotated as viable . Of those , 48 ( 89% ) gave rise to viable adult offspring , one resulted in semi-lethal offspring and 5 ( 9% ) produced no viable offspring . False-positive results might arise due to off-target mutagenesis , mutations that affect neighbouring genes or cis-elements located at the target-locus , or reflect incorrect annotations in the database . Of the five lines giving rise to false-positive results in our screen two target the same gene ( Blos1 ) , arguing against sgRNA-mediated off-target mutagenesis in this case . Screening for lethality not only allowed us to benchmark our sgRNA library , but also revealed multiple lines targeting uncharacterized genes with putative essential functions ( Supplementary file 3 ) . For example , sgRNA line HD_CFD558 targets CG9890 , an evolutionary conserved ( 55% amino acid similarity to the human ortholog ) zinc finger protein of unknown function . Another interesting example is CG6470 , which is targeted by HD_CFD557 and HD_CFD599 with independent sgRNAs . CG6470 encodes an uncharacterized zinc finger protein that despite its essential role during development is evolutionary restricted to the genus Drosophila . These examples highlight the value of our lethality screen beyond benchmarking our technology . To further characterize genes of interest sgRNA lines can then be used for tissue-specific mutagenesis , where genes performing similar cellular functions often give rise to phenotypes with high similarity . To demonstrate this application , we crossed several lines targeting genes associated with dpp/TGFb signalling with nub-Gal4 UAS-uMCas9 flies , which drive CRISPR mutagenesis in selected tissues , including cells giving rise to the adult wing . All these lines result in lethality in combination with a ubiquitous CRISPR system ( Supplementary file 2 ) , but gave rise to viable adults in combination with nub-Gal4 UAS-uMCas9 , highlighting again the tight control of mutagenesis . Moreover , all lines resulted in offspring that had wings of abnormal size and morphology and faithfully recapitulated the known phenotypes of loss-of function mutations of their target genes ( Figure 4E ) . Together these results show that lines of the HD_CFD library can be used for systematic CRISPR screens in vivo and mediate relevant phenotypes with very high penetrance and specificity . Here , we present a large-scale collection of transgenic sgRNA strains for conditional CRISPR mutagenesis in Drosophila . In combination with the associated toolbox of novel Cas9 constructs , the sgRNA lines mediate efficient mutagenesis with precise temporal and spatial control . This allows the rapid targeted disruption of genes in various contexts in the intact organism . The high performance of this resource relies on a ) use of conditional sgRNA constructs to achieve strong dependency of CRISPR mutagenesis on Gal4 , b ) tunable Cas9 expression to achieve high on-target activity with low toxicity , c ) the use of two sgRNAs targeting independent positions in the same gene to increase the fraction of cells that harbor non-functional mutations in both alleles . We validate our library by conducting a fully transgenic CRISPR mutagenesis screen , to our knowledge the so far largest in any multicellular animal , which revealed 259 putative essential genes , of which 56 are poorly characterized . To date RNAi is the most commonly used method to disrupt gene expression in defined cell types or developmental stages in vivo . In Drosophila , transgenic RNAi libraries that cover most protein coding genes have been described ( Dietzl et al . , 2007; Heigwer et al . , 2018; Perkins et al . , 2015 ) . However , a significant number of these lines do not mediate efficient gene knock-down and the majority reduces mRNA levels by less than 75% ( Perkins et al . , 2015 ) . Residual gene expression can therefore mask phenotypes in RNAi experiments , which loss-of function alleles induced by CRISPR mutagenesis might reveal . In support of this notion three recent studies demonstrate that CRISPR mutagenesis in vivo can cause phenotypes that are significantly more penetrant than RNAi ( Meltzer et al . , 2019 ) or are missed altogether in RNAi experiments ( Delventhal et al . , 2019; Schlichting et al . , 2019 ) . Furthermore , our molecular analysis of mutations induced by the CRISPR library described here , as well as the phenotypes arising from them , suggest that the fraction of lines that produce no or only insufficient on-target mutations is less than 10% , which compares favorably to current Drosophila RNAi libraries . Together these observations strongly suggest that screening biological processes of interest by conditional CRISPR mutagenesis can reveal novel gene functions that have so far been missed in RNAi based experiments . For conditional CRISPR mutagenesis to be broadly applicable , it needs to be effective in a wide range of tissues and cell types . We show here that our system works effectively in a number of important tissues in the fly , such as imaginal discs , the adult midgut and the germline . Furthermore , others have shown that CRISPR with transgenic components is also effective in postmitotic neurons ( Delventhal et al . , 2019 ) or in cells of the prothoracic gland , which are endoreplicating and contain multiple copies of the genome ( Huynh et al . , 2018 ) . In addition , our finding that sequencing of PCR amplicons generated from genomic DNA of entire flies frequently indicates mutations in over 90% of the amplicons suggests that CRISPR is effective in most cells of the animal . In parallel to the CRISPR library described here , the National Institute of Genetics ( NIG ) in Japan , the Transgenic RNAi Project ( TRIP ) at Harvard University and the Schuldiner group at the Weizmann Institute are generating collections of transgenic sgRNA lines ( Meltzer et al . , 2019; Zirin et al . , 2020; https://shigen . nig . ac . jp/fly/nigfly/ ) . These projects follow different strategies to prioritise target genes and hence the overlap between different collections is currently limited . Furthermore , there exist significant differences in design between these resources and the library described here . First , the NIG and the majority of TRIP and Weizmann libraries encode a single sgRNAs per transgene , while all HD_CFD lines encode two sgRNAs . Co-expression of more than one sgRNA against the same target leads to more penetrant phenotypes and reduces the number of inactive lines ( Port and Bullock , 2016; Xie et al . , 2015; Yin et al . , 2015 ) . Second , the HD_CFD sgRNAs are encoded in pCFD6 or pCFD6 . FRT , which are conditional UAS vectors , while all other libraries so far used plasmids expressing sgRNAs from ubiquitous U6 promoters . We have previously shown that expression of U6-sgRNA in combination with UAS-Cas9 alone is not sufficient to efficiently restrict mutagenesis to Gal4 expressing cells and that expression of sgRNAs from a UAS vector , such as pCFD6 , results in a significant improvement in spatial and temporal control ( Port and Bullock , 2016 ) . The use of transgenes of the UAS-uCas9 series can reduce , but not prevent , unwanted mutagenesis in combination with U6-sgRNAs , as leaky ( i . e . Gal4 independent ) Cas9 expression is reduced in the presence of a uORF . An advantage of U6-sgRNA vectors is the consistent high sgRNA expression , whereas the level of sgRNAs expressed from UAS promoters depends on the strength of the Gal4 line and can become limiting with weak Gal4 drivers ( Meltzer et al . , 2019 ) . Of note , pCFD6 . FRT can alleviate this problem , as users can easily swap the UAS promoter for a U6:3 promoter in cases where high sgRNA expression is a higher priority than tight conditional mutagenesis . The different sgRNA libraries that are currently being developed are therefore complementary resources for CRISPR mutagenesis . Large-scale screens in different contexts using lines from different libraries will be informative about the optimal use of each resource . Two decades after the publication of the genome sequence of humans , mice , flies , worms and many other organisms , the functional annotation of these genomes are still far from complete . CRISPR-Cas genome editing is accelerating the rate at which new gene functions are described . The resources described here will facilitate context-dependent functional genomics in Drosophila . New insights into the function of the fly genome will inform the functional annotation of the human genome , reveal conserved principles of metazoan biology and suggest control strategies for insect disease vectors . PCRs were performed with the Q5 Hot-start 2x master mix ( New England Biolabs ( NEB ) ) and cloning was performed using the In-Fusion HD cloning kit ( Takara Bio ) or restriction/ligation dependent cloning . Newly introduced sequences were verified by Sanger sequencing . Oligonucleotide sequences are listed in Supplementary file 4 . The UAS-uCas9 series of plasmids was generated using the pUASg . attB plasmid backbone ( Bischof et al . , 2013 ) . The plasmid was linearized with EcoRI and XhoI and sequences coding for mEGFP ( A206K ) and hCas9-SV403’UTR were introduced by In-Fusion cloning using standard procedures . Coding sequences for mEGFP ( A206K ) were ordered as a gBlock from Integrated DNA Technologies ( IDT ) and amplified with primers mEGFPfwd and mEGFPrev ( Supplementary file 4 ) . The sequence coding for SpCas9 and an SV40 3’UTR were PCR amplified from plasmid pAct-Cas9 ( Port et al . , 2014 ) with primers Cas9SV40fwd and Cas9SV40rev . Both PCR amplicons and the linearized plasmid backbone were assembled in a single reaction to generate plasmid UAS-uXXLCas9 . UAS-uCas9 plasmids with shorter uORFs were generated by PCR amplification using UAS-uXXLCas9 as template and the common fwd primer uCas9fwd in combination with rev primers binding at various positions in the mEGFP ORF ( uXSCas9rev for UAS-uXSCas9; uSCas9rev for UAS-uSCas9; uMCas9rev for UAS-uMCas9; uLCas9rev for UAS-uLCas9; uXLCas9rev for UAS-uXLCas9 ) . PCR products were cirularized by In-Fusion cloning and the sequence between the hsp70 promoter and the attP site was verified by Sanger sequencing . The UAS-uCas9 plasmid series and the full sequence of each plasmid will become available from Addgene ( Addgene plasmids 127382–127387 ) . To generate UAS-FRT-GFP-FRT-uMCas9 plasmid UAS-Cas9 . P2 ( Port and Bullock , 2016 ) was digested with EcoRI and the plasmid backbone was gel purified . The FRT-GFP-FRT cassette was ordered as two separate gBlocks from IDT ( GFPflipout5 and GFPflipout3 ) and individually PCR amplified with primers GFPflipout5fwd and GFPflipout5rev or GFPflipout3fwd and GFPflipout3rev and gel purified . The two amplicons were mixed at equalmolar ratios and fused by extension PCR , adding primers GFPflipout5fwd and GFPflipout3rev after 8 PCR cycles for an additional 25 cycles . The final FRT-GFP-FRT cassette was gel purified . The uMCas9EcoRI fragment was PCR amplified from plasmid UAS-uMCas9 with primers uMCas9EcoRIfwd and uMCas9EcoRIrev and gel purified . The plasmid backbone , FRT-GFP-FRT cassette and uMCas9EcoRI fragment were assembled by In-Fusion cloning and sequence from the first FRT site to the end of Cas9 was verified by Sanger sequencing . The UAS-FRT-GFP-FRT-uMCas9 plasmid and the full sequence will become available from Addgene ( Addgene plasmid 127388 ) . pCFD6 . FRT was generated as a derivative of pCFD6 . pCFD6 was linearized by restriction digestion with EcoRI-HF and XbaI . The sgRNA cassette was exchanged with a new cassette encoding ( from 5’ to 3’ ) : 5’UTR spacer , FRT2 site , D . mel . tRNA Gly , BbsI site , sgRNA core , D . mel . tRNA Glu , BbsI site , sgRNA core , Os . sat . tRNA , FRT5 site . The new sgRNA cassette was ordered as a gBlock from IDT and cloned into the linearized pCFD6 plasmid and newly introduced sequences were verified by Sanger sequencing . pCFD6 . FRT will become available from Addgene . All possible sgRNA sequences targeting all transcription factors , kinases , phosphatases and a number of other - mostly disease relevant - genes in the D . melanogaster genome version BDGP6 were identified using the CRISPR library designer ( CLD ) software version 1 . 1 . 2 ( Heigwer et al . , 2016 ) . CLD excludes sgRNA sequences that have predicted off-target sites elsewhere in the genome . The resulting pool of sequences was further filtered according to additional criteria . Specifically , sequences with BbsI and BsaI restriction sites were excluded . In addition , sequences containing stretches of 4 or more identical nucleotides were removed from the pool . Two pairs of sgRNAs targeting each gene were then selected using a random sampling approach . For each gene , up to 10 , 000 pairs of sgRNA sequences were selected at random from the pool of available sequences . Each sequence pair was then evaluated according to a custom scoring function . In order to preferentially select sgRNA pairs that target constitutive exons , the scoring function awarded bonus points for each transcript targeted by either of the sgRNAs . Bonus points were further given to sgRNAs targeting the first half of the gene and small distances to the gene’s transcription start site were awarded additionally . To avoid selecting pairs of overlapping sgRNAs that could potentially interfere with each other’s activity , sgRNA pairs that were less than 75 bp apart from each other were strongly penalized . Further , sgRNAs targeting the gene within 500 bp of each other were penalized . This was done to avoid functional protein products in cases where the second sgRNA might correct an out-of-frame mutation introduced by the first sgRNA . Finally , we penalized sgRNA with predicted off-target effects according to CLD . The two top-scoring pairs for each gene were selected for the HD_CFD library . sgRNA pairs were cloned into BbsI digested pCFD6 ( Port and Bullock , 2016 ) following a two-step pooled cloning protocol . Oligonucleotide pools were ordered from Twist Biosciences and Agilent Technologies . Each oligonucleotide contained two sgRNA protospacer sequences targeting the same gene separated by a BsaI restriction cassette . Furthermore , oligos contained sequences at either end for PCR amplification and BbsI sites at the 5’ end of the first and 3’ end of the second protospacer . An annotated example oligo is shown in Supplementary file 4 . Oligo pools were resuspended in sterile dH2O and amplified by PCR with primers Libampfwd and Libamprev , followed by BbsI digestion and gel purification . Digested oligo pools were then ligated into BbsI digested pCFD6 plasmid backbone , transformed into chemically competent bacteria and plated on agarose plates containing Carbenicillin . After incubation overnight at 37°C transformed bacteria were resuspended and plasmid DNA was extracted and digested with BsaI . Next , the sgRNA core sequence and tRNA required between the two protospacers , but not encoded on the oligos , were introduced . These were PCR amplified from pCFD6 using primers Core_tRNAfwd and Core_tRNArev . PCR amplicons were digested with BsaI and ligated into the BsaI digested pCFD6 plasmid pool containing the library oligos , transformed into chemically competent bacteria and plated on agarose plates containing Carbenicillin . The next day single colonies were picked and used to inoculate liquid cultures . The following day plasmid DNA was extracted and the sgRNA cassette was sequenced with primer pCFD6seqfwd2 to determine which oligo was inserted and to verify the sequence . Individual sequence verified pCFD6-sgRNA2x plasmids were stored at −20°C and make up the HD_CFD plasmid library . Transgenic Drosophila strains used or generated in this study are listed in Supplementary file 5 . Unless specified otherwise flies were kept at 25°C with 50 ± 10% humidity with a 12 hr light/12 hr dark cycle . Transgenesis was performed with the PhiC31/attP/attB system and plasmids were inserted at landing site ( P{y[+t7 . 7]CaryP}attP40 ) on the second chromosome . Additional insertions of UAS-uMCas9 were generated at ( M{3xP3-RFP . attP}ZH-51D ) on the second chromosome and ( M{3xP3-RFP . attP}ZH-86Fb ) and ( PBac{y+-attP-3B}VK00033 ) on the third chromosome . Microinjection of plasmids into Drosophila embryos was carried out using standard procedures either in house , or by the Drosophila Facility , Centre for Cellular and Molecular Platforms , Bangalore , India ( http://www . ccamp . res . in/drosophila ) or by the Fly Facility , Department of Genetics , University of Cambridge , UK ( www . flyfacility . gen . cam . ac . uk/ ) . Transgenesis of sgRNA plasmids was typically performed by a pooled injection protocol , as previously described ( Bischof et al . , 2013 ) . Briefly , individual plasmids were pooled at equimolar ratio and DNA concentration was adjusted to 250 ng/μl in dH2O . Plasmid pools were microinjected into y[1] M{vas-int . Dm}ZH-2A w[*]; ( P{y[+t7 . 7]CaryP}attP40 ) embryos , raised to adulthood and individual flies crossed to P{ry[+t7 . 2]=hsFLP}1 , y[1] w[1118]; Sp/CyO-GFP . Transgenic offspring was identified by orange eye color and individual flies crossed to P{ry[+t7 . 2]=hsFLP}1 , y[1] w[1118]; Sp/CyO-GFP balancer flies . In the very rare case that a plasmid stably inserted at a genomic locus different than the intended attP40 landing site , this typically resulted in a noticeably different eye colouration and such flies were discarded . Transgenic flies from pooled plasmid injections were genotyped to determine which plasmid was stably integrated into their genome . If transgenic flies were male or virgin female , animals were removed from the vials once offspring was apparent and prepared for genotyping . In the case of mated transgenic females , genotyping was performed in the next generation after selecting and crossing a single male offspring , to prevent genotyping females fertilised by a male transgenic for a different construct . Single flies were collected in PCR tubes containing 50 µl squishing buffer ( 10 mM Tris-HCL pH8 , 1 mM EDTA , 25 mM NaCl , 200 µg/ml Proteinase K ) . Flies were disrupted in a Bead Ruptor ( Biovendis ) for 20 s at 30 Hz . Samples were then incubated for 30 min at 37°C , followed by heat inactivation for 3 min at 95°C . 3 µl of supernatant were used in 30 µl PCR reactions with primers pCFD6seqfwd2 and pCFD6seqrev2 . PCR amplicons were analysed by Sanger sequencing with primer pCFD6seqrev2 . Genes considered ‘known lethal’ or ‘known viable’ were chosen based on information available in FlyBase ( release FB2018_1 ) . For each gene report we manually reviewed the lethality information available in the phenotype category . We did not consider information based on RNAi experiments , as these typically were performed with tissue-restricted Gal4 drivers and residual expression might mask gene essentiality . Annotations of viability in FlyBase are heavily skewed towards lethal genes , likely reflecting the uncertainty in many cases whether a viable phenotype reflects residual gene activity of a particular allele . Immunohistochemistry of wing imaginal discs was performed using standard procedures . Briefly , larva were dissected in ice cold PBS and fixed in 4% Paraformaldehyde in phosphate buffered saline ( PBS ) containing 0 . 05% Triton-X100 for 25 min at room temperature . Larva were washed three times in PBS containing 0 . 3% Triton-X100 ( PBT ) and then blocked for 1 hr at room temperature in PBT containing 1% heat-inactivated normal goat serum . Subsequently , larva were incubated with first antibody ( mouse anti-Cas9 ( Cell Signaling ) 1:800; mouse anti-Cut ( DSHB , Gary Rubin ) 1:30; guinea pig anti-Sens ( Boutros lab , unpublished ) 1:300; rabbit anti-Evi [Port et al . , 2008] 1:800 ) in PBT overnight at 4°C . The next day , samples were washed three times in PBT for 15 min and incubated for 2 hr at room temperature with secondary antibody ( antibodies coupled to Alexa fluorophores , Invitrogen ) diluted 1:600 in PBT containing Hoechst dye . Samples were washed three times 15 min in PBT and mounted in Vectashield ( Vectorlabs ) . To visualize apoptotic cells wing discs expressing the apoptosis sensor GC3Ai ( Schott et al . , 2017 ) was fixed in 4% PFA , washed in PBT containing Hoechst and mounted in Vectashield . Images were acquired with a Zeiss LSM800 , Leica SP5 or SP8 or a Nikon A1R confocal microscope in the sequential scanning mode . Samples that were used for comparison of antibody staining intensity were recorded in a single imaging session . Image processing and analysis was performed with FIJI ( Schindelin et al . , 2012 ) . For the comparative analysis of anti-Cas9 , GC3Ai and anti-Evi fluorescent intensities presented in Figure 2 raw image files were used to select the wing pouch area and measure the average fluorescence intensity . Experiments were performed at least twice and more than three samples were analyzed for each experiment . To produce the overlay of several wing imaginal discs shown in Figure 1 the Fiji plug-in bUnwarpJ ( Sorzano et al . , 2005 ) was used . Images were rotated and cropped such that wing discs were oriented dorsal up and anterior left and positioned in the center of the image . A representative image was selected as ‘target’ and all other images registered to this target using bUnwarpJ , selecting ‘mono’ as registration mode and setting landmark weight to 1 . Landmarks were manually selected around the outline of the target wing disc , as well as along the folds in the hinge region of the disc . Registered images were then transformed to a binary image using the Fiji threshold function and assembled to an image stack . Shown are average intensity projections of the indicated number of images using the Fire lookup table . In the resulting image bright areas are CIGAR positive in many discs , while dark areas are devoid of CIGAR signal in most discs . To determine the mutational status at each sgRNA target site the locus was PCR amplified and PCR amplicons were subjected to sequencing . To extract genomic DNA , flies were treated as described above under ‘Genotyping of sgRNA flies’ . Primers to amplify the target locus were designed to hybridize 250–300 bp 5’ or 3’ to the sgRNA target site and are listed in Supplementary file 4 . PCR products were purified using the PCR purification Kit ( Qiagen ) according to the instructions by the manufacturer and sent for Sanger sequencing . While Sanger sequencing is less accurate and quantitative than deep sequencing of amplicons on , for example , the Illumina platform , it typically allows to cover both sgRNA targets on a single amplicon , which is necessary to account for mutations that result in deletions of the intervening sequence . In cases where this was not possible , for example due to the presence of a large intron between the target sites , each site was analysed on a separate PCR amplicon . To account for deletions in these cases additional PCR reactions containing the distal fwd and rev primers were included . Sequencing chromatograms were visually inspected for sequencing quality and presence of the sgRNA target site and analysed by Inference of CRISPR Edits ( ICE ) analysis ( Hsiau et al . , 2019 ) .	Twenty years after the release of the sequence of the human genome , the role of many genes is still unknown . This is partly because some of these genes may only be active in specific types of cells or for short periods of time , which makes them difficult to study . A powerful way to gather information about human genes is to examine their equivalents in ‘model’ animals such as fruit flies . Researchers can use genetic methods to create strains of insects where genes are deactivated; evaluating the impact of these manipulations on the animals helps to understand the roles of the defunct genes . However , the current methods struggle to easily delete target genes , especially only in certain cells , or at precise times . Here , Port et al . genetically engineered flies that carry CRISPR-Cas9 , a biological system that can be programmed to ‘cut’ and mutate precise genetic sequences . The insects were also manipulated in such a way that the CRISPR elements could be switched on at will , and their quantity finely tuned . This work resulted in a collection of more than 1 , 700 fruit fly strains in which specific genes could be deactivated on demand in precise cells . Further experiments confirmed that this CRISPR system could mutate target genes in different parts of the fly , including in the eyes , gut and wings . Port et al . have made their collection of genetically engineered fruit flies publically available , so that other researchers can use the strains in their experiments . The CRISPR technology they refined and developed may also lay the foundation for similar collections in other model organisms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "tools", "and", "resources", "genetics", "and", "genomics" ]	2020	A large-scale resource for tissue-specific CRISPR mutagenesis in Drosophila
Epigenome modulation potentially provides a mechanism for organisms to adapt , within and between generations . However , neither the extent to which this occurs , nor the mechanisms involved are known . Here we investigate DNA methylation variation in Swedish Arabidopsis thaliana accessions grown at two different temperatures . Environmental effects were limited to transposons , where CHH methylation was found to increase with temperature . Genome-wide association studies ( GWAS ) revealed that the extensive CHH methylation variation was strongly associated with genetic variants in both cis and trans , including a major trans-association close to the DNA methyltransferase CMT2 . Unlike CHH methylation , CpG gene body methylation ( GBM ) was not affected by growth temperature , but was instead correlated with the latitude of origin . Accessions from colder regions had higher levels of GBM for a significant fraction of the genome , and this was associated with increased transcription for the genes affected . GWAS revealed that this effect was largely due to trans-acting loci , many of which showed evidence of local adaptation . To better understand how genotype and environment interact to affect DNA methylation and transcription , we grew 150 Arabidopsis thaliana accessions from Sweden ( Long et al . , 2013 ) in two different environments , 10°C and 16°C , chosen because they lead to very different flowering behavior ( Atwell et al . , 2010 ) . Relying on existing genome sequence information ( Long et al . , 2013 ) , methylome- and transcriptome-sequencing data were generated ( see ‘Materials and methods’ ) . In plants , DNA methylation occurs on cytosines in the CG , CHG , and CHH contexts ( where H is any nucleotide except for C ) , each of which is catalyzed by independent pathways ( Finnegan et al . , 1998; Stroud et al . , 2014 ) . Consistent with previous results ( Vaughn et al . , 2007; Eichten et al . , 2013; Schmitz et al . , 2013; Li et al . , 2014; Seymour et al . , 2014; Hagmann et al . , 2015 ) we found considerable variation between accessions regardless of context , even at the level of genome-wide averages ( Figure 1A ) . Temperature , on the other hand , did not appear to affect genome-wide CG or CHG methylation , but had a significant effect on CHH methylation , levels of which were 14% higher at 16°C than at 10°C , on average ( Figure 1A ) . To investigate the genetic basis of DNA methylation , we performed genome-wide association studies ( GWAS ) using different facets of average methylation as the phenotype . For global CG and CHG methylation , no associations reached genome-wide significance , while for CHH methylation a clear peak of association was observed on chromosome 4 ( Figure 1—figure supplement 1 ) . The association was even more significant when restricting attention to average CHH methylation of large transposons ( Figure 1B ) , in agreement with the notion that this type of methylation mostly occurs in transposons in Arabidopsis ( Stroud et al . , 2013 ) . 10 . 7554/eLife . 05255 . 003Figure 1 . The effect of CMT2 on genome-wide CHH methylation levels . ( A ) Genome-wide average methylation level reaction norms for each accession ( 156 samples at 10°C and 125 samples at 16°C ) . Only CHH levels differ significantly between temperatures ( Wilcoxon rank sum test; p = 1 . 7e-16 ) . ( B ) Manhattan plot of genome-wide association studies ( GWAS ) results using average levels of CHH methylation for 151 accessions at 10°C on large transposons as the phenotype ( the peak is also seen at 16°C [not shown] ) . The threshold line indicates a Bonferroni-corrected p-value of 0 . 05 . ( C ) CHH methylation on large ( over 2 kb ) transposons at 10°C by CMT2 two-locus genotype ( population sizes are 36 , 82 , and 24 for CMT2anr/nr/CMT2br/r , CMT2ar/r/CMT2br/r , CMT2ar/r/CMT2bnr/nr , respectively ) . The values plotted are the Best Linear Unbiased Predictor ( BLUP ) estimates after correcting for population structure . Since accessions are homozygous , only four genotypes are possible , of which only three exist due to complete linkage disequilibrium between CMT2a and CMT2b . Figure 1—figure supplement 1 shows Manhattan plots of GWAS results for global methylation averages . Figure 1—figure supplement 2 shows Stepwise GWAS using average CHH methylation of TE's . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 00310 . 7554/eLife . 05255 . 004Figure 1—figure supplement 1 . Manhattan plots of GWAS results for global methylation averages . ( A ) CG methylation at 10°C . ( B ) CHG methylation at 10°C . ( C ) CHH methylation at 10°C . Results for methylation at 16°C were similar ( data not shown ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 00410 . 7554/eLife . 05255 . 005Figure 1—figure supplement 2 . Stepwise GWAS using average CHH methylation of TE's as a phenotype . ( A ) Without a cofactor . ( B ) Including SNP on chr 4 at position 10 , 459 , 127 ( CMT2a ) as a cofactor . ( C ) Including snps on chr 4 at 10 , 459 , 127 ( CMT2a ) and 10 , 454 , 628 ( CMT2b ) as cofactors . The threshold line indicates a Bonferroni-corrected p-value of 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 005 The association centered around a SNP at 10 , 459 , 127 on chromosome 4 , 38 kb downstream from the locus AT4G19020 , which encodes a homolog of the CHG methyltransferase chromo-methylase-3 ( Lindroth et al . , 2001 ) that has recently been shown to catalyze both CHH and CHG methylation on transposons , and is thus an excellent candidate ( Zemach et al . , 2013; Stroud et al . , 2014 ) . A multi-locus mixed model ( Segura et al . , 2012 ) that included the identified SNP ( CMT2a ) as a fixed effect revealed another SNP downstream of CMT2 , at position 10 , 454 , 628 ( CMT2b ) , 4 . 5 kb closer to CMT2 than CMT2a , and in complete linkage disequilibrium with it ( i . e . , the non-reference alleles at CMT2a and CMT2b are never seen together ) . Repeating the GWAS with both CMT2a and CMT2b as cofactors identified no further loci ( Figure 1—figure supplement 2 ) . Both non-reference alleles are common in southern Sweden , but are also found in the north ( 22 . 6% vs 9 . 5% and 30 . 6% vs 7 . 9% for CMT2a and CMT2b , respectively ) . Accessions with the non-reference CMT2a allele have on average more CHH methylation on transposons than those with the reference haplotype ( p = 1 . 1e-03 ) , while those with the non-reference CMT2b allele have lower levels of CHH methylation than the reference haplotype ( p = 8 . 1e-03; Figure 1C ) . The associations were readily confirmed using an F2 population generated by crossing accessions with the CMT2a and CMT2b non-reference alleles ( Figure 2 ) . No significant differences in CMT2 mRNA levels were observed between the alleles in our data and limited Sanger sequencing of cDNA showed no evidence of splicing variants ( although , as will be discussed below , we did detect a putative rare null allele ) . Several non-synonymous polymorphisms in the methyltransferase and BAH domains of CMT2 were detected ( Supplementary files 1 and 2 ) but they do not explain the phenotype as well as the CMT2a and CMT2b SNPs . 10 . 7554/eLife . 05255 . 006Figure 2 . CHH methylation levels in an F2 population map to CMT2 . ( A ) CHH methylation on large transposons by CMT2 genotype in an F2 population of 113 individuals ( population sizes are 19 , 52 , and 38 for CMT2anr/nr/CMT2br/r , CMT2ar/r/CMT2br/r , CMT2ar/r/CMT2bnr/nr , respectively; 4 individuals whose genotype at CMT2 could not be accurately inferred were omitted ) . ( B ) Mapping of CHH methylation of long TEs in the same population . The dotted line indicates a LOD threshold with a genome-wide p-value of 0 . 05 obtained using 1000 permutations , and the vertical blue line shows the marker interval that contains CMT2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 006 The effect of genetic variation on local CHH methylation was examined by calculating the methylation level in 200 bp sliding windows across the genome ( 100 bp overlap between windows ) and running GWAS for the 200 , 000 differentially methylated regions ( DMRs; see ‘Materials and methods’ ) that varied most between individuals . 36023 DMRs had at least one genome-wide significant association ( Bonferroni-corrected p-value of 0 . 05; 7273 remain after correcting for multiple GWAS using an FDR of 0 . 05 ) . 45% ( 15 , 031 ) of the DMRs had a significant cis-association within 100 kb , while the rest showed evidence of trans-regulation , including the dramatic effect of CMT2 on chromosome 4 which accounted for approximately 21% ( 7392 ) of all significant associations ( Figure 3A ) . 10 . 7554/eLife . 05255 . 007Figure 3 . Genetic basis CHH methylation variation . ( A ) GWAS for CHH differentially methylated regions ( DMRs ) at 10°C in 151 accessions , defined using 200 bp sliding windows across the genome and selecting the 200 , 000 most variable ones . For each DMR , SNPs significantly associated at the Bonferroni-corrected 0 . 05-level are plotted . ( B ) Variance-components analysis of the CHH DMRs . For each DMR , a mixed model with cis , CMT2 , and genome-wide trans effects , plus environment and genetic interactions with environment was fitted ( see ‘Materials and methods’ ) . DMRs were binned by the total variance explained by the model . The density of DMRs in each bin is shown at the top , and the bottom shows the average variance-decomposition for each bin . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 007 To quantify the regulation of DMRs , we partitioned the variance of CHH methylation into environmental ( E ) , CMT2 , CMT2 X E , cis , cis X E , trans , and trans X E using a mixed model ( Figure 3B ) . The analysis confirmed substantial cis and trans associations , with the environment modulating the genetic effects rather than having a major direct effect . At least for the cis associations , a possible explanation is that SNPs tag polymorphic TE insertions , with the insertion allele being silenced in a temperature-sensitive manner . The effect of temperature on CHH methylation could also be seen at the local level . We defined ‘temperature DMRs’ by looking for windows that differed significantly between temperatures . Comparing 16°C–10°C , each accession on average gained CHH methylation at ∼400 temperature DMRs and lost it at ∼200 temperature DMRs ( false discovery rate = 0 . 05 ) . CHH methylation is associated with transposable elements ( TEs; Finnegan et al . , 1998 ) , and in agreement with this , 79% of the temperature DMRs where methylation was gained at 16°C were located within 500 bp of an annotated TE ( with 60% directly overlapping one ) . These temperature DMRs were enriched in a small subset of TEs ( 835 , or 2 . 7% of total , permutation based p-value = 0 . 05 ) that were more highly methylated than other transposons , but with lower methylation levels immediately adjacent ( Figure 4A ) . Compared to TEs without temperature DMRs , these ‘variable’ TEs also tended to be euchromatic ( Figure 4B ) , highly expressed ( Figure 4C ) , and recently inserted ( ‘evolutionarily young’ TE insertions for which orthologs are not present in Arabidopsis lyrata [Zhong et al . , 2012] comprised 75% of the variable TEs vs 68% of non-variable TEs ) . At the super-family level , members of the SINE , SINE-like , Helitrons and Mutator-like DNA TE superfamilies were over-represented among the variable transposons , and at the family level , 36 families were over-represented , including the AtREP , Vandal and HAT DNA transposons , as well as COPIA78/ONSEN and META1 retroelements ( Table 1 ) . Interestingly , COPIA78 has been shown to become active in response to heat stress ( Pecinka et al . , 2010; Ito et al . , 2011 ) apparently due to heat-shock promoter elements in its LTR regions ( Cavrak et al . , 2014 ) . 10 . 7554/eLife . 05255 . 008Figure 4 . CHH methylation varies with temperature . ( A ) Average methylation levels over variable transposons at 10°C ( orange ) vs 16°C ( red ) , and over non-variable transposons at 10°C ( purple ) vs 16°C ( dark blue ) . Methylation for variable TEs is significantly higher ( permutation p-value for CHH methylation = 0 . 05 ) . ( B ) The density of variable ( red ) and non-variable TEs along chromosomes in 500 kb windows . Density is defined as the percentage of the total number in either category in each window; pericentromeric regions are shaded grey . ( C ) The expression of TEs at both temperatures . Variable TEs are more highly expressed than non-variable TEs , but the difference is only statistically significant at 16°C ( Wilcoxon: 10°C , p = 0 . 15; 16°C , p = 0 . 023 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 00810 . 7554/eLife . 05255 . 009Table 1 . Super-families ( italics ) and families that are over-represented among ‘variable’ TEsDOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 009TE ( super- ) familyExpectedObservedEnrichment95th QuantileRathE1_cons5264 . 5610RathE3_cons293 . 236RathE2_cons152 . 524SINE372 . 007RC/Helitron3464441 . 28368DNA/MuDR1441841 . 27162ATREP245312 . 078RP1_AT22711 . 595ATTIRX1C11211 . 493ATREP132289 . 876VANDAL221118 . 563SIMPLEHAT11117 . 344VANDAL2N11107 . 323ATREP82136 . 475VANDAL2176 . 083ATREP101105 . 934AT9NMU1175 . 813ATN9_11105 . 754SIMPLEHAT21115 . 634META13205 . 417ATDNAI27T9A3154 . 836ATREP2A3154 . 836ATCOPIA78034 . 672VANDAL18NA034 . 672RathE1_cons5264 . 5610VANDAL14034 . 482SIMPLEGUY13134 . 196ATDNATA1034 . 002TNAT2A143 . 933ATREP74163 . 648RathE3_cons293 . 236ATREP14143 . 183ATREP16143 . 183LIMPET1392 . 906ATREP64142 . 898ARNOLDY27222 . 8513ATSINE4272 . 675ATDNAI27T9C272 . 426ATREP338922 . 3949ARNOLDY16142 . 2111ATREP113231 . 7319HELITRONY337511 . 3648 In order to gain further insight into the mechanisms responsible for variation in CHH methylation , we bisulfite-sequenced knockout lines of CMT2 ( SAIL_906_G03 ) and DCL3 ( dcl3-5 [Daxinger et al . , 2009] , a component of the RdDM pathway ) , and identified 10 , 138 DCL3-dependant DMRs and 33 , 422 CMT2-dependent DMRs as described in section ‘DMR calling on DNA methylation mutants’ of the ‘Materials and methods’ . As expected under the assumption that CMT2 is responsible for the massive GWAS peak on chromosome 4 , the GWAS peak at this locus remains if we consider only the CMT2-dependent DMRs , but not for DCL3-dependent DMRs ( Figure 5—figure supplement 1 ) . Furthermore , while CHH methylation varied with temperature at both DCL3- and CMT2-dependent DMRs ( Figure 5 ) , DCL3-dependent DMRs were much more strongly associated with previously identified temperature DMRs ( 4703 out of 10 , 138 DCL3-dependent DMRs , or 46% , overlapped temperature-sensitive DMRs , whereas the corresponding numbers for CMT2-dependent DMRs were 2299 out of 33 , 422 , or 7%; Fisher's exact p-value < 2 . 2e-16 ) , suggesting that much of the temperature variation in CHH methylation is due to components of the RdDM pathway . This result is consistent with previous findings showing that RNA silencing is less active at lower temperatures ( Romon et al . , 2013 ) . 10 . 7554/eLife . 05255 . 010Figure 5 . Temperature dependent CHH methylation variation at RdDM and CMT2 controlled DMRs . CHH methylation at CMT2- and DCL3-dependent DMRs in natural accessions grown at 10°C and 16°C ( cf . Figure 1A , each population has 110 individuals ) . The difference between temperatures was highly significant for both types of DMR ( Wilcoxon p-value = 9 . 1e-11 and p-value = 5 . 9e-12 respectively ) . Black points/grey lines indicate accessions with the CMT2 reference allele; green , the CMT2a non-reference allele; and orange , the CMT2b non-reference allele . Red is the TAA-03 accession , which has a putative null allele of CMT2 . Average methylation levels for each of the genotypes are shown in bars to the side Figure 5—figure supplement 1 shows GWAS on CMT2 and DCL3 dependant DMRs . Figure 5—figure supplement 2 shows a putative null allele of CMT2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 01010 . 7554/eLife . 05255 . 011Figure 5—figure supplement 1 . GWAS on CMT2 and DCL3 dependent DMRs . ( A ) GWAS for CMT2-dependent DMRs at 10°C . ( B ) GWAS on DCL3-dependent DMRs at 10°C . Results from 16°C were similar in both cases . The threshold line indicates a Bonferoni-corrected p-value of 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 01110 . 7554/eLife . 05255 . 012Figure 5—figure supplement 2 . Putative null allele of CMT2 . A screenshot from a genome browser indicating the lack of read coverage for CMT2 stretching from intron 7 to exon 16 in the accession TAA-03 . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 012 Interestingly , we observed one accession from northern Sweden , TAA-03 , with almost undetectable levels of CHH methylation at CMT2-dependant DMRs ( Figure 5 ) . Further investigation suggested that it has a deletion or rearrangement in CMT2 , as we were unable to map reads between positions 2813 and 4944 ( intron 7 to exon 16 , Figure 5—figure supplement 2 ) . Sanger-sequencing indicates the insertion of a stretch of TC dinucleotide repeats of at least 330 bp . The same deletion appears to be present in three more accessions from northern Sweden ( TAA-14 , TAA-18 , and Gro-3 ) a situation reminiscent of the homologous CMT1 gene , which seems to be non-functional in most Arabidopsis accessions ( Henikoff and Comai , 1998 ) . Although CMT2 null alleles have no obvious phenotype , the gene is highly conserved in plants ( with the exception of maize; Zemach et al . , 2013; West et al . , 2014 ) . It has recently been suggested that natural variation in CMT2 is associated with adaptation to climate ( Shen et al . , 2014 ) , although the alleles identified in that study do not overlap with the ones identified here . Given the sensitivity of CHH methylation to growth temperature observed here , we next investigated the correlation between DNA methylation and the climate of origin ( Hancock et al . , 2011 ) . While CHH methylation was moderately correlated with photosynthetically active radiation ( PAR ) in spring ( Pearson's r = 0 . 38 ) , and CHG showed correlation with aridity ( r = 0 . 35 ) and the number of frost-free days ( Pearson's r = 0 . 30 ) , by far the strongest signal was a strong positive correlation between CG methylation and latitude ( Pearson's r = 0 . 70 ) , as well as with a number of environmental variables that co-vary with latitude in our sample , such as minimum temperature and daytime length ( Table 2 , Figure 6A ) . As a result of the strong latitudinal correlation , accessions originating from northern Sweden ( minimum temperature below −10°C ) had on average 11% higher global CG methylation compared to those from the south ( Figure 6A ) . The correlation appears to be driven by gene body methylation ( GBM ) : as the correlation for CG methylation on transposons was much weaker ( Figure 6A , Figure 6—figure supplement 1 ) . Because the methylation variation observed for genes with average CG methylation below 5% appeared mostly to be noise ( Figure 6—figure supplement 2 , see also the ‘Materials and methods’ section ) , we classified genes into ‘unmethylated’ and ‘having GBM’ using this as a cutoff ( 5% ) . We also eliminated genes showing a transposon-like pattern of methylation in which not only CG , but also the CHH and CHG contexts are highly methylated ( Zemach et al . , 2013 ) . In what follows , we use GBM to refer only to gene body CG methylation for this filtered set . In order to better understand the observed variation in GBM , we examined CG methylation at the single nucleotide resolution within GBM containing genes . Although methylation was detectable ( using a cut-off of 1% ) at a similar number of sites in the north and south ( 1085292 vs 1079443 CG sites ) , those in the north showed a distinct skew towards higher methylation levels ( Figure 6—figure supplement 3 ) . Likewise when the difference between average methylation levels in the north and south was calculated individually for each of the cytosines , the majority of cytosines showed a small increase in the north compared to the south ( Figure 6—figure supplement 4 ) . From this we concluded that there is a general small increase in methylation of most CG dinucleotides in GBM genes , rather than large changes in a specific subset . GBM primarily occurs on long , evolutionary conserved genes that tend to be moderately-to-highly expressed , and is positively correlated with gene expression ( Zilberman et al . , 2007; Takuno and Gaut , 2012 ) . Genes with higher GBM tended to be more highly expressed in our data as well , and—more interestingly—accessions with higher average GBM showed slightly higher average expression of methylated genes ( although the correlation was weak , Figure 6—figure supplement 5 ) . Given that northern accessions had higher GBM , this meant that genes with GBM were on average more highly expressed in northern than in southern accessions , while unmethylated genes showed little difference ( Figure 6B ) . GBM has previously been shown to be anti-correlated with temperature-dependent gene expression ( Kumar and Wigge , 2010 ) . While no large-scale north-south expression differences were observed between 10°C and 16°C in our data , northern accessions showed considerably less variation in expression between the two temperatures for genes with GBM ( Wilcoxon p-value = 1 . 2e-05 ) , while no such difference was observed for genes without it ( Figure 6—figure supplement 6 ) . 10 . 7554/eLife . 05255 . 013Table 2 . Correlation between methylation levels and environment-of-origin variables ( Hancock et al . , 2011 ) DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 013Environmental variableGrowing temp . CGCHGCHHrrhop-valuerrhop-valuerrhop-valueLatitude100 . 690 . 527 . 8E-11−0 . 24−0 . 192 . 7E-020 . 100 . 141 . 1E-01160 . 620 . 473 . 2E-07−0 . 21−0 . 204 . 2E-020 . 04−0 . 112 . 5E-01Longitude100 . 590 . 541 . 2E-11−0 . 14−0 . 093 . 1E-010 . 230 . 287 . 5E-04160 . 550 . 534 . 4E-09−0 . 12−0 . 037 . 4E-010 . 140 . 151 . 2E-01Temperature seasonality100 . 680 . 491 . 6E-09−0 . 27−0 . 244 . 8E-030 . 090 . 092 . 8E-01160 . 620 . 421 . 1E-05−0 . 23−0 . 266 . 6E-030 . 04−0 . 122 . 1E-01Max . temp . ( warmest month ) 10−0 . 140 . 064 . 6E-01−0 . 07−0 . 131 . 3E-010 . 140 . 202 . 0E-0216−0 . 030 . 102 . 9E-01−0 . 10−0 . 203 . 8E-020 . 050 . 037 . 3E-01Min . temp . ( coldest month ) 10−0 . 70−0 . 569 . 1E-130 . 270 . 211 . 2E-02−0 . 07−0 . 064 . 7E-0116−0 . 63−0 . 482 . 7E-070 . 240 . 241 . 4E-020 . 000 . 195 . 6E-02Precipitation ( wettest month ) 100 . 450 . 521 . 2E-10−0 . 25−0 . 271 . 2E-03−0 . 20−0 . 121 . 7E-01160 . 290 . 434 . 0E-06−0 . 26−0 . 241 . 2E-02−0 . 22−0 . 195 . 8E-02Precipitation ( driest month ) 100 . 310 . 401 . 5E-06−0 . 33−0 . 296 . 5E-04−0 . 24−0 . 211 . 6E-02160 . 210 . 327 . 4E-04−0 . 26−0 . 241 . 4E-02−0 . 15−0 . 186 . 0E-02Precipitation seasonality100 . 420 . 447 . 1E-08−0 . 07−0 . 165 . 4E-020 . 050 . 019 . 0E-01160 . 360 . 371 . 2E-04−0 . 13−0 . 161 . 1E-010 . 01−0 . 019 . 1E-01PAR ( spring ) 100 . 040 . 228 . 9E-030 . 200 . 183 . 7E-020 . 240 . 237 . 3E-03160 . 030 . 186 . 6E-020 . 270 . 213 . 5E-020 . 380 . 352 . 8E-04Length of growing season10−0 . 59−0 . 575 . 5E-130 . 240 . 237 . 3E-03−0 . 16−0 . 183 . 3E-0216−0 . 58−0 . 544 . 0E-090 . 230 . 213 . 0E-02−0 . 040 . 018 . 9E-01No . consecutive cold days100 . 600 . 534 . 0E-11−0 . 19−0 . 131 . 2E-010 . 170 . 281 . 1E-03160 . 570 . 534 . 2E-09−0 . 17−0 . 093 . 7E-010 . 100 . 084 . 1E-01No . consecutive frost-free days10−0 . 59−0 . 491 . 2E-090 . 290 . 271 . 5E-030 . 020 . 037 . 1E-0116−0 . 51−0 . 394 . 9E-050 . 300 . 301 . 6E-030 . 070 . 131 . 9E-01Relative humidity ( spring ) 100 . 620 . 475 . 6E-09−0 . 23−0 . 183 . 9E-020 . 090 . 064 . 5E-01160 . 530 . 371 . 2E-04−0 . 20−0 . 267 . 6E-030 . 04−0 . 084 . 3E-01Daylength ( spring ) 100 . 690 . 507 . 2E-10−0 . 27−0 . 211 . 4E-020 . 080 . 055 . 7E-01160 . 630 . 411 . 5E-05−0 . 23−0 . 292 . 7E-030 . 04−0 . 178 . 7E-02Aridity100 . 530 . 498 . 4E-10−0 . 35−0 . 311 . 9E-04−0 . 18−0 . 211 . 3E-02160 . 430 . 428 . 4E-06−0 . 28−0 . 241 . 3E-02−0 . 13−0 . 203 . 8E-02r = Pearson's correlation , rho = Spearman's rank correlation , p-value = significance of rho . PAR = photosynthetically active radiation . 10 . 7554/eLife . 05255 . 014Figure 6 . Latitudinal difference in gene body methylation ( GBM ) and gene expression . ( A ) Global CG methylation levels at 10°C for 151 accessions are strongly correlated with minimum temperature at the location of origin . Results for 16°C are similar . ( B ) Genes with GBM are more highly expressed at 10°C in northern ( blue ) than in southern ( red ) accessions ( wilcoxon rank sum test p = 2 . 1e-03 ) , whereas genes without GBM show little difference ( p = 1 . 9e-02 ) . At 16°C the difference for genes with GBM is more significant ( p = 6 . 4e-05 ) , whereas the difference for genes without GBM is insignificant ( p = 0 . 49 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 01410 . 7554/eLife . 05255 . 015Figure 6—figure supplement 1 . Correlation between CG methylation levels and the minimum temperature at location of origin . Above , GBM at 10°C and 16°C . Below , TE CG methylation at 10°C and 16°C . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 01510 . 7554/eLife . 05255 . 016Figure 6—figure supplement 2 . Filtering of GBM variation data . ( A ) Genes with low or no CHG methylation have variable levels of CG methylation , while genes with appreciable CHG methylation have very high CG ( and CHH ) methylation . ( B ) Among genes with only CG GBM , variance-component analysis reveals a bimodal distribution of the total variance explained: variation in methylation for genes with low levels of methylation typically does not appear to have a genetic basis . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 01610 . 7554/eLife . 05255 . 017Figure 6—figure supplement 3 . Distribution of methylation levels at individual CG dinucleotides within GBM genes . The histogram shows the average methylation level for each individual CG dinucleotide on GBM genes in all accessions in the north ( blue ) or in the south ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 01710 . 7554/eLife . 05255 . 018Figure 6—figure supplement 4 . Distribution of variation in methylation levels between the north and the south for individual CG dinucleotides within GBM genes . The histogram shows the average methylation level for each individual CG dinucleotide on GBM genes in all accessions in the north minus the average methylation level in the south for each dinucleotide . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 01810 . 7554/eLife . 05255 . 019Figure 6—figure supplement 5 . Accessions with higher average GBM tend to have higher average expression ( of genes with GBM , normalized by genes without GBM; r = 0 . 131 , p = 0 . 0386 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 01910 . 7554/eLife . 05255 . 020Figure 6—figure supplement 6 . Genes with GBM show less expression variation between temperatures . Mean per-gene variation in expression between 10°C and 16°C is reduced for GBM containing genes in northern ( blue ) accessions compared to southern ( red ) accessions ( wilcoxon rank sum test p = 1 . 2e-05 ) , whereas for genes without GBM the difference between north ( light blue ) and south ( pink ) is insignificant . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 020 As for CHH DMRs , the genetic basis of GBM was examined using a variance-component approach ( Figure 7A ) . The results were dramatically different: relative to CHH methylation , the trans effects for GBM are massive , and the environment appears to have no effect ( in agreement with the observation that only CHH methylation levels vary with temperature , see Figure 1A ) . To identify the genes responsible , we also performed GWAS for each gene with GBM ( Figure 7B ) . A total of 3241 significant associations were found for 2315 genes . 43% of these genes had a significant cis-association ( within 100 kb of the gene of interest ) , which could represent local variants affecting methylation directly , or indirectly by affecting gene expression ( Gutierrez-Arcelus et al . , 2013 ) . No evidence for major trans-acting loci like CMT2 was found , but 69% of all significant associations were in trans . A comparison of the direction of the effect of GBM-associated SNPs in cis and trans revealed a striking pattern ( Figure 7C ) . While the non-reference alleles of cis-SNPs were 1 . 18 times more likely to be associated with decreased rather than increased GBM ( p = 2 . 01e-04 ) , the non-reference alleles of trans-SNPs were 3 . 48 times more likely to be associated with increased GBM ( p = 2 . 2e-16 ) , and the non-reference alleles at the 15 trans-SNPs that were associated with GBM at five or more genes were always positively correlated ( Figure 7C ) . Furthermore , while cis-SNPs showed a wide distribution of allele frequencies similar to random SNPs , trans-SNPs showed a much more limited distribution of frequencies ( Figure 8A ) and were also much more strongly correlated with latitude ( Figure 8B , C ) . The correlation between GBM and latitude thus appears mostly to be due to trans-acting SNPs . 10 . 7554/eLife . 05255 . 021Figure 7 . The genetic basis of GBM . ( A ) Variance component analysis of GBM . ( B ) Significant associations ( Bonferroni-corrected 0 . 05-level ) from a GWAS of GBM for individual genes . ( C ) Correlation between non-reference allele at associated SNPs and GBM . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 02110 . 7554/eLife . 05255 . 022Figure 8 . Frequency and distribution of GBM associated SNPs . ( A ) Correlation between non-reference allele at associated SNPs and latitude . ( B ) Non-reference allele frequency distribution for cis and trans SNPs compared to random SNPs . ( C ) Accessions carrying the non-reference alleles are limited to northern Sweden ( accessions with the non-reference allele at 8 or more of the 15 loci blue , remaining accessions are red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 02210 . 7554/eLife . 05255 . 023Figure 8—figure supplement 1 . Linkage disequilibrium between the 15 highly associated trans-SNPs . DOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 023 The 15 highly associated trans-SNPs were largely limited to northern Sweden , and were in strong linkage disequilibrium with each other ( Figure 8—figure supplement 1 ) . A . thaliana from northern Sweden show signs of multiple strong selective sweeps ( Long et al . , 2013 ) and harbors many polymorphisms that appear to be involved in local adaptation ( specifically to minimum temperature; Hancock et al . , 2011 ) . The 15 SNPs were more than ninefold over-represented in the previously identified sweep regions ( empirical p-value = 5 . 1e-03 ) and over fivefold over-represented within 2 kb of SNPs in the 1% tail of those associated with minimum temperature ( Hancock et al . , 2011 ) ( empirical p-value = 3 . 1e-04 ) , ( Table 3 ) . The ancestral state could be determined for 10 of the 15 SNPs , and in 8 of these cases , the non-reference allele was derived , further supporting sweeps in northern Sweden . 10 . 7554/eLife . 05255 . 024Table 3 . 15 SNPs associated with gene body methylation ( GBM ) at 5 or more genesDOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 024ChrPositionAssociated with GBM at how many genes ? Non-reference allele countSNP-latitude correlationOverlap with sweep ( Long et al . , 2013 ) Overlap with min . temp . Assoc . SNPs ( Hancock et al . , 2011 ) 19122918420 . 73none1_914088_0 . 21144051035660 . 64nonenone176141015480 . 66nonenone1197559675880 . 75none1_19757140_0 . 24269986316550 . 872_6931030none276550166810 . 612_7613651none276604699550 . 782_76136512_7662427_0 . 30276660595690 . 722_76136512_7665507_0 . 25276808825820 . 632_7613651none279157126510 . 83none2_7913782_0 . 23293824955730 . 71none2_9383856_0 . 34296538789480 . 80nonenone34193098660 . 68nonenone45199828660 . 70nonenone4132900345740 . 74nonenone That the difference in GBM between north and south is likely to reflect local adaptation is also clear from its relative magnitude . The north vs south divide explains a much higher fraction of the additive genetic variance for GBM ( Qst = 0 . 772; see ‘Materials and methods’ ) than of the SNP variance ( Fst = 0 . 187 ) . This strongly suggest that the phenotypic differentiation is driven by selection rather than genetic drift ( Leinonen et al . , 2013 ) . Identifying the causal variants is challenging , a gene-ontology analysis of genes within 100 kb ( the average size of the sweep regions , Long et al . , 2013 ) , of the 15 trans-SNPs found enrichment of loci associated with mRNA transcription ( GO0009299 , p-value = 2 . 62e-03 ) . Genes involved in epigenetic processes are not captured well by standard gene-ontology , but we found that genes from the plant chromatin database ( www . chromdb . org/ ) were significantly overrepresented in these regions as well ( permutation p-value = 0 . 012; Table 4 ) . 10 . 7554/eLife . 05255 . 025Table 4 . Genes in the plant chromatin database that are within 100 kb of one of the 15 SNPs associated with GBM at 5 or more genesDOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 025ChromDBLocusARID3AT2G17410ARP3AT1G13180CHB4AT1G21700CHR9AT1G03750CHR35AT2G16390CONS3AT3G02380DNG12AT1G21710FLCP39AT3G02310FLCP16AT2G22630FLCP9AT2G22540GTI1AT2G22720HMGB4AT2G17560JMJ27AT4G00990NFA1AT4G26110SDG23AT2G22740SDG37AT2G17900YDG2AT2G18000HON3AT2G18050 We also looked for genes whose expression variation is consistent with a causal role . We identified 68 genes within 100 kB of one of the 15-trans SNPs whose expression is highly correlated ( Wilcoxon test p < 0 . 001 ) with the adjacent SNP after correction for population structure ( Table 5 ) . No significant enrichment of Gene Ontology terms was observed among these genes , and manual inspection identified no proteins directly involved in DNA methylation . Instead , a number of proteins involved in the regulation of gene expression and/or chromatin accessibility were present ( Table 5 ) . This may suggest that the increased gene-body methylation observed in the north is not directly due to increased DNA methylation , but may be caused by increases in gene expression driven either by differences in transcription factors networks or chromatin compaction . Identification of the causal variants behind this phenomenon should provide insight into how plants adapt to their local environment . 10 . 7554/eLife . 05255 . 026Table 5 . Genes within 100 kb of the 15 SNPs associated with GBM at 5 or more genes whose expression is also correlated with the SNPDOI: http://dx . doi . org/10 . 7554/eLife . 05255 . 026SNPLocusDesciptionp-value1_19755967AT1G53030Encodes a copper chaperone4 . 72E-071_19755967AT1G52880NO APICAL MERISTEM ( NAM ) Transcription factor with a NAC domain5 . 47E-071_19755967AT1G52990Thioredoxin family protein2 . 36E-051_19755967AT1G52780Protein of unknown function ( DUF2921 ) 1 . 46E-041_4405103AT1G12750RHOMBOID-like protein 6 ( RBL6 ) ; FUNCTIONS IN: serine-type endopeptidase activity3 . 74E-081_4405103AT1G12790RuvA domain 2-like2 . 76E-051_4405103AT1G12730GPI transamidase subunit2 . 81E-051_4405103AT1G13080CYTOCHROME P450 FAMILY 71 SUBFAMILY B POLYPEPTIDE 2 ( CYP71B2 ) 1 . 65E-041_7614101AT1G21790TRAM LAG1 and CLN8 ( TLC ) lipid-sensing domain containing protein1 . 10E-051_7614101AT1G21900Encodes an ER-localized p24 protein8 . 81E-051_7614101AT1G21760F-BOX PROTEIN 7 ( FBP7 ) putative translation regulator in temperature stress response8 . 54E-041_912291AT1G03660Ankyrin-repeat containing protein1 . 26E-101_912291AT1G03770RING1B protein with similarity to polycomb repressive core complex1 ( PRC1 ) 5 . 76E-071_912291AT1G03940HXXXD-type acyl-transferase family protein1 . 18E-061_912291AT1G03610Protein of unknown function ( DUF789 ) 6 . 91E-061_912291AT1G03580Pseudogene with weak similarity to ubiquitin-specific protease 121 . 29E-051_912291AT1G03830Guanylate-binding family protein3 . 50E-052_6998631AT2G16340Unknown protein1 . 35E-082_6998631AT2G16210Transcriptional factor B3 family protein1 . 69E-042_7666059AT2G17630Pyridoxal phosphate ( PLP ) -dependent transferases superfamily protein2 . 47E-182_7660469AT2G17620Cyclin B2;1 ( CYCB2;1 ) 9 . 68E-072_7655016AT2G17740Cysteine/Histidine-rich C1 domain family protein1 . 22E-042_7655016AT2G17420NADPH-DEPENDENT THIOREDOXIN REDUCTASE 2 ( NTR2 ) 9 . 96E-042_7666059AT2G17430MILDEW RESISTANCE LOCUS O 7 ( MLO7 ) 7 . 56E-042_7915712AT2G18100Protein of unknown function ( DUF726 ) 1 . 73E-062_7915712AT2G17980ATSLY member of SLY1 Gene Family1 . 33E-052_7915712AT2G18400Ribosomal protein L6 family protein1 . 26E-042_7915712AT2G18150Haem peroxidase8 . 05E-042_7915712AT2G18050HISTONE H1-3 ( HIS1-3 ) 9 . 47E-042_9382495AT2G22260HOMOLOG OF E . COLI ALKB ( ALKBH2 ) enzyme involved in DNA methylation damage repair1 . 21E-082_9382495AT2G21850Cysteine/Histidine-rich C1 domain family protein5 . 38E-062_9382495AT2G22240MYO-INOSITOL-1-PHOSPHATE SYNTHASE 1 ( MIPS1 ) 8 . 71E-052_9382495AT2G21940SHIKIMATE KINASE 1 ( ATSK1 ) localized to the chloroplast1 . 80E-042_9653878AT2G22660Protein of unknown function ( duplicated DUF1399 ) 2 . 22E-142_9653878AT2G22900Galactosyl transferase GMA12/MNN10 family protein5 . 08E-092_9653878AT2G22830Squalene epoxidase 2 ( SQE2 ) 3 . 91E-062_9653878AT2G22640BRICK1 ( BRK1 ) 6 . 17E-052_9653878AT2G22540SHORT VEGETATIVE PHASE ( SVP ) Floral repressor involved in thermosensory pathway2 . 46E-042_9653878AT2G22570NICOTINAMIDASE 1 ( NIC1 ) 2 . 67E-042_9653878AT2G22770NAI1 Transcription factor7 . 71E-043_419309AT3G02220Protein of unknown function ( DUF2039 ) 2 . 06E-163_419309AT3G02230REVERSIBLY GLYCOSYLATED POLYPEPTIDE 1 ( RGP1 ) 4 . 58E-143_419309AT3G02300Regulator of chromosome condensation ( RCC1 ) family protein1 . 25E-103_419309AT3G02120Hydroxyproline-rich glycoprotein family protein1 . 81E-093_419309AT3G01980Short-chain dehydrogenase/reductase ( SDR ) 3 . 91E-093_419309AT3G02370Unknown protein4 . 53E-083_419309AT3G02020ASPARTATE KINASE 3 ( AK3 ) 4 . 18E-073_419309AT3G02160Bromodomain transcription factor2 . 60E-063_419309AT3G02390Unknown chloroplast protein5 . 60E-063_419309AT3G02050K+ UPTAKE TRANSPORTER 3 ( KUP3 ) 1 . 28E-053_419309AT3G02125Unknown chloroplast protein2 . 12E-053_419309AT3G02200Proteasome component ( PCI ) domain protein1 . 16E-043_419309AT3G02180SPIRAL1-LIKE3 Regulates cortical microtubule organization4 . 56E-043_419309AT3G02250O-fucosyltransferase family protein5 . 31E-043_419309AT3G02110Serine carboxypeptidase-like 25 ( scpl25 ) 6 . 18E-044_13290034AT4G26255Non-coding RNA of unknown function1 . 67E-134_13290034AT4G26450WPP DOMAIN INTERACTING PROTEIN 1 ( WIP1 ) 1 . 13E-044_13290034AT4G26230Ribosomal protein L31e family protein1 . 74E-044_13290034AT4G26160ATYPICAL CYS HIS RICH THIOREDOXIN 1 ( ACHT1 ) 5 . 72E-044_519982AT4G01090Protein of unknown function ( DUF3133 ) 1 . 23E-064_519982AT4G01230Reticulon family protein2 . 33E-054_519982AT4G01410Late embryogenesis abundant ( LEA ) hydroxyproline-rich glycoprotein family5 . 44E-054_519982AT4G01330Serine/threonine-protein kinase2 . 22E-044_519982AT4G01200Calcium-dependent lipid-binding ( CaLB domain ) family protein3 . 93E-044_519982AT4G01390TRAF-like family protein3 . 99E-044_519982AT4G01040Glycosyl hydrolase superfamily protein5 . 66E-044_519982AT4G01000Ubiquitin-like superfamily protein8 . 55E-04 In conclusion , genes with GBM are generally up-regulated and more heavily methylated in northern accessions , and the change appears to be due to trans-acting polymorphisms that have been subject to directional selection . The candidate regions show an overrepresentation of genes involved in transcriptional processes . We also found that CHH methylation of TEs is temperature sensitive , and identified a major trans-acting controller , CMT2 . Taken together , these observations suggest that local adaptation in A . thaliana involves genome-wide changes in fundamental mechanisms of gene regulation , perhaps as a form of temperature compensation .	Organisms need to adapt quickly to changes in their environment . Mutations in the DNA sequence of genes can lead to new adaptations , but this can take many generations . Instead , altering how genes are switched on by changing how the DNA is packaged in cells can allow organisms to adapt within and between generations . One way that genes are controlled in organisms is by a process known as DNA methylation , where ‘methyl’ tags are added to DNA and act as markers for other proteins involved in activating genes . DNA is made of four different molecules called ‘nucleotides’ that are arranged in different orders to produce a vast variety of DNA sequences . One type of DNA methylation can happen at sites where a nucleotide called cytosine is followed by two other non-cytosine nucleotides . Another type of methylation can take place at sites where a cytosine is followed by a guanine nucleotide . However , it is not clear how big a role DNA methylation plays in allowing organisms to adapt to their changing environment . Here , Dubin , Zhang , Meng , Remigereau et al . studied DNA methylation in a plant called Arabidopsis thaliana . Several different varieties of A . thaliana plants from Sweden were grown at two different temperatures . The experiments showed that the A . thaliana plants grown at higher temperatures were more likely to have methyl tags attached to sections of DNA called transposons , which are able to move around the genome . There was a lot of variety in the levels of this DNA methylation in the different plants , and some of it was shown to be associated with variation in a gene that is involved in DNA methylation . However , not all of the DNA methylation in these plants was sensitive to the temperature the plants were grown in . Dubin , Zhang , Meng , Remigereau et al . show that the pattern of a type of DNA methylation that is found within genes depends on how far north in Sweden the plants' ancestors came from rather than the temperature the plants were grown in . Plants that originated from colder regions , farther north , had more DNA methylation within many genes and these genes were more active . These findings suggest that genetic differences in these plants strongly influence the levels of DNA methylation , and they provide the first direct link between DNA methylation and adaption to the environment . Future studies should reveal how DNA methylation is regulated in these plants , and whether it plays a key role in adaptation , or merely reflects other changes in the genome .	[ "Abstract", "Main" ]	[ "plant", "biology", "genetics", "and", "genomics" ]	2015	DNA methylation in Arabidopsis has a genetic basis and shows evidence of local adaptation
Regulation of macromolecular interactions by phosphorylation is crucial in signaling networks . In the spindle assembly checkpoint ( SAC ) , which enables errorless chromosome segregation , phosphorylation promotes recruitment of SAC proteins to tensionless kinetochores . The SAC kinase Mps1 phosphorylates multiple Met-Glu-Leu-Thr ( MELT ) motifs on the kinetochore subunit Spc105/Knl1 . The phosphorylated MELT motifs ( MELTP ) then promote recruitment of downstream signaling components . How MELTP motifs are recognized is unclear . In this study , we report that Bub3 , a 7-bladed β-propeller , is the MELTP reader . It contains an exceptionally well-conserved interface that docks the MELTP sequence on the side of the β-propeller in a previously unknown binding mode . Mutations targeting the Bub3 interface prevent kinetochore recruitment of the SAC kinase Bub1 . Crucially , they also cause a checkpoint defect , showing that recognition of phosphorylated targets by Bub3 is required for checkpoint signaling . Our data provide the first detailed mechanistic insight into how phosphorylation promotes recruitment of checkpoint proteins to kinetochores . Protein kinases are ubiquitous and almost invariably crucial components of cellular signaling networks ( Huse and Kuriyan , 2002; Ubersax and Ferrell , 2007 ) . Kinases transfer a high-energy phosphate group from ATP to the side chains of serine , threonine , tyrosine , and more rarely those of other residues ( Hunter , 2012 ) . The addition of phosphate groups can modify the activity of a target protein directly or indirectly through the modification of its pattern of physical interactions , which in turn might modify the target’s activity , localization , or stability . Phosphorylation is usually a transient state that is reversed by the action of phosphatases . The transient nature of phosphorylation makes it ideally suited for use in signaling networks , where rapid activation and inactivation of defined substrates is important for the networks’ ability to toggle between alternative states . Phosphorylation plays a crucial role also in the spindle assembly checkpoint ( SAC ) , a signaling network required for accurate chromosome segregation during cell division ( Lara-Gonzalez et al . , 2012; Foley and Kapoor , 2013 ) . Checkpoint control creates a dependency between the mechanical aspects of cell division—the complex physical interaction of chromosomes with the mitotic spindle–and the timing of cell cycle progression . To prevent premature sister chromatid separation and mitotic exit in cells whose chromosomes have not yet attained bipolar attachment on the mitotic spindle , the SAC targets the cell cycle machinery required for the metaphase-to-anaphase transition ( Lara-Gonzalez et al . , 2012; Foley and Kapoor , 2013 ) . Work in Saccharomyces cerevisiae originally identified several checkpoint components , including Bub1 , Bub3 , Mad1 , Mad2 , Mad3/BubR1 , and Mps1 ( Hoyt et al . , 1991; Li and Murray , 1991; Hardwick et al . , 1996 ) , which were later found to be de facto ubiquitous in eukaryotes . Within this group , Bub1 and Mps1 are protein kinases . Together with all additional known checkpoint components , Bub1 and Mps1 become highly enriched at kinetochores between mitotic prophase and early prometaphase . Kinetochores are large protein assemblies , built on chromosomal loci known as centromeres . They bind directly to spindle microtubules to ensure the equational and reductional division of chromosomes during mitosis and meiosis , respectively ( Santaguida and Musacchio , 2009 ) . The dynamic interplay between kinetochore attachment to microtubules and checkpoint control is crucial for life in metazoans , but it remains disappointingly poorly understood . Among the targets of the Mps1 kinase activity is a kinetochore protein named Spc105/Knl1 ( also known as Spc7 , Blinkin , CASC5 in different organisms ) ( London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012 ) . Spc105/Knl1 is the largest subunit of a 10-subunit assembly , the KMN network , which is believed to provide the main site of attachment of kinetochores to microtubules ( Figure 1A , B ) ( reviewed in Santaguida and Musacchio , 2009 ) . Within Spc105/Knl1 , Mps1 phosphorylates at least a subset of an array of motifs that are generally referred to as ‘MELT’ and that conform to the consensus M-[E/D]-[L/I/V/M]-T ( Figure 1C; we indicate as MELTP the phosphorylated form of a MELT motif ) . The presence of multiple MELT repeats is an essentially invariant feature of Spc105/Knl1 in evolution ( Cheeseman et al . , 2004 ) . 10 . 7554/eLife . 01030 . 003Figure 1 . Reconstitution of the interaction of Bub1-Bub3 with MELTP motifs . ( A ) Schematic description of the domain and motif organization of the main players discussed in this paper . ( B ) The KMN network ( shown in different tones of blue ) consists of the Ndc80 complex ( NDC80-C ) , the Mis12 complex ( MIS12-C , also known as MIND complex ) , and Spc105/Knl1 ( which also associates with Ydr532cp/Zwint , not shown here ) . Mps1 phosphorylates the MELT repeats of Spc105/Knl1 to promote the recruitment of the Bub1–Bub3 complex . A Bub3-binding domain of Bub1 is shown in orange . ( C ) Sequence of MELT repeats in Spc105/Knl1 of S . cerevisiae . Arrowheads indicate MELT repeats previously shown to be phosphorylated by Mps1 in vitro ( London et al . , 2012 ) . The MELT motifs are shown in red . ( D ) Purified Bub3 , Bub1289–359–Bub3 , and mutants thereof discussed in the text were separated by SDS-PAGE after purification . ( E ) Isothermal titration calorimetry ( ITC ) analysis of the interaction of Bub1289–359–Bub3 with a synthetic peptide corresponding to the phosphorylated version of the second MELTP peptide ( MELT2P ) shown in C . ( F ) ITC analysis , with the unphosphorylation version of the same peptide ( MELT2 ) , shows no binding . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 00310 . 7554/eLife . 01030 . 004Figure 1—figure supplement 1 . Additional calorimetry experiments . ( A ) Isothermal titration calorimetry ( ITC ) analysis of the interaction of Bub1289–359–Bub3 with a synthetic peptide corresponding to the fourth MELTP peptide shown in Figure 1C . ( B ) ITC analysis with the unphosphorylation version of the same peptide shows no binding . ( C ) ITC analysis of the interaction of Bub1289–359–Bub3 with a synthetic peptide with sequence GGGPATPPKKAKKL , which encompasses a segment of histone H1 that is phosphorylated by cyclin-dependent kinase activity . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 004 How the phosphorylation on MELT motifs is interpreted by downstream components of the checkpoint pathway is unclear . Bub1 and Bub3 , a 7-bladed WD40-repeat β-propeller that is constitutively bound to Bub1 ( Figure 1A ) , are robustly recruited to Spc105/Knl1 when the MELT repeats are phosphorylated ( London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012 ) , in line with previous observations linking Mps1 kinase activity to kinetochore recruitment of Bub1 and Bub3 ( Vanoosthuyse et al . , 2004; Vigneron et al . , 2004; Kiyomitsu et al . , 2007 , 2011; Pagliuca et al . , 2009; Schittenhelm et al . , 2009; Maciejowski et al . , 2010; Santaguida et al . , 2010; Ito et al . , 2011; Storchová et al . , 2011; Heinrich et al . , 2012 ) . However , whether Bub1 and Bub3 are sufficient for a tight interaction with MELTP repeats is currently unknown , and so is , therefore , the identity of the binding site for MELTP ( Figure 1B ) . Here , we show that Bub3 binds directly and with high affinity to MELTP motifs . The crucial determinants of this interaction are extremely well conserved in evolution and are required for a functional checkpoint . We discuss the recruitment mechanism of Bub1–Bub3 and its implications for checkpoint signaling . The constellation of Bub3 residues implicated in MELTP binding is perfectly conserved in the nucleoporin Rae1 , suggesting that Rae1 might also be implicated in phosphopeptide binding . Bub1 binds Bub3 through a conserved Bub3-binding domain ( Taylor and McKeon , 1997 ) that is often also referred to as GLEBS motif ( Bailer et al . , 1998; Wang et al . , 2001 ) ( Figure 1A ) . The Bub3-binding domain of Bub1 is necessary for kinetochore recruitment of Bub1 . When expressed in isolation in human cells , this region of Bub1 is sufficient to mediate robust kinetochore recruitment of Bub1 , albeit at partly reduced levels compared to constructs that also include the N-terminal TPR domain of Bub1 ( Taylor and McKeon , 1997; Vanoosthuyse et al . , 2004; Klebig et al . , 2009; Krenn et al . , 2012 ) . The minimal region of human Bub1 , capable of mediating kinetochore targeting , consists of residues 209–270 ( equivalent to residues 289–359 of Bub1 in S . cerevisiae ) . Additional deletions of this segment prevent kinetochore binding ( Krenn et al . , 2012 ) . Because the minimal kinetochore recruitment domain of Bub1 coincides with the Bub3-binding domain , and because it is known that Bub1 and Bub3 reinforce each other in kinetochore localization ( Taylor and McKeon , 1997; Taylor et al . , 1998; Sharp-Baker and Chen , 2001; Millband and Hardwick , 2002; Gillett et al . , 2004; Kadura et al . , 2004; Vanoosthuyse et al . , 2004; Rischitor et al . , 2006; Logarinho et al . , 2008; Klebig et al . , 2009; Windecker et al . , 2009; Krenn et al . , 2012; London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012 ) , it is expected that Bub1 and Bub3 cooperate in the mechanism of kinetochore recruitment . In vitro reconstitution with recombinant purified material is often crucial to address the molecular mechanism of protein interactions . Thus , we attempted the reconstitution of the phosphorylation-dependent recruitment of Bub1–Bub3 to kinetochores . We chose to work with proteins from Saccharomyces cerevisiae because the Bub1–Bub3 complex has already been reconstituted in this organism ( Larsen et al . , 2007 ) and also because the identity of phosphorylated MELT repeats in ScSpc105 ( we will refer to ScSpc105 as Spc105/Knl1 for the remains of this work ) is known from previous work ( London et al . , 2012 ) ( Figure 1C ) . We generated recombinant ScBub1289–359–Bub3 ( where the Bub1 segment is equivalent to the minimal kinetochore targeting region of human Bub1 [Krenn et al . , 2012] ) by bacterial co-expression and purified it to homogeneity ( Figure 1D ) . To assess whether Bub1–Bub3 binds directly to MELTP sequences , we tested its ability to bind MELT sequences in quantitative isothermal titration calorimetry ( ITC ) binding experiments . 19-residue synthetic peptides encompassing the sequences of the second and fourth MELTP motifs of Spc105/Knl1 ( indicated as MELT2 and MELT4 , respectively ) , each flanked by four and eleven residues on the N- and C-terminal ends , respectively , were tested ( Figure 1C ) . ScBub1289–359-Bub3 bound the MELT2P and MELT4P peptides with dissociation constants ( KD ) of 200 nM ( Figure 1E ) and 1 . 3 µM , respectively ( Figure 1—figure supplement 1 ) . Remarkably , no binding was observed with the non-phosphorylated versions of the MELT2 and MELT4 peptides ( Figure 1F and Figure 1—figure supplement 1 ) , indicating exquisite selectivity for the phosphorylated MELT motifs . Conversely , Bub3 did not show any binding affinity for an unrelated phospho-peptide ( Figure 1—figure supplement 1 ) . Thus , ScBub1289–359–Bub3 is sufficient for the reconstitution of tight interactions with two MELTP peptides in vitro that recapitulate a salient feature of this interaction , its dependency on phosphorylation . We crystallized the ScBub1289–359–Bub3-MELT2P ternary complex and determined its structure by X-ray crystallography to a resolution of 1 . 9 Å by molecular replacement with ScBub1315–356–Bub3 as a search model ( PDB ID 2I3S; Larsen et al . , 2007 ) ( Figure 2A–B ) . The model extends to the two ternary complexes in the asymmetric unit and was refined to a ‘free’ R-factor ( Rfree ) of 19 . 2% , with excellent stereochemical parameters ( Table 1 ) . The two trimers in the asymmetric unit are very similar , and their salient features can be described essentially equivalently . 10 . 7554/eLife . 01030 . 005Figure 2 . Structure and conservation of the Bub1–Bub3 complex . ( A ) Top view of the Bub1289–359–Bub3-MELTP ternary complex . N and C indicate the N- and C-terminus , respectively . ( B ) Side view of the ternary complex . ( C ) Sequence alignment of Bub3 from the indicated species . The presented alignment was extracted from a much larger alignment consisting of more than 40 Bub3 sequences from distant eukaryotes ( Vleugel et al . , 2012 ) . The indicated levels of conservation were derived from the larger alignment . Green asterisks indicate residues predicted , on structural ground , to be important for the stability of the Bub3 propeller . Orange asterisks point to residues that interact with Bub1 . Red asterisks point to residues that interact with the MELTP peptide . Black asterisks point to conserved residues of uncertain function . BL = blade . ( D ) Mapping of secondary structure elements on the sequence of the Bub3-binding domain of Bub1 . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 00510 . 7554/eLife . 01030 . 006Figure 2—figure supplement 1 . Comparison of Bub3–Bub1 and Bub3–Mad3 structures with Rae1-Nup98 . ( A ) Cartoon representations of the Bub1–Bub3-MELTP complex ( the MELT peptide was omitted for clarity ) shows that the β1 strand of the Bub3-binding domain of Bub1 pairs with the β2 strand in the ‘roof’ domain . ( B ) Crystal structure of the Bub1315–356:Bub3 complex from Saccharomyces cerevisiae ( Larsen et al . , 2007 ) . The region corresponding to the β1 strand was absent from this construct ( see panel E ) . ( C ) Crystal structure of the Mad3354–401:Bub3 complex ( Larsen et al . , 2007 ) . Also in this case , the sequence encoding the β1 strand was missing . ( D ) Crystal structure of the Nup98-Rae1 complex shows a striking similarity with the Bub1–Bub3 structure . A sequence alignment for the Rae1 and Bub3 sequences is shown in Figure 2C . ( E ) Sequence alignment of GLEBS motifs is included in the different crystal structures . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 00610 . 7554/eLife . 01030 . 007Figure 2—figure supplement 2 . Composite omit maps of the region corresponding to the phospho-MELT peptide . ( A ) The panel shows the region around chain C . ( B ) The panel shows the region around chain F . In both cases , maps were calculated with Fourier coefficients 2mF0-DFc and contoured to 1 . 5σ , and clipped to limit the map to the region around the respective chain . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 00710 . 7554/eLife . 01030 . 008Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 008Data collectionWavelength ( Å ) 1 . 21Resolution range ( Å ) 46 . 8–1 . 95 ( 2–1 . 95 ) Space groupC2Unit cella = 138 . 7; b = 57 . 9; c = 118 . 7; α = 90 β = 102 . 5 γ = 90 α = γ = 90° β = 102 . 5°Total reflections438568Unique reflections66488Multiplicity6 . 6 ( 6 . 3 ) Completeness ( % ) 98 . 70 ( 88 . 81 ) Mean I/sigma ( I ) 14 . 15 ( 3 . 09 ) Wilson B-factor26 . 16Rsym0 . 069 ( 1 . 023 ) CC ( 1/2 ) 99 . 9 ( 87 . 3 ) RefinementR-factor0 . 1726 ( 0 . 2552 ) R-free0 . 1916 ( 0 . 2864 ) Number of atoms6412 Macromolecules6052 Metal ions2 Water358Protein residues775Average B-factor36 . 40 Macromolecules36 . 00 Solvent42 . 20GeometryRMS ( angles , Å ) 0 . 89RMS ( bonds , ° ) 0 . 006Ramachandran favored ( % ) 97Ramachandran outliers ( % ) 0MolProbity score†1 . 29 ( 99th percentile ) Statistics for the highest-resolution shell are shown in parentheses . Percentage of correlation between intensities from random half-datasets ( Karplus and Diederichs , 2012 ) . †MolProbity score combines the clashscore , rotamer , and Ramachandran evaluations into a single score , normalized to be on the same scale as X-ray resolution ( Chen et al . , 2010 ) . Each blade of the 7-bladed Bub3 β-propeller consists of four β-strands , with the innermost and outermost strands referred to as βA and βD , respectively . Most intra- and inter-blade loops in Bub3 are short , giving rise to a rather regular toroid . The two notable exceptions are the βD5-βA6 and βB7-βC7 loops ( Figure 2A–C ) , both of which interact extensively with Bub1 . As shown previously ( Larsen et al . , 2007 ) , Bub1 meanders on the top surface of the Bub3 β-propeller ( defined as the surface that contains the βD–βA loops that connect consecutive blades ) . The fragment of ScBub1 contained in our crystals , however , is 29 residues longer ( residues 289–315 ) at its N-terminus relative to the one previously co-crystallized with ScBub3 ( Larsen et al . , 2007 ) ( Figure 2—figure supplement 1 ) . Residues in this extension contribute to the formation of a β-hairpin ( β1-β2 , Figure 2A and 2D ) , which pairs , via β2 , with a β-hairpin within the extended βD5-βA6 loop of Bub3 . Together , the β-hairpins from Bub1 and Bub3 form a joint 4-stranded β-sheet that creates a ‘roof’ on the MELTP peptide . Bub3 plays a dominant role at the interface with the MELT2P peptide ( Figures 2 and 3 ) . The latter ( for which there is excellent electron density between residues 166–176 [Figure 2—figure supplement 2] ) docks on blades 4–6 of the Bub3 β-propeller , with its main chain oriented almost orthogonally to the vertical axis of the propeller’s toroid . This docking mode , which is unprecedented in β-propeller-peptide interactions ( Figure 3—figure supplements 1 and 2 ) , is accompanied by the formation of at least five hydrogen bonds between the main chain atoms of the peptide and of the βD-strands of blades five and six and of the βD4-βA5 loop ( Figure 3A ) . 10 . 7554/eLife . 01030 . 009Figure 3 . The interface between Bub1–Bub3 and MELTP . ( A ) The MELTP peptide ( here shown with carbon atoms in light gray color ) orients transversally to the blades but its main chain amide and carbonyl groups form several hydrogen bonds with the main chain of the outermost strands of three consecutive blades , blades four to six . ( B ) Details of the interaction around MELTP sequence . The boxed regions are enlarged in panels C and D . ( E–H ) Surface representation of sequence conservation ( resulting from the alignment discussed in the legend of Figure 2C ) shows a dramatic concentration at the interface with MELTP . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 00910 . 7554/eLife . 01030 . 010Figure 3—figure supplement 1 . A collection of modes of ligand binding by β-propellers . ( A–H ) We manually scanned the PDB for β-propeller structures in complex with peptide ligands . Each interaction is shown with a side and a top view . The molecular species involved and the corresponding PDB code are indicated . Of note , the Cdc4-Skp1-Cyclin E complex contains a phosphorylated Cyclin E peptide . It binds to the top of the Cdc4 β-propeller domain , not on the side as in Bub3 . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 01010 . 7554/eLife . 01030 . 011Figure 3—figure supplement 2 . Mode of binding of phosphopeptides from Cyclin E and β-catenin to the β-propellers of Fbw7 and β-TrCP . ( A and B ) Top and side view of the Fbw7/Skp1/Cyclin E2P complex . The di-phosphorylated peptide is accommodated on a binding site that involves the top surface of the 8-bladed WD40 β-propeller domain of Fbw7 ( Hao et al . , 2007 ) . ( C and D ) Top and side view of the β-TrCP/Skp1/β-catenin2P complex . Also in this case , a di-phosphorylated peptide is accommodated on a binding site that involves the top surface of the 7-bladed β-propeller domain of β-TrCP ( Wu et al . , 2003 ) . The scale of figures in panels B and D is smaller in comparison to the models in A and C . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 01110 . 7554/eLife . 01030 . 012Figure 3—figure supplement 3 . Electrostatics on the Bub3 surface at the MELTP interface . ( A ) The charge distribution around the MELTP binding site of Bub3 is shown to be highly positive . The electrostatic potential at the surface ( in units of Volts ) was displayed within boundaries of −1 . 0 to 1 . 0 . ( B ) A close-up of the MELT-binding region boxed in A . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 012 The side chains of the MEMTP motif are also extensively involved in the interaction with Bub3 . The binding site on Bub3 is essentially bipartite , with a highly hydrophobic ‘south’ interface interacting with the hydrophobic side chains of Met169MELT and Met171MELT , and a highly positively charged ‘north’ interface contacting the acidic side chains of Glu170MELT and P-Thr172MELT ( Figure 3B ) Specifically , at the south end , the side chains of Met169MELT and of Met171MELT are embedded in a deep hydrophobic pocket lined up by the side chains of Phe236Bub3 , Phe238Bub3 , Trp278Bub3 , and by the aliphatic portion of the side chain of Arg283Bub3 ( Figure 3C ) . At the north end , the side chains of Glu170MELT and P-Thr172MELT face Arg217Bub3 , Arg239Bub3 , and Arg242Bub3 , which are therefore ideally positioned to compensate the negative charge of the phosphopeptide ( Figure 3D , Figure 3—figure supplement 3 ) and are at the core of a complex network of hydrogen bonds that also engages Glu317Bub1 . Within this array of residues , Arg242Bub3 , in the βD5-βA6 loop , faces Glu170MELT and has additional stabilizing effects on the side chains of Arg217Bub3 and Arg239Bub3 , which face P-Thr172MELT ( Figure 3D ) . Taken together , these interactions explain the positive discrimination by Bub1289–359–Bub3 for phosphorylated versions of the MELT peptides . Finally , at the ‘west’ end of the binding site , the aromatic side chain of Phe175MELT2 ( +3 position relative to P-Thr ) stacks against the side chain of Lys193Bub3 , which is held in position by the side chain of Tyr194Bub3 ( not shown ) . Both Bub3 residues are exposed and invariable in evolution ( Figure 2C ) , suggesting that they play an important functional role , but there is no strong preference for phenylalanine or other hydrophobic residues in the sequence of MELT repeats at the +3 position ( Figure 1C ) . It is also possible that Lys193Bub3 and Tyr194Bub3 are required to stabilize the interaction with Bub1 , whose Ile309Bub1 , Ile319Bub1 , and Phe323Bub1 are in direct van der Waals contact with the side chain of Tyr194Bub3 ( not shown ) . A plot of sequence conservation on the surface of the Bub1–Bub3 complex ( Figure 3E–H ) shows an extreme concentration of conserved residues at the interface with the Spc105/Knl1 peptide . The level of conservation at this site even exceeds the conservation of Bub1-binding residues ( Figure 2C ) . Thus , binding to phosphorylated sequences is a crucial property of Bub3 . Overall , the pattern of sequence conservation strongly suggests that binding to MELTP and Bub1 might be the only two widely conserved functions of Bub3 . Because the presence of a phosphate on Thr172Spc105/Knl1 is essential for high-affinity binding of ScBub1289–359–Bub3 to MELTP peptides ( Figure 1 ) , we concentrated our mutational analysis on two residues that are directly implicated in the recognition of the peptide’s phospho-threonine , Arg217Bub3 and Arg239Bub3 . Positively charged residues are invariant at these positions of the Bub3 alignment ( Figure 2C ) . We generated single or double alanine mutants of Arg217Bub3 and Arg239Bub3 in the context of ScBub1289–359–Bub3 and tested their binding affinity for the MELT2P peptide by ITC . Importantly , the mutant complexes were expressed and purified essentially like the wild type complex and did not suffer obvious losses of stability ( Figure 1D ) . Replacement of Arg217 or Arg239 with alanine caused a 13- to 25-fold reduction in the binding affinity for the MELT2P peptide , with KDs of 2 . 7 µM and 5 µM for the R217A and R239A mutants ( Figure 4A , B ) , respectively , compared with 200 nM for the wild type interaction ( Figure 1E ) . Thus , neither arginine side chain is completely indispensable for binding , but each is required for high-affinity binding . When the mutations were combined in the Bub3R217A–R239A double mutant , no significant residual binding to the MELTP peptide was observed ( Figure 4C ) . Collectively , the mutational analysis is in line with the observation that the binding of ScBub1289–359–Bub3 to the MELT2P peptide is exquisitely phosphorylation-sensitive . 10 . 7554/eLife . 01030 . 013Figure 4 . Biochemical validation of the interaction . ( A ) ITC analysis of the interaction of Bub1289–359–Bub3R217A with a synthetic peptide encompassing the MELT2P sequence . ( B–D ) ITC experiments with the MELT2P peptide and Bub1289–359–Bub3R239A , Bub1289–359–Bub3R217A–R239A double mutant , and Bub3 , respectively . ( E ) Close-up of the MELTP binding site indicating possible roles of the Bub3-binding motif of Bub1 . The amino acid sequence of the Bub3-binding domain of Bub1 is reported in Figure 2D . ( F ) ITC experiment with the MELT2P peptide and the Bub1289–359–R314A–Bub3 complex . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 013 The majority of residues involved in the interaction with the MELTP motifs of Spc105/Knl1 are located in Bub3 , indicating that the latter plays the prominent role in kinetochore recruitment of Bub1 . However , as already anticipated in the Introduction , several lines of evidence indicate that Bub1 contributes to this interaction ( Discussion ) . To test this possibility formally , we measured the binding of purified Bub3 ( Figure 1D ) to the MELT2P peptide in the absence of Bub1 . Remarkably , we observed a 10-fold reduction in the binding affinity of Bub3 for the MELT2P peptide ( KD = 2 µM; Figure 4D ) compared to the Bub1–Bub3 complex ( Figure 1E ) , indicating that Bub1 does indeed positively contribute to the interaction . Such function of Bub1 is probably exerted primarily through its structuring effects on the 4-stranded β-sheet ‘roof’ that dominates the peptide-binding region of Bub3 and which restrains the position of the positively charged residues in the ‘north’ area of the MELTP-binding site ( Figure 4E ) . Additionally , we observe that Arg314Bub1 , in the β1-β2 loop , contributes to the interaction with the phosphate group of P-Thr172Spc105/Knl1 ( the interaction , however , is only observed in one of the two complexes in the asymmetric unit , and in proximity of a crystal contact ) and is therefore directly engaged in the interaction with the MELT2P peptide . In ITC measurements , we observed a ∼fourfold reduction in the binding affinity of the Bub3-Bub1R314A mutant for the MELT2P peptide , in agreement with a role of Arg314Bub1 is in the interaction of the Bub1-Bub3 complex with MELT2P ( Figure 4F ) . Similarly , Bub3–Bub1R314A bound the MELT4P with ∼threefold decreased affinity compared to wild type Bub3–Bub1 ( data not shown ) . Next , we asked if mutations in Bub3 that prevent its interaction with MELT2P motifs in vitro also affected its recruitment to kinetochores . To this end , we inserted three copies in tandem of the coding sequence for mCherry in frame at the 3′ end of the coding sequence of S . cerevisiae BUB3 or of the bub3R217A–R239A mutant . The transgenes were inserted at the TRP1 locus of a bub3Δ strain . The resulting strains were viable and expressed similar levels of wild type or mutant Bub3 ( Figure 5—figure supplement 1 ) . To assess if the Bub3-mCherry localized to kinetochores , we tested its co-localization with Mtw1 , a subunit of the MIS12 kinetochore complex ( MIS12-C , also known as MIND complex , Figure 1B ) . Unsynchronized S . cerevisiae cells expressing tagged versions of Bub3 and of the kinetochore subunit Mtw1 ( Bub3-mCherry and Mtw1-GFP ) were imaged in a flow cell by wide-field fluorescence microscopy . Bub3-mCherry appeared to co-localize with Mtw1 shortly before budding and until approximately metaphase ( Figure 5A , Video 1 ) , in agreement with a previous study ( Gillett et al . , 2004 ) . This behavior of Bub3-mCherry was formalized , for each video frame , through computation of a ‘localization index’ whose peaks coincide with kinetochore recruitment ( Figure 5B , ‘Materials and methods’ , Figure 5—figure supplement 2 for details ) . Thus , as shown previously ( Gillett et al . , 2004 ) , Bub3-mCherry localizes to kinetochores during each cell cycle in unperturbed S . cerevisiae cells . In contrast to Bub3-mCherry , Bub3R217A–R239A-mCherry never co-localized with Mtw1-GFP during the cell cycle , indicative of defective kinetochore recruitment ( Figure 5C–D , Video 2 ) . This behavior of Bub3R217A–R239A-mCherry agrees with the inability of the recombinant Bub3 mutant to interact with phosphorylated MELT repeats in vitro ( Figure 4C ) . 10 . 7554/eLife . 01030 . 014Figure 5 . Mutant Bub3 does not localize to kinetochores and mislocalizes Bub1 . ( A ) Live S . cerevisiae cells expressing wild type Bub3-mCherry and Mtw1-GFP where filmed to assess kinetochore localization of the fluorescent proteins ( Video 1 ) . Selected frames are shown . ( B ) A localization index was calculated as discussed in ‘Materials and Methods’ . High values of the index indicate recruitment of Bub3-mCherry to kinetochores . ( C ) As in panel A , but using cells expressing Bub3R217A–R239A-mCherry . ( D ) Localization index for Bub3R217A–R239A-mCherry . The localization index for the Bub3 mutant fluctuates around the value of 3 . 7 , which we identify as corresponding to ‘perfect delocalization’ ( ‘Materials and methods’ ) . ( E ) Selected frames from Video 3 demonstrating kinetochore localization of Bub3-mCherry and Bub1–GFP . ( F ) Peaks in the localization index indicate the timing of kinetochore recruitment of Bub3-mCherry and Bub1–GFP during subsequent cell cycles . Red and blue curves report localization of Bub3-mCherry and Bub1–GFP , respectively . The time of initiation of budding is marked by black squares . The diagram extends to three budding events . There is excellent correlation of the Bub1 and Bub3 signal , indicative of co-localization . ( G ) Bub3R217A–R239A-mCherry does not localize to kinetochores ( Video 2 ) . Bub1–GFP fails to localize to kinetochores in Bub3R217A–R239A-mCherry cells ( selected frames from Video 4 ) , in agreement with the role of Bub3 in kinetochore recruitment of Bub1 ( Gillett et al . , 2004 ) . ( H ) The localization index for Bub3R217A–R239A-mCherry and Bub1–GFP is flat , close the numerical value corresponding to delocalization in wild type cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 01410 . 7554/eLife . 01030 . 015Figure 5—figure supplement 1 . Expression levels of Bub3 and Bub3 mutants in bub3Δ Saccharomyces cerevisiae’s cells . Western blotting of total cell lysates with an anti-mCherry antibody was used to detect the expression of the Bub3-mCherry construct in bub3Δ cells . TCA extracts where carried out according to ( Mariani et al . , 2012 ) . Pgk1 was used as loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 01510 . 7554/eLife . 01030 . 016Figure 5—figure supplement 2 . Validation of the localization index . ( A ) The overall intensity distribution for a diffuse fluorescent protein describes a Gaussian . In this case , the localization index defined in ‘Materials and methods’ adopts a value of ∼3 . 7 . ( B ) When the fluorescent protein is kinetochore-localized , the distribution is skewed-Gaussian , with a more extended right tail . In this case , our measure for localization is larger than ∼3 . 7 . When the localization index assumes values that are significantly higher than ∼3 . 7 , we observed the fluorescent protein to be localized to kinetochores , as confirmed by co-localization with Mtw1 ( Figure 5 ) . ( C–E ) distribution of maxima ( C ) , minima ( D ) , and standard deviation ( E ) of the localization index of Bub3wt-mCherry or Bub3R217A–R239A-mCherry over an entire cell cycle ( i . e . , the time between two budding events ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 01610 . 7554/eLife . 01030 . 017Video 1 . Localization of Mtw1-GFP ( left ) and Bub3wt-mCherry ( right ) in replicating S . cerevisiae’s cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 01710 . 7554/eLife . 01030 . 018Video 2 . Localization of Mtw1-GFP ( left ) and diffuse localization of Bub3R217A–R239A-mCherry ( right ) in replicating S . cerevisiae’s cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 018 Collectively , these results indicate that the integrity of the MELTP binding site of Bub3 is essential for its kinetochore recruitment . Because Bub1 interacts with Bub3 , we asked if its pattern of kinetochore localization was similar to that of Bub3-mCherry ( Figure 5E , Video 3 ) . Indeed , the kinetochore localization indexes for Bub3-mCherry and of a Bub1–GFP construct peaked at the same time ( Figure 5F ) . Thus , also Bub1–GFP localizes to kinetochores during an unperturbed cell cycle in S . cerevisiae . 10 . 7554/eLife . 01030 . 019Video 3 . Localization of Bub1-GFP ( left ) and Bub3wt-mCherry ( right ) in replicating S . cerevisiae’s cells . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 019 Next , we tested if Bub1–GFP localized to kinetochores in cells expressing Bub3R217A–R239A-mCherry in bub3Δ cells . Kinetochore localization of Bub1–GFP was completely suppressed in these cells ( Video 4 , Figure 5G ) , and the kinetochore localization index was correspondingly flat ( Figure 5H ) . In summary , these observations provide a clear demonstration of the fact that the interaction of Bub3 with MELTP motifs is crucial for the kinetochore recruitment of the Bub1–Bub3 complex . 10 . 7554/eLife . 01030 . 020Video 4 . Diffuse localization of Bub1–GFP ( left ) in replicating S . cerevisiae’s cells expressing Bub3 R217A–R239A-mCherry ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 020 Next , we asked if the mutations in Bub3 that affect Spc105/Knl1 binding in vitro or in vivo also affect the ability of S . cerevisiae cells to activate the spindle checkpoint . Cells arrested in G1 with α-factor were released in the cell cycle in the presence of nocodazole to activate the spindle checkpoint . To assess checkpoint proficiency , we monitored the ability of cells to arrest in mitosis and to prevent re-replication , as well as lack of rebudding . Wild type cells and bub3Δ expressing Bub3-mCherry cells completed DNA replication at ∼60 min after release from the G1 block in nocodazole and arrested as budded cells with 2C DNA content ( Figure 6 , panels A , C and E ) , indicative of a functional SAC . Conversely , bub3Δ cells and bub3Δ expressing Bub3R217A–R239A-mCherry cells were unable to arrest , re-replicated their DNA , and re-budded , indicative of a disrupted SAC ( Figure 6 , panels B , D and E ) . These observations demonstrate that the spindle checkpoint is disrupted when the ability of Bub3 to interact with MELTP sequences is impaired . 10 . 7554/eLife . 01030 . 021Figure 6 . Mutant Bub3 cannot sustain the checkpoint . ( A ) G1-arrested wild type S . cerevisiae cells were released in the cell cycle in the presence of nocodazole . FACS analysis at the indicated time points shows that cells first undergo DNA replication and subsequently arrest with 2C DNA content , indicative of mitotic checkpoint arrest . ( B ) bub3Δ cells are checkpoint deficient , fail to arrest , and re-replicate they DNA . ( C ) A functional checkpoint is re-established upon expression of wild type Bub3 in bub3Δ cells . ( D ) Bub3R217A–R239A is unable to restore a functional checkpoint when expressed in bub3Δ cells . Panels A–D report experiments that were carried out at the same time and at least twice . ( E ) Re-budding in the presence of nocodazole was taken as an independent indication of checkpoint deficiency . Wild type cells , and bub3Δ cells reconstituted with wild type Bub3 were able to maintain the checkpoint arrest and did not re-bud during the time of observation . Conversely , bub3Δ cells and cells reconstituted with Bub3R217A–R239A re-budded , indicative of checkpoint failure . ( F ) The binding affinity of Bub1–Bub3 for individual MELTP is high . This predicts that multiple Bub1–Bub3 complexes may become bound to a single Spc105/Knl1 molecule . Mad3/BubR1 requires Bub1 for kinetochore recruitment , indicating that it is not able to target autonomously to kinetochores . Because Mad3/BubR1 is , like Bub1 , constitutively bound to Bub3 , it is plausible that Mad3/BubR1 suppresses the MELTP-binding activity of Bub3 . Whether this occurs , and how , are purely speculative at this time . DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 021 Detailed information on the mechanisms of recruitment to , activation at , and release from kinetochores of the checkpoint proteins has been missing . As a consequence , it has been hard to design targeted experiments aiming to dissect the role of specific binding interactions in the regulation of checkpoint proteins . Phosphorylation regulates many interactions at the kinetochore ( Lara-Gonzalez and Taylor , 2012; Foley and Kapoor , 2013 ) . In a few well-characterized cases , phosphorylation negatively regulates protein interactions at the kinetochore ( e . g . , Cheeseman et al . , 2006; DeLuca et al . , 2006; Meadows et al . , 2011; Rosenberg et al . , 2011 ) . The kinase activity of Mps1 , on the other hand , has a positive role in the recruitment of other checkpoint proteins to the kinetochore ( Lara-Gonzalez and Taylor , 2012; Foley and Kapoor , 2013 ) . The recent discovery that MELTP motifs are required for the recruitment of the Bub1–Bub3 complex represented an important advancement ( London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012 ) . Here , we have taken a considerable step forward by identifying Bub3 as the MELTP reader and by probing its importance for checkpoint signaling . Specifically , ours is the first detailed mechanistic description of how the phosphorylation of a kinetochore subunit promotes recruitment of downstream elements . Our studies identify Bub3 as a new member of a family of protein domains involved in the recognition of phosphorylated sequence motifs , which includes , among others , the SH2 , PTB , BRCT , FHA and polo-box domains ( Seet et al . , 2006; Pawson and Kofler , 2009 ) . Interestingly , the structure of the complex of the GLEBS motif of the nucleoporin and proto-oncogene Nup98 with the WD40-repeat nuclear transport factor Rae1/Gle2 ( PDB ID 3MMY ) ( Ren et al . , 2010 ) is very closely related to that of the Bub3–Bub1 complex ( Figure 2C , Figure 2—figure supplement 1 ) . Most of the residues involved in MELTP binding on Bub3 are perfectly conserved in Rae1 ( Figure 2C ) , where they form a hydrophobic-basic bipartite interface that is almost indistinguishable from that in Bub3 ( Ren et al . , 2010 ) . The striking conservation at this interface , and our realization that the interface is implicated in MELTP binding , suggests that Rae1 , or possibly its complex with Nup98 , might also be a receptor for phosphorylated motifs . The kinetochore levels of SAC proteins are dynamically regulated during the process of attachment of kinetochores to microtubules , with maximal levels being reached at unattached kinetochores and minimal levels being reached at metaphase ( Lara-Gonzalez et al . , 2012; Foley and Kapoor , 2013 ) . It is plausible that such dynamic behavior reflects a requirement for specific steps of activation and inactivation of the checkpoint proteins at kinetochores in response to the status of kinetochore-microtubule attachment . For instance , forced retention of the Mad1-Mad2 ‘template’ complex ( De Antoni et al . , 2005 ) at kinetochores results in a protracted checkpoint-dependent arrest despite all chromosomes being bipolarly aligned at the metaphase plate ( Gassmann et al . , 2010; Maldonado and Kapoor , 2011 ) , indicating that removal of Mad1-Mad2—which is normally mediated by the Dynein-Spindly complex ( Griffis et al . , 2007; Yamamoto et al . , 2008; Chan et al . , 2009; Barisic et al . , 2010; Gassmann et al . , 2010 ) —is necessary for the inactivation of the checkpoint signal before anaphase . The importance of kinetochore recruitment of Bub1–Bub3 in the spindle checkpoint , on the other hand , has been controversial ( Klebig et al . , 2009; Vanoosthuyse et al . , 2009; Windecker et al . , 2009; Shepperd et al . , 2012; Yamagishi et al . , 2012 ) . Recently , however , it was shown that mutation of the phosphorylated Thr residue in the MELT repeats of Spc105/Knl1 prevents kinetochore recruitment of Bub1–Bub3 and results in a checkpoint defect in several species ( London et al . , 2012; Shepperd et al . , 2012; Yamagishi et al . , 2012 ) . Formally , checkpoint deficiency in the presence of mutants that prevent the phospohorylation of the MELT motifs might be due to the impaired binding of MELTP-binding proteins other than Bub1–Bub3 . In this study , we demonstrate that Bub1–Bub3 binds directly to MELTP sequences and recapitulate the checkpoint defect by mutating residues within the MELTP-binding site of Bub3 that impair its own kinetochore recruitment and that of Bub1 . Overall , these observations indicate that Bub1 localization to kinetochores is required for checkpoint function . We suspect that the recruitment of Bub1–Bub3 to Spc105/Knl1 unleashes Bub1’s role in the checkpoint . Because the kinase domain of Bub1 is not required for checkpoint signaling ( Sharp-Baker and Chen , 2001; Fernius and Hardwick , 2007; Perera et al . , 2007; Klebig et al . , 2009; Ricke et al . , 2012 ) , Bub1’s role in the checkpoint is probably entirely exerted through macromolecular interactions . For instance , Bub1 is required for kinetochore recruitment of Mad1-Mad2 and of Mad3/BubR1 ( Millband and Hardwick , 2002; Liu et al . , 2006; Rischitor et al . , 2006; Perera et al . , 2007; Logarinho et al . , 2008; Klebig et al . , 2009 ) ( Figure 6F ) . Mad3/BubR1 is , like Bub1 , constitutively bound to Bub3 . Thus , a requirement on Bub1 for kinetochore recruitment of Mad3/BubR1 is unexpected and suggests that when bound to Mad3/BubR1 , Bub3 might be unable to perform an equivalent function in kinetochore recruitment to the one it performs when bound to Bub1 . Indeed , our studies indicate that when bound to Bub3 , Bub1 plays an important positive role in the interaction with MELTP ( Figure 4D–F ) . Although the presence of a positively charged residue at position 314 is not a conserved feature of Bub1 ( not shown ) , it is plausible that differences in the sequence of the β1-β2-loop in Bub1 and Mad3/BubR1 ( Figure 2D ) might account for the different behavior of Bub3 in its respective complexes . Multisite phosphorylation is commonly used in signaling networks for its potential to generate non-linear responses to stimuli ( Kapuy et al . , 2009; Salazar and Höfer , 2009 ) . Multisite phosphorylation of Sic1 and Cyclin E by Cyclin-dependent kinases , for instance , mediates their interaction with the Cdc4 and Fbw7 ubiquitin ligases and their subsequent ubiquitination and destruction ( Nash et al . , 2001; Orlicky et al . , 2003; Hao et al . , 2007; Kõivomägi et al . , 2011 ) . Similar to Bub3 , also Cdc4 and Fbw7 are WD40 β-propeller proteins , but the position and organization of the phosphopeptide binding sites in Cdc4 and Fbw7 and in Bub3 are distinct . In Cdc4 and Fbw7 ( as well as in β-TrCP , another WD40 β-propeller Ub-ligase that interacts with phosphorylated motifs in target proteins ) the phosphopeptide-binding site is located on the top surface of the propeller domain ( Figure 3—figure supplements 1 and 2 ) ( Orlicky et al . , 2003; Wu et al . , 2003; Hao et al . , 2007 ) . The binding affinity of Bub1–Bub3 for individual MELTP repeats of Spc105/Knl1 is high , matching the highest binding affinities measured for SH2-phosphopeptide interactions ( Seet et al . , 2006 ) . This suggests that each MELTP sequence has the potential to act as a docking site for an individual Bub1–Bub3 complex and that multiple Bub1–Bub3 complexes may bind to Spc105/Knl1 concomitantly if multiple phosphorylated MELT repeats are present ( Figure 6F ) . Understanding how variations in the levels of phosphorylation of MELT motifs reflect on the strength of the checkpoint response is a crucial question for future studies . The high affinity of the Bub1–Bub3 complex for MELTP sequences might explain why Bub1 is quite stably bound to kinetochores during checkpoint activation , as shown by fluorescence recovery after photobleaching ( FRAP ) experiments ( Howell et al . , 2004; Shah et al . , 2004 ) . It is possible that additional interactions of the Bub1–Bub3 complex with Spc105/Knl1 or other checkpoint components at the kinetochore further increase the binding affinity of the Bub1–Bub3 complex for kinetochores . The TPR domain of Bub1 , whose function is essential for the spindle checkpoint ( Kadura et al . , 2004; Vanoosthuyse et al . , 2004; Kiyomitsu et al . , 2007; Klebig et al . , 2009 ) , might be playing a crucial role in such additional interactions of Bub1 ( Brady and Hardwick , 2000; Kiyomitsu et al . , 2007; Klebig et al . , 2009; Bolanos-Garcia et al . , 2011; Kiyomitsu et al . , 2011; Kim et al . , 2012; Krenn et al . , 2012 ) . In this study , we have clarified how Bub1 becomes recruited to kinetochores and shown that such recruitment is crucial for the spindle checkpoint . Future studies will have to shed light on the molecular mechanisms subtending the function of Bub1 at kinetochores and shift the focus to asking how the recruitment of Bub1 leads to the recruitment of Mad1 and Mad2 and the generation of a diffusible signal that emanates from the kinetochore . cDNA sequences coding for S . cerevisiae Bub1289–359 and for full length Bub3 were subcloned in the first and second cassettes of pGEX-6P-2rbs vector ( Ciferri et al . , 2005 ) . In this construct , the cDNA encoding Bub1289–359 was sub-cloned in frame with the gene encoding GST and a cleavage site for PreScission protease , and was translated from the first ribosome-binding site ( rbs ) . Untagged Bub3 was subcloned downstream of the second rbs in the same vector . Expression was carried out in BL21 DE3 plysS cells in LB medium by auto-induction with 0 . 3% lactose at 18°C for approximately 16 hr . Cells were harvested by centrifugation and resuspended in Buffer A ( 20 mM Tris pH 7 . 5 , 300 mM NaCl , 10% Glycerol , 1 mM DTE , 1 mM EDTA pH 8 . 0 , 1 mM PMSF ) typically at a dilution of 3 ml lysis buffer per ml of bacterial pellet . Cells were lysed by sonication , and the lysates were cleared by centrifugation at 75000 × g for 45 min . The resulting clear supernatant was incubated with gentle rotation with 1/50 ( vol/vol ) of GSH-sepharose slurry ( GE Healthcare ) for 1–2 hr at 4°C . Beads were washed with 150 vol of Buffer A . To elute the Bub1–Bub3 complex from beads , GST-PreScission protease ( 0 . 02 mg/mg of protein target ) was added for 16 hr at 4°C . The eluate was concentrated on Amicon Ultra Centrifugal Filters MwCO 3000 ( Merck Millipore , Billerica , MA ) and further purified by size exclusion chromatography ( SEC ) on a Superdex 75 16/60 column using GF buffer ( 20 mM Tris pH 7 . 5 , 150 mM NaCl , 1 mM DTE ) . The complex eluted in a single peak with apparently stoichiometric amounts of Bub1289–359 and Bub3 and was subsequently concentrated to ∼10 mg/ml for crystallization or to 2–4 mg/ml for ITC experiments . The MELT2P ( sequence DPTSMEM{PTHR}EVFPRSIRQKN ) , MELT2 ( DPTSMEMTEVFPRSIRQKN ) , MELT4P ( DTVEGEPIDL{PTHR}EYESKPYVPN ) , and MELT4 ( DTVEGEPIDLTEYESKPYVPN ) peptides ( 95% purity ) were custom made by GeneScript ( Piscataway , NJ ) . Histone H1-derived phosphorylated peptide ( GGGPA{pTHR}PKKAKKL , 95% purity ) was purchased from AnaSpec ( Catalog number 61741 ) . Binding isotherm was measured at 25°C by isothermal titration calorimetry on a MicroCal ITC200 device ( GE Healthcare , Piscataway , NJ ) . All samples were extensively dialysed into fresh GF buffer ( 20 mM Tris pH7 . 5 , 150 mM NaCl , 1 mM DTE ) . In each titration , the Bub1289–359–Bub3 in the cell ( at a 30 µM concentration ) was titrated with thirty-five 2-µl injections ( at 90 s intervals ) of the indicated synthetic MELTP peptides ( at a concentration of 400 µM ) . The injections were continued beyond saturation levels to allow for determination of heats of ligand dilution . Data were fitted by least-square procedures to a single-site binding model using ORIGIN 5 . 0 software package ( MicroCal , Northampton , MA ) . All yeast strains ( Table 2 ) were derivatives of , or were backcrossed at least three times to , W303 ( ade2-1 , trp1-1 , leu2-3 , 112 , his3-11 , 15 , ura3 , ssd1 ) . Cells were grown in YEP medium ( 1% yeast extract , 2% bactopeptone , 50 mg/l adenine ) supplemented with 2% glucose ( YEPD ) . α-factor and nocodazole were used at 3 µg/ml and 15 µg/ml , respectively . 150 minutes after α-factor release , nocodazole was re-added to the cultures at 7 . 5 µg/ml . Synchronization experiments were carried out at 30°C . 10 . 7554/eLife . 01030 . 022Table 2 . Strains used in this study ( all in W303 background ) DOI: http://dx . doi . org/10 . 7554/eLife . 01030 . 022NameRelevant genotypeyAC1MATayAC411MATa , bub3::LEU2yAC1990MATa , bub3::LEU2 trp1::BUB3R127A/R239A-3Cherry::TRP1yAC2036MATa , bub3::LEU2 trp1::BUB3R239A-3Cherry::TRP1yAC2048MATa , bub3::LEU2 trp1::BUB3R127A/R239A-3Cherry::TRP1yAC2072MATa , bub3::LEU2 trp1::BUB3R127A/R239A-3Cherry::TRP1 , MTW1-GFP::TRP1yAC2110MATa , bub3::LEU2 trp1::BUB3-3Cherry::TRP1 , MTW1-GFP::TRP1yAC2090MATa , bub3::LEU2 trp1::BUB3-3Cherry::TRP1 , BUB1-GFP::TRP1yAC2091MATa , bub3::LEU2 trp1::BUB3R127A/R239A-3Cherry::TRP1 , BUB1-GFP::TRP1 To obtain strains expressing Bub3-mCherry or Bub3R127A–R239A-mCherry , plasmids AC122 and AC125 were created by subcloning in Yiplac204 cDNA fragments encoding wild type Bub3 or the double mutant ( with 250 bp upstream of ATG and 200 bp downstream of the stop codon ) . The Bub3 sequences were fused to three copies of mCherry that had been amplified from pCM79-pFA6a::3mcherry::hphNT1 ( Maeder et al . , 2007 ) . The plasmids were integrated at the TRP1 locus by digestion with Bsu36I , and the copy number of the integrated plasmids was verified by Southern blotting . Flow cytometric DNA quantitation was performed on a Becton-Dickinson ( Franklin Lakes , NJ ) FACScalibur device and analysed with CellQuest software . Kinetics of re-budding was scored on ethanol-fixed cells . Time lapse videos were performed at 30°C using CELLASIC microfluidic chambers and recorded using a Delta Vision Elite imaging system ( Applied Precision , Issaquah , WA ) based on an IX71 inverted microscope ( Olympus , Shinjuku , Tokyo , Japan ) with a CoolSNAP HQ2 camera ( Photometrics , Tucson , AZ ) and a UPlanApo 60 × ( 1 . 4 NA ) oil immersion objective ( Olympus ) . Prior to crystallization , the Bub1289–359–Bub3 complex was mixed at a 1:2 ratio with a synthetic Spc105/Knl1 phosphopeptide ( sequence DPTSMEM{TP}EVFPRSIRQKN , with N-terminal amide and C-terminal acetyl groups ) and subjected to crystallization by the sitting drop method with a Mosquito nanodrop dispenser . Initial crystals were obtained with the G9 condition of Qiagen ( Venlo , The Netherlands ) PACT Suite screen ( 0 . 2 M K/Na tartrate , 0 . 1 M Bis Tris propane pH 7 . 5 , 20% PEG 3350 ) and did not require further optimization . The crystals grew to a typical size of ∼50 µm in each direction . X-ray diffraction data were collected at the PXII–X10SA beamline at the Swiss Light Source ( SLS ) ( Villigen , Switzerland ) and processed using XDS ( Kabsch , 2010 ) . Due to anisotropic diffraction , the data were subject to anisotropy correction using the UCLA diffraction anisotropy server ( Strong et al . , 2006 ) . Model refinement against the corrected data resulted in final maps of significantly better quality compared to maps obtained with uncorrected data . Initial phases were obtained by the molecular replacement method , which was carried out using the program PHASER ( McCoy et al . , 2007 ) and the structure of S . cerevisiae Bub3–Bub1 ( PDB ID 2I3S ) ( Larsen et al . , 2007 ) as a search model . Two copies of the Bub3–Bub1 dimer were placed in the asymmetric unit . Model building and refinement were carried out using Coot ( Emsley et al . , 2010 ) and phenix . refine from the PHENIX suite ( Adams et al . , 2010 ) , respectively . The final model contains two Bub3 monomers ( including residues 1–223 and 233–340 in chain A and residues 1–223 and 234–340 for chain D ) , two Bub1 fragments ( including residues 302–311 and 314–347 for chain B and residues 309–347 for chain E ) , and two Spc105/Knl1 MELTP peptides ( including residues 165–176 in chain C and residues 166–177 in chain F ) . Simulated annealing composite omit maps were produced using phenix . autobuild ( Terwilliger et al . , 2008 ) . Figures were generated using either CCP4MG ( McNicholas et al . , 2011 ) or Pymol ( Schrödinger LLC , Portland , OR ) . The final model and the structure factor amplitudes have been submitted to the Protein Data Bank under the accession numbers 4bl0 and r4bl0sf , respectively . Segmentation and fluorescence analysis for single cell images was performed with software written in MATLAB . For the segmentation and tracking of yeast cells , we used the program ‘phyloCell’ , written by Gilles Charvin ( unpublished results ) . To quantify the localization of proteins , we focused on the brightest pixels within each segmented area ( i . e . , within each cell ) . We observed that the area where the brightest pixels are typically localized amounts to roughly 1% of the area of the whole cell . Therefore , to compute an index for localization , we calculated the average of the brightest 1% of pixels . The value of this average depends on properties of the overall intensity distribution . When confronting two Gaussian distributions of pixel intensities , it is expected that the value of the average of the brightest intensity will be higher for the distribution whose mean intensity is higher or alternatively , for distributions with similar mean intensity , for the distribution whose standard deviation is higher . To correct for these factors , we used the following measure to quantify localization: Localization index = ( ( average 1% brightest pixels ) − mean ) /std This definition allows us to distinguish quantitatively between localization and delocalization . In case of perfect delocalization , the overall intensity will adopt a gaussian distribution , and in this case our measure of localization adopts a value of ∼3 . 7 ( Figure 5—figure supplement 2A for an example of overall intensity distribution when the protein is delocalized and for the corresponding localization index ) . When the protein is kinetochore-localized ( as shown by co-localization with Mtw1 ) , the distribution is skewed-Gaussian , with a more extended right tail ( Figure 5—figure supplement 2B for an example of overall intensity distribution when the protein is kinetochore localized and for the corresponding localization index ) . In this case , our measure for localization is larger than ∼3 . 7 . When the localization index assumes values that are significantly higher than ∼3 . 7 , we observed the fluorescent protein to be localized to kinetochores , as confirmed by co-localization with Mtw1 ( Figure 5 ) . The four plots in Figure 5—figure supplement 2C–E report the distribution of maxima ( C ) , minima ( D ) , and standard deviation ( E ) of the localization index of Bub3wt-mCherry or Bub3R217A-R239A-mCherry over an entire cell cycle ( i . e . , the time between two budding events ) . For Bub3wt-mCherry , we measured 55 cycles in 37 cells . For Bub3R217A–R239A-mCherry , we measured 45 cycles in 38 cells . The distribution of minima of the localization index is similar for the two Bub3 species , corresponding to delocalization . Wild type and mutant , however , differ clearly for the distribution of maxima , where the wild type only shows localization ( i . e . , localization index >3 . 7 ) . The higher amplitude of the localization index in wild types gives rise to a larger standard deviation of the localization index when compared to the wild type ( C ) .	The cell cycle is the process by which a cell divides to produce two near-identical daughter cells . Two crucial parts of the cell cycle are the duplication of the chromosomes in the original cell , and the segregation of these chromosomes between the two daughter cells . These and other parts of the cell cycle are strictly regulated to prevent errors , which can lead to cancer and other diseases . After chromosome duplication has taken place , the pairs of identical chromosomes , known as sister chromatids , remain tightly bound to each other . These sister chromatids line up in the middle of the cell , with protein filaments called microtubules connecting them to a bipolar structure called the spindle . For the cell to divide correctly , the sister chromatids in each pair must be connected to opposite poles of the spindle . A signalling network known as the spindle assembly checkpoint ( SAC ) ensures that the sister chromatids have enough time to line up correctly and to correct possible problems . Once everything is in place , the SAC releases its ‘break’ , and the microtubules then pull the sister chromatids away from each other . This way , each daughter cell receives the same complement of chromosomes that was present in the mother cell . The microtubules are not directly attached to the sister chromatids but to protein complexes called kinetochores that assemble on each sister chromatid . In particular , each microtubule binds to a very large protein complex called the KMN network . Knl1 , which is part of this network , recruits two SAC proteins–Bub1 and Bub3–to the kinetochore . It is known that a phosphate group is added to Knl1 when the SAC is active , and that Knl1 can only recruit Bub1 and Bub3 after it has been phosphorylated . However , the details of the interactions between Knl1 , Bub1 and Bub3 are not understood , and it is not clear whether these interactions are essential for the SAC . Now Primorac et al . have shown that Bub3 binds directly to Knl1 through a region that contains multiple MELT motifs ( where M , E , L and T are all amino acids ) , and that this interaction only happens if these ‘MELT repeats’ have been phosphorylated . Moreover , once bound to the Knl1 , Bub3 then recruits Bub1 to the kinetochore . By showing that the recognition of phosphorylated Knl1 by the Bub1-Bub3 complex has a central role in the spindle assembly checkpoint , these results highlight the importance of phosphorylation as a way of regulating the timing of events during the cell cycle .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2013	Bub3 reads phosphorylated MELT repeats to promote spindle assembly checkpoint signaling
Endothelial cells ( ECs ) in the central nervous system ( CNS ) acquire their specialized blood–brain barrier ( BBB ) properties in response to extrinsic signals , with Wnt/β-catenin signaling coordinating multiple aspects of this process . Our knowledge of CNS EC development has been advanced largely by animal models , and human pluripotent stem cells ( hPSCs ) offer the opportunity to examine BBB development in an in vitro human system . Here , we show that activation of Wnt signaling in hPSC-derived naïve endothelial progenitors , but not in matured ECs , leads to robust acquisition of canonical BBB phenotypes including expression of GLUT-1 , increased claudin-5 , decreased PLVAP , and decreased permeability . RNA-seq revealed a transcriptome profile resembling ECs with CNS-like characteristics , including Wnt-upregulated expression of LEF1 , APCDD1 , and ZIC3 . Together , our work defines effects of Wnt activation in naïve ECs and establishes an improved hPSC-based model for interrogation of CNS barriergenesis . In the central nervous system ( CNS ) , vascular endothelial cells ( ECs ) are highly specialized , with complex tight junctions , expression of a spectrum of nutrient and efflux transporters , low rates of vesicle trafficking , no fenestrae , and low expression of immune cell adhesion molecules ( Reese and Karnovsky , 1967; Obermeier et al . , 2013 ) . ECs bearing these attributes , often referred to as the blood–brain barrier ( BBB ) , work in concert with the other brain barriers to facilitate the tight regulation of the CNS microenvironment required for proper neuronal function ( Daneman and Engelhardt , 2017; Profaci et al . , 2020 ) . During development , the Wnt/β-catenin signaling pathway drives both CNS angiogenesis , during which vascular sprouts originating from the perineural vascular plexus invade the developing neural tube , and the coupled process of barriergenesis by which resulting ECs begin to acquire BBB properties ( Liebner et al . , 2008; Stenman et al . , 2008; Daneman et al . , 2009; Engelhardt and Liebner , 2014; Umans et al . , 2017 ) . Specifically , neural progenitor-derived Wnt7a and Wnt7b ligands signal through Frizzled receptors and the obligate co-receptors RECK and GPR124 ( ADGRA2 ) on ECs ( Kuhnert et al . , 2010; Cullen et al . , 2011; Vanhollebeke et al . , 2015; Cho et al . , 2017; Eubelen et al . , 2018; Vallon et al . , 2018 ) . Other ligands function analogously in other regions of the CNS , including Norrin in the retina and cerebellum ( Ye et al . , 2009; Wang et al . , 2012 ) and potentially Wnt3a in the dorsal neural tube ( Daneman et al . , 2009 ) . Furthermore , Wnt/β-catenin signaling is required for maintenance of CNS EC barrier properties in adulthood ( Tran et al . , 2016 ) , with astrocytes as a major source of Wnt7 ligands ( He et al . , 2018; Vanlandewijck et al . , 2018; Guérit et al . , 2021 ) . Molecular hallmarks of Wnt-mediated CNS EC barriergenesis are ( i ) acquisition of glucose transporter GLUT-1 expression , ( ii ) loss of plasmalemma vesicle-associated protein ( PLVAP ) , and ( iii ) upregulation of claudin-5 ( Daneman et al . , 2009; Kuhnert et al . , 2010; Cho et al . , 2017; Umans et al . , 2017; Wang et al . , 2019 ) . Notably , the Wnt-mediated switch between the ‘leaky’ EC phenotype ( GLUT-1– PLVAP+ claudin-5low ) and the barrier EC phenotype ( GLUT-1+ PLVAP– claudin-5high ) correlates with reduced permeability to molecular tracers ( Wang et al . , 2012; Cho et al . , 2017 ) and is conserved in multiple contexts . For instance , medulloblastomas that produce Wnt-inhibitory factors have leaky vessels ( Phoenix et al . , 2016 ) . Moreover , vasculature perfusing circumventricular organs is leaky due to low levels of Wnt signaling ( Benz et al . , 2019; Wang et al . , 2019 ) . Notably , ectopic activation of Wnt in ECs of circumventricular organs induces GLUT-1 and suppresses PLVAP ( Benz et al . , 2019; Wang et al . , 2019 ) . However , similar ectopic activation of Wnt in liver and lung ECs produces only very minor barriergenic effects ( Munji et al . , 2019 ) , and Wnt activation in cultured primary mouse brain ECs does not prevent culture-induced loss of barrier-associated gene expression ( Sabbagh and Nathans , 2020 ) . The reasons for the apparent context-dependent impacts of Wnt activation in ECs remain unclear and motivate systematic examination of this process in a simplified model system . Further , given species differences in brain EC transporter expression ( Uchida et al . , 2011 ) , drug permeability ( Syvänen et al . , 2009 ) , and gene expression ( Song et al . , 2020 ) , this process warrants investigation in human cells to complement mouse in vivo studies . Prior studies have evaluated the impact of Wnt activation in immortalized human brain ECs and observed only modest effects on barrier phenotype ( Paolinelli et al . , 2013; Laksitorini et al . , 2019 ) . Combined with the aforementioned deficits observed in primary adult mouse brain ECs that are not rescued by ectopic Wnt activation ( Sabbagh and Nathans , 2020 ) , one possibility is that mature , adult endothelium is largely refractory to Wnt activation , and that Wnt responsiveness is a property of immature ECs analogous to those in the perineural vascular plexus . Human pluripotent stem cells ( hPSCs ) offer an in vitro human model system for systematic investigation of molecular mechanisms of BBB phenotype acquisition , especially given their ability to model early stages of endothelial specification and differentiation . However , currently available hPSC-based models of CNS endothelial-like cells are not well suited for modeling the BBB developmental progression as they do not follow a developmentally relevant differentiation trajectory , lack definitive endothelial identity , or have been incompletely characterized with respect to the role of developmental signaling pathways ( Lippmann et al . , 2020; Workman and Svendsen , 2020; Lu et al . , 2021 ) . As a potential alternative , hPSCs can also be used to generate immature , naïve endothelial progenitors ( Lian et al . , 2014 ) that could be used to better explore the induction of BBB phenotypes . For example , we recently reported that extended culture of such hPSC-derived endothelial progenitors in a minimal medium yielded ECs with improved BBB tight junction protein expression and localization , which led to improved paracellular barrier properties ( Nishihara et al . , 2020 ) . However , as shown below , these cells exhibit high expression of PLVAP and little expression of GLUT-1 , indicating the need for additional cues to drive CNS EC specification . In this work , we aimed to define the effects of activating Wnt/β-catenin signaling in hPSC-derived , naïve endothelial progenitors and assess the extent to which this strategy would drive development of a CNS EC-like phenotype . We found that many aspects of the CNS EC phenotype , including the canonical GLUT-1 , claudin-5 , and PLVAP expression effects , were regulated by CHIR 99021 , a small molecule agonist of Wnt/β-catenin signaling . CHIR treatment in matured ECs produced a more limited response . Whole-transcriptome analysis revealed definitive endothelial identity of the resulting cells and CHIR-upregulated expression of known CNS EC transcripts , including LEF1 , APCDD1 , AXIN2 , SLC2A1 , CLDN5 , LSR , ABCG2 , SOX7 , and ZIC3 . We also observed an unexpected CHIR-mediated upregulation of caveolin-1 , which did not , however , correlate with increased uptake of a dextran tracer . Thus , we provide evidence that Wnt activation in hPSC-derived naïve endothelial progenitors is sufficient to induce many aspects of the CNS barrier EC phenotype , and we establish a model system for further systematic investigation of putative barriergenic cues . We adapted an existing protocol to produce endothelial progenitor cells ( EPCs ) from hPSCs ( Lian et al . , 2014; Bao et al . , 2016; Figure 1A ) . To achieve mesoderm specification , this method employs an initial activation of Wnt/β-catenin signaling with CHIR 99021 ( CHIR ) , a small molecule inhibitor of glycogen synthase kinase-3 ( GSK-3 ) , which results in inhibition of GSK-3β-mediated β-catenin degradation . After 5 days of expansion , the resulting cultures contained a mixed population of CD34+CD31+ EPCs and CD34–CD31– non-EPCs ( Figure 1B and C ) . We used magnetic-activated cell sorting ( MACS ) to isolate CD31+ cells from this mixed culture and plated these cells on collagen IV-coated plates in a minimal EC medium termed hECSR ( Nishihara et al . , 2020 ) . We first asked whether Wnt3a , a ligand widely used to activate canonical Wnt/β-catenin signaling ( Kim et al . , 2005; Kim et al . , 2008; Liebner et al . , 2008; Cecchelli et al . , 2014; Praça et al . , 2019 ) , could induce GLUT-1 expression in the resulting ECs . After 6 days of treatment , we observed a significant increase in the fraction of GLUT-1+ ECs in Wnt3a-treated cultures compared to controls ( Figure 1D and E ) . Consistent with previous observations ( Nishihara et al . , 2020 ) , we also detected a population of calponin+ smooth muscle protein 22-⍺+ putative smooth muscle-like cells ( SMLCs ) outside the endothelial colonies ( Figure 1—figure supplement 1 ) , and these SMLCs expressed GLUT-1 in both control and Wnt3a-treated conditions ( Figure 1D ) . Based on these promising results with Wnt3a , we next tested a low concentration ( 4 µM ) of the GSK-3 inhibitor CHIR because of its ability to activate Wnt signaling in a receptor/co-receptor-independent manner . In addition to GLUT-1 , we evaluated expression of two other key proteins: claudin-5 , which is known to be upregulated in CNS ECs in response to Wnt ( Benz et al . , 2019 ) , and caveolin-1 , given the low rate of caveolin-mediated transcytosis in CNS compared to non-CNS ECs ( Reese and Karnovsky , 1967; Andreone et al . , 2017; Figure 2A ) . 4 µM CHIR robustly induced GLUT-1 expression in approximately 90% of ECs while increasing EC number and increasing EC purity to nearly 100% ( Figure 2B ) . Furthermore , CHIR led to an approximately 1 . 5-fold increase in average claudin-5 abundance and a 10- to 30-fold increase in GLUT-1 abundance , but also a 2- to 4-fold increase in caveolin-1 ( Figure 2B ) . We therefore titrated CHIR to determine an optimal concentration for EC expansion , purity , GLUT-1 induction , and claudin-5 upregulation while limiting the undesirable non-CNS-like increase in caveolin-1 abundance . Although 2 µM CHIR did not lead to increased caveolin-1 expression compared to vehicle control ( DMSO ) , it also did not elevate claudin-5 or GLUT-1 expression compared to control and was less effective in increasing EC number and EC purity than 4 µM CHIR ( Figure 2—figure supplement 1 ) . On the other hand , 6 µM CHIR further increased GLUT-1 abundance but also further increased caveolin-1 abundance and did not improve EC number , EC purity , or claudin-5 expression ( Figure 2—figure supplement 1 ) . Therefore , we conducted further experiments using 4 µM CHIR . We confirmed that the CHIR-mediated increases in EC purity , EC number , and caveolin-1 and GLUT-1 expression were conserved in an additional hPSC line , although claudin-5 upregulation was not apparent ( Figure 2—figure supplement 2 ) . We also used two hPSC lines with doxycycline-inducible expression of short hairpin RNAs targeting CTNNB1 ( β-catenin ) to confirm that CHIR-mediated upregulation of GLUT-1 in ECs was β-catenin-dependent . Indeed , doxycycline treatment in combination with CHIR significantly reduced GLUT-1 abundance in ECs derived from these hPSC lines ( Figure 2—figure supplement 3 ) . Finally , we confirmed that increased EC number was the result of increased EC proliferation in CHIR-treated cultures ( Figure 2—figure supplement 4 ) . Together , these results suggest that activation of the Wnt/β-catenin pathway is capable of inducing CNS-like phenotypes in hPSC-derived endothelial progenitors . Since CHIR elicited a robust Wnt-mediated response , we next asked whether other aspects of the CNS EC barrier phenotype were CHIR-regulated . PLVAP , a protein that forms bridges across both caveolae and fenestrae ( Herrnberger et al . , 2012 ) , is one such canonically Wnt-downregulated protein . We therefore first evaluated PLVAP expression in Passage 1 control ( DMSO ) or CHIR-treated ECs using confocal microscopy ( Figure 3A ) . We observed numerous PLVAP+ punctate vesicle-like structures in both conditions , with CHIR treatment reducing PLVAP abundance by approximately 20% ( Figure 3A and B ) . This effect was not apparent in western blots of Passage 1 ECs , likely due to the relatively modest effect ( Figure 4A and B ) . However , after two more passages ( Figure 1A ) , Passage 3 ECs demonstrated a robust downregulation of PLVAP in CHIR-treated cells compared to controls ( Figure 4C and D ) . We also used western blotting to confirm CHIR-mediated upregulation of GLUT-1 and claudin-5 both at Passage 1 and Passage 3 ( Figure 4A–D ) . We next evaluated expression of the tricellular tight junction protein LSR ( angulin-1 ) because of its enrichment in CNS versus non-CNS ECs , and the temporal similarity between LSR induction and the early stage of Wnt-mediated CNS barriergenesis ( Sohet et al . , 2015 ) . We found that CHIR treatment led to a strong increase in LSR expression in both Passage 1 and Passage 3 ECs ( Figure 4A–D ) , suggesting that Wnt signaling upregulates multiple necessary components of the CNS EC bicellular and tricellular junctions . CHIR treatment produced two apparently competing changes in ECs related to vesicular transport: an expected downregulation of PLVAP and an unexpected upregulation of caveolin-1 . We therefore asked whether the rate of total fluid-phase endocytosis differed between CHIR-treated and control ECs using a fluorescently labeled 10 kDa dextran as a tracer . After incubating Passage 1 cultures with dextran for 2 hr at 37°C , we used flow cytometry to gate CD31+ ECs and assess total dextran accumulation ( Figure 5A and B ) . In ECs incubated at 37°C , CHIR treatment did not change the geometric mean dextran signal compared to DMSO ( Figure 5B and C ) , but did cause a broadening of the distribution of dextran intensities as quantified by the coefficient of variation ( CV ) , indicative of subpopulations of cells with decreased and increased dextran uptake ( Figure 5B and D ) . We confirmed that the dextran signal measured by this assay was endocytosis-dependent by carrying out the assay at 4°C and with inhibitors of specific endocytic pathways ( Figure 5—figure supplement 1A–C ) . Compared to vehicle control , chlorpromazine ( inhibitor of clathrin-mediated endocytosis ) and rottlerin ( inhibitor of macropinocytosis ) both decreased dextran uptake , while nystatin ( inhibitor of caveolin-mediated endocytosis ) did not significantly affect uptake ( Figure 5—figure supplement 1B and C ) , consistent with the very small number of dextran+ caveolin-1+ puncta observed by confocal imaging ( Figure 5—figure supplement 1D ) . Thus , despite the generally uniform elevation of caveolin-1 and decrease of PLVAP observed by immunocytochemistry in CHIR-treated ECs , our functional assay suggests neither an overall increase nor decrease in total fluid-phase endocytosis . Instead , it indicates that CHIR increases the heterogeneity of the EC population with respect to the rate of endocytosis . We also compared the paracellular barrier properties of DMSO- and CHIR-treated ECs . Because Passage 1 cultures contain SMLCs that preclude formation of a confluent endothelial monolayer , we evaluated paracellular barrier properties of Passage 3 ECs that had undergone selective dissociation and replating ( see Materials and methods ) , a strategy that effectively purifies the cultures ( Nishihara et al . , 2020 ) . CHIR-treated Passage 3 ECs had elevated transendothelial electrical resistance ( TEER ) ( Figure 5E ) and decreased permeability to the small molecule tracer sodium fluorescein ( Figure 5F ) . Together , these results are consistent with CHIR-mediated increases to tight junction protein expression ( e . g . , claudin-5 and LSR ) and suggest that Wnt activation leads to functional improvements to paracellular barrier in this system . Given the relatively weak responses to Wnt activation in adult mouse liver ECs in vivo ( Munji et al . , 2019 ) and adult mouse brain ECs cultured in vitro ( Sabbagh and Nathans , 2020 ) , we sought to determine whether the immature , potentially more plastic state of hPSC-derived endothelial progenitors contributed to the relatively robust CHIR-mediated response we observed . To test this hypothesis , we matured hPSC-derived ECs in vitro for four passages ( until approximately day 30 ) prior to initiating CHIR treatment for 6 days and compared the resulting cells to differentiation-matched samples treated with CHIR immediately after MACS ( Figure 6A ) . Both Passage 1 DMSO-treated ECs and Passage 5 DMSO-treated ECs , which are analogous to EECM-BMEC-like cells we previously reported ( Nishihara et al . , 2020 ) , did not have detectable GLUT-1 expression ( Figure 6B ) . Compared to DMSO controls , the CHIR-treated Passage 5 ECs exhibited no increase in GLUT-1 abundance ( Figure 6B–D ) , which contrasts with the marked increase observed when CHIR treatment was initiated immediately after MACS ( Figure 6B–D ) . Furthermore , CHIR treatment in matured ECs did not increase claudin-5 expression and did not increase EC number ( Figure 6B–D ) , in contrast to the increases observed in both properties when treatment was initiated immediately after MACS ( Figure 6B–D ) . We observed a similar lack of robust GLUT-1 induction in an additional differentiation and an additional hPSC line in which CHIR treatment was carried out at Passage 4 ( Figure 6—figure supplement 1 ) . Together , these data suggest that early , naïve endothelial progenitors are more responsive to Wnt activation than more mature ECs derived by the same differentiation protocol . We turned next to RNA-sequencing as an unbiased method to assess the impacts of Wnt activation on the EC transcriptome . We performed four independent differentiations and analyzed Passage 1 ECs treated with DMSO or CHIR using fluorescence-activated cell sorting ( FACS ) to isolate CD31+ ECs from the mixed EC/SMLC cultures . We also sequenced the SMLCs from DMSO-treated cultures at Passage 1 from two of these differentiations . DMSO- and CHIR-treated ECs at Passage 3 from three of these differentiations were also sequenced . Principal component analysis of the resulting whole-transcriptome profiles revealed that the two cell types ( ECs and SMLCs ) segregated along principal component ( PC ) 1 , which explained 57% of the variance . In ECs , the effects of passage number and treatment were reflected in PC 2 , which explained 21% of the variance ( Figure 7A ) . We next validated the endothelial identity of our cells; we observed that canonical endothelial marker genes ( including CDH5 , CD34 , PECAM1 , CLDN5 , ERG , and FLI1 ) were enriched in ECs compared to SMLCs and had high absolute abundance , on the order of 100–1000 transcripts per million ( TPM ) ( Figure 7B , Supplementary file 1 ) . SMLCs expressed mesenchymal ( mural/fibroblast ) -related transcripts ( including PDGFRB , CSPG4 , PDGFRA , TBX2 , CNN1 , and COL1A1 ) , which ECs generally lacked , although we did observe slight enrichment of some of these genes in Passage 1 DMSO-treated ECs , likely reflective of a small amount of SMLC contamination despite CD31 FACS ( Figure 7B ) . SMLCs also expressed SLC2A1 ( Supplementary file 1 ) consistent with protein-level observations ( Figure 1D ) . We also observed little to no expression of the epithelial genes CDH1 , EPCAM , CLDN1 , CLDN3 ( Castro Dias et al . , 2019 ) , CLDN4 , and CLDN6 , reflecting the definitive endothelial nature of the cells ( Figure 7B , Supplementary file 1 ) . First comparing CHIR- and DMSO-treated ECs at Passage 1 , we identified 1369 significantly upregulated genes and 2037 significantly downregulated genes ( Figure 7C , Supplementary file 2 ) . CHIR-upregulated genes included SLC2A1 , CLDN5 , LSR , and CAV1 , consistent with protein-level assays . PLVAP was downregulated , as were a number of mesenchymal genes ( TAGLN , COL1A1 ) , again reflective of slight contamination of SMLC transcripts in the DMSO-treated EC samples ( Figure 7C and D ) . Additionally , important downstream effectors of Wnt signaling were upregulated , including the transcription factors LEF1 and TCF7 , the negative regulator AXIN2 , and the negative regulator APCDD1 , which is known to modulate Wnt-regulated barriergenesis in retinal endothelium ( Mazzoni et al . , 2017; Figure 7C and D ) . We also identified upregulated transcription factors: ZIC3 , which is highly enriched in brain and retinal ECs in vivo and downstream of Frizzled4 signaling ( Wang et al . , 2012; Sabbagh et al . , 2018 ) , and SOX7 , which acts cooperatively with SOX17 and SOX18 in retinal angiogenesis ( Zhou et al . , 2015 ) , were upregulated by CHIR in our system ( Figure 7D ) . MSX1 and EBF1 , which are expressed by murine brain ECs in vivo ( Vanlandewijck et al . , 2018 ) , were also CHIR-upregulated ( Figure 7D ) . Additional CHIR-upregulated genes included ABCG2 ( encoding the efflux transporter breast cancer resistance protein [BCRP] ) , APLN , a tip cell marker enriched in postnatal day 7 murine brain ECs compared to those of other organs , and subsequently downregulated in adulthood ( Sabbagh et al . , 2018; Sabbagh and Nathans , 2020 ) , and FLVCR2 , a disease-associated gene with a recently identified role in brain angiogenesis ( Santander et al . , 2020; Figure 7C and D ) . Finally , we detected CHIR-mediated downregulation of the fatty acid-binding protein-encoding FABP4 , which is depleted in brain ECs compared to those of peripheral organs ( Sabbagh et al . , 2018 ) . We also observed similar downregulation of SMAD6 , which is depleted in brain ECs compared to lung ECs and is a putative negative regulator of BMP-mediated angiogenesis ( Mouillesseaux et al . , 2016; Vanlandewijck et al . , 2018; Figure 7D ) . In Passage 3 ECs , many of the CHIR-mediated gene expression changes observed at Passage 1 persisted , including SLC2A1 , LSR , LEF1 , AXIN2 , APCDD1 , ZIC3 , EBF1 , FLVCR2 , and ABCG2 upregulation and PLVAP downregulation ( Figure 7E , Figure 7—figure supplement 1 ) . Additional concordantly CHIR-upregulated genes encoding secreted factors , transcription factors , and transmembrane proteins are shown in Figure 7—figure supplement 2 and include REEP1 , a gene enriched in brain versus non-brain ECs ( Sabbagh et al . , 2018; Vanlandewijck et al . , 2018 ) that encodes a regulator of endoplasmic reticulum function and the Notch ligand-encoding gene JAG2 . On the other hand , at Passage 3 , CLDN5 was not upregulated in CHIR-treated cells compared to DMSO-treated cells , but was highly expressed ( ~2500 TPM ) . Similarly , CAV1 abundance remained high , but was not CHIR-upregulated in Passage 3 cells ( Figure 7—figure supplement 1 ) . Conversely , JAM2 , which encodes junctional adhesion molecule 2 , a component of EC tight junctions ( Aurrand-Lions et al . , 2001; Tietz and Engelhardt , 2015 ) , was upregulated by CHIR at Passage 3 , but not at Passage 1 , as was the retinol-binding protein-encoding gene RBP1 ( Figure 7—figure supplement 1 ) . We used weighted gene correlation network analysis ( WGCNA ) ( Zhang and Horvath , 2005; Langfelder and Horvath , 2008 ) to identify modules containing genes with highly correlated expression across the 14 EC samples ( Figure 7—figure supplement 3A , Supplementary file 3 ) . One such module ( the green module , containing 441 genes ) had a representative gene expression profile ( module eigengene ) with a strong , positive correlation with CHIR treatment ( Figure 7—figure supplement 3B ) . Importantly , genes central to this module included canonical transcriptional targets of Wnt/β-catenin signaling , including AXIN2 and APCDD1 , further supporting the key role of β-catenin signaling in transcriptional changes observed in CHIR-treated ECs . Additional central ( highly correlated ) genes within the green module included SLC2A1 , ZIC3 , and FLVCR2 , consistent with pairwise differential expression analysis , transcription factors ( CASZ1 , PRRX1 ) , and genes with putative roles in vesicle trafficking ( SNX4 , ARL8B , AP1AR , VTI1A , VPS41 ) and lipid metabolism ( AGPAT5 , ASAH1 ) ( Figure 7—figure supplement 3C ) . To determine the effects of extended culture , we next compared control ( DMSO-treated ) ECs at Passage 3 versus Passage 1 ( Figure 7—figure supplement 4 , Supplementary file 2 ) . Extended culture to Passage 3 in the absence of exogeneous Wnt activation led to 1521 upregulated genes , including CLDN5 and CAV1 , consistent with previously reported protein-level observations in EECM-BMEC-like cells ( Nishihara et al . , 2020 ) , which are analogous to Passage 3 DMSO-treated cells . We also observed 1625 downregulated genes , including marked downregulation of PLVAP ( Figure 7—figure supplement 4 ) . SLC2A1 , however , was not upregulated at Passage 3 ( Figure 7—figure supplement 4 ) , concordant with absence of GLUT-1 protein expression in the control ECs ( Figure 6B ) , nor was LSR . Further , despite some similarly regulated genes between the passage number and CHIR treatment comparisons ( e . g . , CLDN5 , CAV1 , PLVAP ) , the transcriptional responses to these two experimental variables were globally distinct as assessed by gene correlation network analysis ( Figure 7—figure supplement 3B ) . We also evaluated transcript-level expression of components of the Wnt signaling pathway in Passage 3 control ( DMSO-treated ) ECs as a first step towards understanding the relative lack of responsiveness observed when CHIR treatment was initiated in matured ( Passage 4 ) ECs ( Figure 7—figure supplement 4 ) . While CTNNB1 , GSK3B , and genes encoding components of the destruction complex were not significantly different between Passage 3 and Passage 1 , LEF1 and TCF7 were strongly downregulated in Passage 3 cells ( Figure 7—figure supplement 4 ) . Finally , to further understand the strengths and limitations of this model system both as a readout of early developmental changes in CNS ECs ( Passage 1 cells ) or as a source of CNS-like ECs for use in downstream modeling applications , we evaluated absolute transcript abundance and effects of treatment or passage number on 53 characteristic CNS EC genes encompassing tight junction components , vesicle trafficking machinery , solute carriers , and ATP-binding cassette ( ABC ) efflux transporters selected based on high expression in human brain ECs from a meta-analysis of single-cell RNA-seq data ( Gastfriend et al . , 2021; Figure 7—figure supplement 5 ) . While ECs expressed CLDN5 , TJP1 , TJP2 , OLCN , and LSR , they lacked MARVELD2 ( encoding tricellulin ) under all conditions . ECs under all conditions also lacked MFSD2A and , despite CHIR-mediated downregulation of PLVAP , retained high absolute expression of this and other caveolae-associated genes . Finally , while many solute carriers and ABC transporters were expressed ( SLC2A1 , SLC3A2 , SLC16A1 , SLC38A2 , ABCG2 ) , others expressed at the in vivo human BBB were not ( SLC5A3 , SLC7A11 , SLC38A3 , SLCO1A2 , ABCB1 ) ( Figure 7—figure supplement 5 ) . Thus , while CHIR treatment yields ECs with certain elements of CNS-like character , additional molecular signals are likely necessary to impart other aspects of the in vivo CNS EC phenotype . To globally assess whether CHIR-mediated gene expression changes in our system are characteristic of the responses observed in ECs in vivo and similar to those observed in other in vitro contexts , we compared our RNA-seq dataset to those of studies that employed a genetic strategy for β-catenin stabilization ( the Ctnnb1flex3 allele ) in adult mouse ECs in several contexts: ( i ) pituitary ECs , which acquire some BBB-like properties upon β-catenin stabilization Wang et al . , 2019; ( ii ) liver ECs , which exhibit little to no barriergenic response to β-catenin stabilization ( Munji et al . , 2019 ) ; ( iii ) brain ECs briefly cultured in vitro , which rapidly lose their BBB-specific gene expression profile even with β-catenin stabilization ( Sabbagh and Nathans , 2020 ) , and offer the most direct comparison to our in vitro model system . Upon recombination , the Ctnnb1flex3 allele produces a dominant mutant β-catenin lacking residues that are phosphorylated by GSK-3β to target β-catenin for degradation ( Harada et al . , 1999 ) ; as such , this strategy for ligand- and receptor-independent Wnt activation by β-catenin stabilization is similar to CHIR treatment , although GSK-3 phosphorylates targets other than β-catenin ( discussed below ) . We first used literature RNA-seq data from postnatal day 7 murine brain , liver , lung , and kidney ECs ( Sabbagh et al . , 2018 ) to define core sets of genes in brain ECs that are differentially expressed compared to all three of the other organs ( Figure 8A and B ) . Using the resulting sets of 1094 brain-enriched and 506 brain-depleted genes , we asked how many genes in our Passage 1 ECs were concordantly regulated by CHIR: 130 of the brain-enriched genes were CHIR-upregulated and 116 of the brain-depleted genes were CHIR-downregulated ( Figure 8C ) . At Passage 3 , 61 genes were concordantly upregulated and 46 downregulated ( Figure 8—figure supplement 1 ) . In pituitary ECs with β-catenin stabilization , 102 of the brain-enriched genes were upregulated and 48 of the brain-depleted genes were downregulated ( Figure 8D ) . Compared with the pituitary ECs , there were far fewer concordantly regulated genes in liver ECs with β-catenin stabilization , with 25 upregulated and 1 downregulated ( Figure 8E ) . Finally , cultured primary mouse brain ECs with β-catenin stabilization exhibited 72 concordantly upregulated and 16 downregulated genes ( Figure 8F ) . The only gene concordantly regulated in all four comparisons was the canonical Wnt target AXIN2 . Several additional genes were concordantly upregulated in three of four , including TCF7 , FAM107A , NKD1 , TNFRSF19 , GLUL , SLC30A1 , and ABCB1 , which was the only gene concordantly regulated in all comparisons except the hPSC-derived ECs ( Figure 8G ) . Several canonical target genes were shared by the hPSC-derived EC and pituitary EC systems , including APCDD1 , LEF1 , CLDN5 , and SLC2A1; also in this category were LSR , the zinc/manganese transporter SLC39A8 , and 12 additional genes ( Figure 8G ) . Notably , the caveolae inhibitor MFSD2A was robustly upregulated by β-catenin in pituitary ECs , but not in any other context ( Figure 8C–F ) , suggesting that other brain-derived factors may cooperate with Wnt to regulate expression of this important inhibitor of caveolin-mediated transcytosis . Complete gene lists from this comparative analysis are provided in Supplementary file 4 . In sum , the data suggest that the hPSC-derived ECs responded to Wnt activation in a fashion that led to modest induction of CNS transcriptional programs and that the response was most similar to the pituitary β-catenin stabilization model . Importantly , this analysis also supports the hypothesis that immature endothelium is highly responsive to Wnt activation where mature ( adult ) endothelium is largely refractory except in regions proximal to barrier-forming regions . Last , because GSK-3 is a component of numerous signaling pathways in addition to Wnt/β-catenin ( Eto et al . , 2005; Beurel et al . , 2015; Hermida et al . , 2017 ) , we used RNA-seq data to infer pathways that might be differentially regulated by the two strategies for activating Wnt/β-catenin signaling employed in the experiments above: CHIR treatment , which increases β-catenin stability by inhibiting GSK-3 , or direct stabilization of β-catenin . We tested lists of upregulated genes in ( i ) our Passage 1 ECs treated with CHIR versus DMSO , ( ii ) Passage 3 ECs treated with CHIR versus DMSO , and ( iii ) pituitary ECs with β-catenin stabilization versus controls ( Wang et al . , 2019 ) , against the Hallmark gene set collection ( Liberzon et al . , 2015; Figure 8—figure supplement 2 , Supplementary file 5 ) . In all three comparisons , the Wnt/β-catenin signaling gene set was significantly enriched ( Figure 8—figure supplement 2A ) . Similarly , the Notch signaling , TNFα signaling via NF-κB , KRAS signaling up , and several additional gene sets were consistently enriched in all three comparisons ( Figure 8—figure supplement 2A and B , Supplementary file 5 ) , suggesting similar regulation by GSK-3 inhibition and direct β-catenin stabilization . In contrast , the PI3K AKT mTOR signaling gene set was enriched in Passage 1 ECs , but not in Passage 3 ECs or pituitary ECs . Similarly , the gene set mTORC1 signaling was enriched in Passage 1 ECs and pituitary ECs , but genes driving this enrichment were distinct ( Figure 8—figure supplement 2C ) , and this gene set was not enriched in Passage 3 ECs . Thus , given the known , bidirectional interactions of GSK-3 and AKT/mTOR pathway components ( Hermida et al . , 2017 ) , these results suggest that CHIR-mediated inhibition of GSK-3 may transiently activate this pathway in Passage 1 ECs . Conversely , the gene set TGF-β signaling was enriched only in pituitary ECs with β-catenin stabilization ( Figure 8—figure supplement 2 , Supplementary file 5 ) . Taken together , these results , coupled with those of our CTNNB1 knockdown experiments and gene correlation network analysis , suggest a central role for β-catenin as a key effector of CHIR-mediated signaling , but also highlight some potential differences in the pathways activated in response to CHIR treatment versus β-catenin stabilization . Differences in other aspects of these two experimental paradigms ( in vitro versus in vivo , naïve versus CNS-proximal , human versus mouse ) , however , caution against overinterpretation of these results . The Wnt/β-catenin signaling pathway plays a central role in CNS angiogenesis and in establishing the unique properties of CNS ECs ( Liebner et al . , 2008; Stenman et al . , 2008; Daneman et al . , 2009; Kuhnert et al . , 2010; Cullen et al . , 2011; Vanhollebeke et al . , 2015; Cho et al . , 2017 ) . In this work , we investigated the role of Wnt/β-catenin signaling in induction of BBB properties in a human EC model using naïve endothelial progenitors derived from hPSCs . We reasoned that these immature EPCs ( Lian et al . , 2014 ) would be similar to the immature endothelium in the perineural vascular plexus and thus competent to acquire CNS EC phenotypes in response to Wnt activation . To activate Wnt signaling , we evaluated the widely used ligand Wnt3a ( Liebner et al . , 2008 ) and the GSK-3 inhibitor CHIR . We found that CHIR treatment robustly induced several canonical CNS EC molecular phenotypes , including a marked induction of GLUT-1 , upregulation of claudin-5 , and downregulation of PLVAP , which correlated with differential gene expression in RNA-seq data . We also observed a functional decrease in paracellular permeability . Further , using RNA-seq and western blotting , we identified LSR ( angulin-1 ) as CHIR-induced in this system , supporting the notion that this highly CNS EC-enriched tricellular tight junction protein ( Daneman et al . , 2010a; Sohet et al . , 2015 ) is Wnt-regulated . In RNA-seq data , we observed differential expression of known CNS EC-enriched/depleted and Wnt-regulated genes including upregulated LEF1 , AXIN2 , APCDD1 , ABCG2 , SOX7 , ZIC3 , FLVCR2 , JAM2 , and RBP1 , and downregulated PLVAP , FABP4 , SMAD6 , and SLIT2 . These RNA-seq data should therefore be useful in generating hypotheses of BBB-associated genes regulated by Wnt activation in ECs for future functional studies . Our work also defines an important set of phenotypes for which Wnt activation in ECs is not sufficient in our system: in the context of vesicle trafficking , we observed caveolin-1 ( CAV1 ) upregulation , no change in mean functional endocytosis , virtually no expression of MFSD2A , and high absolute PLVAP abundance in RNA-seq data despite CHIR-mediated downregulation . Given roles of brain pericytes in regulating PLVAP , MFSD2A , and functional transcytosis ( Armulik et al . , 2010; Daneman et al . , 2010b; Ben-Zvi et al . , 2014; Stebbins et al . , 2019 ) , and the observation that MFSD2A is Wnt-regulated in pituitary ECs in vivo ( Wang et al . , 2019 ) , where pericytes are present , it is plausible that pericyte-derived cues are necessary in addition to Wnts to achieve the characteristically low rate of CNS EC pinocytosis . Next , while ABCG2 ( BCRP ) was Wnt-induced in our system , other hallmark efflux transporters were not Wnt-regulated and either expressed at low levels ( e . g . , ABCC4 , encoding MRP-4 ) or not expressed ( e . g . , ABCB1 , encoding P-glycoprotein ) . Notably , however , Abcb1a was Wnt-regulated in the three other β-catenin stabilization experiments from the literature that we evaluated ( Munji et al . , 2019; Wang et al . , 2019; Sabbagh and Nathans , 2020 ) . Thus , pericyte-derived cues , astrocyte-derived cues , and/or activation of the pregnane X or other nuclear receptors may be important for complete acquisition of the complement of CNS EC efflux transporters ( Bauer et al . , 2004; Berezowski et al . , 2004; Praça et al . , 2019 ) . CHIR is widely used to activate Wnt/β-catenin signaling in cell culture ( Lian et al . , 2012; Lian et al . , 2014; Patsch et al . , 2015; Sakaguchi et al . , 2015; Gomez et al . , 2019; Pellegrini et al . , 2020; Guo et al . , 2021 ) . It remains unknown , however , to what extent CHIR-mediated inhibition of GSK-3 in ECs mimics the effects of Wnt ligand-induced inhibition of GSK-3 or direct stabilization of β-catenin . In our system , although the GLUT-1-inductive effect of CHIR was partially inhibited by β-catenin knockdown and our RNA-seq data revealed a transcriptional response characteristic of canonical Wnt signaling , it is possible that CHIR affects other signaling pathways , as suggested by pathway enrichment analysis . Thus , employing ligand-based strategies to activate Wnt signaling will be an important next step . Our RNA-seq data suggest that the receptors and coreceptors necessary to transduce Wnt7 and Norrin signaling ( e . g . , FZD4 , LRP6 , RECK , ADGRA2 [GPR124] , TSPAN12 , DVL2 ) are expressed by hPSC-derived ECs ( Figure 7—figure supplement 6 ) . Given evidence that Wnt ligands have poor solubility ( Janda et al . , 2012 ) and our preliminary data suggesting that supplementation of culture medium with Wnt7a and Wnt7b is largely ineffective in activating Wnt/β-catenin signaling in this system , special emphasis should be placed on strategies that present Wnt7a , Wnt7b , and/or Norrin in a manner that concentrates ligands at the cell surface , for example , by using direct cocultures of endogenously Wnt-producing cells ( neural progenitors or astrocytes ) or Wnt-overexpressing cells . Importantly , neural progenitor cells and astrocytes likely would also contribute other yet-unidentified ligands important for acquisition of CNS EC phenotype . Finally , it would also be informative to directly compare CHIR and/or Wnt ligand treatment to direct stabilization of β-catenin in this system , for example , by generating an hPSC line with inducible expression of a dominant active β-catenin . We also directly addressed the hypothesis that immature ECs are more plastic , that is , more competent to acquire BBB properties upon Wnt activation than mature ECs . This hypothesis is supported by existing observations that ectopic expression of Wnt7a is sufficient to induce GLUT-1 expression in non-CNS regions of the mouse embryo ( Stenman et al . , 2008 ) , but β-catenin stabilization in adult mouse liver and lung ECs produces only a slight effect ( Munji et al . , 2019 ) . We repeated our CHIR treatment paradigm in hPSC-derived ECs after an extended period of in vitro culture ( Passage 4 ECs ) and observed much weaker induction of GLUT-1 and no pro-proliferative effect . Thus , our results support this hypothesis and suggest that the loss of BBB developmental plasticity in ECs is an intrinsic , temporally controlled process rather than a result of the peripheral organ environment . The molecular mechanisms underlying this loss of plasticity remain poorly understood . While previous studies have demonstrated that the level of Wnt/β-catenin signaling in CNS ECs peaks early in development and subsequently declines ( Corada et al . , 2019; Hübner et al . , 2018 ) , this finding does not address mechanisms underlying the competence of ECs ( CNS and non-CNS ) to respond to Wnt signals . In RNA-seq data of Passage 3 control ( DMSO-treated ) ECs , LEF1 and TCF7 were strongly downregulated compared to Passage 1 cells . This result suggests that low baseline expression of these transcription factors , which form a complex with nuclear β-catenin to regulate Wnt target genes , may partially explain the poor efficacy of CHIR in matured ECs , although additional work is necessary to assess the functional relevance of these differences . Interestingly , ECs in non-BBB-forming regions of the CNS ( i . e . , circumventricular organs ) , and in the anterior pituitary , which is directly proximal to the CNS , retain some of their plasticity in adulthood ( Wang et al . , 2019 ) , possibly as the result of a delicate balance between Wnt ligands and Wnt-inhibitory factors in these regions . Our model should facilitate additional systematic examination of factors that may enhance or attenuate EC Wnt responsiveness . Finally , our work establishes an improved hPSC-based model for investigating mechanisms of BBB development in naïve ECs . hPSCs are an attractive model system to complement in vivo animal studies because they ( i ) are human , ( ii ) permit investigation of developmental processes in contrast to primary or immortalized cells , ( iii ) are highly scalable , ( iv ) can be derived from patients to facilitate disease modeling and autologous coculture systems , and ( v ) are genetically tractable . While widely used hPSC-based BBB models are useful for measuring molecular permeabilities and have been employed to understand genetic contributions to barrier dysfunction ( Vatine et al . , 2017; Vatine et al . , 2019; Lim et al . , 2017 ) , they have not been shown to proceed through a definitive endothelial progenitor intermediate ( Lippmann et al . , 2012; Lu et al . , 2021 ) and express epithelial-associated genes ( Qian et al . , 2017; Delsing et al . , 2018; Vatine et al . , 2019; Lu et al . , 2021 ) . Thus , new models with developmentally relevant differentiation trajectories and definitive endothelial phenotype are needed for improved understanding of developmental mechanisms . Motivated in part by prior use of ECs derived from hematopoietic progenitors in human cord blood to generate BBB models ( Boyer-Di Ponio et al . , 2014; Cecchelli et al . , 2014 ) , we and others recently showed that hPSC-derived naïve endothelial progenitors or ECs are good candidates for such a system ( Praça et al . , 2019; Nishihara et al . , 2020; Roudnicky et al . , 2020a; Roudnicky et al . , 2020b ) . For example , Praça et al . showed that a combination of VEGF , Wnt3a , and retinoic acid directed EPCs to brain capillary-like ECs with moderate TEER similar in order of magnitude to that reported here . We previously showed that BBB-like paracellular barrier characteristics are induced in hPSC-EPC-derived ECs after extended culture in a minimal medium . These so-called EECM-BMEC-like cells had TEER and small molecule permeability similar to primary human brain ECs , well-developed tight junctions , and an immune cell adhesion molecule profile similar to brain ECs in vivo ( Nishihara et al . , 2020 ) . In this study , we showed it was possible to use the small molecule Wnt agonist CHIR to induce additional hallmarks of CNS EC phenotype in hPSC-EPC-derived ECs , including canonical GLUT-1 , claudin-5 , and PLVAP effects ( both Passage 1 and 3 CHIR-treated ECs ) . However , it is important to note that despite the improvements in CNS EC character with CHIR treatment , further improvements to functional endocytosis , and efflux transporter and solute carrier phenotype should be targets of future study and may be facilitated by cocultures and/or additional molecular factors . Along these lines , the Passage 1 CHIR-treated CNS-like ECs would be at a differentiation stage well suited to investigate cues subsequent to Wnt signaling that may be key for the induction of additional CNS EC properties . Alternatively , the Passage 3 CHIR-treated CNS-like ECs may be suitable for other BBB modeling applications . In summary , our work has defined the EC response to Wnt activation in a simplified , human system and established a new hPSC-derived in vitro model that will facilitate improved understanding of endothelial barriergenesis . Tissue culture plates were coated with Matrigel , Growth Factor Reduced ( Corning , Glendale , AZ ) . A 2 . 5 mg aliquot of Matrigel was thawed and resuspended in 30 mL DMEM/F-12 ( Life Technologies , Carlsbad , CA ) , and the resulting solution used to coat plates at 8 . 7 µg/cm2 ( 1 mL per well for 6-well plates; 0 . 5 mL per well for 12-well plates ) . Plates were incubated at 37°C for at least 1 hr prior to use . hPSCs were maintained on Matrigel-coated plates in E8 medium ( STEMCELL Technologies , Vancouver , Canada ) at 37°C , 5% CO2 . hPSC lines used were IMR90-4 iPSC , WTC11 iPSC , H9-CDH5-eGFP hESC , H9-7TGP-ishcat2 hESC , and 19-9-11-7TGP-ishcat3 iPSC . Medium was changed daily . When hPSC colonies began to touch , typically at approximately 70–80% confluence , cells were passaged using Versene ( Life Technologies ) . Briefly , cells were washed once with Versene , then incubated with Versene for 7 min at 37°C . Versene was removed and cells were dissociated into colonies by gentle spraying with E8 medium . Cells were transferred at a split ratio of 1:12 to a new Matrigel-coated plate containing E8 medium . hPSC cultures were routinely tested for mycoplasma contamination using a PCR-based assay performed by the WiCell Research Institute ( Madison , WI ) . EPCs were differentiated according to previously published protocols ( Lian et al . , 2014; Bao et al . , 2016; Nishihara et al . , 2020 ) with slight modifications . On day –3 ( D-3 ) , hPSCs were treated with Accutase ( Innovative Cell Technologies , San Diego , CA ) for 7 min at 37°C . The resulting single-cell suspension was transferred to 4× volume of DMEM/F-12 ( Life Technologies ) and centrifuged for 5 min , 200× g . Cell number was quantified using a hemocytometer . Cells were resuspended in E8 medium supplemented with 10 µM ROCK inhibitor Y-27632 dihydrochloride ( Tocris , Bristol , UK ) and seeded on Matrigel-coated 12-well plates at a density of ( 1 . 5–2 . 5 ) × 104 cells/cm2 , 1 mL per well . Cells were maintained at 37°C , 5% CO2 . On the following two days ( D-2 and D-1 ) , the medium was replaced with E8 medium . The following day ( D0 ) , differentiation was initiated by changing the medium to LaSR medium ( Advanced DMEM/F-12 [Life Technologies] , 2 . 5 mM GlutaMAX [Life Technologies] , and 60 µg/mL l-ascorbic acid 2-phosphate magnesium [Sigma-Aldrich , St . Louis , MO] ) supplemented with 7–8 µM CHIR 99021 ( Tocris ) , 2 mL per well . The following day ( D1 ) , medium was replaced with LaSR medium supplemented with 7–8 µM CHIR 99021 , 2 mL per well . On the following three days ( D2 , D3 , and D4 ) , the medium was replaced with pre-warmed LaSR medium ( without CHIR ) , 2 mL per well . On D5 , EPCs were isolated using CD31 MACS . Cells were treated with Accutase for 15–20 min at 37°C . The resulting cell suspension was passed through a 40 µm cell strainer into an equal volume of DMEM ( Life Technologies ) supplemented with 10% FBS ( Peak Serum , Wellington , CO ) and centrifuged for 5 min , 200× g . Cell number was quantified using a hemocytometer . Cells were resuspended in MACS buffer ( Dulbecco’s phosphate buffered saline without Ca and Mg [DPBS; Life Technologies] supplemented with 0 . 5% bovine serum albumin [Sigma-Aldrich] and 2 mM EDTA [Sigma-Aldrich] ) at a concentration of 107 cells per 100 µL . The CD31-FITC antibody ( Miltenyi Biotec , Auburn , CA ) was added to the cell suspension at a dilution of 1:50 . The cell suspension was incubated for 30 min at room temperature ( RT ) , protected from light . The cell suspension was brought to a volume of 15 mL with MACS buffer and centrifuged for 5 min , 200× g . The supernatant was aspirated and the pellet resuspended in MACS buffer at a concentration of 107 cells per 100 µL . The FITC Selection Cocktail from the EasySep Human FITC Positive Selection Kit ( STEMCELL Technologies ) was added at a dilution of 1:10 , and the cell suspension was incubated for 20 min at RT , protected from light . The Dextran RapidSpheres ( magnetic particles ) solution from the Selection Kit was added at a dilution of 1:20 , and the cell suspension was incubated for an additional 15 min at RT . The cell suspension was brought to a total volume of 2 . 5 mL with MACS buffer ( for total cell number less than 2 × 108 , the approximate maximum yield from two 12-well plates; for a larger number of plates/cells , a total volume of 5 mL was used ) . 2 . 5 mL of cell suspension was transferred to a sterile 5 mL round-bottom flow cytometry tube and placed in the EasySep magnet ( STEMCELL Technologies ) for 5 min . The magnet was inverted to pour off the supernatant , the flow tube removed , the retained cells resuspended in 2 . 5 mL of MACS buffer , and the flow tube placed back in the magnet for 5 min . This step was repeated three times , and the resulting cell suspension transferred to a centrifuge tube , and centrifuged for 5 min , 200× g . Cell number was quantified using a hemocytometer . Resulting EPCs were used directly for experiments as described below or cryopreserved in hECSR medium supplemented with 30% FBS and 10% DMSO for later use . hECSR medium is Human endothelial serum-free medium ( Life Technologies ) supplemented with 1× B-27 supplement ( Life Technologies ) and 20 ng/mL FGF2 ( Waisman Biomanufacturing , Madison , WI ) . Collagen IV ( Sigma-Aldrich ) was dissolved in 0 . 5 mg/mL acetic acid to a final concentration of 1 mg/mL . Collagen IV-coated plates were prepared by diluting a volume of this stock solution 1:100 in water , adding the resulting solution to tissue culture plates , or #1 . 5 glass-bottom plates ( Cellvis , Sunnyvale , CA ) for cells intended for confocal imaging ( 1 mL per well for 6-well plates , 0 . 5 mL per well for 12-well plates , 0 . 25 mL per well for 24-well plates ) , and incubating the plates for 1 hr at RT . Collagen IV coating solution was removed , and EPCs obtained as described above were suspended in hECSR medium and plated at approximately 3 × 104 cells/cm2 . In some experiments , ligands and small molecules were added to hECSR medium: CHIR 99021 ( Tocris ) was used at 4 µM except where indicated; DMSO ( Sigma-Aldrich ) was used as a vehicle control for CHIR; Wnt3a ( R&D Systems ) was used at 20 ng/mL; doxycycline was used at 1 , 2 , or 4 µg/mL . The hECSR medium , including any ligands or small molecules , was replaced every other day until confluent ( typically 6 days ) . We denote this time point as ‘Passage 1 . ’ For extended culture , ECs were selectively dissociated and replated as previously described ( Nishihara et al . , 2020 ) . Cells were incubated with Accutase until ECs appeared round , typically 2–3 min at 37°C . The plate was tapped to release the ECs while SMLCs remained attached , and the EC-enriched cell suspension transferred to 4× volume of DMEM/F-12 and centrifuged for 5 min , 200× g . Cells were resuspended in hECSR medium and seeded on a new collagen IV-coated plate at approximately 3 × 104 cells/cm2 . hECSR medium was replaced every other day until confluent ( typically 6 days ) . The selective dissociation and seeding described above was repeated , and hECSR medium was again replaced every other day until confluent ( typically 6 days ) . We denote this time point as ‘Passage 3 . ’ In one experiment , these steps were repeated for another two passages . Except where indicated , CHIR 99021 or vehicle ( DMSO ) was included in the hECSR medium for the entire duration of culture . RNA-seq was performed on ECs and SMLCs from the IMR90-4 hPSC line . Four independent differentiations were performed , with DMSO- and CHIR-treated ECs at Passage 1 analyzed from all four differentiations . DMSO- and CHIR-treated ECs at Passage 3 were analyzed from three of the four differentiations . DMSO-treated SMLCs at Passage 1 were analyzed from two of the four differentiations . FACS was used to isolate CD31+ ECs and CD31– SMLCs from mixed Passage 1 cultures . Cells were incubated with Accutase for 10 min at 37°C , passed through 40 µm cell strainers into 4× volume of DMEM/F-12 , and centrifuged for 5 min , 200× g . Cells were resuspended in MACS buffer and incubated with CD31-APC antibody ( Miltenyi Biotec ) for 30 min at 4°C , protected from light . The cell suspension was brought to a volume of 15 mL with MACS buffer and centrifuged at 4°C for 5 min , 200× g . Cells were resuspended in MACS buffer containing 2 µg/mL 4′ , 6-diamidino-2-phenylindole ( DAPI; Life Technologies ) . A BD FACSAria III Cell Sorter ( BD Biosciences , San Jose , CA ) was used to isolate DAPI–CD31+ cells ( live ECs ) and DAPI–CD31– cells ( live SMLCs ) . The resulting cell suspensions were centrifuged at 4°C for 5 min , 200× g , and cell pellets immediately processed for RNA extraction as described below . RNA was isolated using the RNeasy Plus Micro Kit ( Qiagen , Germantown , MD ) . Buffer RLT Plus supplemented with 1% β-mercaptoethanol was used to lyse cells ( pellets from FACS of Passage 1 cells , or directly on plates for Passage 3 ECs ) . Lysates were passed through gDNA Eliminator spin columns , loaded onto RNeasy MinElute spin columns , washed with provided buffers according to the manufacturer’s instructions , and eluted with RNase-free water . Sample concentrations were determined using a NanoDrop spectrophotometer ( Thermo Scientific , Waltham , MA ) and RNA quality assayed using an Agilent 2100 Bioanalyzer with Agilent RNA 6000 Pico Kit ( Agilent , Santa Clara , CA ) . First-strand cDNA synthesis was performed using the SMART-Seq v4 Ultra Low Input RNA kit ( Takara Bio , Mountain View , CA ) with 5 ng input RNA followed by nine cycles of PCR amplification and library preparation using the Nextera XT DNA Library Prep Kit ( Illumina , San Diego , CA ) . Sequencing was performed on a NovaSeq 6000 ( Illumina ) , with approximately 40–60 million 150 bp paired-end reads obtained for each sample . FASTQ files were aligned to the human genome ( hg38 ) and transcript abundances quantified using RSEM ( v1 . 3 . 3 ) ( Li and Dewey , 2011 ) calling bowtie2 ( v2 . 4 . 2 ) ( Langmead and Salzberg , 2012 ) . Estimated counts from RSEM were input to DESeq2 ( v1 . 26 . 0 ) ( Love et al . , 2014 ) implemented in R ( v3 . 6 . 3 ) for differential expression analysis . Elsewhere , transcript abundances are presented as TPM . Differentiation pairing as described above was included in the DESeq2 designs . The Wald test with Benjamini–Hochberg correction was used to generate adjusted p-values . Principal component analysis was performed on counts after the DESeq2 variance stabilizing transformation . Transcription factor annotations were based on the list available at http://humantfs . ccbr . utoronto . ca/ ( Lambert et al . , 2018 ) ; secreted and transmembrane annotations were based on the UniProt database ( UniProt Consortium , 2021 ) . WGCNA ( v1 . 70–3 ) ( Zhang and Horvath , 2005; Langfelder and Horvath , 2008 ) was performed on the 14 EC datasets . Genes with an average of fewer than 50 estimated counts across these datasets were excluded , and the DESeq2 variance stabilizing transformation was used to generate the expression matrix for input to WGCNA . The topological overlap matrix ( TOM ) was constructed using the signed network type and a power of 20 . Hierarchical clustering was performed on dissimilarity ( 1 – TOM ) with average linkage . Gene modules were detected by a constant height ( 0 . 99 ) cut of the hierarchical clustering dendrogram with a minimum module size of 30 genes . Module eigengenes ( the first principal component of the expression matrix for genes in each module ) were computed as described , and the Pearson correlation between module eigengenes and experimental variables ( CHIR vs . DMSO: CHIR = 1 , DMSO = 0; Passage 3 vs . Passage 1: Passage 3 = 1 , Passage 1 = 0 ) was used to identify modules of interest . Cytoscape ( v3 . 8 . 3 ) ( Shannon et al . , 2003 ) was used to visualize the 30 genes in the green module ( strong positive correlation with CHIR treatment ) with the highest intramodular connectivity . The list of genes , corresponding modules , and correlations to experimental variables and module eigengenes is provided in Supplementary file 3 . Bulk RNA-seq data from the literature ( FASTQ files; see ‘Previously published datasets used’ ) were obtained from the Gene Expression Omnibus ( GEO ) . These FASTQ files were aligned to the mouse genome ( mm10 ) and transcript abundances quantified as described above . DESeq2 was used for differential expression analysis as described above . For direct comparison of human and mouse data , the biomaRt package ( v2 . 42 . 1 ) ( Durinck et al . , 2009 ) and Ensembl database ( Yates et al . , 2020 ) were used to map human gene names to mouse homologs . Venn diagrams were generated using the tool available at http://bioinformatics . psb . ugent . be/webtools/Venn/ . To identify solute carrier and efflux transporter genes highly expressed at the human BBB in vivo , we used five human brain scRNA-seq datasets ( Han et al . , 2020; Hodge et al . , 2019; La Manno et al . , 2016; Polioudakis et al . , 2019; Zhong et al . , 2020; see ‘Previously published datasets used’ ) integrated in a previous meta-analysis ( Gastfriend et al . , 2021 ) . SLC and ABC genes with average expression greater than 100 TPM in ECs across the five independent datasets were selected . For pathway enrichment analysis , lists of upregulated genes ( log2 ( fold change ) > 0 , adjusted p-value<0 . 05 , DESeq2 Wald test with Benjamini–Hochberg correction ) were tested against the Hallmark gene sets collection ( Liberzon et al . , 2015 ) using the tool available at http://www . gsea-msigdb . org/gsea/msigdb/annotate . jsp . Immunocytochemistry was performed in 24-well plates . Cells were washed once with 500 µL DPBS and fixed with 500 µL cold ( –20°C ) methanol for 5 min , except cells intended for calponin/SM22a and CD31/Ki67 detection , which were fixed with 500 µL of 4% paraformaldehyde for 15 min . Cells were washed three times with 500 µL DPBS and blocked in 150 µL DPBS supplemented with 10% goat serum ( Life Technologies ) for 1 hr at RT , except cells intended for calponin/SM22⍺ detection , which were blocked and permeabilized in DPBS supplemented with 3% BSA and 0 . 1% Triton X-100 , or cells intended for CD31/Ki67 detection , which were blocked and permeabilized in DPBS supplemented with 5% non-fat dry milk and 0 . 4% Triton X-100 . Primary antibodies diluted in 150 µL of the above blocking solutions ( see Key resources table for antibody information ) were added to cells and incubated overnight at 4°C on a rocking platform . Cells were washed three times with 500 µL DPBS . Secondary antibodies diluted in 150 µL of the above blocking solutions ( see Key resources table for antibody information ) were added to cells and incubated for 1 hr at RT on a rocking platform , protected from light . Cells were washed three times with 500 µL DPBS , followed by 5 min incubation with 500 µL DPBS plus 4 µM Hoechst 33342 ( Life Technologies ) . Images were acquired using an Eclipse Ti2-E epifluorescence microscope ( Nikon , Tokyo , Japan ) with a 20× or 30× objective or an A1R-Si+ confocal microscope ( Nikon ) with a 100× oil objective . Confocal images were acquired with 1 µm slice spacing . Images were analyzed using FIJI ( ImageJ ) software . For epifluorescence images , five fields ( 20× or 30× ) were analyzed per well , with 3–4 wells per treatment condition . For quantification of cell number , EC colonies were manually outlined , and the Analyze Particles function was used to estimate the number of nuclei within the EC colonies . Nuclei outside the EC colonies were manually counted . EC purity ( % EC ) was calculated as the number of nuclei within EC colonies relative to total nuclei . To estimate % GLUT-1+ ECs , cells within the EC colonies with membrane-localized GLUT-1 immunoreactivity were manually counted . To estimate % Ki67+ ECs , cells within the EC colonies with at least one nuclear-localized Ki67 punctum were manually counted . For quantification of fluorescence intensity in epifluorescence images , EC colonies were manually outlined , and the Measure function was used to obtain the mean fluorescence intensity for each image channel ( fluorophore ) . A cell-free area of the plate was similarly quantified for background subtraction . Following background subtraction , the mean fluorescence intensity of each protein of interest was normalized to the mean fluorescence intensity of Hoechst to correct for effects of cell density . For confocal images , 3–4 fields ( 100× ) containing only VE-cadherin+ ECs were analyzed per well , with four wells per treatment condition . The first slice with visible nuclei ( closest to glass ) was defined as Z = 0 , and the Measure function was used to obtain the mean fluorescence intensity for each image channel ( fluorophore ) in each slice from Z = 0 to Z = 7 µm . A cell-free area of the plate was similarly quantified for background subtraction . After background subtraction , to approximate total abundance ( area under the fluorescence versus Z curve [AUC] ) for each channel , mean fluorescence intensities were summed across all slices . AUCs for the proteins of interest were normalized to Hoechst AUC . Passage 1 cultures were dissociated by treatment with Accutase for 10 min at 37°C . Cell suspensions were passed through 40 µm cell strainers into 4× volume of DMEM/F-12 and centrifuged for 5 min , 200× g . Approximately 5 × 105 cells per replicate were resuspended in MACS buffer and incubated with the CD31-APC antibody ( Miltenyi Biotec ) for 30 min at 4°C , protected from light . Cell suspensions were brought to a volume of 5 mL with MACS buffer and centrifuged at 4°C for 5 min , 200× g . Cells were resuspended in 500 µL MACS buffer containing 2 µg/mL DAPI and 0 . 5 µL Vybrant DyeCycle Green Stain ( Invitrogen ) and incubated at RT for 1 hr , protected from light . Cells were analyzed on an Attune NxT flow cytometer ( Invitrogen ) . FlowJo software ( BD Biosciences ) was used to gate CD31+ cells and quantify the percentage of S/G2/M phase cells . To enrich samples from Passage 1 cultures for ECs , the Accutase-based selective dissociation method described above was employed . Dissociated cells were centrifuged for 5 min , 200× g , and resulting cell pellets were lysed in RIPA buffer ( Rockland Immunochemicals , Pottstown , PA ) supplemented with 1× Halt Protease Inhibitor Cocktail ( Thermo Scientific ) . Passage 3 cells were lysed with the above buffer directly on plates . Lysates were centrifuged at 4°C for 5 min , 14 , 000× g , and protein concentration in supernatants quantified using the Pierce BCA Protein Assay Kit ( Thermo Scientific ) . Equal amounts of protein were diluted to equal volume with water , mixed with sample buffer , and heated at 95°C for 5 min , except lysates intended for GLUT-1 western blotting , which were not heated . Samples were resolved on 4–12% Tris-Glycine gels and transferred to nitrocellulose membranes . Membranes were blocked for 1 hr in Tris-buffered saline plus 0 . 1% Tween-20 ( TBST ) supplemented with 5% non-fat dry milk . Primary antibodies ( see Key resources table for antibody information ) diluted in TBST plus 5% non-fat dry milk were added to membranes and incubated overnight at 4°C on a rocking platform . Membranes were washed five times with TBST . Secondary antibodies ( see Key resources table for antibody information ) diluted in TBST were added to membranes and incubated for 1 hr at RT on a rocking platform , protected from light . Membranes were washed five times with TBST and imaged using an Odyssey 9120 ( LI-COR , Lincoln , NE ) . Band intensities were quantified using Image Studio software ( LI-COR ) . A fixable , Alexa Fluor 488-conjugated dextran with an average molecular weight of 10 kDa ( Invitrogen ) was used as a tracer to estimate total fluid-phase endocytosis . Dextran was added at 10 µM to the medium of Passage 1 cultures . Plates were incubated on rotating platforms at 37 or 4°C for 2 hr . For inhibitor experiments , 20 µM chlorpromazine ( Sigma ) , 100 U/mL nystatin ( Sigma ) , or 2 µM rottlerin ( Tocris ) were added to the medium 30 min prior to addition of dextran . Medium was removed and cells were washed once with DPBS , and then incubated with Accutase for 10 min at 37°C . Cell suspensions were passed through 40 µm cell strainers into 4× volume of DMEM/F-12 and centrifuged for 5 min , 200× g . Cells were resuspended in MACS buffer and incubated with the CD31-APC antibody ( Miltenyi Biotec ) for 30 min at 4°C , protected from light . Cell suspensions were brought to a volume of 5 mL with MACS buffer and centrifuged at 4°C for 5 min , 200× g . Pellets were resuspended in DPBS supplemented with 4% paraformaldehyde and incubated for 15 min at RT , protected from light . Cells were centrifuged for 5 min , 200× g . Pellets were resuspended in MACS buffer and analyzed on a BD FACSCalibur flow cytometer ( BD Biosciences ) . FlowJo software was used to gate CD31+ cells and quantify geometric mean fluorescence intensity and CV of dextran . For imaging , the dextran accumulation assay was performed on cells cultured on #1 . 5 glass-bottom plates . After 2 hr of dextran treatment , medium was removed and cells washed with DPBS . Cells were fixed with 4% paraformaldehyde for 15 min . Cells were washed three times with 500 µL DPBS and blocked and permeabilized with DPBS supplemented with 10% goat serum and 0 . 1% Triton X-100 for 1 hr at RT . Cells were stained with the caveolin-1 primary antibody and imaged on a confocal microscope as described above . Transwell inserts ( 6 . 5 mm diameter with 0 . 4 µm pore polyester filters ) ( Corning ) were coated with 50 µL of a solution of collagen IV ( 400 µg/mL ) and fibronectin ( 100 µg/mL ) in water for 4 hr at 37°C . Passage 3 DMSO- and CHIR-treated ECs were seeded on Transwell inserts at 105 cells/cm2 in hECSR medium supplemented with DMSO or CHIR . Medium volumes were 200 µL for the apical chamber and 800 µL for the basolateral chamber . Beginning the day after seeding , TEER was measured daily for 6 days using an EVOM2 epithelial voltohmmeter with STX2 chopstick electrodes ( World Precision Instruments , Sarasota , FL ) . Medium was replaced every other day . TEER values were corrected by subtracting the resistance of a collagen IV/fibronectin-coated Transwell insert without cells and multiplying by the filter surface area of 0 . 33 cm2 . Permeability of endothelial monolayers to sodium fluorescein was assessed 6 days after seeding cells on Transwell inserts . Medium in both apical and basolateral chambers was replaced and cells returned to the incubator for 1 hr . Medium in apical chambers , including the apical chamber of a collagen IV/fibronectin-coated Transwell insert without cells , was then replaced with medium supplemented with 10 µM sodium fluorescein ( Sigma-Aldrich ) , and plates placed on an orbital platform in an incubator . At 15 , 30 , 45 , and 60 min , an 80 µL sample of the basolateral chamber medium was withdrawn from each Transwell , transferred to a 96-well plate , and 80 µL fresh medium replaced in the basolateral chamber of each Transwell . At 60 min , an 80 µL sample of apical chamber medium was also withdrawn from each Transwell and transferred to the 96-well plate . 80 µL of medium lacking sodium fluorescein was also transferred to the 96-well plate for background subtraction . Fluorescence intensity of all samples was measured using an Infinite M1000 PRO plate reader ( Tecan , Männedorf , Switzerland ) with 485 nm excitation and 530 nm emission wavelengths . Background-subtracted fluorescence intensity values at the 30 , 45 , and 60 min timepoints were corrected for sampling-induced dilution as previously described ( Stebbins et al . , 2016 ) . The endothelial permeability coefficient ( Pe ) , which is a concentration-independent parameter corrected for the permeability of a cell-free Transwell insert , was calculated as previously described ( Stebbins et al . , 2016 ) . Individual wells of cultured cells that underwent identical experimental treatments are defined as replicates , and all key experiments were repeated using multiple independent hPSC differentiations . Detailed information about replication strategy is provided in figure legends . Student’s t test was used for comparison of means from two experimental groups . One-way analysis of variance ( ANOVA ) was used for comparison of means from three or more experimental groups , followed by Dunnett’s post-hoc test for comparison of multiple treatments to a single control , or Tukey’s honest significant difference ( HSD ) post-hoc test for multiple pairwise comparisons . When data from multiple differentiations were combined , two-way ANOVA ( one factor being the experimental treatment and one factor being the differentiation ) was used for comparison of means to achieve blocking of differentiation-based variability , followed by post-hoc tests as described above if more than two experimental treatments were compared . For fluorescence intensities ( a . u . ) , two-way ANOVA was performed prior to normalization of these values to the control group within each differentiation ( for visualization in plots ) . Statistical tests were performed in JMP Pro ( v15 . 0 . 0 ) . For RNA-seq differential expression analysis , the DESeq2 Wald test with Benjamini–Hochberg correction was used to calculate p-values . Descriptions of the statistical tests used are provided in figure legends .	The cells that line the inside of blood vessels are called endothelial cells . In the blood vessels of the brain , these cells form a structure called the ‘blood-brain barrier’ , which allows nutrients to pass from the blood into the brain , while at the same time preventing harmful substances like toxins from crossing . Faults in the blood-brain barrier can contribute to neurological diseases , but the blood-brain barrier can also restrict drugs from accessing the brain , making it difficult to treat certain conditions . Understanding how the endothelial cells that form the blood-brain barrier develop may offer insight into new treatments for neurological diseases . During the development of the embryo , endothelial cells develop from stem cells . They can also be generated in the laboratory from human pluripotent stem cells or ‘hPSCs’ , which are cells that can produce more cells like themselves , or differentiate into any cell type in the body . Scientists can treat hPSCs with specific molecules to make them differentiate into endothelial cells , or to modify their properties . This allows researchers to monitor how different types of endothelial cells form . Endothelial cells at the blood-brain barrier are one of these types . During their development , these cells gain distinct features , including the production of proteins called GLUT-1 , claudin-5 and LSR . GLUT-1 transports glucose across endothelial cells’ membranes , while claudin-5 and LSR tightly join adjacent cells together , preventing molecules from leaking into the brain through the space between cells . In mouse endothelial cells , a signaling protein called Wnt is responsible for turning on the genes that code for these proteins . But how does Wnt signaling impact human endothelial cells ? Gastfriend et al . probed the effects of Wnt signaling on human endothelial cells grown in the lab as they differentiate from hPSCs . They found that human endothelial cells developed distinct blood-brain barrier features when Wnt signaling was activated , producing GLUT-1 , claudin-5 and LSR . Gastfriend et al . also found that human endothelial cells were more responsive to Wnt signaling earlier in their development . Additionally , they identified the genes that became activated in human endothelial cells when Wnt signaling was triggered . These findings provide insight into the development and features of the endothelial cells that form the human blood-brain barrier . The results are a first step towards a better understanding of how this structure works in humans . This information may also allow researchers to develop new ways to deliver drugs into the brain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2021	Wnt signaling mediates acquisition of blood–brain barrier properties in naïve endothelium derived from human pluripotent stem cells
Anticipating the future is a key motif of the brain , possibly supported by mental simulation of upcoming events . Rodent single-cell recordings suggest the ability of spatially tuned cells to represent subsequent locations . Grid-like representations have been observed in the human entorhinal cortex during virtual and imagined navigation . However , hitherto it remains unknown if grid-like representations contribute to mental simulation in the absence of imagined movement . Participants imagined directions between building locations in a large-scale virtual-reality city while undergoing fMRI without re-exposure to the environment . Using multi-voxel pattern analysis , we provide evidence for representations of absolute imagined direction at a resolution of 30° in the parahippocampal gyrus , consistent with the head-direction system . Furthermore , we capitalize on the six-fold rotational symmetry of grid-cell firing to demonstrate a 60° periodic pattern-similarity structure in the entorhinal cortex . Our findings imply a role of the entorhinal grid-system in mental simulation and future thinking beyond spatial navigation . Anticipation of the future is a central adaptive function of the brain and enables adequate decision-making and planning . Simulating or imagining future events and scenarios relies on a network of brain regions known to be involved in episodic memory , navigation and prediction ( Buckner , 2010; Byrne et al . , 2007; Hassabis and Maguire , 2007; Hasselmo , 2009; Schacter et al . , 2012 ) . For instance , before leaving your favorite cafe , you may picture the scenery in front of the cafe in your mind’s eye to determine whether to take a left or a right turn to get home . To accomplish this you have to recall both the location of the cafe as well as the direction you are facing when leaving the building . Electrophysiological recordings in freely moving rodents have demonstrated that positional information during navigation is represented by place cells in the hippocampus ( O’Keefe and Dostrovsky , 1971 ) and grid cells in entorhinal cortex ( Hafting et al . , 2005 ) . Place cells typically exhibit one firing field ( O’Keefe and Dostrovsky , 1971 ) , while grid cells are characterized by multiple firing fields arranged in a regular hexagonal pattern tessellating the entire environment ( Hafting et al . , 2005 ) . Complementarily , directional information is carried by head direction cells , which increase their firing rate as a function of the animal’s directional heading irrespective of its location ( Taube et al . , 1990; Taube , 2007 ) . Intracranial recordings in patients exploring virtual-reality ( VR ) environments demonstrated the existence of place and grid cells in the human hippocampus and entorhinal cortex , respectively ( Ekstrom et al . , 2003; Jacobs et al . , 2010 , 2013 ) . A 60° directional periodicity of BOLD-signal modulations in the entorhinal cortex during virtual navigation indicates that grid-like entorhinal signals can also be detected with fMRI ( Doeller et al . , 2010; Kunz et al . , 2015; Horner et al . , 2016 ) . Notably , place cell activity can also represent locations other than the one currently occupied by the animal as illustrated by activation sequences corresponding to upcoming trajectories during rest periods ( Dragoi and Tonegawa , 2011 ) . Intriguingly , these ‘preplay’ sequences preferentially represent paths leading up to motivationally relevant locations ( Ólafsdóttir et al . , 2015 ) . These observations support the notion that prospective coding of hippocampal place cells relates to the well-established role of the human hippocampus in mental simulation and imagination ( Buckner , 2010; Byrne et al . , 2007; Hassabis and Maguire , 2007; Hasselmo , 2009; Schacter et al . , 2012 ) . Akin to firing rate increases of neurons in the human medial temporal lobe specific to the content of imagination ( Kreiman et al . , 2000 ) , firing patterns of spatially tuned cells might be reinstated to imagine the view from a certain location during mental simulation ( Bird et al . , 2012; Byrne et al . , 2007; Hasselmo , 2009 ) . Prospective coding properties of grid cells ( De Almeida et al . , 2012; Kropff et al . , 2015 ) and recent evidence for spatial coherence of grid with place cell activity during replay ( Ólafsdóttir et al . , 2016 ) further suggest a similar involvement of the entorhinal grid system in future anticipation and prediction . This is in line with the observation of grid-like representations during imagined movement through an environment ( Horner et al . , 2016 ) . However , hitherto it remains unknown if grid-like representations support mental simulation independent of imagined movement , which could suggest a more general role of grid cell computations in navigational planning , future anticipation and cognition . We combined fMRI with multi-voxel pattern analysis and VR to investigate whether the entorhinal grid system contributes to the imagination of directions from stationary viewpoints ( Figure 1a , b ) . After extensive navigation training ( see Materials and methods and Figure 1—figure supplement 1 ) , participants were asked to imagine directions between pairs of buildings in ‘Donderstown’ , a large-scale realistic VR city ( http://www . doellerlab . com/donderstown/ ) . In a carefully counterbalanced design , we probed the fine-grained representations of twelve equally spaced directions ( see Materials and methods and Figure 1—figure supplement 2 ) . Imagined directions had to be indicated ( Figure 1b ) and participants successfully performed this task ( mean error 33 . 68° ± 19 . 09° SD; Figure 1c ) . Behavioral performance in the direction-imagination task was highly correlated with navigation success ( r = 0 . 85 , p<0 . 001 ) and the accuracy of direction estimates ( r = 0 . 94 , p<0 . 001 ) during training and performance in a post-scan map test ( r = 0 . 95 , p<0 . 001 ) , indicating successful translation of the acquired representation of Donderstown to the imagination task ( Figure 1—figure supplement 3 ) . 10 . 7554/eLife . 17089 . 003Figure 1 . Direction-imagination task . ( a ) Twelve evenly spaced directions were sampled using 18 buildings distributed regularly across Donderstown . We sampled each direction ( indicated by black arrows ) from different start locations ( yellow circles ) , which dissociated the directions from visual features of imagined views ( Figure 1—figure supplement 2 ) , and employed a counterbalancing regime ensuring equal sampling of directions and start locations throughout the experiment ( see Materials and methods ) . Buildings marked with a green circle served as target locations only . Importantly , the regular arrangement of building locations did not correspond to the street layout and was not revealed to participants , who experienced Donderstown only from a first-person perspective ( see also Figure 1—figure supplement 1d ) . ( b ) Trials began with a cue indicating start ( top building name ) and target ( bottom building name ) location and thereby defining the relevant direction ( black arrow ) . During an imagination period the screen was black and participants were instructed to imagine the view they would encounter when standing in front of the start building facing the direction of the target building . An auditory signal terminated the imagination period and participants indicated the imagined direction ( red arrow ) in a sparse VR environment , followed by a confidence judgment . Performance was measured as the absolute angular difference between the correct and the indicated direction ( red arc ) . Note that only the bottom row of images was presented to participants , top row for illustration only . ( c ) Circular histogram of average absolute angular difference between correct and indicated directions across participants ( mean error 33 . 68° ± 19 . 09° SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 00310 . 7554/eLife . 17089 . 004Figure 1—source data 1 . Average absolute angular errors . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 00410 . 7554/eLife . 17089 . 005Figure 1—figure supplement 1 . Overview of behavioral training . ( a ) To familiarize themselves with the controls of the computer game and the layout of the city , participants explored Donderstown for 10 min and searched for a set of landmarks irrelevant for the direction imagination task . This exploration phase was omitted in the second training session . The black circle and arrow on the map in the top panel indicate the participants’ position and orientation when first encountering Donderstown . ( b ) Subsequently , participants learned the names of 18 task-relevant buildings ( top ) to criterion . Knowledge of the building names was assessed in test blocks during which participants had to select the building belonging to the presented name from a display of three buildings by pressing one of three buttons ( bottom ) . ( c ) For the remainder of the session , participants were trained on the building locations in Donderstown . Bottom row shows the trial structure as presented to the participants , top row for illustration only . Participants were instructed to navigate to the building whose name was presented on the screen . Once the building was located , participants encoded the position and were then asked to estimate the direction to the following target building . Performance was measured as the number of buildings located during the training session and the absolute angular difference between the estimated direction and the correct direction as defined by the current location and the new target building . ( d ) Overview of Donderstown highlighting the task-relevant buildings , which largely differed in features salient from the first-person perspective such as size , shape and rotation with respect to the hexagonal building layout ( red arrows ) , which makes an influence of the regular arrangement of their entrances on participants’ cognitive representation of the city unlikely . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 00510 . 7554/eLife . 17089 . 006Figure 1—figure supplement 2 . Sampling of directions in the imagination task . ( a–f ) From each of the six start locations ( yellow circles ) ten directions were sampled . Directions ( black arrows ) were defined based on the angle of the vector connecting the start and the target locations ( green circles ) . Screenshots show view from Donderstown corresponding to direction indicated by dashed arrow . Note that start locations could also be goal locations . The building combinations used in the direction imagination task were carefully counterbalanced so that in a task block of 24 trials each direction was sampled twice , each start building served as a start location four times and each building combination did not occur more than twice throughout the experiment ( see Materials and methods ) . Trials sampling directions using buildings located on the same street ( purple in b and c ) were subject to an additional control analysis ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 00610 . 7554/eLife . 17089 . 007Figure 1—figure supplement 3 . Accuracy of direction judgments during imagination task is related to behavioral performance during training and a post-scan map test . ( a ) Across subjects , the accuracy of direction judgments in the direction imagination task ( π minus the mean angular difference between the correct and indicated directions in radians ) correlated significantly with navigation success indexed by the number of buildings found during training . ( b ) Accuracy of direction judgments during the training sessions also correlated highly with performance during the imagination task . ( c ) Additionally , performance during the imagination task was correlated with z-scored accuracy in the post-scan map test . All correlations remained significant when controlling for variability in the time spent navigating the VR city during training using partial correlations ( partial correlations coefficients r>0 . 84 , all p<0 . 001 ) . Correlations were also significant when calculated between the accuracy of direction judgments and training performance measures separately for the two training sessions ( all correlations coefficients r>0 . 80 , all p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 007 The contribution of spatial representations to imagination was assessed using representational similarity analysis ( Kriegeskorte and Kievit , 2013 ) , which compares activation patterns across voxels to estimate neural similarity . In line with the suggested role for grid cell computations in vector navigation ( Bush et al . , 2015 ) , we expected the grid-cell system to be involved in determining the vector comprising the direction and distance from the start to the target building in our task . Our approach focused on the direction to the target to track putative grid-cell representations during imagination by systematically comparing neural similarity of imagined directions with varying angular differences ( see Materials and methods ) . We predicted that an involvement of the grid system in mental simulation should be reflected in a 60° periodic pattern-similarity structure in the entorhinal cortex , consistent with the hexagonal firing properties of grid cells ( Hafting et al . , 2005 ) and the hexadirectional fMRI signal in entorhinal cortex observed during virtual navigation ( Doeller et al . , 2010; Horner et al . , 2016; Kunz et al . , 2015 ) . It is important to note that we did not rely on the building layout as an absolute reference frame in our analyses , but rather assessed pattern similarity based on the relative angle between the directions sampled in a trial pair , see below . In a first step , we ascertained that absolute directional representations are detectable with our novel imagination task . We expected increased neural similarity during imagination of similar directions , consistent with previous findings of absolute directional coding during navigation in parahippocampal cortex ( Doeller et al . , 2010 ) and two recent studies reporting directional representations during imagination in a local reference frame in the retrosplenial complex ( Marchette et al . , 2014 ) and coarse representations of directions to a goal in the entorhinal/subicular region ( Chadwick et al . , 2015 ) . However , it remains unclear whether global spatial representations are involved in human imagination in the absence of visual input . Here , we compared pattern similarity during imagination of directions in pairs of trials sampling similar directions ( angular difference ≤ 30° ) to pairs of trials sampling dissimilar directions ( Figure 2a ) in brain regions representing facing direction ( Baumann and Mattingley , 2010; Chadwick et al . , 2015; Marchette et al . , 2014; Vass and Epstein , 2016 , 2013 ) . We observed the predicted one-fold symmetric pattern-similarity structure in a cluster of voxels in the left posterior parahippocampal gyrus ( T23 = 4 . 82 , p = 0 . 024 , FWE-corrected for multiple comparisons using small volume correction; Figure 2b , c; see Materials and methods ) . Increased pattern similarity for similar directions was not due to trial comparisons with identical building combinations ( Figure 2—figure supplement 1 ) . Further , this effect was not driven by the specific locations used to sample directions or the distances between these locations in Donderstown ( Figure 2—figure supplement 2 , see Materials and methods ) . 10 . 7554/eLife . 17089 . 008Figure 2 . Absolute directional coding in posterior parahippocampal gyrus . ( a ) Analysis logic of the one-fold directional analysis for three example trials . High pattern similarity was predicted for pairs of trials sampling similar directions with a maximum angular difference of 30° ( red ) compared to trials sampling directions 60° or more apart ( blue ) . Note that for illustration purposes the predicted similarity matrix is shown for comparisons across conditions , not single trials . ( b ) Searchlight results show a significant cluster of voxels in the posterior parahippocampal gyrus ( peak voxel MNI coordinates: 34 -34 -10; T23 = 4 . 82 , p = 0 . 024 corrected for multiple comparisons using small-volume correction ) with higher pattern similarity for trials sampling similar directions compared to trials sampling dissimilar directions . Results are shown on the structural MNI template . For display purposes , the statistical map is thresholded at p<0 . 001 uncorrected . ( c ) Sagittal and coronal view of the mask used to correct for multiple comparisons ( see Materials and methods ) displayed on the MNI template brain . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 00810 . 7554/eLife . 17089 . 009Figure 2—source data 1 . Searchlight results for absolute directional coding analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 00910 . 7554/eLife . 17089 . 010Figure 2—figure supplement 1 . Increased pattern similarity for similar directions after excluding trial pairs sampling a direction with the same combination of buildings . To control for the possibility of increased pattern similarity in the parahippocampal gyrus for similar direction pairs being due to the imagination of identical scenes , we excluded trial pairs from the analysis in which the same combination of start and target building was used . This revealed a significant cluster of voxels ( peak voxel MNI coordinates: 34 -36 -10 , T23 = 4 . 54 , p = 0 . 042 corrected for multiple comparisons using small volume correction ) very similar to the one observed in the main analysis ( see Figure 2b ) . For display purposes , the statistical map is thresholded at p<0 . 001 uncorrected . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01010 . 7554/eLife . 17089 . 011Figure 2—figure supplement 2 . Absolute directional coding during imagination is independent of locations and distances in Donderstown . ( a ) Pattern similarity difference between trial pairs sampling similar and dissimilar directions in the peak voxel of the main absolute directional coding analysis ( Figure 2 ) after exclusion of comparisons with trials using the same start ( bar I ) and the same target location ( bar II ) . Both T23 > 3 . 00 , both p<0 . 007 . ( b ) We considered the distances between start and target locations in a trial pair by controlling for three distance measures , which differed between pairs of trials sampling similar and dissimilar directions . The distance measures are illustrated based on two example trials . For each trial pair , we calculated ( I ) the mean length of the vectors connecting start and target location , ( II ) the difference in length of the two vectors and ( III ) the mean length of the vectors connecting all four relevant locations of a pair . In separate GLMs we used the distance measures as predictors of pairwise pattern similarity . ( c ) Mean pattern similarity difference in peak voxel of cluster from main absolute directional coding analysis ( Figure 2 ) between trial pairs sampling similar directions and pairs sampling dissimilar directions computed on the residuals of the GLMs . With this approach we controlled for pattern similarity due to ( I ) the average distance from start to target location in a trial pair , ( II ) the difference in distance from start to target location in a trial pair and ( III ) the average distance between all four buildings in a trial pair ( see Materials and methods ) . All T23 > 3 . 60 , all p<0 . 001 . Bars in a and c show mean pattern similarity difference with error bars reflecting S . E . M . , dashed line shows mean pattern similarity difference in peak voxel from main analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 011 Having verified that we can detect directional representations in our novel imagination paradigm , we tested , in a next step , whether activation patterns during imagination follow a six-fold rotational symmetry , akin to the six-fold symmetric firing pattern of grid cells ( Hafting et al . , 2005 ) and the six-fold modulation of entorhinal fMRI signals during virtual ( Doeller et al . , 2010; Kunz et al . , 2015 ) and imagined ( Horner et al . , 2016 ) navigation in humans . The rationale underlying our analysis is that activation patterns during directional imagination should exhibit the highest neural similarity for directions that are ( multiples of ) 60° apart from each other ( see Figure 3a–d and Figure 3—figure supplement 1 for details of analysis logic ) . Because grid cells are most abundant in the medial entorhinal cortex in rodents ( Hafting et al . , 2005 ) , we predicted the effect to be present in posterior medial entorhinal cortex ( pmEC ) , the likely homologue region of the rodent medial entorhinal cortex in the human brain ( Navarro Schröder et al . , 2015 ) ( Figure 3e ) . 10 . 7554/eLife . 17089 . 012Figure 3 . Grid-like representations during imagination . ( a ) Six-fold symmetric firing fields of a hypothetical grid cell ( dark blue dotted circles ) superimposed on an aerial view of Donderstown . Black arrows indicate the twelve sampled directions; light and dark shading highlights directions ( multiples of ) 60° apart . For illustration purposes , the grid orientation is aligned to the sampled directions; see Figure 3—figure supplement 1 for a different example . ( b ) The firing rate of the hypothetical response of the grid-cell system as a function of direction , showing a 60° modulation . Shading displays sampling of directions and red and blue markers indicate the two conditions . Note that the oscillatory firing pattern is sampled at the same phase in the 0° modulo 60° condition , but at different phases in the 30° modulo 60° condition . ( c ) Based on this , we expected a 60° modulation of fMRI pattern similarity values when comparing trial pairs based on the angular difference of their sampled directions . Red and blue shading illustrates the two conditions . ( d ) Specifically , we predicted higher pattern similarity for trial pairs with a remainder of 0° ( 0° modulo 60° condition , red ) compared to trial pairs with a remainder of 30° ( 30° modulo 60° condition , blue ) , when dividing the angular difference of the pair’s sampling directions by 60° . Note that for illustration purposes the predicted similarity matrix is shown for comparisons across conditions , not single trials . ( e ) ROI mask for posterior medial entorhinal cortex ( pmEC ) from previous report ( Navarro Schröder et al . , 2015 ) . ( f ) Pattern similarity difference ( mean and S . E . M . ) between the two conditions . The left pmEC exhibited a significant 60° modulation of pattern similarity . No significant differences in pattern similarity were observed in the right pmEC ( T23 = 0 . 57 , p = 0 . 58 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01210 . 7554/eLife . 17089 . 013Figure 3—source data 1 . Pattern similarity difference between 0° modulo 60° and 30° modulo 60° condition in left and right posterior medial entorhinal cortex . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01310 . 7554/eLife . 17089 . 014Figure 3—figure supplement 1 . Rationale of 60° modulation analysis . ( a ) Six-fold symmetric firing fields of a hypothetical grid cell ( dark blue dotted circles ) superimposed on a top-down view of Donderstown . Black arrows indicate the twelve sampled directions . Light and dark shading indicates directions ( multiples of ) 60° apart . ( b ) The firing rate of the hypothetical grid cell as a function of sampling direction exhibits a 60° modulation . Shading shows sampled directions with red and blue markers illustrating the two conditions . Note that the oscillatory firing pattern is sampled at the same phase in the 0° modulo 60° condition , but at different phases in the 30° modulo 60° condition . ( c ) Based on this , we expected increased pattern similarity when comparing trial pairs from the 0° modulo 60° condition to trial pairs from the 30° modulo 60° condition . The difference between the conditions is smaller than in Figure 3c due to the different sampling of directions with respect to the grid orientation . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01410 . 7554/eLife . 17089 . 015Figure 3—figure supplement 2 . Pattern similarity difference between 0° modulo 60° condition and 30° modulo 60° condition in pmEC and alEC . ( a ) ROI mask for anterior lateral entorhinal cortex ( alEC ) based on our previous report ( Navarro Schröder et al . , 2015 ) . ( b ) Pattern similarity did not differ between the 0° modulo 60° and the 30° modulo 60° condition in alEC ( T23 = 0 . 04 , p = 0 . 97 and T23 = − 0 . 08 , p = 0 . 94 for left and right alEC , respectively ) . ( c ) Colored markers show pattern similarity difference for each participant in pmEC and alEC . Boxplots indicate 25th and 75th percentile with the middle line representing median pattern similarity difference across participants . Whiskers extend to most extreme data points not considered outliers . Data points defined as outliers ( values more than 1 . 5 times the interquartile range above the 75th percentile or more than 1 . 5 times the interquartile range below the 25th percentile ) are represented by square markers . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01510 . 7554/eLife . 17089 . 016Figure 3—figure supplement 3 . Signal quality in the entorhinal cortex . ( a ) To assess signal quality in pmEC and alEC , we computed the temporal signal-to-noise ratio ( tSNR , see Materials and methods ) . A repeated-measures ANOVA revealed neither a main effect of region ( F1 , 23 = 0 . 60 , p = 0 . 448 ) or hemisphere ( F1 , 23 = 0 . 00 , p = 0 . 953 ) nor an interaction between the factors region and hemisphere ( F1 , 23 = 0 . 97 , p = 0 . 336 ) . Bars represent mean and S . E . M . ( b ) Left slice shows the mean functional scan averaged across participants . The mean functional images from the four fMRI runs were averaged for each participant before averaging the resulting mean images across participants . Right slice shows the corresponding section of the MNI template . Note that for some participants the edge of the superior parietal lobe was outside the field of view . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01610 . 7554/eLife . 17089 . 017Figure 3—figure supplement 4 . 60° periodicity of pattern similarity is consistent across angular differences only in left posterior medial entorhinal cortex . ( a ) Pattern similarity was analyzed based on the angular differences of the directions sampled in a trial pair . High pattern similarity was predicted for pairs in the 0° modulo 60° condition ( red ) in contrast to pairs in the 30° modulo 60° condition ( blue ) . Filled pattern and color indicates angular differences and corresponds to bars in ( b–e ) , which visualize average pattern similarity for all possible angular differences for exploratory purposes in the entorhinal ROIs . Note the consistent 60° periodicity of the pattern similarity profile across angular differences in left pmEC . A statistical test was performed on the within-subject difference between the two conditions and was significant in left pmEC only ( T23 = 2 . 37 , p = 0 . 027 , for all other ROIs , p>0 . 5 ) . Error bars for each angular difference would not reflect the statistical test performed and are therefore omitted . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01710 . 7554/eLife . 17089 . 018Figure 3—figure supplement 5 . Pattern similarity structure across pair-wise comparisons of trials for entorhinal ROIs . ( a–d ) Matrices show the pair-wise correlations between voxel patterns in the subregions of the entorhinal cortex ( a: left pmEC; b: right pmEC; c: left alEC; d: right alEC ) across all possible trial comparisons averaged across participants . Arrows signal sampled directions in a given pair of trials . Colorbar indicates size of the correlations in panels ( a–d ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01810 . 7554/eLife . 17089 . 019Figure 3—figure supplement 6 . 60° modulation of pattern similarity during imagination is not driven by specifics of task design . ( a ) Pattern similarity difference between the 0° modulo 60° and the 30° modulo 60° condition remained significant in left pmEC after controlling for specifics of the design . Bars show the mean pattern similarity difference after excluding trial pairs with ( I ) the same start location ( T23 = 2 . 39 , p = 0 . 025 ) , ( II ) the same target location ( T23 = 2 . 57 , p = 0 . 017 ) , ( III ) the same combination of start and target location ( T23 = 2 . 45 , p = 0 . 022 ) , ( IV ) pairs from the same run ( T23 = 2 . 08 , p = 0 . 049 ) and ( V ) pairs with target locations in the inner ring of buildings ( T23 = 5 . 29 , p<0 . 001; see Materials and methods ) . This excludes potential influences of imagining the same start or same target location , the same combination of start and target location and temporal auto-correlations on the effect . Error bars indicate S . E . M . , dashed line shows mean pattern similarity difference in left pmEC from main analysis ( Figure 3b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 01910 . 7554/eLife . 17089 . 020Figure 3—figure supplement 7 . 60° modulation of pattern similarity during imagination after controlling for distance measures . We controlled for the distances between start and target locations in a trial pair using three distance measures , which differed between the 0° modulo 60° and the 30° modulo 60° condition ( see Materials and methods ) . The distance measures are illustrated in Figure 2—figure supplement 2b . In separate GLMs we used the distance measures as predictors of pairwise pattern similarity and computed the mean pattern similarity difference between the 0° modulo 60° and the 30° modulo 60° condition on the residuals of these GLMs . With this approach we controlled for pattern similarity due to ( I ) the average distance from start to target location in a trial pair , ( II ) the difference in distance from start to target location in a trial pair and ( III ) the average distance between all four buildings in a trial pair ( see Materials and methods ) . All T23 > 2 . 36 , all p<0 . 03 , error bars indicate S . E . M . , dashed line shows mean pattern similarity difference in left pmEC from main analysis ( Figure 3b ) . The effect also remained significant when using binary ( high vs . low ) distance predictors ( all T23 > 2 . 44 , all p<0 . 03 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 02010 . 7554/eLife . 17089 . 021Figure 3—figure supplement 8 . Behavioral performance for trial pairs in the 0° modulo 60° and the 30° modulo 60° condition . Error values were multiplied for the two trials of each pair and averaged for the two conditions . Boxplots indicate 25th and 75th percentile with the middle line representing median combined error across participants . Whiskers extend to most extreme data points not considered outliers . Data points connected by lines show combined errors for all subjects in the two conditions ( no difference between conditions , T23 = 1 . 24 , p = 0 . 227 ) . Data points defined as outliers ( values more than 1 . 5 times the interquartile range above the 75th percentile or more than 1 . 5 times the interquartile range below the 25th percentile ) are represented by square markers . Note that the participant shown here as an outlier performed above chance and that there were no outliers in our main pattern similarity analysis ( see Figure 3—figure supplement 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 02110 . 7554/eLife . 17089 . 022Figure 3—figure supplement 9 . No evidence for representations of cardinal directions or 90° modulation of pattern similarity in the entorhinal cortex . ( a ) Structure of pattern similarity values used to test for coding of cardinal directions in the entorhinal cortex . If entorhinal cortex activity would be sensitive to cardinal directions , high pattern similarity would be expected for pairs of trials sampling cardinal directions in comparison with trials sampling other directions . Note that for illustration purposes the tested similarity matrix is shown for comparisons across conditions , not single trials . ( b ) Pattern similarity did not differ between pairs of trials sampling cardinal directions and pairs of trials sampling other directions in pmEC ( differences scores: left: T23 = -0 . 136 , p = 0 . 893; right: T23 = −0 . 449 , p = 0 . 658 ) or alEC ( differences scores: left: T23 = 0 . 266 , p = 0 . 793; right: T23 = 0 . 530 , p = 0 . 601 ) . ( c ) To corroborate the specificity of the 60° modulation of pattern similarity in pmEC , we examined a possible , yet biologically implausible four-fold symmetry in entorhinal pattern similarity values . We tested for increased pattern similarity for pairs of trials sampling directions 90° or multiples thereof apart ( 0° modulo 90° against 30° or 60° modulo 90° ) , using the same analysis logic as for the main analysis ( 0° modulo 60° against 30° modulo 60° ) but now with a 90° periodicity . ( d ) Pattern similarity values did not differ between these conditions in pmEC ( differences scores: left: T23 = −0 . 48 , p = 0 . 637; right: T23 = −1 . 81 , p = 0 . 084 ) or alEC ( differences scores: left: T23 = −0 . 83 , p = 0 . 413; right: T23 = 0 . 50 , p = 0 . 618 ) . Bars in ( b and d ) represent mean and S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 02210 . 7554/eLife . 17089 . 023Figure 3—figure supplement 10 . Searchlight analysis for 60° modulation of pattern similarity during imagination . For each search sphere , the difference in pattern similarity for trial pairs in the 0° modulo 60° condition and the 30° modulo 60° condition was calculated . ( a ) One entorhinal cluster exhibited increased pattern similarity for the 0° modulo 60° condition ( peak MNI coordinates: −18 − 20 −22; T23 = 4 . 04 , p = 0 . 046 , FWE-corrected for multiple comparisons in bilateral pmEC and alEC using small volume correction ) and is shown together with the masks of pmEC and alEC , outlined in dark and light green , respectively . This result confirms our finding from the ROI analysis . Statistical map is thresholded at p < 0 . 005 uncorrected and masked to show entorhinal cortex only . ( b ) Further exploratory whole-brain searchlight analysis showed greater pattern similarity for the 0° modulo 60° condition in the lingual gyrus ( MNI coordinates: 8 −62 6 , T23 = 6 . 11 ) , precuneus ( −20 −60 4 , T23 = 5 . 35 ) , cuneal cortex ( 10 −74 26 , T23 = 5 . 26 ) , lateral occipital cortex ( −42 −66 6 , T23 = 5 . 06 ) , occipital pole ( −32 −90 10 , T23 = 4 . 93 ) , supplementary motor cortex ( 0 −14 56 , T23 = 4 . 54 ) , central opercular cortex ( −50 −16 12 , T23 = 4 . 46 ) , occipital fusiform gyrus ( 38 −74 −18 , T23 = 4 . 36 ) , angular gyrus ( 54 −56 26 , T23 = 4 . 34 ) , superior parietal lobule ( 30 −52 58 , T23 = 4 . 12 ) and supramarginal gyrus ( −64 −50 22 , T23 = 4 . 09 ) . Slices show the statistical image at 1 mm resolution at a significance threshold of p<0 . 001 . Labels were obtained from the Harvard-Oxford Structural Cortical Structural Atlas available in FSL . For each region peak voxel MNI coordinates and statistics are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 17089 . 023 We observed pattern similarity increases with a 60° periodicity in the left pmEC ( T23 = 2 . 37 , p = 0 . 027; one-tailed test , Bonferroni corrected for test in both hemispheres; Cohen’s d = 0 . 48; Figure 3f and Figure 3—figure supplement 2c ) . The effect was further confirmed using permutation-based significance testing ( pseudo T23= 2 . 89 , p = 0 . 008; see Materials and methods ) . A control analysis showed that the effect was not present in the anterior lateral entorhinal cortex ( p>0 . 9; Figure 3—figure supplement 2; see Figure 3—figure supplement 3 for information on signal quality in the entorhinal cortex ) , the human homologue of lateral EC , which does not contain grid cells ( Hafting et al . , 2005 ) . The 60° periodicity in left pmEC was consistent across all angular differences ( Figure 3—figure supplement 4 and Figure 3—figure supplement 5 ) and the effect was not driven by the specifics of our design and the VR town used . Specifically , the effect remained significant after excluding combinations of trial pairs ( Figure 3—figure supplement 6 ) with the same start ( T23 = 2 . 39 , p = 0 . 025 ) or target location ( T23 = 2 . 57 , p = 0 . 017 ) , the same combination of start and target location ( T23 = 2 . 45 , p = 0 . 022 ) and comparisons from the same task block ( T23 = 2 . 08 , p = 0 . 049 ) . Further control analyses demonstrated that the effect was independent of the mean distance between start and target locations in a trial pair ( T23 = 2 . 37 , p = 0 . 027; Figure 3—figure supplement 7 ) , the difference of this distance within a pair ( T23 = 4 . 32 , p<0 . 001 ) and the mean distance between all four buildings in a given trial pair ( T23 = 2 . 37 , p = 0 . 027 ) . Behavioral performance did not differ between the conditions ( T23 = 1 . 24 , p = 0 . 227 , Figure 3—figure supplement 8 ) . Furthermore , the effect was specific to a 60° modulation of pattern similarity values and there was no evidence for coding of cardinal directions in the entorhinal cortex ( Figure 3—figure supplement 9; see also Materials and methods ) . Results of a whole-brain searchlight analysis confirmed the 60° periodicity of pattern similarity increases in pmEC observed in the ROI analysis ( T23 = 4 . 04 , p = 0 . 046 , FWE-corrected for multiple comparisons using small volume correction; Figure 3—figure supplement 10 ) . A similar pattern similarity structure was observed in regions in parietal and visual cortices ( Figure 3—figure supplement 10 ) , which might reflect reactivation of egocentric and visual representations associated with imagined directions , possibly modulated by entorhinal representations in line with a model of mental imagery ( Bird et al . , 2012; Byrne et al . , 2007 ) . Future research will need to investigate these putative interactions in more detail . In sum , we report two important findings: Firstly , pattern similarity values in the parahippocampal gyrus exhibited a one-fold symmetry congruent with fine-grained representations of imagined facing direction , reflecting the role of this brain region - which has been implicated in spatial processing in the absence of visual input ( Wolbers et al . , 2011 ) - in representing the directional aspect of the imagined views . An alternative explanation of this effect through visual similarity of the imagined views appears unlikely due to the complex nature of the task in which each direction was sampled from multiple locations in our large-scale environment ( Figure 1—figure supplement 2 ) and buildings served as cues for a wider range of sampling directions . Therefore , our finding provides the first evidence for fine-grained coding of absolute direction at an unprecedented angular resolution of 30° , consistent with the characteristics of the head direction system in rodents ( Taube et al . , 1990; Taube , 2007 ) , and constitutes a three-fold increase in resolution of the directional representations observed in humans compared to previous studies ( Marchette et al . , 2014; Chadwick et al . , 2015; Baumann and Mattingley , 2010; Vass and Epstein , 2013 , 2016; Shine et al . , 2016 ) . Secondly , the structure of pattern similarity in entorhinal cortex was characterized by a six-fold rotational symmetry akin to the firing properties of grid cells ( Hafting et al . , 2005 ) . Our findings provide evidence for an involvement of grid-like representations in mental simulation in the absence of imagined movement . Crucially , participants imagined directions from stationary viewpoints in a realistic , large-scale virtual city and were not re-exposed to the virtual town during the imagination task . Therefore , our findings provide novel evidence , complementary to a recent report ( Horner et al . , 2016 ) showing evidence for grid-like entorhinal processing during imagined movement through a simple virtual arena . Furthermore , in contrast , we investigated spatial processing in a large-scale , urban environment ( Stokes et al . , 2015 ) and , moreover focused on multi-voxel patterns . In particular , we demonstrate that this novel analysis approach , which does not rely on the estimation of the orientation of the hexadirectional signal in entorhinal cortex in an independent data set ( Doeller et al . , 2010; Kunz et al . , 2015; Horner et al . , 2016; Constantinescu et al . , 2016 ) , is sensitive to grid-like entorhinal signals by capitalizing on the six-fold symmetry of grid cell firing patterns . Contrary to the previously employed approach relying on the estimation of the orientation of the hexadirectional signal for each participant ( Doeller et al . , 2010; Kunz et al . , 2015; Horner et al . , 2016; Constantinescu et al . , 2016 ) , the individual grid orientation is not approximated using the multivariate analysis . Yet , the grid orientation might influence the strength of the grid-like entorhinal signal observed in a given participant because the sampled directions might be more or less aligned with this individual’s grid orientation ( Figure 3a–c and Figure 3—figure supplement 1 for illustration ) . This needs to be taken into consideration when aiming to relate grid-like signals to behavior . However , only the multivariate approach enabled us to investigate the six-fold rotational symmetry in our large-scale environment , in which a continuous sampling of directions as required for the estimation of the orientation of the hexadirectional signal would not have been feasible . This parsimonious approach might prove valuable for future studies investigating the role of grid-like signals in human cognition , in particular in studies with children ( Bullens et al . , 2010 ) or older participants ( Schuck et al . , 2015 ) and in clinical settings ( Hartley et al . , 2007; Maguire et al . , 2001 ) , where time for data acquisition is typically limited and could for instance help to further understand the putative link between the entorhinal grid system and Alzheimer’s disease ( Kunz et al . , 2015 ) . On a theoretical level , our findings are consistent with accounts of imagination positing medial-temporal-lobe involvement in the reactivation and recombination of prior experiences ( Buckner , 2010; Byrne et al . , 2007; Hassabis et al . , 2007; Hassabis and Maguire , 2007; Hasselmo , 2009; Schacter et al . , 2012 ) . The hippocampal formation and grid cells in particular have been implicated in path integration ( Hafting et al . , 2005; Wolbers et al . , 2007 ) , for which computing a homing vector based on translations from a given starting point is central ( Vickerstaff and Cheung , 2010 ) . Notably , the grid system is well-suited to also perform the inverse operation of calculating relative vectors between known positions in the service of navigational planning ( Bush et al . , 2015 ) . Hence , it is plausible that the grid-cell system contributes to the calculation of vectors between start and target location during imagination ( Bird et al . , 2012; Bush et al . , 2015; Hasselmo , 2009; Horner et al . , 2016 ) , while the head direction system ( Taube , 2007; Taube et al . , 1990 ) processes the absolute direction between the two locations ( Bird et al . , 2012; Byrne et al . , 2007; Hasselmo , 2009 ) in our task . Our findings suggest an involvement of the entorhinal grid system in calculating vectors to target locations during navigational planning , in line with a theoretical account of vector navigation ( Bush et al . , 2015 ) . Functional neuroimaging can measure the firing pattern of specific cell types only indirectly ( Logothetis , 2008 ) . However , intracranial recordings in patients exploring virtual-reality environments demonstrated the existence of place ( Ekstrom et al . , 2003; Jacobs et al . , 2010 ) and grid ( Jacobs et al . , 2013 ) cells in the human hippocampus and entorhinal cortex , respectively . Importantly , our results are in line with single-cell recordings in rodents that suggest a possible contribution of spatially tuned cells to future anticipation via place cell preplay of upcoming trajectories ( Dragoi and Tonegawa , 2011 ) and preferential preplay of firing sequences of paths leading to motivationally relevant locations ( Ólafsdóttir et al . , 2015 ) . Prospective coding properties of grid cells ( De Almeida et al . , 2012; Kropff et al . , 2015 ) and recent evidence for grid cell replay ( Ólafsdóttir et al . , 2016 ) further suggest a similar involvement of the entorhinal grid system in future anticipation and prediction . By translating these ideas to human imagination , during which content-specific firing rate increases of neurons in the human medial temporal lobe have been observed ( Kreiman et al . , 2000 ) , it is conceivable that spatially tuned cells provide the machinery for the flexible recombination of spatial and mnemonic details necessary for the construction of mental simulations ( Bird et al . , 2012; Brown et al . , 2016; Buckner , 2010; Byrne et al . , 2007; Eichenbaum and Cohen , 2014; Hassabis et al . , 2007; Hassabis and Maguire , 2007; Hasselmo , 2009; Schacter et al . , 2012 ) and the representation of conceptual knowledge ( Constantinescu et al . , 2016 ) . In concert with the recent report of grid-like processing in the entorhinal cortex during imagined navigation ( Horner et al . , 2016 ) our findings provide a substantial advancement for the field . Importantly , grid-like entorhinal signals during imagined navigation were observed with a similar orientation as during actual navigation through the VR environment ( Horner et al . , 2016 ) . This finding strengthens our interpretation of the six-fold symmetric pattern similarity structure in the entorhinal cortex during imagination of directions from stationary viewpoints observed in this study as reflecting computations of the entorhinal grid system operating similarly in our realistic large-scale VR city as during navigation in smaller and simpler environments typically used in electrophysiological recording studies in rodents ( Hafting et al . , 2005 ) or fMRI experiments in humans ( Doeller et al . , 2010; Horner et al . , 2016; Kunz et al . , 2015 ) . Importantly , the interpretation of our results as a global grid signal coding for space beyond boundaries and obstacles is in line with the report of a global grid pattern emerging with experience in rodents exploring an environment divided into two connected compartments ( Carpenter et al . , 2015 ) . In conclusion , we show involvement of both absolute directional parahippocampal and grid-like entorhinal signals in imagination , which provides important evidence for these representations in the absence of sensory input or imagined movement . This might suggest a more fundamental role of spatial computations in the grid-cell system during mental simulation and possibly other forms of prospective coding and future thinking in the service of goal-directed decision-making ( Bird et al . , 2012; Buckner , 2010; Byrne et al . , 2007 ) . 32 male participants were recruited via the online recruitment system of Radboud University Nijmegen . The study was approved by the local ethics committee ( CMO Arnhem-Nijmegen , the Netherlands ) and participants gave their written informed consent prior to the experiment . All participants had normal or corrected to normal vision and were compensated for their participation . Eight participants were excluded from the analysis because of motion sickness during the VR navigation training ( n = 2 ) , technical problems with the MRI scanner ( n = 1 ) or chance-level performance during the direction imagination task ( n = 5; median absolute angular error not significantly smaller than 90° as determined by Wilcoxon signed-rank test ) . Thus , 24 participants ( age range 18–29 years , mean age 24 . 52 years , standard deviation 2 . 91 years ) entered the analysis . The experiment was conducted on two days and consisted of an extensive behavioral training in our virtual-reality city ‘Donderstown’ ( http://www . doellerlab . com/donderstown/ ) and a direction imagination task in the MRI scanner . Additionally , participants’ ability to locate the task-relevant buildings on a map was assessed . On the first day , participants learned the names and locations of 18 buildings in Donderstown in a two-hour training session in the virtual city . Before the fMRI session on the following day , participants were trained for an additional hour . Functional images were acquired at a 3 T Siemens Trio MRI system ( Siemens , Erlangen , Germany ) using a 3D EPI sequence with an isotropic voxel size of 2 mm and a TR of 1800 ms ( TE = 25 ms , 64 slices , distance factor 50% , flip angle 15° , field of view 224 × 224 × 128 mm ) . Four runs corresponding to the four task blocks were collected . The duration of each run depended on the time required by the participant to complete the task block . On average , a run lasted 11 . 76 min ( ± 1 . 63 SD min ) . T1-weighted structural images were acquired with an MPRAGE sequence ( TR = 2300 ms , TE = 3 . 03 ms , flip angle 8° , in-plane resolution = 256 × 256 mm , voxel size 1 mm isotropic ) . We investigated whether a 60° modulation of absolute angular error values was also present in the behavioral data . We multiplied angular error values to combine performance for all pairs of trials and compared the resulting values as a function of angular difference of the directions sampled in a pair . Analogous to the 60° modulation analysis , we calculated the difference between combined error values in the 0° modulo 60° and the 30° modulo 60° condition for each participant on the first level and tested for a difference on the second level using a one-sample t-test . The combined error values were not different between the conditions ( T23 = 1 . 24 , p = 0 . 227 , Figure 3—figure supplement 8 ) . We further explored the behavioral data obtained during the direction imagination task . To assess whether the distance between the start and target building of a trial was predictive of the angular error of that trial , we calculated Pearson correlations between the absolute angular error and the distance from start to target building for each participant . This relationship did not reach statistical significance in any of the participants ( mean r = −0 . 08 ± 0 . 09 standard deviation , range −0 . 24–0 . 10; all Bonferroni-adjusted p>0 . 454 ) .	Recordings of brain activity in moving rats have found neurons that fire when the rat is at specific locations . These neurons are known as grid cells because their activity produces a grid-like pattern . A separate group of neurons , called head direction cells , represents the rat’s facing direction . Functional magnetic resonance imaging ( fMRI ) studies that have tracked brain activity in humans as they navigate virtual environments have found similar grid-like and direction-related responses . A recent study showed grid-like responses even if the people being studied just imagined moving around an arena while lying still . Theoretical work suggests that spatially tuned cells might generally be important for our ability to imagine and simulate future events . However , it is not clear whether these location- and direction-responsive cells are active when people do not visualize themselves moving . Bellmund et al . used fMRI to track brain activity in volunteers as they imagined different views in a virtual reality city . Before the fMRI experiment , the volunteers completed extensive training where they learned the layout of the city and the names of its buildings . Then , during the fMRI experiment , the volunteers had to imagine themselves standing in front of certain buildings and facing different directions . Crucially , they did not imagine themselves moving between these buildings . By using representational similarity analysis , which compares patterns of brain activity , Bellmund et al . could distinguish between the directions the volunteers were imagining . Activity patterns in the parahippocampal gyrus ( a brain region known to be important for navigation ) were more similar when participants were imagining similar directions . The fMRI results also show grid-like responses in a brain area called entorhinal cortex , which is known to contain grid cells . While participants were imagining , this region exhibited activity patterns with a six-fold symmetry , as Bellmund et al . predicted from the characteristic firing patterns of grid cells . The findings presented by Bellmund et al . provide evidence that suggests that grid cells are involved in planning how to navigate , and so support previous theoretical assumptions . The computations of these cells might contribute to other kinds of thinking too , such as remembering the past or imagining future events .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Grid-cell representations in mental simulation
Rough-skinned newts ( Taricha granulosa ) use tetrodotoxin ( TTX ) to block voltage-gated sodium ( Nav ) channels as a chemical defense against predation . Interestingly , newts exhibit extreme population-level variation in toxicity attributed to a coevolutionary arms race with TTX-resistant predatory snakes , but the source of TTX in newts is unknown . Here , we investigated whether symbiotic bacteria isolated from toxic newts could produce TTX . We characterized the skin-associated microbiota from a toxic and non-toxic population of newts and established pure cultures of isolated bacterial symbionts from toxic newts . We then screened bacterial culture media for TTX using LC-MS/MS and identified TTX-producing bacterial strains from four genera , including Aeromonas , Pseudomonas , Shewanella , and Sphingopyxis . Additionally , we sequenced the Nav channel gene family in toxic newts and found that newts expressed Nav channels with modified TTX binding sites , conferring extreme physiological resistance to TTX . This study highlights the complex interactions among adaptive physiology , animal-bacterial symbiosis , and ecological context . Coevolutionary interactions among species are a central force driving the origin of novel , adaptive phenotypes , yet the traits under selection are often complex and arise from multifaceted interactions among genetic , physiological , and environmental forces that are not well understood ( Ehrlich and Raven , 1964; Futuyma and Agrawal , 2009; Schoener , 2011 ) . Chemical interactions among species in the form of defensive compounds have evolved across all domains of life , and these toxins often target evolutionarily conserved proteins in potential predators ( Brodie and Ridenhour , 2003; Hodgson , 2012; Whittaker and Feeny , 1971 ) . For example , tetrodotoxin ( TTX ) , the primary neurotoxin found in poisonous pufferfishes ( Tsuda and Kawamura , 1952 ) , has been discovered across a broad phylogenetic distribution of animals ( Chau et al . , 2011; Hanifin , 2010 ) . The unusual molecular structure of this toxin serves to selectively target voltage-gated sodium ( Nav ) channels , which are critical for generating action potentials in neurons , muscles , and other excitable cells ( Hille , 2001 ) . Thus , TTX toxicity can have substantial impacts on eco-evolutionary interactions among species , impacting both the toxic species and potential predators . Rough-skinned newts ( Taricha granulosa ) are among the most poisonous TTX-producing animals and serve as an excellent model system for understanding ecological influences on toxin production and predation ( Figure 1A ) . This species is endemic to the Pacific Northwest of North America , where certain populations possess high quantities of TTX relative to other TTX-laden species including pufferfishes and blue-ringed octopuses ( Hanifin , 2010; Williams , 2010 ) . In some populations , individual newts possess enough TTX to kill several adult humans ( Brodie et al . , 2005; Hanifin , 2010; Hanifin et al . , 1999 ) . Variation in newt toxicity is driven in part by the evolution of TTX resistance in predatory garter snakes ( Thamnophis sirtalis ) , as TTX toxicity and resistance in newts and snakes are strongly correlated geographically , suggesting that these two phenotypes are coevolving ( Brodie et al . , 2005; Brodie et al . , 2002; Hanifin et al . , 2008 ) . Furthermore , TTX resistant Nav channels have evolved independently across different garter snake populations , suggesting multiple independent origins of TTX resistance in predatory snakes ( Feldman et al . , 2009; Geffeney , 2002; Geffeney et al . , 2005 ) . Despite the central role of TTX toxicity in coevolutionary interactions between newts and snakes , the origin of TTX in newts and other freshwater animals is unknown ( Daly , 2004; Hanifin , 2010 ) . In TTX-bearing marine species , toxicity is derived either from dietary accumulation from TTX-laden prey , or from symbiotic interactions with TTX-producing bacteria ( Chau et al . , 2011; Miyazawa and Noguchi , 2001 ) . Pufferfishes harbor numerous TTX-producing bacteria symbionts in toxic tissues including skin , liver , intestines , and ovaries , and cultured non-toxic pufferfishes are able to sequester dietary-administered TTX under laboratory conditions ( Jal and Khora , 2015 ) . TTX-producing bacteria have also been isolated from xanthid crabs , horseshoe crabs , starfish , chaetognaths , nemerteans , gastropods , and blue-ringed octopuses ( Jal and Khora , 2015; Magarlamov et al . , 2017 ) . However , the origin of TTX in rough-skinned newts has been more controversial ( Hanifin , 2010 ) . Newts raised in long-term captivity on artificial diets maintain their TTX toxicity ( Hanifin et al . , 2002 ) , and captive newts forced to secrete their TTX by electric shock are able to slowly regenerate their toxicity over time ( Cardall et al . , 2004 ) . Additionally , one group attempted to amplify bacterial DNA from various newt tissues using 16S rRNA gene primers , but failed to recover any PCR products aside from the intestine , which contains low levels of TTX ( Lehman et al . , 2004 ) . These studies suggest that the source of TTX in newts is not dietary , but whether newts have acquired the ability to produce TTX endogenously via convergent evolution or horizontal gene transfer , or from symbiosis with TTX-producing bacteria remains unclear . Furthermore , despite the extreme toxicity of some newt populations , the molecular basis of TTX resistance in this species is not well understood . A previous study identified amino acid replacements in the highly conserved pore-loop ( P-loop ) region of the skeletal muscle isoform Nav1 . 4 and found that skeletal muscle fibers were considerably resistant to TTX ( Hanifin and Gilly , 2015 ) . Amphibians , however , possess six Nav channel isoforms that are differentially expressed across excitable tissues ( Zakon et al . , 2011 ) . Unlike pufferfishes in which TTX is sequestered in certain tissues , newts possess TTX throughout their bodies ( Mebs et al . , 2010; Wakely et al . , 1966; Yotsu et al . , 1990 ) , indicating that the central and peripheral nervous systems are exposed to TTX . Thus , the evolution of whole animal resistance should necessarily involve all Nav channel subtypes , providing an opportunity to examine molecular evolution in response to a specific selective pressure ( i . e . , TTX ) across an entire gene family . In this study , we investigated the source of TTX toxicity and the molecular basis for TTX resistance in rough-skinned newts . We re-examined the hypothesis that newts derive their TTX from symbiosis with TTX-producing bacteria , focusing on the bacterial communities inhabiting the skin of T . granulosa , as this organ possesses specialized granular glands for storing toxins and contains the highest quantities of TTX in the animal ( Daly et al . , 1987; Hamning et al . , 2000; Hanifin et al . , 2004; Santos et al . , 2016; Toledo and Jared , 1995; Tsuruda et al . , 2002 ) . We took advantage of the natural variation in TTX toxicity across newt populations to characterize the skin-associated microbiota in newts from a highly toxic and a non-toxic population and applied an unbiased cultivation-based approach to isolate numerous bacterial symbionts from the skin of toxic newts . Subsequent LC-MS/MS screening of bacterial cultivation media revealed TTX production in eleven bacterial strains from four genera: Aeromonas , Pseudomonas , Shewanella , and Sphingopyxis . Furthermore , to determine the molecular and physiological basis of extreme TTX resistance observed in this species , we cloned and sequenced five uncharacterized Nav channel paralogs ( Nav1 . 1 , Nav1 . 2 , Nav1 . 3 , Nav1 . 5 , Nav1 . 6 ) from a highly toxic population of newts in Oregon . We identified amino acid substitutions in all five genes , many of which have been observed in other TTX-possessing species . To test whether Nav channel mutations impact TTX resistance in newts , we used site-directed mutagenesis to insert three newt-specific replacements identified in Nav1 . 6 into the TTX-sensitive Nav1 . 6 ortholog from Mus musculus . We found that each amino acid replacement reduced TTX sensitivity compared to wild-type M . musculus channels , but these mutations had the greatest effect when combined together . Overall , our results suggest that host-associated bacteria may underlie the production of a critical defensive compound in a vertebrate host , impacting predator-prey coevolution and potentially shaping the evolution of TTX resistance in a well-known ecological model system . To investigate whether bacterial symbionts produce TTX in newts , we leveraged natural variation in toxicity across newt populations to characterize skin-associated microbiota and determine whether highly toxic newts harbored candidate TTX-producing bacteria . We compared two populations previously reported to differ substantially in TTX levels ( Hanifin et al . , 2008; Hanifin et al . , 1999 ) , a toxic population in Lincoln County , OR and a non-toxic population in Latah County , ID ( Figure 1A–B ) . As expected , skin biopsies collected from the dorsal skin of adult newts confirmed that Oregon newts ( n = 5 ) possessed high TTX concentrations ( 126 . 5 ± 42 . 1 ng mL−1 per mg skin ) while Idaho newts ( n = 17 ) lacked detectable levels of TTX ( Figure 1C and Figure 1—source data 1 ) . T . granulosa have particularly enlarged granular glands , which amphibians use to store and secrete noxious or toxic compounds ( Hanifin , 2010; Santos et al . , 2016; Toledo and Jared , 1995 ) . Interestingly , scanning electron micrographs of dorsal granular glands revealed the presence of bacteria inhabiting the surface and outer pore of TTX-sequestering granular glands , and mixed bacterial communities including rod and coccus-shaped bacterial cells were present within the ducts of these glands ( Figure 1D–E ) . We therefore focused our sequencing and cultivation efforts on skin microbiota as a potential source of TTX in this species . Bacterial communities inhabiting the dorsal and ventral skin , cloacal gland , and submandibular gland of T . granulosa were compared by culture-independent 16S rRNA gene sequencing targeting the V4 hypervariable region . Bacterial samples were collected by swabbing wild newts captured in the field at each site ( Oregon n = 12 and Idaho n = 16 ) . Bacterial samples were collected from the Oregon and Idaho populations at separate times , in June 2013 and September 2016 , respectively . However , all bacterial DNA samples were extracted , amplified , and sequenced together on the same run . In total , we identified 4160 operational taxonomic units ( OTUs ) with an average Good’s coverage ( Good , 1953 ) of 0 . 9454 ± 0 . 0067 ( mean ± SEM ) across samples from both populations . 614 OTUs were unique to toxic newts , 1943 were unique to non-toxic newts , and 1603 were shared between the two populations . Among the 20 most abundant OTUs , 8 OTUs were shared between both populations while 12 were present in only one population ( Figure 1—figure supplement 1 and Figure 1—source data 2 ) . These highly abundant and conserved OTUs may represent core skin microbiota of T . granulosa . Idaho newts possessed a greater number of distinct bacterial types with a more even distribution across their microbiota , reflected in a higher number of observed OTUs ( t unequal var . = 7 . 70 , p<0 . 0001 ) and higher OTU richness ( Chao1 index , t unequal var . = 7 . 90 , p<0 . 0001 ) on average than in Oregon newts . Bacterial alpha diversity was also significantly greater in Idaho than Oregon newts ( Simpson 1-D index [t unequal var . = 4 . 11 , p<0 . 0001] ) ( Figure 1F–G and Figure 1—source data 3 ) . Phylum-level divisions show that newt microbiota consists primarily of Proteobacteria , Bacteroidetes , and Firmicutes , which together comprise 76 . 2–83 . 5% of the average bacterial community across all four body sites in both populations ( Figure 1—figure supplement 2 ) . At the genus level , the relative abundance of each bacterial OTU differed markedly between the two populations , as well as from soil samples collected from their respective habitats ( Figure 1H ) . The composition and relative abundances of OTUs ( i . e . beta diversity ) also differed significantly between the two newt populations ( Figure 2 ) . Principal coordinates analysis ( PCoA ) shows a distinct clustering based on geographic location in both composition ( Jaccard index ) and structure ( Bray-Curtis index ) of skin microbiota from each population ( Figure 2A–B ) . Permutational multivariate analysis of variance ( PERMANOVA ) tests revealed that different skin sites across the animal harbored similar communities within a population ( Jaccard , p=0 . 375; Bray-Curtis , p=0 . 065 ) , but that community composition ( Jaccard index , F = 18 . 12 , p<0 . 0001 ) and structure ( Bray-Curtis index , F = 40 . 40 , p<0 . 0001 ) differed significantly between populations . The skin communities of Idaho newts were also more variable than those of Oregon newts ( Permutational test for multivariate dispersion , PERMDISP , p=0 . 0053 ) ( Figure 2—figure supplement 1 ) . Interestingly , toxic Oregon newts maintain a high relative abundance of Pseudomonas OTUs relative to non-toxic Idaho newts ( Figure 2C ) . Three Pseudomonas OTUs ( 00042 , 00224 , and 00485 ) were present in greater relative abundance in toxic newts , and OTU00042 was a significant driver of the beta diversity differences observed between these two populations . Indeed , linear discriminant analysis effect size ( LEfSE ) indicates that Pseudomonas OTU00042 is among the top 10 most differentially abundant OTUs in Oregon newts ( Figure 2D ) . In subsequent non-targeted cultivation of newt skin bacteria , we isolated numerous culturable TTX-producing Pseudomonas spp . strains from toxic newts ( below ) . To determine whether newt skin microbes produce TTX , we employed an unbiased , cultivation-based small molecule screen to examine bacterial culture media for the presence of TTX production in vitro . This approach was necessary because the genetic basis of TTX biosynthesis is unknown , preventing application of metagenomic or other sequencing approaches to determine whether newts or their microbiota possess the genetic toolkit for TTX production . Mixed bacterial communities were collected by swabbing the dorsal skin of toxic newts and cultured on nutrient-limited minimal media ( Reasoner’s 2A agar ) or blood agar supplemented with defibrinated sheep’s blood ( 10% v/v ) . Individual colonies were re-streaked , isolated in pure culture , and taxonomically identified by 16S rRNA gene sequencing . In total , we generated a culture collection of 355 strains representing 65 bacterial genera ( summarized in Figure 3—figure supplement 1 ) . Isolated strains were subsequently grown in 10% Reasoner’s 2 broth ( R2B ) for 2 weeks at 20 °C; 1 mL of each culture was then removed and examined for TTX using mixed cation exchange solid-phase extraction ( SPE ) for sample purification and liquid chromatography tandem mass spectrometry ( LC-MS/MS ) for molecular screening ( Figure 3A ) . Applying this strategy , we detected TTX in cultures from four distinct genera: Aeromonas , Pseudomonas , Shewanella , and Sphingopyxis ( Table 1 and Figure 3B ) . LC-MS/MS screening confirmed the presence of product ions at 162 . 1 and 302 . 1 m/z , corresponding to the primary and secondary product ions formed by TTX fragmentation , respectively ( Jen et al . , 2008 ) . For quantification , we ran standard curves of 0 . 01 , 0 . 05 , 0 . 1 , 0 . 5 , 1 , 2 . 5 , 5 , 10 , and 25 ng mL−1 pure TTX and used linear regression to estimate TTX concentrations in bacterial culture media from the base peak intensity ( BPI ) signal in each LC-MS/MS run . Across all TTX-positive samples , TTX concentrations produced were on average 0 . 236 ± 0 . 087 ng mL−1 ( mean ± SEM , n = 14 ) . TTX was occasionally detected in samples with signals clearly above background noise and our limit of detection ( LOD ) , but below our lower limit of quantification ( LLOQ; 0 . 01 ng mL−1 ) . These samples were not included in our quantitative analyses , but this observation suggests that TTX production could be enhanced with strain-specific optimization of bacterial culture conditions . TTX was detected in seven independent isolates of Pseudomonas spp . cultured from newt skin . Pairwise alignment of 16S rRNA gene sequences suggests that these isolates may represent four bacterial strains ( Figure 3—figure supplement 2 ) . Strains TX111003 , TX174011 , and TX180010 shared >99% nucleotide identity across homologous bases , and TX135003 and TX135004 also shared >99% sequence identity , but these two groups appeared to be distinct from each other ( maximum similarity is 96 . 1% ) ; these two groups were also isolated on different cultivation media , blood agar and R2A agar , respectively . 16S rRNA gene sequences for the remaining two Pseudomonas spp . , strains TX111008 and TX111009 , were unique from each other and the other two groups . Furthermore , we identified two TTX-producing Shewanella spp . strains that were isolated on different media and shared 94 . 2% 16S sequence identity . Thus , it appears several strains of Pseudomonas and Shewanella can produce TTX in lab culture . We also identified one individual strain each of Aeromonas and Sphingopyxis that produced TTX under these culture conditions ( Table 1 ) . Identification of numerous TTX-producing symbionts from distinct genera present on newt skin is consistent with observations in other toxic animal hosts such as the pufferfish and blue-ringed octopus , from which numerous distinct TTX-producing strains have been isolated ( Magarlamov et al . , 2017 ) . However , we note that the vast majority of bacterial isolates screened in this study did not produce TTX under our culture conditions . Rough-skinned newts are the most toxic of TTX-producing animals ( Hanifin , 2010 ) , but the molecular basis of their TTX resistance has not been characterized . To determine the basis of TTX resistance in T . granulosa , we sequenced the Nav channel gene family and investigated the TTX-binding site , the S5-6 pore loop ( P-loop ) , to determine if they possessed adaptive mutations that affect TTX binding and resistance . We generated transcriptomes from two excitable tissues ( brain and nose ) from a toxic newt and obtained partial sequences for five SCN genes that encode Nav1 . 1 , Nav1 . 2 , Nav1 . 3 , Nav1 . 5 , and Nav1 . 6 proteins . We then cloned and sequenced the DI-DIV transmembrane sequences of each gene for verification , including the Nav1 . 6 channel of both toxic and non-toxic newts . SCN4A ( Nav1 . 4 ) was obtained from GenBank for sequence comparison ( accession number KP118969 . 1 ) . Although S5-S6 pore-loop ( P-loop ) sequences are highly conserved across the vertebrate Nav gene family , several amino acid substitutions were present in the P-loops across all six Nav channels in T . granulosa ( Figure 4 ) . In DI , Tyr-371 was replaced independently across four of the six channels; this parallel substitution involves a replacement from an aromatic Tyr or Phe to a non-aromatic amino acid , either Cys , Ser , or Ala . In the mammalian Nav1 . 5 channel , this site is also replaced with a Cys and has been shown to underlie the classic TTX resistance of the cardiac Na+ current ( Satin et al . , 1992 ) . An additional DI difference was found at N374T in Nav1 . 5; this site is also altered in amniotes , but not in Xenopus frogs , indicating that these mutations may be convergent . In DII , only one substitution is present at T938S in Nav1 . 3 , which is adjacent to the electronegative Glu-937 that directly binds the positively charged guanidinium group of TTX ( Shen et al . , 2018 ) . This region is otherwise well-conserved across tetrapods , suggesting that the DII P-loop sequence is under strong purifying selection . Three sites differed in DIII including V1407I , M1414T , and A1419P in Nav1 . 6 , Nav1 . 4 , and Nav1 . 1 , respectively . The DIII M1414T substitution is present in at least five Nav channel paralogs in TTX-laden pufferfishes and increases TTX resistance at least 15-fold ( Jost et al . , 2008 ) ; thus , T . granulosa and pufferfish have converged on the identical molecular solution to reduce TTX sensitivity . The other two differences have not been previously characterized , but we subsequently tested the effects of Nav1 . 6 V1407I on TTX binding ( below ) . Finally , replacements in DIV occur across four sites , including a substitution of A1703G in the selectivity filter DEKA motif in Nav1 . 2 , Ile-1699 in Nav1 . 4 and Nav1 . 6 , D1706S in Nav1 . 4 , and Gly-1707 in Nav1 . 1 , Nav1 . 2 , and Nav1 . 4 . To determine whether TTX resistance was conferred by the P-loop mutations in T . granulosa , we focused on the neural subtype Nav1 . 6 , which is widely expressed in both the central and peripheral nervous system ( Caldwell et al . , 2000; Hu et al . , 2009; Lorincz and Nusser , 2010; Mercer et al . , 2007 ) . We identified three amino acid replacements in the Nav1 . 6 channel of both toxic and non-toxic newts ( Figure 5A ) and used site-directed mutagenesis to insert each mutation ( DI Y371A , DIII V1407I , and DIV I1699V ) , as well as all three mutations , into the TTX-sensitive Nav1 . 6 ortholog from mouse . We found that TTX sensitivity was greatly reduced in triple mutant channels ( Figure 5B–C ) , and that each individual substitution contributed to TTX resistance ( Figure 5—figure supplement 1 ) . Estimated half-maximal inhibitory concentrations ( IC50 ) confirmed that DIII and DIV mutations provided 1 . 5-fold and 3-fold increases in resistance , respectively , while the DI and triple mutant channels were estimated to provide a > 600 fold increase in resistance ( Table 2 and Figure 5D ) . Thus , while all three mutations impact TTX resistance , the DI Y371A replacement provides considerable resistance independently . These results show that the three P-loop modifications in newt Nav1 . 6 provide resistance to even extremely high concentrations of TTX , and comparison of Nav sequences from toxic and non-toxic newts revealed identical substitutions in both populations , suggesting that newts are broadly TTX-resistant regardless of toxicity . In this study , we found that bacterial isolates from four genera , Aeromonas , Pseudomonas , Shewanella , and Sphingopyxis , cultured from the skin of T . granulosa produce TTX under laboratory conditions . Although TTX-producing symbionts have been identified in marine animals ( Chau et al . , 2011 ) , this is the first identification of TTX-producing bacteria associated with a freshwater or terrestrial animal . The origin of TTX in rough-skinned newts and other amphibians has been controversial: wild-caught toxic newts maintain their toxicity in long-term laboratory captivity ( Hanifin et al . , 2002 ) , and newts forced to secrete their TTX by electric shock regenerate their toxicity after nine months , despite laboratory conditions that prevented access to dietary sources of TTX ( Cardall et al . , 2004 ) . Such results demonstrate that newts do not derive TTX from their natural diet , but the results do not explicitly rule out a symbiotic origin for TTX toxicity . A subsequent investigation attempted to amplify 16S rRNA genes from DNA extracted from newt tissues by PCR , but the authors were unable to amplify bacterial DNA from any tissue except the gut ( Lehman et al . , 2004 ) . This result has been widely cited to claim that newts lack symbiotic bacteria altogether , thus supporting an endogenous origin for TTX ( Cardall et al . , 2004; Gall et al . , 2011; Gall et al . , 2014; Hanifin , 2010; Hanifin and Gilly , 2015; Williams , 2010 ) . However , sequencing-based approaches for characterization of microbial communities were limited at that time , and it is increasingly clear that most , if not all , animals possess cutaneous bacterial communities on their external epithelium ( McFall-Ngai et al . , 2013 ) . Thus , our results strongly suggest that symbiotic bacteria are the ultimate source TTX toxicity in rough-skinned newts . Surprisingly , many of the TTX-producing strains isolated from newts are from the same genera as those previously identified in marine animals . TTX-producing Pseudomonas spp . have been isolated from toxic pufferfish , blue-ringed octopus , and sea snails ( Cheng et al . , 1995; Hwang et al . , 1989; Yotsu et al . , 1987 ) , and TTX-producing Aeromonas spp . and Shewanella spp . have both been isolated from pufferfish and sea snails ( Auawithoothij and Noomhorm , 2012; Cheng et al . , 1995; Simidu et al . , 1990; Wang et al . , 2008; Yang et al . , 2010 ) . TTX-producing Sphingopyxis spp . have not been identified in host animals or environmental samples , and this strain may be unique to freshwater or terrestrial environments . Interestingly , several other newt species from diverse genera are known to possess TTX , including Notophthalmus , Triturus , Cynops , Paramesotriton , Pachytriton , and Laotriton ( Brodie et al . , 1974; Yotsu-Yamashita and Mebs , 2001; Yotsu-Yamashita et al . , 2007; Yotsu-Yamashita et al . , 2017 ) . Frogs and toads from the genera Atelopus , Brachycephalus , Colostethus , and Polypedates also possess TTX ( Daly et al . , 1994; Kim et al . , 2003; Mebs et al . , 1995; Tanu et al . , 2001; Yotsu-Yamashita and Tateki , 2010 ) , as well as two species of freshwater flatworms ( Stokes et al . , 2014 ) . Thus , the TTX toxicity observed in other amphibians and freshwater animals could be derived from bacterial sources similar to those identified in this study . One of the most interesting insights to arise from this work is the possibility that the skin microbiome contributes to the predator-prey arms race between toxic newts and TTX-resistant garter snakes . Populations of garter snakes sympatric with TTX-laden newts possess several amino acid replacements in their Nav channels that prevent TTX binding , allowing resistant snakes to prey on highly toxic newts ( Feldman et al . , 2009; Geffeney , 2002; Geffeney et al . , 2005 ) . As snake populations accumulate stepwise adaptive mutations in their Nav channels , selection drives increasing levels of toxicity in newts . Reciprocal selection for elevated toxicity and resistance in newt and snake populations , respectively , leads to an asymmetric escalation of these two traits , or a ‘coevolutionary arms race’ ( Brodie and Brodie , 1999; Brodie et al . , 2005; Dawkins and Krebs , 1979 ) . If selection by predatory garter snakes favors increasing levels of toxicity in newt populations , selection may be acting not only on genetic variation in the host species , but also potentially on variation across the skin microbiome . Selection could also act by increasing the relative abundance of TTX-producing symbionts in the skin ( Bordenstein and Theis , 2015; Theis et al . , 2016 ) . Consistent with this hypothesis , we found that three abundant Pseudomonas OTUs were present in greater relative abundance in the microbiota of toxic newts compared to non-toxic newts ( Figure 2C–D ) . Pseudomonas OTU00042 was particularly abundant in toxic newts and a significant driver of beta diversity between the toxic and non-toxic populations . Numerous TTX-producing Pseudomonas strains were also isolated in our cultivation assay , suggesting that this differential abundance may contribute to observed variation in TTX toxicity across newt populations . However , we did not observe a differential abundance of Aeromonas OTUs , which were abundant in both populations , nor of Shewanella or Sphingopyxis OTUs , which were found only on toxic newts , but were only present in a few samples and in very low abundance ( Figure 2—figure supplement 2 ) . These results may also reflect more favorable culture conditions for TTX-producing Pseudomonas spp . than for the other genera . Thus , further population-level comparisons across toxic and non-toxic newts are needed to determine whether the composition and/or structure of the microbiome directly influences newt toxicity . Additionally , if variation in TTX toxicity is subject to selective forces , TTX-producing symbionts would need to be heritable , directly or indirectly , across generations . The mechanisms underlying microbiome heritability vary from environmental acquisition of microbes across each generation to direct vertical transmission from parent to offspring ( Mandel , 2010 ) . The development of skin-associated microbial communities in newts , and amphibians more broadly , is not clear , as both host species identity and habitat appear to play important roles across different amphibian taxa ( Ellison et al . , 2019; Ross et al . , 2019 ) . In newts , one possibility is that TTX-producing bacteria are vertically transferred from females to their eggs , as newt eggs contain TTX and egg toxicity is correlated with the toxicity of the mother ( Gall et al . , 2012; Hanifin et al . , 2003 ) . Another possibility is that newts possess adaptive traits to facilitate the acquisition and proliferation of TTX-producing bacteria anew from the environment through each generation . Host factors impacting the microbiome may include anti-microbial peptide expression ( SanMiguel and Grice , 2015 ) or the production of metabolites that favor TTX-producing microbes . Other traits may influence interspecific interactions within the microbiome to promote colonization and proliferation of TTX-producing symbionts . These traits may be under selective pressure to ultimately benefit TTX-producing symbionts and increase TTX toxicity across newt populations ( Carroll et al . , 2003; Magarlamov et al . , 2017 ) . Further investigations comparing toxic and non-toxic newts through developmental stages in the wild and in captivity may begin to shed light on this complex process . Furthermore , because of the challenges of in vitro cultivation and characterization of microbial physiology in symbiotic microbes isolated from their hosts , it is difficult to determine how the dynamics of TTX production are regulated within the in vivo host-associated communities ( Magarlamov et al . , 2017 ) . Under our culture conditions in the lab , we observed TTX production that was typically less than 0 . 5 ng mL−1 . However , given that the TTX-producing bacteria identified in this study and in other toxic animals were grown under artificial lab conditions independent of host factors and interactions with other host-associated microbes , estimating the true biosynthetic potential of these TTX-producing bacteria poses a major technical challenge . Identifying the genetic basis of TTX production may help circumvent this problem and allow future researchers to apply sequencing-based metagenomic approaches to determine which organisms are capable of producing TTX ( Chau and Ciufolini , 2011; Chau et al . , 2011 ) . These efforts may also facilitate the development of targeted cultivation strategies to better replicate the host environment and more accurately measure TTX production in vitro . Our results also show that toxic newts possess adaptations in their Nav channels that confer TTX resistance . The presence of parallel mutations across the Nav channel family of newts and other TTX-resistant animals suggests that the evolution of resistance involves a highly constrained walk through a narrow adaptive landscape . For example , studies of the skeletal muscle isoform Nav1 . 4 across a variety of TTX-resistant snake species identify numerous convergent substitutions in the P-loop regions of DIII and DIV , but never in DI or DII ( Feldman et al . , 2012 ) . The Nav1 . 4 subtype of TTX-resistant newts , including T . granulosa , also possess several mutations in DIV and one in DIII , but none in DI or DII . Conversely , mutations in the DI Y/F-371 site are often seen in neural subtypes of TTX-resistant pufferfishes , and we found that this mutation was present in three of the four neural subtypes of newts . Furthermore , when comparing Nav channel sequences in newts and other TTX-resistant animals , we found that Nav1 . 6 sequences in newts and garter snakes share two identical substitutions in the P-loops of DIII V1407I and DIV 1699V ( Figure 4—figure supplement 1 ) . Both newt and snake Nav sequences were derived from individuals caught in Benton Co . , OR , where newts are highly toxic and snakes are highly resistant . These mutations may reflect convergent molecular evolution between predators and prey responding to the same selection pressure . Whether or not these patterns have arisen by chance or through Nav subtype-specific constraints on P-loop evolution would be interesting to explore in future studies . Given the potential strength of selection on interactions between newts and their symbiotic microbiota with regard to TTX toxicity , it may be more appropriate to consider the effects of selection across the hologenome , the collective genetic variation present in both host and symbionts ( Bordenstein and Theis , 2015; Rosenberg and Zilber-Rosenberg , 2013 ) . Many recent studies emphasize the critical importance of host-associated microbes in basic animal physiology , development , nutrition , nervous system function , and even behavior ( Archie and Theis , 2011; Eisthen and Theis , 2016; McFall-Ngai et al . , 2013; Shropshire and Bordenstein , 2016; Theis et al . , 2016; van Opstal and Bordenstein , 2015 ) . In the coevolutionary arms race between toxic newts and resistant snakes , selection may act upon the phenotype that emerges from the collective interactions between the newt host and bacterial symbionts , termed the holobiont . One prediction of the hologenome theory is that adaptive evolution can occur rapidly by increasing the relative abundance of specific symbionts if the metabolites derived from that symbiont are critical for holobiont fitness ( Theis et al . , 2016 ) . This potential evolutionary force would avoid a long and winding road through a complex adaptive landscape for the host , particularly for epistatic traits such as TTX biosynthesis , which is predicted to involve a dozen or more enzymes ( Chau and Ciufolini , 2011; Chau et al . , 2011 ) . Future studies exploring the relationship between newt host toxicity and the composition of newt skin microbiota could provide a mechanistic basis for the observed variation in newt toxicity across different populations , revealing potentially interesting cases of parallel evolution occurring at the hologenomic level . Overall , chemical defenses such as neurotoxins provide excellent models for investigating adaptive evolution , as these toxins often target evolutionarily conserved proteins in animal nervous systems , revealing mechanistic associations among protein sequence , physiology , and evolution . Adult male rough-skinned newts ( Taricha granulosa ) were collected in Oregon , USA ( January Pond; 44°36'13 . 8"N 123°38'12 . 1"W ) under Oregon Department of Fish and Wildlife permit number 104–15 . Animals were housed in glass aquaria containing Holtfreter’s solution ( 60 mM NaCl , 0 . 67 mM KCl , 0 . 81 mM MgSO4 , and 0 . 68 mM CaCl2; pH 7 . 2–7 . 6 ) . Floating platforms in each aquarium provided terrestrial refuges , and newts were maintained at 20°C with a 14:10 light-dark cycle and fed blackworms ( Lumbriculus variegatus ) 2–3 times weekly . To collect bacterial samples , newts were first rinsed in reverse osmosis ( RO ) H2O for 5 s to remove transient bacteria and swabbed 10 times ( down and back ) each on the dorsal and ventral skin surfaces using a sterile cotton swab ( Puritan Medical Products , Guilford , ME ) . The sample swab was then placed in 1 mL Hank’s Buffered Salt Solution ( HBSS; 0 . 137 M sodium chloride , 5 . 4 mM potassium chloride , 0 . 25 mM disodium phosphate , 0 . 56 M glucose , 0 . 44 mM monopotassium phosphate , 1 . 3 mM calcium chloride , 1 . 0 mM magnesium sulfate , 4 . 2 mM sodium bicarbonate ) and diluted ten-fold over four serial dilutions: 10−1 , 10−2 , 10−3 , and 10−4 . 100 µL of each dilution was then plated on either R2A agar ( 0 . 5 g casein hydrolysate , 0 . 5 g dextrose , 0 . 5 g soluble starch , 0 . 5 g yeast extract , 0 . 3 g potassium phosphate , 0 . 3 g sodium pyruvate , 0 . 25 g casein peptone , 0 . 25 g meat peptone , 0 . 024 g magnesium sulfate , 15 g agar , final volume 1 L ) or blood agar ( 10 g peptone , 10 g meat extract , 5 g sodium chloride , 15 g agar , final volume 1 L ) infused with defibrinated sheep’s blood ( 10% v/v ) ( Fisher Scientific , Hampton , NH ) . Petri dishes containing these mixed community cultures were wrapped in Parafilm to prevent desiccation and incubated at room temperature ( 20°C ) for 1–2 weeks . The combination of nutrient-limited media , cool temperatures , and relatively long incubation periods has been shown to promote microbial diversity and the growth of previously uncultivated microbes ( Sommer , 2015; Stevenson et al . , 2004; Stewart , 2012 ) . Following cultivation of mixed communities , individual bacterial colonies were picked and streaked onto new plates to establish pure cultures . Plates were then wrapped in Parafilm and allowed to incubate at 20°C until colonies appeared . Bacterial stocks were generated by collecting bacterial samples from each streaked plate and submerging in 0 . 5 mL HBSS with 10% dimethyl sulfoxide ( DMSO ) for cryoprotection . Samples were then stored at −80°C . To identify bacterial isolates , we performed colony PCR using the 16S rRNA gene universal primers 8F ( 5’—AGAGTTTGATCCTGGCTCAG—3’ ) and 1492R ( 5’—CGGTTACCTTGTTACGACTT—3’ ) . Bacterial colonies were picked with sterile toothpicks and submerged directly into a PCR master mix ( final concentration: 1X PCR buffer , 1 . 5 mM MgCl2 , 0 . 2 mM dNTPs , 0 . 25 µM forward and reverse primer , 0 . 05% NP-40 , 1 . 25U Taq polymerase , and nuclease-free H2O ) . PCR reactions were performed using the following conditions: 3 min at 95°C; 30 s at 95°C , 30 s at 45°C , 1 . 5 min at 72°C repeated 30 times; and a final elongation for 5 min at 72°C . PCR products were analyzed by gel electrophoresis and samples yielding products were cleaned using ExoSAP-IT ( Affymetrix , Santa Clara , CA ) following manufacturer’s instructions . DNA samples were submitted to Michigan State University’s Genomics Core ( East Lansing , MI ) for Sanger sequencing using 16S rRNA 8F universal primer ( 5’—AGAGTTTGATCCTGGCTCAG—3’ ) . Sequences were screened for quality using 4Peaks ( Nucleobytes , Amsterdam , Netherlands ) and sequences with at least 400 bp of unambiguous base calls after quality trimming were assigned genus-level classifications using the Ribosomal Database Project ( RDP ) Classifier tool and an 80% confidence threshold ( Cole et al . , 2014 ) . Evolutionary relationships among cultured bacteria were inferred by constructing maximum-likelihood phylogenetic trees . Multiple sequence alignments were generated by aligning 16S rRNA gene sequences with the SILVA ribosomal RNA reference database ( Quast et al . , 2013 ) . Gaps and non-informative sites were trimmed to generate the final alignment . Trees were constructed using randomized axelerated maximum-likelihood ( RAxML ) with 1000 bootstrap replicates ( Stamatakis , 2014 ) in Geneious v11 . 0 . 5 ( Kearse et al . , 2012 ) and edited in FigTree v1 . 4 . 3 ( https://github . com/rambaut/figtree/ ) . To estimate TTX concentrations in newt skin , we followed the non-lethal sampling technique described by Bucciarelli and coworkers ( Bucciarelli et al . , 2014 ) . Animals were first anesthetized in pH-corrected 0 . 1% tricaine-S ( MS-222 ) dissolved in Holtfreter’s solution . Two skin biopsies were then collected from symmetrical sites on the dorsal skin surface , approximately 1 cm laterally from the vertebrae and 1 cm anterior to the hind limbs , using sterile , disposable 2 mm skin biopsy punches ( Acu-Punch , Acuderm Inc , Fort Lauderdale , FL ) . The two skin biopsies from each individual were weighed and then combined in 300 µL 0 . 1 M acetic acid . Each sample was then placed into a boiling water bath for 5 min followed by an ice bath for an additional 5 min . Subsequent steps were carried out at room temperature . To minimize protein and macromolecular debris , samples were centrifuged at 13 , 000 x g for 20 min and the supernatant transferred to an Amicon Ultra 10 , 000 MWCO centrifugal filter ( Sigma-Aldrich , St . Louis , MO ) followed by a second centrifugation at 13 , 000 x g for 20 min . Finally , 100 µL 0 . 1 M acetic acid was added to the filter and a third centrifugation at 13 , 000 x g for 20 min was performed to wash any remaining TTX . The final sample volume was adjusted to 1 mL before proceeding to solid-phase extraction ( below ) . To identify TTX-producing bacteria , isolated bacterial strains were revived from frozen stocks and inoculated in 5 ml of R2B broth ( 0 . 5 g casein hydrolysate , 0 . 25 g casein peptone , 0 . 25 g meat peptone , 0 . 5 g dextrose , 0 . 5 g soluble starch , 0 . 5 g yeast extract , 0 . 3 g potassium phosphate , 0 . 3 g sodium pyruvate , 0 . 024 g magnesium sulfate , final volume 1 L ) diluted to either 10% or 50% strength in reverse osmosis ( RO ) H2O . The use of dilute broth was intended to encourage the production of secondary metabolites . Cultures were grown at room temperature 20°C on a tissue culture rotator for 1 or 2 weeks . After cultivation , each culture was centrifuged at 13 , 000 x g for 5 min at room temperature , and 1 mL of supernatant was used in solid-phase extraction . TTX extractions were performed using a modified solid-phase extraction ( SPE ) protocol based on that described by Jen et al . ( 2008 ) . Each skin or bacterial sample was loaded onto a mixed cation exchange cartridge ( Oasis MCX cartridges , Waters , MA ) previously regenerated with 1 mL of methanol and equilibrated with 1 mL RO H2O . Samples were drawn through the cartridge over 30 s using a Vac-Man laboratory vacuum manifold ( Promega , Madison , WI ) coupled with VacConnectors ( Qiagen , Germantown , MD ) . Each cartridge was then washed with 1 mL acetonitrile , 1 mL methanol , and 1 mL distilled H2O . TTX was eluted twice from the cartridge with 0 . 125 mL 0 . 2 M HCl in 20% methanol . Both eluates were combined and dried in a SpeedVac vacuum centrifuge ( Savant SpeedVac SC110 , Thermo Fisher Scientific , Waltham , MA ) , then resuspended in 0 . 2 mL 0 . 5% acetic acid in water . 50 µL aliquots of each sample were prepared for LC-MS/MS analysis . TTX analyses were performed using a Waters TQ-D mass spectrometer coupled to a Waters ACQUITY UPLC system with a binary solvent manager . Chromatographic separations were performed on a Waters ACQUITY UPLC BEH amide column ( 2 . 1 × 100 mm; 1 . 7 µm particles; Waters Co . , Milford , MA ) ; column temperature was held at 40°C . For liquid chromatography , we used 0 . 1% formic acid in water ( mobile phase A ) and acetonitrile ( mobile phase B ) with a flow rate of 0 . 4 mL/min . The injection volume was set to 10 µL . The linear gradient elution program was as follows ( A/B ) : 0–1 . 0 min ( 5/95 ) , 1 . 0–1 . 5 min ( 50/50 ) , 1 . 5–2 . 0 min ( 55/45 ) , 2 . 0–3 . 5 min ( 60/40 ) , 3 . 5–4 . 0 min ( 65/35 ) before the gradient returned to the initial condition ( 5/95 ) . TTX was analyzed in positive electrospray ionization mode using multiple reaction monitoring with a transition of 320 . 1 > 162 . 1 ( cone voltage: 50 eV; collision energy: 40 eV ) as the primary channel for quantification and 320 . 1 > 302 . 1 ( cone voltage: 50 eV; collision energy: 40 eV ) as the secondary channel for confirmation . The capillary voltage was 3 . 0 kV . Source and desolvation temperatures were 130°C and 500°C , respectively; cone gas and desolvation gas flows were 40 and 700 L/hr , respectively . Data were acquired using MassLynx 4 . 1 software ( Waters Co . ) . Extracts from bacterial and skin samples were compared with TTX analytical standards acquired from Sigma-Aldrich ( St . Louis , MO ) . A calibration curve was included in each LC-MS/MS run with the following concentrations: 0 . 01 , 0 . 05 , 0 . 1 , 0 . 5 , 1 , 2 . 5 , 5 , 10 , and 25 ng/ml . Concentrations of TTX quantified from skin biopsies were normalized relative to tissue mass . The presence of TTX in skin samples and bacterial cultures was confirmed by a retention time identical to that of authentic TTX as well as the presence of both primary and secondary ion transitions . All chromatograms were plotted in R v3 . 4 . 1 . 3 × 3 mm skin samples were dissected from the dorsal region of a euthanized newt . Each sample was fixed in 4% glutaraldehyde in 0 . 1 M sodium phosphate buffer ( pH 7 . 4 ) overnight at 4°C . Following fixation , samples were briefly rinsed in 0 . 1 M sodium phosphate buffer and dehydrated in an ethanol gradient ( 25 , 50 , 75 , 95 , 100 , 100 , 100% ) for 10 min each . Any remaining liquid in the samples was removed by critical point drying in a Balzers Model 010 critical point dryer ( Balzers Union Ltd . , Balzers , Liechtenstein ) using carbon dioxide as the transitional fluid . Each skin sample was then mounted on an aluminum stub using carbon suspension cement ( SPI Supplies , West Chester , PA ) and coated with platinum ( 8 nm thickness ) using a Q150T turbo pumped sputter coater ( Quorum Technologies , Laughton , East Sussex , England ) purged with argon gas . Samples were examined and images obtained using a JEOL JSM-7500F cold field emission scanning electron microscope ( JEOL Ltd , Tokyo , Japan ) . Skin bacterial samples were collected from two populations of rough-skinned newts , one in Oregon ( January Pond; 44°36'13 . 8"N 123°38'12 . 1"W ) and one in Idaho ( Virgil Phillips Farm Park , Idaho; 46°48'49 . 9"N 117°00'57 . 2"W ) under Oregon Department of Fish and Wildlife permit number 104–15 and Idaho Department of Fish and Game Wildlife Bureau permit number 150521 , respectively . Microbial samples were collected from January Pond , Oregon in Summer 2013 and Virgil Philips Farm Park , Idaho in Fall 2016 . Animals were caught in ponds with dipnets or minnow traps , and each animal was handled with a fresh pair of nitrile gloves . Bacterial samples were collected from two skin sites ( dorsal and ventral ) and from the surfaces of two external glands ( submandibular gland and cloaca ) for a total of four samples per animal . Sterile cotton-tipped swabs were dipped into fresh aliquots of filter-sterilized wetting solution ( 0 . 15M NaCl and 0 . 1% Tween-20 ) and stroked across each body surface 20 times . Each swab was then placed into a sterile 1 . 5 mL conical tube and kept on dry ice until transported to the lab , where they were stored at −80°C . In addition to swabs from newts , we also collected soil samples from pond sediment and pond water samples from each site in sterile 50 mL conical tubes . Total DNA from swab samples was extracted using a QIAamp DNA Mini Kit ( Qiagen ) as follows . First , 500 µl TE buffer ( 10 mM Tris-HCl , 50 mM EDTA , pH 8 , 0 . 2 µm filter-sterilized ) was added to each cotton swab sample and pulse vortexed for 15 s . The buffer was then transferred to a sterile bead-beating tube containing 750 mg zirconia silica beads ( 0 . 1 mm , BioSpec , Bartlesville , OK ) and each sample underwent bead-beating for 60 s on a Thermo Savant FastPrep FP120 ( Thermo Fisher , Waltham , MA ) at setting 5 . Samples were briefly centrifuged and the lysate transferred to a new 2 mL tube . 25 µL proteinase K and 500 µL kit buffer AL were added to each sample , and samples were then pulse vortexed for 15 s and incubated at 56°C for 10 min on a heat block . Each lysate was then acidified by adding 100 µL sodium acetate ( 3M , pH 5 . 5 ) , followed by 500 µL 100% ethanol . Samples were pulse vortexed for 15 s and applied to QIAamp mini spin columns attached to a vacuum manifold via a sterile VacConnector ( Qiagen ) to a Luer valve . The entire lysate was pulled through the column by application of a vacuum and then each column was washed with 750 µL Buffer AW1 and Buffer AW2 , respectively . Next , the spin column was transferred to a clean collection tube and centrifuged at 6000 x g for 1 min in a bench-top microcentrifuge to dry the membrane . After drying , the spin column was placed into a clean 1 . 5 mL microcentrifuge tube , 50 µL nuclease-free H2O was applied to the membrane , and the column was incubated for 5 min at 20°C . Each tube was then centrifuged at 10 , 000 rpm for 1 min to elute the DNA . For soil and water samples , DNA extraction was performed using the MoBio DNeasy PowerSoil Kit ( Qiagen ) per manufacturer’s instruction . For soil samples , 0 . 2 g of pond sediment was directly added to the PowerBead tubes provided by the kit; for pond water , we centrifuged 15 mL pond water at 10 , 000 x g for 10 min at 4°C and resuspended the bacterial cell pellet in 500 µl TE buffer , which was then transferred to a bead beating tube . For negative controls , we performed DNA extractions and PCR reactions on cotton swab samples prepared in the field . We included PCR products from these negative controls with each batch of bacterial samples and included the resulting products in our 16S rRNA gene amplicon library preparation and sequencing . Illumina paired-end reads overlap in the V4 region , allowing for poor quality base calls to be discarded in favor of higher quality sequence on the opposite strand . A dual-barcoded two-step PCR was therefore conducted to amplify the V4 hypervariable regions of the bacterial 16S rRNA gene . V4 primers were designed based on those provided by Kozich et al . ( 2013 ) with the addition of adapter sequences for dual-index barcodes ( 515F–ACACTGACGACATGGTTCTACAGTGCCAGCMGCCGCGGTAA; 806R–TACGGTAGCAGAGACTTGGTCTTGGACTACHVGGGTWTCTAAT ) . In a dedicated PCR hood , 2 µl DNA extract was added to a PCR mixture containing 0 . 05 µM primers ( Integrated DNA Technologies , Coralville , IA ) , and Q5 Hot Start High Fidelity 2X Master Mix ( New England Biolabs , Ipswich , MA ) diluted with nuclease-free water to a 1X final concentration ( 25 µl final volume ) . PCR was conducted using a Veriti thermal cycler ( Applied Biosystems , Foster City , CA ) under the following conditions: 98°C for 30 s; then 98°C for 10 s , 51°C for 20 s , and 72°C for 20 s for 15 cycles . The machine was then paused and 2 µl primers ( 2 µM ) with dual-index barcodes and Illumina sequencing adapters ( University of Idaho IBEST Genomics Resources Core Facility ) were added to each reaction , bringing the final reaction volume to 25 µl . Amplification resumed with 98°C for 30 s; then 98°C for 10 s , 60°C for 20 s , and 72°C for 20 s for 15 cycles; then a final extension step of 72°C for 2 min . Samples were held at 4°C in the thermocycler until being stored at −20°C . Quality of PCR amplicons was evaluated using a QIAxcel DNA screening cartridge ( Qiagen ) and DNA quantified using a Qubit fluorometer ( Invitrogen , Carlsbad , CA ) and the Qubit dsDNA High Sensitivity Assay ( Thermo Fisher Scientific , Waltham , MA ) . Equimolar volumes of each PCR product containing 50 ng DNA were pooled to create a composite sample for high-throughput sequencing and submitted to the University of Idaho IBEST genomics core . Amplicon pools were size-selected using AMPure beads ( Beckman Coulter , Brea , CA ) . The cleaned amplicon pool was quantified using the KAPA Illumina library quantification kit ( KAPA Biosciences , Roche , Basel , Switzerland ) and StepOne Plus real-time PCR system ( Applied Biosystems , Foster City , CA ) . Sequences were obtained using an Illumina MiSeq ( San Diego , CA ) v3 paired-end 300 bp protocol for 600 cycles . Raw DNA sequence reads were processed using the Python application dbcAmplicons ( https://github . com/msettles/dbcAmplicons ) , which was designed to process Illumina double-barcoded amplicons generated in the manner described above . For sequence pre-processing , barcodes were allowed to have at most 1 mismatch ( Hamming distance ) and primers were allowed to have at most 4 mismatches ( Levenshtein distance ) as long as the final 4 bases of the primer perfectly matched the target sequence . Reads lacking a corresponding barcode and primer sequence were discarded . V4 sequences were processed in mothur ( v 1 . 39 . 5 ) following the MiSeq protocol ( Kozich et al . , 2013 ) . Paired sequence reads were joined into contigs , screened for quality and removal of chimeras , then aligned to the SILVA 16S ribosomal RNA database ( Quast et al . , 2013 ) and clustered into operational taxonomic units ( OTUs ) based on 97% nucleotide identity . Taxonomic assignment of OTUs was then performed using the RDP classifier ( Cole et al . , 2014 ) . Prior to analysis , each 16S rRNA gene amplicon profile was subsampled to 5000 sequences . Rarefaction and Good’s coverage analyses were conducted using the rarefaction . single ( ) and summary . single ( ) commands in mothur , respectively . Relative abundances of bacterial OTUs were calculated and visualized using the Phyloseq package in R ( v3 . 4 . 1 ) ( McMurdie and Holmes , 2013 ) . All subsequent microbial ecology analyses were all conducted in R using the vegan package ( v2 . 5–3 ) ( Oksanen et al . , 2014 ) , and plots were produced using the ggplot package ( Wickham and Chang , 2007 ) . Beta diversity matrices were produced using Jaccard and Bray-Curtis dissimilarity indices ( Whittaker , 1972 ) . Principal coordinates analyses ( PCoA ) were conducted on each dissimilarity matrix , and significant differences between groups were determined using a permutational multivariate analysis of variance ( PERMANOVA ) with 9999 permutations and a p<0 . 05 cutoff . A permutation test for multivariate dispersion ( PERMDISP ) was conducted to test for differences in variance among community samples . Linear discriminant analysis effect size ( LEfSe ) was performed on the Galaxy server ( http://huttenhower . sph . harvard . edu/galaxy/ ) . Newts were euthanized by immersion in pH-corrected 0 . 1% MS-222 , and tissue samples including brain , nose , heart , and skeletal muscle were collected and stored in RNAlater ( ThermoFisher Scientific , Waltham , MA ) at −20 °C . Total RNA was extracted from newt tissues using TRIzol ( ThermoFisher Scientific ) . Briefly , each tissue was aseptically dissected and placed into a sterile tube containing 1 mL TRIzol reagent . Tissues were homogenized in TRIzol using a TissueRuptor ( Qiagen , Hilden , Germany ) , and total RNA was extracted following manufacturer’s instructions . The clean RNA pellet was resuspended in 50 µL tris-EDTA buffer ( pH 8 . 0 ) and total RNA yield was quantified by fluorescence using a Qubit fluorometer ( ThermoFisher Scientific ) . We generated reference transcriptomes from the brain and nose of T . granulosa for identification of SCN genes . Non-excitable tissues including liver and skin were included in the sequencing run but were not used for analysis in this study . Poly-adenylated RNA was purified from total RNA samples ( previous section ) using the NEXTflex PolyA Bead kit ( Bioo Scientific , Austin , TX ) according to the manufacturer’s instructions . Lack of contaminating ribosomal RNA was confirmed using the Agilent 2100 Bioanalyzer . Strand-specific libraries for each sample were prepared using the dUTP NEXTflex RNAseq Directional kit ( Bioo Scientific ) , which includes magnetic bead-based size selection , resulting in an average library size of 462 bp . Libraries were pooled in equimolar amounts after quantification using the fluorometric Qubit dsDNA high sensitivity assay kit ( Life Technologies ) according to the manufacturer’s instructions . Libraries were sequenced on an Illumina HiSeq 2000 ( Harvard University , Cambridge , MA ) in one lane to obtain 503 , 241 , 123 paired-end 100 bp reads . We first corrected errors in the Illumina reads using Rcorrector ( Song and Florea , 2015 ) ; parameters: run_rcorrector . pl -k 31 ) and then applied quality and adaptor trimming using Trim Galore ! ( http://www . bioinformatics . babraham . ac . uk/projects/trim_galore/; parameters: trim_galore --paired --phred33 --length 36 -q 5 --stringency 5 --illumina -e 0 . 1 ) . After filtering and trimming , a total of 500 , 631 , 191 reads remained for de novo assembly . We created the newt transcriptome de novo assembly using Trinity , ( parameters: --seqType fq --SS_lib_type RF ) . The raw Trinity assembly produced 2 , 559 , 666 contigs ( N50: 399 bp ) . To reduce redundancy in the assembly , we ran cd-hit-est ( 50 ) ( parameters: -c 0 . 97 ) , resulting in 2 , 208 , 791 contigs ( N50: 390 bp ) . As the sequences were obtained from a wild-caught newt , we next filtered the assembly to remove parasites , microbes , and other contaminants . To accomplish this , we used BLAST to compare each contig with proteins in the Uniprot SwissProt database ( e-value threshold of 1e-5 ) ; specifically , we used non-vertebrate reference genomes , including those of arthropods ( Drosophila ) , microbes ( fungi , Saccharomyces; bacteria , Pseudomonas ) , and parasites ( Caenorhabditis ) to identify potential contaminants , resulting in the removal of 61 , 185 contigs . For the purposes of our study , we only retained contigs with homologs to vertebrate proteins based on this BLAST search of the Swiss Prot database; our final draft assembly of the newt transcriptome contained 77 , 535 contigs with an N50 of 3025 bp . We assessed the completeness of this final assembly by examining vertebrate ortholog representation using BUSCO ( Simão et al . , 2015 ) , which showed 86% of BUSCO groups represented in the assembly . To evaluate SCN gene sequence assembly and confirm the presence of P-loop mutations , we used sequence-specific PCR primers to amplify the DI-DIV transmembrane sequences of each gene from newt cDNA , followed by Sanger sequencing . cDNA templates were synthesized from newt brain and nose RNA using the SuperScript III First-Strand Synthesis kit following manufacturer’s instructions ( Invitrogen , Carlsbad , CA ) . RNA ( 0 . 5–1 µg ) was primed with oligo ( dT ) 20 primers , targeting the mRNA poly ( A ) + tail to enhance synthesis of expressed mRNA transcripts . cDNA samples were stored at −20 °C until use . PCR reactions were performed using Q5 High-Fidelity 2X master mix ( New England Biolabs , Ipswich , MA ) and analyzed by gel electrophoresis on 0 . 8% w/v agarose gel in tris-acetate-EDTA buffer ( pH 8 . 0 ) . Amplified DNA was either sequenced directly from PCR products or cloned into the pGEM-T DNA vector ( Promega , Madison , WI ) . In the latter case , PCR products were first purified by spin column using the DNA Clean and Concentrator kit ( Zymo Research , Irvine , CA ) , then A-tailed using GoTaq ( Promega ) by combining 10 µL purified PCR product , 2 . 5 µL 10X buffer , 5 µL dATP ( 1 mM ) , 0 . 2 µL Taq polymerase , and 7 . 3 µL nuclease-free water to a total volume of 25 µL , and then incubated at 72 °C for 20 min . A-tailed products were used for TA cloning using the pGEM Easy Vector system ( Promega ) . Ligated PCR products were transformed into STBL2 competent E . coli cells by heat shock at 42 °C for 45 s , and 950 µL of SOC media was added to each sample . The following procedures were then adjusted specifically for SCN gene cloning based on published recommendations ( Feldman and Lossin , 2014 ) : samples were incubated at 30 °C for 60 min and plated on ½ strength antibiotic Luria-Bertani ( LB ) agar plates ( 50 µg/mL ampicillin or 7 . 5 µg/mL tetracycline ) . Plates were incubated at 30 °C for two days , and smaller colonies were preferentially selected over large colonies . Plasmid DNA was recovered using the QIAprep Spin Miniprep kit ( Qiagen ) and quantified using a Qubit fluorometer . Aliquots of transformed competent cells in LB were combined with equal volumes of 50% glycerol and stored at −80 °C . PCR products or cloned PCR amplicons were submitted for Sanger Sequencing at the Michigan State University Genomics Core Facility ( East Lansing , MI ) . Sanger sequences from SCN gene PCR products were analyzed in Geneious v11 . 0 . 5 ( Biomatters Inc , Newark , NJ ) . We identified full length coding sequences for all genes except SCN2A ( encodes all four transmembrane domains and P-loop regions ) and SCN1A ( only encodes DIII and DIV of the channel ) . The sequence for T . granulosa Nav1 . 4 was obtained from GenBank ( KP118969 . 1 ) for comparison with other newt channels . All other GenBank accession numbers for vertebrate Nav channel sequences are shown in Figure 4—source data 1 . Consensus sequences for each T . granulosa SCN gene are available on GenBank ( accession numbers MT125668-72 ) . We assessed the quality of sequence base calls by peak shape in the sequence electropherogram files . To identify mutations in newt Nav channels , sequences were translated and aligned with orthologous Nav sequences from representative vertebrate taxa: human ( Homo sapiens ) , mouse ( Mus musculus ) , chicken ( Gallus gallus ) , green anole ( Anolis carolinensis ) , and the two amphibians for which complete SCN gene sequence data were available in GenBank , the Western clawed frog ( Xenopus tropicalis ) and Tibetan plateau frog ( Nanorana parkeri ) . We aligned all Nav protein sequences using MUSCLE ( v3 . 8 . 425 ) and extracted the P-loop regions for analysis of T . granulosa mutations . Mutations were annotated and numbered by reference to the homologous amino acid site in the mouse Nav1 . 6 channel ( GenBank accession: U26707 . 1 ) . Characterization of the effects of three individual mutations in T . granulosa Nav1 . 6 on TTX binding were examined by heterologous expression and electrophysiological recording in Xenopus laevis oocytes . We introduced these mutations into an orthologous M . musculus SCN8A construct ( mSCN8A ) , kindly provided by Dr . Al Goldin ( Smith et al . , 1998 ) , to create a chimeric newt-mouse SCN8A construct . The mSCN8A construct contained an upstream T7 promotor and a downstream NotI restriction enzyme site for plasmid linearization . Site-directed mutagenesis ( SDM ) was performed using the Q5 SDM kit ( New England Biolabs ) . SDM primers containing the target mutation were produced using the NEBase Changer tool ( https://nebasechanger . neb . com ) . After PCR amplification , PCR products were treated with kinase-ligase-DpnI enzyme mix ( New England Biolabs ) , then purified by spin column using the DNA Clean and Concentrator kit ( Zymo Research ) . Mutated plasmid DNA was cloned into STBL2 E . coli competent cells by heat shock at 42 °C for 45 secs . Incubation and colony selection was performed following the protocol of Feldman and Lossin ( 2014 ) ; specifically , all incubations were performed at 30 °C using ½ antibiotic ( ampicillin: 50 mg L−1; tetracycline: 5 mg L−1 ) and two-day incubation periods . Colonies were picked and submerged in LB for overnight incubation , and plasmid DNA was recovered by mini-prep ( Qiagen ) . Samples of each culture were combined with an equal volume of 50% glycerol and stored at −80 °C . Because rearrangements and other replication errors are common with sodium channel sequences ( 52 ) , each plasmid was screened by restriction enzyme ( RE ) digest using BamH1 and IgIII ( New England Biolabs ) and run on a 0 . 8% w/v agarose gel to ensure the correct fragmentation pattern was present . Samples with the correct RE pattern were inoculated in 400 mL of LB and incubated overnight , and plasmid DNA was recovered using the Qiagen plasmid maxi-prep kit ( Qiagen ) . Maxi-prepped DNA was quantified using a Qubit fluorometer and the mSCN8A reading frame was sequenced at the MSU Genomics Core Facility to ensure the correct substitution was made and that no other mutations were introduced into the construct . Capped mRNA ( cRNA ) was synthesized from linearized DNA templates . Plasmid DNA containing the unmutated mSCN8A , individual mutations , or the triple-mutant construct was linearized by overnight digestion using the NotI restriction enzyme ( New England Biolabs ) . 10% SDS and proteinase K were added to each reaction and incubated at 50 °C for 1 hr . Two volumes of phenol were added and mixed into each sample prior to centrifugation at 12 , 000 x g at 4 °C for 10 min , and the upper aqueous layer was transferred to a new tube . Linearized DNA was precipitated by the addition of two volumes of ice cold 100% ethanol , 20 µL of 3M sodium acetate , and 1 µL of glycogen , followed by overnight incubation at −20 °C . Samples were then centrifuged at 12 , 000 x g at 4 °C for 20 min , the supernatant discarded , and the DNA pellet washed with 0 . 5 mL 75% ethanol by brief vortexing and re-centrifugation . The supernatant was discarded , and the DNA pellet air-dried and resuspended in 10 µL nuclease-free water . cRNA was produced using the T7 mMessage mMachine kit following manufacturer’s instructions ( ThermoFisher Scientific ) . Reaction components were combined with 250 ng of linearized template DNA and incubated at 37 °C for 2–8 hr . Synthesized RNA was recovered by lithium chloride precipitation with the addition of 1 µL glycogen . RNA pellets were resuspended in 20 µL nuclease-free water and aliquots at 50 ng/µL concentration were produced for injection into oocytes . Aliquots were stored at −80 °C . Ovaries of adult Xenopus laevis were purchased from Xenopus 1 ( Dexter , MI ) for electrophysiological recordings . Individual oocytes were collected from the ovary by enzymatic digestion using collagenase ( 0 . 4 mg mL−1 , type II activity 255 µ/mg ) in Ca2+-free ND96 solution ( in mM: 96 NaCl , 2 KCl , 1 . 8 CaCl2 , 1 MgCl2 , and 5 HEPES adjusted to pH 7 . 5 and supplemented with 0 . 1 mg/ml gentamycin , 0 . 55 mg/ml pyruvate , and 0 . 5 mM theophylline ) . After 90 min incubation , treated ovaries were washed 5X with normal ND96 , and Stage 5 and 6 oocytes were selected for injection . Oocytes were injected with cRNA samples using a Nanoject III ( Drummond Scientific , Broomall , PA ) . Nanoject glass capillaries ( Drummond , 3-000-203-G/X ) were pulled into pipettes using a Sutter P-97 micropipette puller ( Sutter Instruments Co . , Novato , CA ) using the following conditions: Heat = ramp+5 , Pull = 100 , Velocity = 50 , Delay = 50 , Pressure = 500 . Pipettes were backfilled with mineral oil and placed onto the Nanoject . A 4 µL droplet of each cRNA sample ( 50 ng/µL ) was front loaded into the pipette , and oocytes were injected with either 10 , 25 , or 50 nL of cRNA ( 0 . 5–2 . 5 ng/oocyte ) . Oocytes were incubated at 14 °C in ND96 and used for recording within 2–10 days . Macroscopic sodium currents were measured by two-electrode voltage clamp using a Warner Instruments Oocyte Clamp ( model OC-725C ) . Borosilicate glass pipettes ( 1B120F-4 , World Precision Instruments , Sarasota , FL ) pulled to a 1 or 2 MΩ tip ( Heat = Ramp + 5 , Pull = 100 , Vel = 50 , Time = 50 , on a Sutter Instruments P-97 puller ) served as current and voltage electrodes , respectively . Pipettes were filled with 3M KCl and 0 . 5% agarose . Oocytes were recorded in a RC-26Z diamond bath recording chamber with a chamber volume of 350 µL ( Warner Instruments , Hamden , CT ) in filter-sterilized ND96 recording solution ( 96 mM NaCL , 2 mM KCl , 1 . 8 mM CaCal2 , 1 mM MgCl2 , and 10 mM HEPES; pH 7 . 5 ) at room temperature ( 20°−22 °C ) . Purified TTX ( Sigma Aldrich or Abcam , Cambridge , UK ) was diluted in recording solution and perfused through the chamber for experimental applications . Na+ current traces were digitized at 10 kHz using a Digidata 1550B ( Molecular Devices , San Jose , CA ) and recorded in pCLAMP v10 . 7 ( Molecular Devices ) . Leak currents were subtracted by P/4 correction . The electrophysiological properties measured from each construct ( wildtype , triple mutant , and three individual mutants of mSCN8A ) include peak Na+ current ( Imax ) , conductance , and the voltage-dependence of fast inactivation . Current-voltage ( I/V ) relationships for each mSCN8A construct were determined using an activation protocol in which each oocyte was clamped to a membrane potential of −100 mV and depolarized from −80 to +65 mV in 5 mV steps . The pulse duration was 50 ms with an inter-pulse interval of 5 s . Fast inactivation was measured by clamping the membrane at increasingly depolarized membrane potentials from −100 mV to +10 mV in 5 mV steps for 100 ms followed by a 50 ms test pulse at 0 mV . In such a protocol , the current generated during the test pulse is inversely related to the proportion of channels that are inactivated . To measure the effects of TTX on Imax , 10 chamber volumes ( 3 . 5 mL ) of ND96 recording solution containing experimental concentrations of TTX ( 100 nM , 1 µM , and 10 µM ) were perfused over the oocyte at a flow rate of 5 mL per min . TTX block was monitored by delivering a 50 ms test pulse of 0 mV from a holding potential of −100 mV every 10 secs . Currents were considered to have reached steady-state block when we observed no change in peak current for 10 consecutive pulses , approximately 5 mins after the application of TTX . Step activation and fast inactivation were then measured in the presence of TTX using the protocols described above . To determine whether TTX was responsible for reductions in Imax , all oocytes were then washed for 5–10 mins with normal ND96 and re-recorded by the same protocols . Recordings were discontinued if the membrane leak potential increased more than 0 . 1 mV during the recording . Data were extracted from Na+ current traces in Clampfit v10 . 7 ( Molecular Devices ) and exported for analysis in R Studio v3 . 6 . 0 . Peak Na+ currents elicited by each voltage step in the presence of TTX were normalized relative to the maximum peak current ( Imax ) recorded for each oocyte prior to the addition of TTX . Statistical differences in peak current in the presence and absence of TTX were determined using one-way repeated measures analysis of variance ( ANOVA ) at the −20 mV depolarization step , at which Imax was typically largest , followed by a post-hoc Tukey’s test with Bonferroni correction . Normalized conductance curves for each oocyte were determined by GNa = Imax/ ( V-VNa ) , where Imax is the peak current , V is the voltage step , and VNa is the Na+ reversal potential . VNa was assessed empirically for each oocyte from the corresponding I/V curve . Conductance-voltage plots were fit with a single Boltzman equation , GNa = 1/ ( 1 + exp[- ( V – V1/2 ) /k] ) , where V is the voltage step , V1/2 is the voltage required for half-maximal activation , and k is the slope of the Boltzmann fit . The IC50 for TTX binding was determined from the ratio of peak currents in the presence and absence of TTX by a single-site Langmuir equation , IC50 = [TTX] ( ITTX/Imax ) / ( 1 – ( ITTX/Imax ) ) , where Imax is the peak current recorded under control conditions and ITTX is the current recorded for a given concentration of TTX . The ratio of ITTX/Imax across concentrations of 0 . 1 , 1 , 10 , and 30 µM were fit with a single Hill equation to generate IC50 values using the drc package in R ( Ritz et al . , 2015 ) .	Rough-skinned newts produce tetrodotoxin or TTX , a deadly neurotoxin that is also present in some pufferfish , octopuses , crabs , starfish , flatworms , frogs , and toads . It remains a mystery why so many different creatures produce this toxin . One possibility is that TTX did not evolve in animals at all , but rather it is made by bacteria living on or in these creatures . In fact , scientists have already shown that TTX-producing bacteria supply pufferfish , octopus , and other animals with the toxin . However , it was not known where TTX in newts and other amphibians comes from . TTX kills animals by blocking specialized ion channels and shutting down the signaling between neurons , but rough-skinned newts appear insensitive to this blockage , making it likely that they have evolved defenses against the toxin . Some garter snakes that feed on these newts have also evolved to become immune to the effects of TTX . If bacteria are the source of TTX in the newts , the emergence of newt-eating snakes resistant to TTX must be putting evolutionary pressure on both the newts and the bacteria to boost their anti-snake defenses . Learning more about these complex relationships will help scientists better understand both evolution and the role of beneficial bacteria . Vaelli et al . have now shown that bacteria living on rough-skinned newts produce TTX . In the experiments , bacteria samples were collected from the skin of the newts and grown in the laboratory . Four different types of bacteria from the samples collected produced TTX . Next , Vaelli et al . looked at five genes that encode the channels normally affected by TTX in newts and found that all them have mutations that prevent them from being blocked by this deadly neurotoxin . This suggests that bacteria living on newts shape the evolution of genes critical to the animals’ own survival . Helpful bacteria living on and in animals have important effects on animals’ physiology , health , and disease . But understanding these complex interactions is challenging . Rough-skinned newts provide an excellent model system for studying the effects of helpful bacteria living on animals . Vaelli et al . show that a single chemical produced by bacteria can impact diverse aspects of animal biology including physiology , the evolution of their genes , and their interactions with other creatures in their environment .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2020	The skin microbiome facilitates adaptive tetrodotoxin production in poisonous newts
Membrane attack complex/perforin/cholesterol-dependent cytolysin ( MACPF/CDC ) proteins constitute a major superfamily of pore-forming proteins that act as bacterial virulence factors and effectors in immune defence . Upon binding to the membrane , they convert from the soluble monomeric form to oligomeric , membrane-inserted pores . Using real-time atomic force microscopy ( AFM ) , electron microscopy ( EM ) , and atomic structure fitting , we have mapped the structure and assembly pathways of a bacterial CDC in unprecedented detail and accuracy , focussing on suilysin from Streptococcus suis . We show that suilysin assembly is a noncooperative process that is terminated before the protein inserts into the membrane . The resulting ring-shaped pores and kinetically trapped arc-shaped assemblies are all seen to perforate the membrane , as also visible by the ejection of its lipids . Membrane insertion requires a concerted conformational change of the monomeric subunits , with a marked expansion in pore diameter due to large changes in subunit structure and packing . The bacterial CDCs and ubiquitous MACPF proteins are expressed as soluble monomers but assemble on membranes to form large , oligomeric pores . They form two branches of the largest superfamily of pore-forming proteins . Proteins of this MACPF/CDC superfamily share a common core topology of a highly bent and twisted β-sheet flanked by two α-helical regions , though lacking any detectable sequence homology between the two branches ( Rosado et al . , 2008 ) . Crystal structures of CDCs in their soluble , monomeric form ( perfringolysin , Rossjohn et al . , 1997; anthrolysin , Bourdeau et al . , 2009; suilysin , Xu et al . , 2010; listeriolysin , Köster et al . , 2014 ) revealed extended , key-shaped molecules . Pore-forming domains 1 and 3 ( see also below ) are linked by a long thin β-sheet ( domain 2 ) to an immunoglobulin fold domain ( 4 ) which can bind to the membrane via a tryptophan-rich loop . CDCs form heterogeneous rings and arcs ( Dang et al . , 2005; Tilley et al . , 2005; Sonnen et al . , 2014 ) on cholesterol-rich liposomes and lipid monolayers ( for example , the CDC perfringolysin O hardly binds to membranes with <30% molar concentration of cholesterol , Johnson et al . , 2012 ) . Extensive biophysical and molecular analysis of CDCs established that , on CDC binding to the membrane ( Ramachandran et al . , 2004; Hotze et al . , 2012 ) , α-helical regions in domain 3 unfurl to form transmembrane β-hairpins , denoted as TMH1 and TMH2 ( Shepard et al . , 1998; Shatursky et al . , 1999 ) . If the TMH regions are trapped by introducing a disulphide bond ( Hotze et al . , 2001 ) , prepore oligomers are formed on the membrane surface . Cryo-EM and single particle analysis of liposome-bound CDCs led to low-resolution 3D structures of prepore and pore forms of pneumolysin , a major virulence factor of Streptococcus pneumoniae ( Tilley et al . , 2005 ) . These structures , as well as an AFM study of perfringolysin ( Czajkowsky et al . , 2004 ) , established that the 11 nm high molecule must collapse to a height of 7 nm above the membrane in order to insert the TMH regions . Simple pseudo-atomic models were obtained by fitting domains ( broken at plausible hinge points ) into the EM density maps . It was proposed that the long , thin β-sheet domain 2 collapses after the molecule opens up to release the TMH regions . However , because of the heterogeneity of the oligomeric assemblies and aggregation of the liposomes upon pore formation , resolution has been limited by the difficulty of obtaining sufficiently large data sets . After comparing several CDCs ( pneumolysin , suilysin , anthrolysin , and listeriolysin ) , we found that suilysin was less susceptible to these problems and we chose it to pursue new structural and dynamic studies . A disulphide-locked double cysteine mutant of suilysin , designed to prevent TMH1 insertion , enabled us to trap an active prepore state as well as to visualize the pore formation process by AFM in solution . Cryo-EM reconstruction and fitting revealed new details of the β-sheet unbending and changes in subunit packing upon conversion of prepores to pores . AFM images reveal that the prepore state is highly mobile . Following the addition of DTT to trigger insertion of the disulphide-locked prepore , time-lapse AFM yielded real-time movies of its conversion to ring and crescent-shaped pores . The observed distributions of rings and arcs can be explained by a theoretical model for kinetically trapped , noncooperative assembly , fully determined by the relative kinetics of monomer binding to the membrane and monomer assembly on the membrane surface . Together these studies provide substantial new understanding of the structure and dynamics of CDC pore formation . Negative stain EM and rotational symmetry analysis of complete rings of disulphide locked ( Gly52Cys/Ser187Cys ) suilysin prepores and wild-type pores formed on lipid monolayers revealed that most rings contain 37 subunits ( Figure 1—figure supplement 1 ) . Unexpectedly , the diameter of the 37-fold suilysin prepore was smaller than the diameter of the 37-fold pore ( see below for quantification ) , indicating that conformational changes during pore formation are accompanied by changes in subunit packing . 3D reconstruction of suilysin prepores and pores in liposomes was performed using a pseudo single-particle approach ( Tilley et al . , 2005 ) , yielding a 15 Å cryo-EM map of the prepore using the disulphide-locked construct ( Figure 1A , D ) , and a 15 Å cryo-EM map of a wild-type suilysin pore ( Figure 1B , F , see also Figure 1—figure supplement 2 ) . The 3D maps , both of 37-mers , confirmed a significant expansion in ring diameter upon pore formation . In order to interpret the maps , we performed flexible fitting of suilysin domains from the crystal structure ( Figure 1C; Xu et al . , 2010; PDB:3hvn ) . Domain deformations and hinge movements were identified by normal mode analysis ( Lindahl et al . , 2006 ) . Additional evidence for the correctness of the fits followed from electrostatic potential maps and analysis of interacting residues at the interfaces of domain 1 ( Figure 1—figure supplement 3 ) . The results show that both models have extended regions of complementary charge on the predicted interacting surfaces . Although the extent of complementary charge is less in the pore model , in this case the oligomer is stabilized by the β-barrel of the pore . Measured from the fitted position of the base of domain 4 , the prepore and pore diameters are 296 Å and 319 Å , respectively ( Figure 1E ) , an expansion of 8% . In addition to the 4 nm reduction in height , each pore subunit approximately doubles in width . 10 . 7554/eLife . 04247 . 003Figure 1 . Structural transitions during pore formation . 3D cryo-EM maps of 37-mer prepore and pore forms of suilysin are shown with fitted atomic structures . ( A ) Density map of prepore , surrounded by the extracted disk of membrane , with domains 1 and 4 fitted . ( B ) Density map of pore with all domains fitted , including the β-barrel with strands at 20° tilt . ( C ) Suilysin crystal structure with the domains labelled , showing positions of cysteines introduced in the locked form ( black circles ) and a helical domain adjacent to the bend in the central β-sheet ( dashed oval ) . ( D ) Cross-section through one side of the prepore map with the partial fit of atomic structures . ( E ) Overlay of one side of the prepore ( blue ) and pore maps ( red ) , aligned to the same centre , showing the displacement of domain 4 ( arrow ) . ( F ) Pore section with fit . ( G ) View of 4 subunits from outside the prepore . ( H ) View of 4 subunits from outside the pore . ( I ) Cartoons of domain packing in prepore and pore . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 00310 . 7554/eLife . 04247 . 004Figure 1—figure supplement 1 . Symmetry of suilysin prepores and pores , determined by negative-stain EM . ( A ) Averaged views of 36-fold ( upper row ) and 37-fold ( lower row ) symmetric prepores with their rotational autocorrelations . ( B ) Averaged views of 36-fold ( first row ) , 37-fold ( second row ) , 38-fold ( third row ) , and 39-fold ( fourth row ) symmetric pores with their rotational autocorrelations . ( C ) Plot of pore distribution vs symmetry . Scale bar ( see A , upper row ) : 10 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 00410 . 7554/eLife . 04247 . 005Figure 1—figure supplement 2 . Resolution curves for EM maps . ( A and B ) Fourier-shell correlation curves reporting 15 Å resolution at 0 . 5 correlation for the EM maps of prepore and pore . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 00510 . 7554/eLife . 04247 . 006Figure 1—figure supplement 3 . Electrostatic potential maps and interacting residues . ( A and B ) Electrostatic potential mapped onto the surfaces in crystal structures of domain 1 fitted into the prepore ( A ) and pore ( B ) maps respectively . Red and blue colored regions denote negative and positive , colored by charges , respectively according to the color scale bar . ( C and D ) Interacting residues between a given dimer of domain 1 ( shown in orange ) mapped onto the crystal structure in the prepore ( C ) and pore ( D ) respectively . We consider two residues as interacting ( interface residue ) if their corresponding Cβ atoms were found within the distance of 7 Å ( Malhotra et al . , 2014 ) . 32 common interacting residues ( shown in magenta ) for domain 1 were identified by comparing the interacting residues of the corresponding prepore and pore dimers , suggesting that approximately 50% of the interface is preserved between the prepore and pore fit . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 00610 . 7554/eLife . 04247 . 007Figure 1—figure supplement 4 . Comparison between prepore and crystal structure conformations . Overlays of the monomer structure in the crystal ( yellow ) and in the partial model of the prepore ( blue ) , seen from the oligomer interface ( A ) and from the outside of the ring ( B ) , showing a sideways tilt and inward rotation of domain 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 007 The prepore is distorted from the crystal structure , with some collapse of domain 2 and opening of the β-sheet ( Figure 1—figure supplement 4 ) , despite the presence of the disulphide bridge ( Figure 1D ) . However , the map features are not sufficiently defined to guide fitting , most likely owing to the greater flexibility of the prepore state , as described below . The pore structure is similar to that observed with pneumolysin , but with improved resolution , and showing significant differences in the hinge bending and domain movements . The β-strands are tilted by 20° , in agreement with previous results ( Reboul et al . , 2012; Sato et al . , 2013 ) . The distortion to domain 2 differs from that proposed in the earlier model ( Tilley et al . , 2005 ) . Seen from outside the ring , domain 2 collapses sideways , to the right , such that domain 1 is aligned above the adjacent domain 4 , with a sideways tilt that expands the ring . The expansion is clearly seen in the wider spacing between subunits at domain 4 ( Figure 1G , H , I ) . Domain 2 must bend at a central hinge point to fit into the EM density ( Figure 1H , I; Reboul et al . , 2014 ) . As expected , there is a major opening of the bent β-sheet . In addition , a helical subdomain flanking the bend of the central β-sheet ( residues 335–347 , dashed oval in domain 3 , Figure 1C ) moves as a separate rigid body , as also shown by a spectroscopic study ( Ramachandran et al . , 2004 ) . Notably , the equivalent region has been implicated in the triggering mechanism for unbending in a recent EM study of a remotely related MACPF protein ( Lukoyanova et al . , in press ) . When imaged by negative-stain EM , both the prepore and pore states appeared in heterogeneous ring- and arc-shaped assemblies ( Figure 2A , B; Sonnen et al . , 2014; Köster et al . , 2014 ) . The expansion in ring diameter upon membrane insertion was confirmed by a statistical analysis of the radius of curvature of the arc assemblies ( Figure 2—figure supplement 1 ) , which also revealed significantly larger variations in arc curvature , that is , larger flexibility , for the prepore than for the pore state . 10 . 7554/eLife . 04247 . 008Figure 2 . Negative-stain EM and AFM of disulphide-locked suilysin . ( A ) Negative-stain EM disulphide-locked suilysin ( ds-SLY ) on egg PC:cholesterol monolayers ( 45:55% ) , locked in the prepore state ( −DTT ) . ( B ) as ( A ) , for disulphide-locked suilysin incubated in the presence of 5 mM DTT in solution to reduce the disulphide bridge , so that the suilysin is rapidly converted to the pore conformation . ( C ) AFM of densely packed suilysin prepores , confined to the egg PC-rich domain of a phase-separated egg PC:DDAB:Cholesterol ( 33:33:33% ) supported lipid bilayer , with its corresponding height distribution ( D ) referenced to the membrane surface . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 00810 . 7554/eLife . 04247 . 009Figure 2—figure supplement 1 . Radius of curvature for arc-shaped suilysin assemblies in the prepore and pore states . ( A ) For wild-type suilysin ( WT-SLY ) , the curvature distribution of the arc-shaped assemblies shows a sharp peak close to the radius of the complete ring with 37-fold symmetry . ( B ) For disulphide-locked suilysin in the prepore state ( ds-SLY , −DTT ) , the curvature distribution peaks at slightly lower radius but also shows a larger spread to radii of curvature far exceeding that of the complete 37-mer ring . ( C ) When the disulphide bridge is unlocked by DTT ( ds-SLY , +DTT , cf . Figure 2B ) , the insertion of the transmembrane hairpins in the lipids and formation of the β-barrel leads to a shaper distribution of the arc-shaped oligomers , similar to the wild-type suilysin . Corresponding negative stain EM views are shown under each plot . These observations are further evidence that the prepore intermediate is a structurally flexible state . The arrows refer to the circular fit radius to a 37-mer suilysin ring , which corresponds to 13 . 9 nm for the prepore assembly and 15 . 1 nm for the pore state assembly . Scale bars: 25 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 009 To facilitate AFM imaging of the disulphide-locked suilysin prepores , the protein was confined to well-defined domains on phase-separated lipid membranes ( Connell et al . , 2013 ) , showing densely packed suilysin rings and arcs that extended 10–11 nm above the membrane surface ( Figure 2C , D ) , consistent with the structural data for the prepore state ( Figure 1A , D ) . At lower packing density on the membrane , suilysin prepores were only resolved when the temperature was lowered to 15°C . Cooling appeared to reduce the prepore mobility such that individual prepore assemblies could be observed while diffusing over the membrane ( Video 1 ) . When imaged at room temperature , suilysin prepores appeared as streaks in the AFM images—as can be expected for highly mobile proteins—with slightly improved contrast at the lipid phase boundaries ( Figure 3A ) . As demonstrated by real-time AFM images of the same area on the membrane , the disulphide-locked suilysin reproducibly converted from the prepore to the pore state upon exposure to DTT ( Figure 3A–E , Figure 3—figure supplement 1 ) : in less than a minute , exposure to DTT triggered the appearance of diffuse rings and arcs that became progressively clearer and more prominent ( bottom half of Figure 3A ) , adopting the reduced height typical of the pore state ( Figure 3B–C ) . This process was accompanied by a gradual disappearance of the diffuse streaks ( i . e . , suilysin prepores ) , while additional high ( white ) features appeared on the surface . We identify these features as lipid micelles or fragments being ejected from the membrane . This interpretation is supported by the subsequent appearance of larger plateaus that were consistent in height with the collapse of newly formed lipid layers on top of the membrane ( Figure 3D ) . After about 20 min , the membrane was cleared of these features , leaving a heterogeneous population of suilysin pore assemblies perforating the membrane ( Figure 3E ) . The transition from prepore to pore , as well as the emergence and clearance of lipid aggregates , could also be observed via height profiles taken along various topographic features in these images ( Figure 3F , G ) . 10 . 7554/eLife . 04247 . 010Video 1 . Mobile disulphide-locked suilysin ( prepore ) assemblies diffusing on the membrane . At a temperature of 15°C , the mobility of disulphide-locked suilysin is sufficiently reduced for the assemblies to be resolved by real-time AFM at 15 s/frame . This sequence of images was captured ∼30 min after protein injection and at 384 pixels per line . The timing of the video is accelerated by a factor of ∼100 . Full z-colour scale = 20 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 01010 . 7554/eLife . 04247 . 011Figure 3 . Real-time imaging of the prepore-to-pore transition and membrane perforation by suilysin . Subsequent AFM frames of the same area were alternatively recorded from top to bottom and from bottom to top , as indicated by white arrows . Frame time: 4 min , colour scale: 35 nm . ( A ) Loosely bound to sphingomyelin-rich domains in the phase-separated lipid mixture ( DOPC:sphingomyelin:cholesterol , 33:33:33% ) , the prepore intermediates of disulphide-locked suilysin appear as diffuse streaks . 5 mM of DTT is injected on about 50% completion of the scan . ( B ) On consecutive scanning , the streaks become more clearly defined as arc-shaped oligomers and complete rings . Towards the top end of the scan , clusters of arc-shaped complexes , mostly in the prepore intermediate ( ∼10 . 5 nm high ) , can be distinguished ( Δ ) . ( C ) With the scan direction reversed , and the same area scanned again , the cluster of prepore complexes has converted into the pore state ( ∼7 . 5 nm high ) , within ∼2 min . The prepore to pore transition is followed by the ejection of globular features of varying dimensions exceeding 15 nm above the suilysin in the pore state . We interpret these as ejected lipids . ( D ) These lipids gradually detach from the surface on the pore state suilysin assemblies and can be observed as patches of lipids condensing back onto the membrane . The prepore to pore transition of the suilysin is now complete . ( E ) After ∼20 min , the surface is almost clear of the ejected lipids . ( F ) Cross-sectional line profile extracted as indicated ( Δ in B–C ) , illustrating the prepore to pore transition . ( G ) Cross-sectional line profile extracted as indicated ( ◊ in D–E ) , illustrating the lipid ejection and eventual formation of an aqueous pore in the membrane ( * ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 01110 . 7554/eLife . 04247 . 012Figure 3—figure supplement 1 . Reproducibility of real-time imaging of the prepore-to-pore transition and membrane perforation by suilysin . Recorded in a separate experiment but otherwise similar to Figure 3 , a sequence of AFM images was captured from top to bottom and from bottom to top , as indicated by white arrows . ( A ) Prepore intermediates of ds-SLY appear as diffuse streaks on the sphingomyelin-rich phase of the membrane . ( B ) Upon injection of 5 mM of DTT at about 25% of the way through the scan , the diffuse streaks gradually disappear to be replaced by more clearly resolved arcs and rings of SLY . The brighter assemblies measure ∼10 . 5 nm in height corresponding to the prepore intermediate state , next to the pore state suilysin which is ∼7 . 5 nm in height . ( C ) In the next scan , ∼6 min after DTT injection , large globular features are observed , as in Figure 3C , D , which we interpret as ejected lipids . ( D ) Some lipids detach into the supernatant revealing the pore state suilysin assemblies . Frame time: 5 min , full z-colour scale: 35 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 012 The heterogeneity of ring- and arc-shaped assemblies was confirmed for wild-type suilysin by negative-stain EM on egg PC:cholesterol monolayers and by AFM in solution on supported bilayers ( Figure 4A , B ) . EM and AFM yielded quantitatively similar arc-length distributions for identical lipid composition , protein concentration , and incubation temperature ( Figure 4—figure supplement 1 ) . Qualitatively similar behaviour could be observed by negative-stain EM on lipid vesicles ( Figure 4A , inset ) . Unlike the disulphide locked construct , wild-type suilysin was converted from its soluble , monomeric state ( Figure 4—figure supplement 2 ) to the pore state without the appearance of prepore intermediates , within the time resolution of our AFM experiments . Rings and arcs had a height of 7–8 nm in AFM ( Figure 4B , inset ) , in agreement with the cryo-EM data on the pore state ( Figure 1B , F ) . 10 . 7554/eLife . 04247 . 013Figure 4 . Suilysin assembles into ring- and arc-shaped oligomers that perforate the membrane . ( A ) Negatively stained EM of arc- and ring-shaped assemblies of wild-type suilysin on an egg PC:cholesterol ( 45:55% ) lipid monolayer , and ( inset ) on a liposome of egg PC:cholesterol ( 45:55% ) . ( B ) AFM topography of wild-type suilysin on a supported egg PC:cholesterol ( 67:33% ) lipid bilayer . The wild-type suilysin extends 7–8 nm above the lipid bilayer background , as indicated by the height histogram for 402 individual particles ( inset ) . ( C ) The AFM topography of a complete suilysin ring reveals a circular hole ( dark ) in its lumen , whereas the lipid bilayer surrounding the ring remains intact ( green ) . ( D ) The topography of a suilysin arc shows a hole ( dark ) in the membrane only partially enclosed by the suilysin assembly . Images in C and D are shown in a 15° tilted representation , and height profiles measured across the ring/arc confirm membrane perforation . ( E ) Examples of wild-type suilysin arcs of different lengths . Transmembrane holes are consistently observed . ( F ) Examples of interlocked-arc assemblies . As shown in the right image , the membrane area removed by the two arcs is larger than the hole in the complete ring ( C ) . ( G ) Sequence of AFM images of the same interlocked-arc assembly , stable for at least 50 min . Scale bars A–B: 50 nm , C–G: 15 nm , full z colour scale B–G: 12 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 01310 . 7554/eLife . 04247 . 014Figure 4—figure supplement 1 . Suilysin pore assemblies by EM and AFM . The arc-length distributions for wild-type suilysin as measured by negative-stain EM on monolayers ( A ) and the corresponding AFM data on supported lipid bilayers ( B ) . The grey dashed curves denote the fits of the experimental data with the oligomerization model with ka/kb = 3 . 461 ± 0 . 019 µm2 ( A ) , and 3 . 468 ± 0 . 004 µm2 ( B ) . The numbers in brackets in ( A ) and ( B ) denote the number of monomers per square micron . When using similar lipid compositions ( egg PC:Cholesterol 67:33% ) and incubation conditions ( 27°C ) , both experiments yield very similar distributions . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 01410 . 7554/eLife . 04247 . 015Figure 4—figure supplement 2 . Suilysin is a monomer in solution . Negative-stain EM of a carbon grid after incubation with wild-type suilysin at a concentration of 10 µg/ml . In the absence of lipids , only monomers are observed . Inset: crystal structure of suilysin ( Xu et al . , 2010 ) , for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 01510 . 7554/eLife . 04247 . 016Figure 4—figure supplement 3 . AFM assays of wild-type suilysin ( WT-SLY ) doped with disulphide-locked suilysin ( ds-SLY ) . AFM topographic images of WT-SLY doped with an equimolar amount of ds-SLY ( A–C ) in solution and incubated on egg PC: cholesterol ( 67:33% ) lipid bilayers . ( A ) The presence of ds-SLY largely traps the WT-SLY in the prepore conformation with only few pores observed ( blue and green arrows ) . Addition of ∼5 mM DTT unlocks the disulphide bond and AFM imaging in the same area ( B ) reveals more arcs and rings of suilysin in the pore state . ( C ) Line profile from a suilysin arc after addition of DTT , demonstrating that the membrane is perforated ( * ) . ( D–F ) AFM images of WT-SLY doped with decreasing amounts of ds-SLY in solution and incubated on DOPC:sphingomyelin:cholesterol , 33:33:33% . ( D ) At 1:1 ratios of WT-SLY:ds-SLY , the result is similar to ( A ) with very few suilysin pores observed and prepore locked oligomers prevalent and confined by and at the lipid boundaries . ( E ) At lower amounts of dopant ( WT-SLY:ds-SLY = 4:1 ) , more suilysin pores become visible , with some remaining prepore suilysin oligomers observed as higher arcs and rings and as diffuse streaks . This demonstrates that on reducing the proportion of ds-SLY , the WT-SLY recovers its effectiveness in forming pores in the membrane . ( F ) With the relative ds-SLY proportion reduced even further ( WT-SLY:ds-SLY = 8:1 ) , mostly suilysin pores are prevalent . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 016 To verify if suilysin rings and arcs perforate the membrane , we used high-aspect ratio AFM tips to probe the membrane in the pore lumen ( Figure 3G ) . Both rings ( Figure 4C ) and arcs ( Figure 4D ) enclosed local depressions in the membrane , the depth of which was tip-dependent , but in many cases exceeded the 2 . 2 nm length of an extended lipid molecule , indicating that the lipid bilayer was locally removed . These depressions were reproducible between the trace and retrace versions of the AFM line scans and were observed for various lengths and orientations of the arcs ( Figure 4E ) . We therefore conclude that suilysin can perforate the membrane irrespective of the completion of the ring assembly , locally removing the lipids to create partial β-barrel pores with an unsealed edge of the lipid bilayer . This is in agreement with recent cryo electron tomography data on pneumolysin assemblies ( Sonnen et al . , 2014 ) . Besides isolated rings and arcs , we observed interlocked arcs ( Figure 4A , B , arrows ) , in which one or both ends of the arc contacted another arc or ring . As was the case for the isolated arcs , we found these interlocked arcs capable of perforating the membrane and form membrane lesions that can be smaller , but also larger than those in closed rings ( Figure 4F ) . Once assembled in the pore conformation , the arcs were stable and did not evolve further; even interlocked arcs did not merge into complete rings ( Figure 4G ) . Interestingly , the prepore-to-pore transition and membrane perforation by the wild-type suilysin was largely prevented , in a dose-dependent manner , by adding the disulphide-locked mutant in the incubation process ( Figure 4—figure supplement 3 ) . Subsequent exposure to DTT restored normal pore formation . To analyze the oligomerization process , we measured the arc-length distributions of wild-type suilysin in the pore state and of the disulphide-locked mutant in prepore ( −DTT ) and reduced , pore ( +DTT ) configurations , based on negative stain EM analysis under the same conditions ( Figure 5A–C ) . The distributions show a broad peak centred between lengths of 10–30 monomers and a smaller , narrow peak corresponding to completed rings of about 37 monomers . The arc-length distributions for disulphide-locked prepores and pores after reduction by DTT are practically identical . Combined with the observation that suilysin does not oligomerize before binding to the cholesterol-containing membrane ( Figure 4—figure supplement 2 ) , this demonstrates that assembly is completely determined and terminated in the prepore state , i . e . , is not affected by the prepore-to-pore transition . This conclusion is further confirmed by the lack of growth of individual arcs in the pore state upon subsequent , further addition of wild-type suilysin ( Figure 5—figure supplement 1 ) , and greatly simplifies the interpretation of the arc-length distributions . 10 . 7554/eLife . 04247 . 017Figure 5 . Oligomerization states for arc- and ring-shaped assemblies of suilysin . ( A ) The arc-length distribution of wild-type suilysin displays a broad peak for arcs that contain between 15 and 30 monomers , and a smaller , sharp peak for complete rings ( 37-mers ) . ( B ) Arc-length distribution for the disulphide-locked suilysin prepore intermediate . ( C ) For the disulphide-locked mutant incubated in the presence of DTT ( pore-state ) , the arc-length distribution is practically identical to the distribution for the prepore-locked intermediate . ( D ) Calculated arc-length distributions for a simple model of kinetically trapped oligomerization , with C = 2000 monomers per square micron ( see ‘Materials and methods’ ) . The peak of the arc-length distribution shifts from smaller to larger oligomers on increasing the ratio between the rate constants for monomer association ( ka ) and monomer binding to the membrane ( kb ) . Vertical scale bar: 40 counts . Grey , dashed lines in A–C denote fits of the experimental data with the oligomerization model , yielding ka/kb = 0 . 893 ± 0 . 008 µm2 ( A ) ; 0 . 438 ± 0 . 012 µm2 ( B ) ; 0 . 425 ± 0 . 012 µm2 ( C ) . Numbers in brackets in A–C indicate the estimated total number of monomers per square micron . The experimental data here are based on negative-stain EM images on monolayers of egg PC:cholesterol ( 45:55% ) , incubated at 37°C . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 01710 . 7554/eLife . 04247 . 018Figure 5—figure supplement 1 . Sequential addition of wild-type suilysin in the pore state . ( A–D ) Sequence of AFM images in the same area , showing the effect of sequential injections of wild-type suilysin ( WT-SLY ) in the solution above the supported lipid bilayer ( egg PC:cholesterol , 67:33% ) at 27°C . The overall increase in the number of arcs and rings in the pore state can be readily observed while individual arcs in the pore state can be tracked and characterized following each injection . ( E ) After the third injection , the open-ended arc still persists and the length of the arc does not increase further even as more arcs and rings have assembled in the vicinity of the open-ended arc . ( F ) After the fourth addition of WT-SLY , the open-ended arcs of SLY are still prevalent and the newly formed arc is interlocked with the arc already present from the previous addition of toxin . ( G ) The results show that while the total number of monomers in the pore state increases after each WT-SLY addition , the arc length distribution remains unchanged . This implies that after each monomer addition , new arcs ( and ring complexes ) are formed that do not oligomerize with arcs already in the pore state . Thus suilysin oligomers can only assemble in the prepore state . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 018 To explain these distributions , we calculated the oligomeric populations for a model in which monomers from the solution irreversibly bind to the membrane with a rate constant kb , and in which irreversible oligomerization on the membrane occurs only by monomer addition , with a rate constant ka ( Figure 5D ) . In such a simple model , the oligomerization reaction is arrested by depletion of monomers , yielding kinetically trapped assembly intermediates . Since the resulting populations only depend on the ratio ka/kb and on the ( experimentally known ) total number of monomers per unit area of membrane ( C ) , this model can be used to fit the experimental data with a single free parameter ( ka/kb ) . The broad peak for intermediate arc-lengths ( Figure 5A–C and Figure 4—figure supplement 1 ) can thus be explained by a ratio ka/kb that is sufficiently large to ensure a steady supply of monomers to sustain the oligomerization reaction , but not large enough to yield only completed assemblies ( i . e . , rings ) . CDCs are protein toxins that are potent virulence factors in bacteria . They are part of the major MACPF/CDC superfamily of pore-forming proteins . Our data map the structure ( Figure 1 ) and assembly pathways of membrane pore formation by the CDCs in unprecedented detail and accuracy , as summarized in Figure 6 . Using the disulphide-locked suilysin variant , we have resolved the initial membrane binding of suilysin monomers ( Figure 4—figure supplement 2 ) , oligomerization ( Figure 5 ) , and membrane insertion stages ( Figures 2–4 ) in pore formation . 10 . 7554/eLife . 04247 . 019Figure 6 . Schematic representation of suilysin membrane binding , assembly , and pore formation . From left to right: monomers bind to the membrane and oligomerize . The assembly of monomers proceeds in the prepore intermediate and results in either complete rings or kinetically trapped arc-shaped oligomers . The arc- and ring-shaped assemblies subsequently collapse to the pore configuration with the transmembrane β-hairpins unfurled and inserted into the lipid bilayer in a concerted conformational change . Lipids are subsequently ejected from the membrane ( shown as grey spheres ) and aqueous pores of different sizes are formed in the membrane . The inset shows a possible configuration of lipids at the unsealed edges of the bilayer . DOI: http://dx . doi . org/10 . 7554/eLife . 04247 . 019 Suilysin oligomers exhibit a broad distribution of arc- and ring-shaped assemblies ( Figures 4 and 5 ) . Under given conditions such as incubation temperature , lipid composition , protein sequence and concentration , these distributions are reproducible between EM and AFM experiments ( Figure 4—figure supplement 1 ) . The similarity between prepore and pore distributions ( Figure 5B , C ) implies that oligomerization is arrested before pore insertion . Therefore , our work rigorously establishes that the whole CDC assembly takes place in the prepore state , as was previously suggested for perfringolysin ( Hotze et al . , 2001; Hotze and Tweten , 2012 ) . As can be deduced from the prevalence of larger but incomplete assemblies of the disulphide-locked suilysin , the association of such larger oligomers ( ≳5 subunits ) is not a determining factor in membrane pore formation by CDCs . On the contrary , the oligomerization appears to be dominated by monomer addition ( Figure 5 ) , although the addition of smaller oligomers ( ≲5 subunits ) cannot be fully excluded . As demonstrated by our oligomerization model , the measured arc-length distributions are consistent with the kinetically trapped product of a two-stage irreversible and noncooperative reaction . The outcome is largely determined by the ratio of the corresponding rate constants and the density of monomers per unit area of membrane ( in addition to the effects of steric hindrance at higher surface densities ) . The first stage can be the binding of monomers to the membrane , as assumed here , or a rate-limiting nucleation step that triggers the oligomerization reaction , as assumed elsewhere ( Hotze et al . , 2001 ) . The second stage of this reaction is oligomerization by addition of monomers or very small oligomers . As the rate constants can be expected to vary from one protein to another , this model implies that the various CDCs can yield differing oligomeric populations . Suilysin prepore intermediates appear both more mobile ( Figure 3A–E ) and more flexible than pore assemblies ( Figure 2—figure supplement 1 ) . This is consistent with the observation that the β-sandwich membrane-binding domain 4 does not significantly penetrate the membrane ( Nakamura et al . , 1995; Ramachandran et al . , 2002 ) . In contrast to the earlier observations on pneumolysin ( Tilley et al . , 2005 ) , prepore assembly in suilysin is accompanied by some opening of the central β-sheet . A possible explanation for this difference is that the wild-type pneumolysin preparation in that study formed many stable prepores , apparently in an inactive , dead-end state , whereas wild-type suilysin is extremely active and is not observed in a prepore state . The activity of wild-type suilysin was greatly impaired , however , by incubating it in the presence of an equal concentration of disulphide-locked suilysin ( Figure 4—figure supplement 3 ) and recovered on unlocking the mutant . These results imply that wild-type and mutant co-assemble as expected , and that the prepore-to-pore transition requires a concerted conformational change of all subunits in the suilysin assemblies , suggesting a cooperative insertion of subunits into the membrane . The mobility of prepore assemblies on the membrane surface makes it possible for the subunits to slide apart upon pore formation , as seen in the 8% diameter expansion . This size difference was seen in the earlier work , but the symmetry measurement was less clear , and it was explained by assigning a lower symmetry to the prepore ( 31 vs 38 for the pore ) ( Tilley et al . , 2005 ) . In view of the present observations , it seems likely that pneumolysin rings also expand and that the previous assignment of different symmetries to pneumolysin pores and prepores was most likely incorrect . We observed arc-shaped assemblies as small as 5 subunits , with heights corresponding to the suilysin pore state . This gives an estimate of the minimum oligomer size required for membrane insertion . The size of the AFM tip was too large to probe the membrane perforation in arc-shaped complexes smaller than about 15 monomers ( see e . g . , Figure 4E ) . We have observed membrane perforation for arcs at any size between 15 and 37 monomers , which unambiguously demonstrates that a fully enclosed β-barrel is not essential for the insertion of the transmembrane β-hairpins and membrane perforation , as was previously presumed ( Hotze and Tweten , 2012 ) . Unsealed lipid edges and incomplete β-barrels are a surprising by-product of this membrane perforation by incomplete CDC oligomers ( Figure 4D , E ) , as was recently suggested based on electron tomography of various pneumolysin assemblies ( Sonnen et al . , 2014 ) . The size of the membrane lesions can thus vary between less than half the lumen in a completed SLY pore to the larger pores formed by interlocked arcs ( Figure 4F , G ) , as suggested by conductance measurements on black lipid membranes ( Marchioretto et al . , 2013 ) . Regardless of the extent of oligomerization , both arc and ring prepores can convert directly to the pore configuration as the β-hairpins unfurl and insert into the lipid bilayer . This is followed by the ejection of lipids from the membrane as the pore is formed ( Figure 3C , D and Figure 3—figure supplement 1C , D ) . The results imply that the hydrophilic inner surface of the partially or fully completed β-barrel is sufficient to destabilize the lipid membrane in the pore lumen , leading to ejection of lipid micelles from the pore . The approach of identifying hinge regions for domain fitting to the suilysin pore map has yielded a pseudo-atomic model that goes beyond the earlier rigid body fitting to pneumolysin ( Tilley et al . , 2005 ) . The mechanism of collapse through buckling of domain 2 involves a sideways movement around the ring , and the bending is in the opposite direction to that proposed in the earlier model . The resulting tilt of domain 1 results in the 8% radial expansion of the ring , clearly seen by the displacement of domain 4 ( Figure 1E ) . A helical subdomain thought to be involved in triggering of sheet opening in a MACPF protein ( dashed oval in domain 3 , Figure 1C; Lukoyanova et al . , in press ) is also likely to move in suilysin , strengthening the notion that the mechanism of unfolding and pore formation is conserved between the remotely related MACPF and CDC subfamilies . In summary , we have visualized the various stages of membrane pore formation by a CDC at greatly improved spatial and temporal resolution , to provide new insights in domain movements and pathways of assembly for a major superfamily of pore-forming proteins . All lipid materials ( cholesterol , egg PC , DOPC , sphingomyelin , DDAB ) were purchased from Avanti Polar Lipids ( Alabaster , AL ) . Small unilamellar lipid vesicles were prepared by the extrusion method ( Hope et al . , 1985 ) . Briefly , lipids in powdered form were weighed and dissolved in chloroform to produce a homogeneous mixture with a lipid concentration of ∼1 mg/ml . The solvent was then slowly evaporated for at least 5 hr by passing a steady stream of argon in a fume hood , yielding a dry lipid film . The lipid film was resuspended , by vigorous vortexing for 5 min , in 1 ml of 20 mM Tris , 150 mM NaCl , pH 7 . 8 , to form large , multilamellar vesicles . This solution was transferred to a Fisherbrand FB11201 bath sonicator ( Fisher Scientific , Loughborough , UK ) , maintained above the gel–liquid transition temperature of the constituent lipids . The large multilamellar vesicles were disrupted by 15-min sonication treatments at frequencies between 40 and 80 kHz , interspersed by two freeze/thaw cycles . The solution containing the lipid dispersion was loaded into an Avanti mini-extruder kit ( Avanti Polar Lipids ) and kept above the transition temperature of the lipids . The lipid solution was forced through a Whatman Nucleopore polycarbonate filter ( GE Healthcare Lifesciences , Buckinghamshire , UK ) with an 80 nm nominal pore diameter . The extrusion process was repeated at least 30 times to yield small unilamellar vesicles with a diameter near the pore size of the filter used , as verified by negative stain EM . For EM experiments , liposomes were prepared from 5 mM lipids containing ∼45 mol% of egg PC and ∼55 mol% cholesterol resuspended in 100 mM NaCl , 50 mM HEPES , pH 7 . 5 by extrusion through an 80-nm filter as previously described ( Tilley et al . , 2005 ) . Ring images were centered and analysed by multivariate statistical analysis ( MSA; van Heel , 1984 ) for classification into subsets of homogeneous diameter and subunit number . Suilysin arc length distributions were determined from negative-stain EM images at 1 . 85 Å pixel size and AFM images acquired at 26 . 8 Å pixel size . Using DNA Trace software ( Mikhaylov et al . , 2013 ) , individual suilysin arcs were manually traced with a step size of 25 Å for both EM and AFM images . The number of monomers within each arc was then calculated by dividing the manually traced arc length by the average size of a monomer in the prepore and pore states , 23 . 6 Å and 25 . 7 Å , respectively , as estimated from rotationally averaged negative-stain EM images . This approach yielded an error within ±2 monomers as estimated from averages of rings from the EM monolayer data . The defocus of the cryo-EM images was determined by CTFFIND3 ( Mindell and Grigorieff , 2003 ) and phases were corrected using SPIDER ( Frank et al . , 1996 ) . Side-view images of prepores ( 1374 ) and pores ( 2700 ) were extracted using Boxer ( EMAN 1 . 9; Ludtke et al . , 1999 ) . Images were aligned in SPIDER to reprojections of pneumolysin prepore and pore maps ( Tilley et al . , 2005 ) and the aligned images were sorted by diameter with MSA . Initial reconstructions were calculated by back-projection of either of class sums or aligned raw images , up to 35° from the side view plane , and symmetry estimated by maximising density variance within the maps . These estimates were consistent with the outcomes of statistical analysis for negatively stained prepores and pores formed on lipid monolayers . Most of the pores ( ∼60% ) and prepores exhibited 37-fold symmetry . These 37-fold maps were further refined by projection matching with up to 20° out-of-plane tilt . MSA was used to detect and correct for misalignments . Reconstructions were calculated by back-projection in SPIDER . 450 prepore and 600 pore views were selected for the final reconstructions . The final resolution was estimated by 0 . 5 FSC ( Figure 1—figure supplement 2 ) . Small unilamellar vesicles were injected onto a freshly cleaved mica surface at a concentration between 5 and 25 nM in the presence of 60 µl of 20 mM Tris , 150 mM NaCl , 20 mM MgCl2 , pH 7 . 8 . Incubation of the vesicles on the mica for 30 min at room temperature allowed them to rupture and adsorb onto the surface , yielding an extended lipid bilayer film . Any remaining vesicles were removed by gently rinsing with 80 µl of the adsorption buffer . The rinsing process was repeated 3–7 times to ensure a clean and uniform surface conducive for AFM imaging . Wild-type and disulphide-locked suilysin were injected into a 150 µl fluid cell containing the supported lipid bilayers and allowed to equilibrate for ∼10 min prior to imaging . The concentration of suilysin in the various AFM experiments was 12–180 nM . For the doping assays ( Figure 4—figure supplement 1 ) , wild-type and locked suilysin were mixed in the desired molar ratios and 60 nM of the protein mixture was incubated on the lipid bilayers for 10 min . Real-time topographic images of suilysin on the supported lipid bilayers were collected on a Multimode 8 system ( Bruker , Santa Barbara , CA ) by performing rapid force-distance ( PeakForce Tapping ) curves . The PeakForce method continuously records force-distance curves with a user-defined force set-point ( here about 50 pN ) that is referenced to a continuously adjusted baseline . Typically , these force-distance curves were recorded at a frequency of 2 kHz with a maximum tip-sample separation between 5 and 20 nm . The topographic features were verified for consistency between trace and retrace images , as well as for their reproducibility in subsequent scan frames . For imaging , the vertical scan limit was reduced to ∼1 . 5 μm . Typically , images were recorded at 0 . 2–1 frames/min . Suilysin prepores and pores were also imaged at rates of up to 10 frames/min using a home-built AFM system and miniaturized cantilevers ( Leung et al . , 2012 ) , but this did not yield information additional to the data presented here . Suilysin prepores were only resolved at high concentration on the membrane ( Figure 2C ) , or when the temperature was lowered to 15°C ( Video 1 ) . The real-time , low-temperature measurements were carried out on a Dimension FastScan AFM system in tapping mode with images acquired at 4 frames/min using FastScan Dx probes ( Bruker ) . The AFM probes used for suilysin imaging had nominal spring constants ranging from 0 . 1 to 0 . 7 N/m and resonance frequencies between 10 and 130 kHz in liquid . We used silicon nitride AFM probes with batch-processed silicon tips including MSNL E and F ( Bruker ) , ScanAsyst Fluid+ ( Bruker ) , and cantilevers with individually grown carbon tips , for example , Biotool ( Nanotools , Munich , Germany ) . Batches of AFM probes were screened for tip sharpness and appropriate tilt angles prior to data collection . All AFM imaging was performed in the presence of 20 mM Tris , 150 mM NaCl , 20 mM MgCl2 , pH 7 . 8 with either an E or a J scanner with an integrated temperature control . Images were analysed by either the Nanoscope Analysis software package ( Bruker ) or using the open-source SPM analysis software , Gwyddion ( www . sourceforge . net ) . The raw AFM images were plane-levelled and subsequently line-by-line flattened using the lipid membrane as reference . A Gaussian filter with a full-width-half-maximum of 2-pixels was applied to smooth out high frequency noise where necessary . The assembly of suilysin ( SLY ) in the prepore state was described by the irreversible reactions SLYn−1 ( pre ) +SLY1 ( pre ) →SLYn ( pre ) for oligomerization via monomer-association with a rate constant ka . Here n denotes the number of monomeric subunits in an oligomer , ranging from 1 to the maximum number of monomers in a complete ring , N = 37 . The prepore monomers originated from the binding of soluble suilysin monomers to the membrane , SLY1 ( sol ) →SLY1 ( pre ) , here assumed to occur with a rate constant kb . σn ( t ) was defined as the number of suilysin prepore n-mers per unit area on the membrane , and C as the number of monomers in solution above a unit membrane area , immediately after injection of suilysin at time t = 0 . With these definitions , the oligomerization reactions can be modelled by the rate equationsdσn ( t ) dt=δn , 1kbCe−kb t+∑m=1n−112ka ( δm , 1+δn−m , 1 ) σm ( t ) σn−m ( t ) −∑m=1N−nka ( δm , 1+δn , 1 ) σm ( t ) σn ( t ) . These reactions lead to kinetically trapped prepore assemblies on depletion of free monomers on the membrane , that is , when σ1 ( t ) →0 as time evolves . With the substitutions t = τ/kb and σn ( t ) = sn ( τ ) kb/ka , the rate equations can be rewritten in terms of a dimensionless surface density sn ( τ ) and a dimensionless time τ , to demonstrate that the shape of the solution for sn ( τ→∞ ) versus n , and thus of the resulting arc length distribution ( σn ( t→∞ ) versus n ) , is a function of the parameter Cka/kb only . The coupled and nonlinear differential equations for sn ( τ ) were integrated numerically for different Cka/kb using the Runge-Kutta method , until a stationary solution was reached . For fitting experimental data , C was determined from the accumulated length of all measured oligomers , normalized to the measured membrane area . The best ka/kb then followed from the numerical solution that yielded the lowest sum of squared residues .	Many disease-causing bacteria secrete toxic proteins that drill holes into our cells to kill them . Cholesterol-dependent cytolysins ( CDCs ) are a family of such toxins , and are produced by bacteria that cause pneumonia , meningitis , and septicaemia . The bacteria release CDC toxins as single protein molecules , which can bind to the membrane that surrounds the host cell . After binding to the membrane , the toxin molecules assemble in rings to form large pores in the host membrane . There are several stages to this process , but our understanding of what happens at the molecular level is incomplete . Leung et al . studied suilysin , a CDC toxin produced by a bacterium that has a big impact on the pig farming industry because it causes meningitis in piglets . The bacterium can also cause serious diseases in humans through exposure to contaminated pigs or pig meat . Leung et al . used a technique called electron microscopy to obtain atomic-scale snapshots of the toxin structures before and after the toxins were inserted into the membrane . In addition , real-time movies of the process were gathered using another technique called atomic force microscopy . The experiments show that suilysin forms assemblies on the membrane that grow by one molecule at a time , rather than by the merging of larger assemblies of molecules . This results in a mixture of ring-shaped and arc-shaped toxin assemblies on the membrane . The arcs of suilysin are incomplete ring assemblies , but they are still able to make holes in the cell membrane . In order to insert into the membrane , the toxin molecules in the arcs and rings undergo a dramatic change in shape . Understanding how CDCs assemble in membranes will guide further work into the development of new vaccines that can target these proteins to reduce the damage caused by bacterial infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2014	Stepwise visualization of membrane pore formation by suilysin, a bacterial cholesterol-dependent cytolysin
Tapeworms grow at rates rivaling the fastest-growing metazoan tissues . To propagate they shed large parts of their body; to replace these lost tissues they regenerate proglottids ( segments ) as part of normal homeostasis . Their remarkable growth and regeneration are fueled by adult somatic stem cells that have yet to be characterized molecularly . Using the rat intestinal tapeworm , Hymenolepis diminuta , we find that regenerative potential is regionally limited to the neck , where head-dependent extrinsic signals create a permissive microenvironment for stem cell-driven regeneration . Using transcriptomic analyses and RNA interference , we characterize and functionally validate regulators of tapeworm growth and regeneration . We find no evidence that stem cells are restricted to the regeneration-competent neck . Instead , lethally irradiated tapeworms can be rescued when cells from either regeneration-competent or regeneration-incompetent regions are transplanted into the neck . Together , the head and neck tissues provide extrinsic cues that regulate stem cells , enabling region-specific regeneration in this parasite . Tapeworms are parasitic flatworms that infect humans and livestock , causing lost economic output , disease , and in rare cases , death ( Del Brutto , 2013 ) . These parasites are well known for their ability to reach enormous lengths . For example , humans infected with the broad or fish tapeworm , Diphyllobothrium latum , harbor parasites that average 6 m in length ( Craig and Ito , 2007 ) . It is less commonly appreciated that tapeworms can regenerate to accommodate their multi-host life cycle . Adult tapeworms in their host intestines develop proglottids ( segments ) that are gravid with embryos . Tapeworms pinch off the posterior and gravid sections of their body , which exit with the host excrement , to be eaten by a suitable intermediate host that supports larval tapeworm development . Despite losing large body sections , the tapeworm does not progressively shorten; instead , it regenerates proglottids , allowing the worms to maintain an equilibrium length . Despite this remarkable biology , tapeworms are an unexplored animal model in the study of regenerative behaviors . Up to the 1980s the rat intestinal tapeworm , Hymenolepis diminuta , had been a favorite model organism among parasitologists . H . diminuta grows rapidly–within the first 15 days of infection , it produces up to 2200 proglottids , increases in length by up to 3400 times , and weight by up to 1 . 8 million times ( Roberts , 1980 ) –and is easily propagated in the laboratory . Foundational work on their biochemistry , ultrastructure , and developmental biology enriched our understanding of these tapeworms ( Arai , 1980 ) . However , with the dawn of the molecular age and the rise of genetic model organisms , H . diminuta was essentially left behind . Here , we show that H . diminuta is an excellent , tractable model for the study of stem cells and regeneration , with the power to inform us about parasite physiology . As an obligate endoparasite , adult H . diminuta will expire once its host rat dies . However , the lifespan of H . diminuta can be greatly increased via regeneration . A single adult tapeworm can be serially amputated and transplanted into a new host intestine , where the fragment can regenerate into a mature tapeworm even after 13 rounds of amputation over 14 years ( Read , 1967 ) . These observations have led to speculation that H . diminuta may be inherently immortal . This situation is reminiscent of the free-living cousins of tapeworms: freshwater planarians like Schmidtea mediterranea , which reproduce indefinitely by fission , and can regenerate their whole body from tiny fragments ( Newmark and Sánchez Alvarado , 2002 ) . Planarian immortality and regeneration are enabled by adult somatic stem cells called neoblasts ( Newmark and Sánchez Alvarado , 2002; Reddien , 2018; Baguñà , 2012 ) . These stem cells are the only dividing undifferentiated cells within the soma . Like planarians , H . diminuta maintains a population of neoblast-like adult somatic stem cells ( Roberts , 1980 ) that are likely responsible for their growth and regenerative ability . Recently , stem cells of multiple species of parasitic flatworms have been described ( Collins et al . , 2013; Koziol et al . , 2014; Koziol et al . , 2015; Wang et al . , 2013; Koziol et al . , 2010 ) . Stem cells play crucial roles in parasite development , transmission , homeostasis , and even disease . For example , stem cells enable prolific reproduction and longevity ( Collins , 2017 ) , mediate host-parasite interactions ( Collins et al . , 2016 ) , and allow metastatic parasite transmission in host tissues ( Brehm and Koziol , 2014 ) . How stem cells may regulate regeneration in parasites such as tapeworms is largely unexplored and the subject of this study . We use H . diminuta , to investigate the molecular basis of tapeworm regeneration . We have established and refined experimental tools such as transcriptomics , in vitro parasite culture , whole-mount and fluorescent RNA in situ hybridization ( WISH and FISH ) , cycling-cell tracing with thymidine analogs , RNA interference ( RNAi ) , and cell transplantation , all described in this work . We determine that the ability to regenerate is regionally limited to the neck of adult H . diminuta . However , regeneration from the neck is finite without signals from the tapeworm head . Using RNA sequencing ( RNA-seq ) , we identify and characterize various markers of the somatic cycling-cell population , which includes tapeworm stem cells . Using RNAi , we functionally validate molecular regulators of growth and regeneration . However , our analyses failed to uncover a neck-specific stem cell population that explains the regional regenerative ability displayed by H . diminuta . Instead , we show that cells from both regeneration-competent and regeneration-incompetent regions of H . diminuta have stem cell ability and can restore viability to lethally irradiated tapeworms . Our results show that extrinsic signals present in the tapeworm neck , rather than specialized stem cells , confer region-specific regenerative ability in this tapeworm . The anatomy of adult H . diminuta consists of a head with four suckers , an unsegmented neck , and a body with thousands of proglottids/segments that grow and mature in an anterior-to-posterior direction ( Roberts , 1980; Rozario and Newmark , 2015 ) ( Figure 1a ) . What regions of the tapeworm body are competent to regenerate ? In order to test regeneration competency , it is necessary to grow tapeworms in vitro instead of in the intestine , where the suckers are required to maintain parasites in vivo . We established H . diminuta in vitro culture conditions modified from Schiller's method ( Schiller , 1965 ) and tested the regeneration competence of 1 cm amputated fragments ( Figure 1b–c ) . The anterior-most fragments ( head+neck+body ) were competent to regenerate , confirming in vivo observations using amputation and transplantation ( Read , 1967; Goodchild , 1958 ) . Anterior fragments that were first decapitated ( neck+body ) were also competent to regenerate . In contrast , ‘body only’ fragments failed to regenerate proglottids . All amputated fragments could grow in length ( Figure 1d ) , differentiate mature reproductive structures , and mate . Despite the failure to regenerate , ‘body only’ fragments could grow because each existing proglottid increased in length as it progressively matured ( Figure 1—figure supplement 1a–b ) . However , only fragments that retained the neck were able to regenerate new proglottids over time . The neck of 6-day-old tapeworms used in this study is typically 2–3 mm long when observed after DAPI staining and widefield fluorescent microscopy . By amputating 2 mm ‘neck only’ fragments , we find that the neck is sufficient to regenerate an average of 383 proglottids ( SD = 138 , N = 4 , n = 20 ) after 12 days in vitro ( Figure 1e ) . In no case did we observe head regeneration . Furthermore , amputated heads alone could not regenerate in vitro ( Figure 1—figure supplement 1c ) nor in vivo ( Read , 1967 ) . Thus , neither the head nor body can regenerate proglottids , but the neck is both necessary and sufficient for proglottid-specific regeneration in H . diminuta . Previous in vivo studies have shown that H . diminuta can regenerate after serial rounds of amputation and transplantation for over a decade ( Read , 1967 ) and perhaps indefinitely . Using in vitro culture , we confirmed that anterior fragments of H . diminuta can regenerate after at least four rounds of serial amputation ( Figure 1f–g ) . Decapitated ( -head ) fragments regenerated proglottids after the first amputation; however , re-amputation abrogated regeneration ( Figure 1f–g ) . After decapitation , a definitive neck could not be maintained and eventually , the whole tissue was comprised of proglottids ( Figure 1—figure supplement 2 ) . Without the head , proglottid regeneration from the neck is finite . Thus , while the neck is necessary and sufficient for proglottid regeneration , the head is required to maintain an unsegmented neck and for persistent regeneration . If signals from the head regulate regeneration , is regenerative potential asymmetric across the anterior-posterior ( A-P ) axis of the neck ? We subdivided the neck into three 1 mm fragments and found that the most-anterior neck fragments regenerated more proglottids than the middle or posterior neck fragments ( Figure 1h–i ) . Thus , regeneration potential is asymmetric across the neck A-P axis with a strong anterior bias . Since the neck is the only region competent to regenerate , are stem cells preferentially confined to the neck ? In lieu of specific molecular markers for stem cells , we examined the distribution of all cycling cells in adult tapeworms . In flatworms , it has been repeatedly shown that the only proliferative somatic cells are undifferentiated cells with stem cell morphology and/or function; these cells have been termed neoblasts , adult somatic stem cells , or germinative cells , depending on the organism ( Collins et al . , 2013; Koziol et al . , 2014; Baguñà et al . , 1989; Newmark and Sánchez Alvarado , 2000; Ladurner et al . , 2000 ) . In H . diminuta , proliferation does not occur in regions comprised solely of differentiated cells ( muscle and tegument/parasite skin at the animal edge ) ( Bolla and Roberts , 1971 ) . Instead , proliferation is only detected in regions where undifferentiated cells with the typical morphology of stem cells can be distinguished ( Bolla and Roberts , 1971; Sulgostowska , 1972 ) . Thus , cycling somatic cells in H . diminuta would not include differentiated cells , but would include stem cells and any dividing progeny . To label cycling cells , we used two methods: ( i ) uptake of the thymidine analog F-ara-EdU ( Neef and Luedtke , 2011 ) to mark cells in S-phase and ( ii ) FISH to detect cell cycle-regulated transcripts , such as the replication licensing factor minichromosome maintenance complex component 2 ( mcm2 ) and histone h2b ( h2b ) , which are conserved cycling-cell markers in free-living and parasitic flatworms ( Collins et al . , 2013; Solana et al . , 2012 ) . We detected cycling somatic cells throughout the tapeworm body ( Figure 2a–b ) . Contrary to previous results ( Bolla and Roberts , 1971 ) , we also detected cycling cells in the head , though in small numbers ( Figure 2a ) . The scarcity of these cells may be the reason they were originally missed . Taken together , cycling cells are present in all regions , regardless of regeneration competence . To further our understanding of how tapeworm stem cells are distributed and regulated , we sought to identify stem cell markers . Stem cell genes have been discovered previously in flatworms by identifying transcripts downregulated after exposure to irradiation , which depletes cycling cells ( Collins et al . , 2013; Solana et al . , 2012; Eisenhoffer et al . , 2008 ) . Exposing H . diminuta to 200 Gy x-irradiation reduced the number of cycling cells by 91 ± 6% after 3 days ( Figure 2c–d ) and abrogated both growth and regeneration ( Figure 2—figure supplement 1a–b ) . This dosage is lethal; all fragments from worms exposed to 200 Gy x-irradiation degenerated after 1 month ( Figure 2—figure supplement 1c–d ) . We leveraged the sensitivity of H . diminuta to lethal irradiation in order to identify new molecular markers of cycling somatic cells by RNA-seq ( Figure 2e ) . A de novo transcriptome of 14 , 346 transcripts was assembled ( see Materials and methods ) to which sequencing reads were mapped . We identified 683 transcripts that were irradiation sensitive ( downregulated; FDR ≤ 0 . 05 ) ( Supplementary file 1a ) . Expression of irradiation-sensitive transcripts by WISH was indeed reduced after exposure to irradiation , validating our RNA-seq approach ( Figure 2—figure supplement 2 ) . Two rounds of expression screening were then applied to hone in on cycling-cell transcripts from our irradiation-sensitive dataset ( Figure 2e ) . The position of cycling cells in the neck is spatially restricted in a conserved pattern ( Koziol and Castillo , 2011 ) ( Figure 3a ) : cycling cells reside in the neck parenchyma bounded by the nerve cords and are absent from the animal edge where muscle and tegument are located ( Bolla and Roberts , 1971 ) . Among 194 irradiation-sensitive transcripts that displayed clear WISH patterns , 63% were expressed in the neck parenchyma , though in a variety of patterns ( Figure 3—figure supplement 1 ) . 13% showed similar patterns to h2b and mcm2 ( Figure 3b–c , Figure 3—figure supplement 1b ) . These include the predicted nucleic acid binding factors Zn finger MYM type 3 ( zmym3 ) and pogo transposable element with ZN finger domain-like ( pogzl ) , as well as NAB co-repressor domain two superfamily member ( nab2 ) and nuclear lamina component laminB receptor ( lbr ) . 25% of irradiation-sensitive transcripts , were expressed in a minority of cells in the neck parenchyma ( Figure 3—figure supplement 1c ) . 24% were expressed within the parenchyma and more broadly toward the animal edge ( Figure 3—figure supplement 1d ) . The remainder represented transcripts expressed at segment boundaries or in differentiated tissues ( Figure 3—figure supplement 1e–f ) . All transcripts that were expressed in the neck parenchyma were also found throughout the worm body , even in the most posterior proglottids ( Figure 3—figure supplement 1b–c ) . In conclusion , irradiation-sensitive transcripts identified by RNA-seq likely represent markers for stem cells , progenitors , and even differentiated cells that were lost or compromised following irradiation . To focus on transcripts with enriched expression in cycling cells , we performed double FISH ( dFISH ) with irradiation-sensitive candidates and either h2b or mcm2 , which we used interchangeably as they are co-expressed in the neck parenchyma ( Figure 3—figure supplement 2 ) . After dFISH for 53 candidates , 72% of transcripts tested were co-expressed in cycling cells ( Figure 3—figure supplement 3a , Supplementary file 1b ) . The irradiation-sensitive transcripts from Figure 3c were indeed colocalized in cycling somatic cells ( Figure 3d ) . One transcript , the homeobox factor prospero ( prox1 ) , was expressed exclusively in a subset of cycling cells ( Figure 3—figure supplement 3b ) . We confirmed that genes with expression that only partially overlapped in the neck parenchyma , such as the Zn finger-containing gene HDt_10981 and palmitoyl-protein thioesterase 1 ( ppt1 ) , were expressed in both cycling cells and non-cycling cells ( Figure 3—figure supplement 3c ) . We propose that these genes likely represent lineage-committed stem cells or progenitors for tissues such as muscle , neurons , tegument , or protonephridia . 28% of irradiation-sensitive transcripts were predominantly expressed in non-cycling cells that were juxtaposed to cycling cells ( Figure 3—figure supplement 3d ) . The transcriptional heterogeneity detected in the cycling-cell compartment is reminiscent of similar observations made in the regenerative planarian S . mediterranea ( Reddien , 2018 ) . A comparative analysis between verified tapeworm cycling-cell transcripts and their putative planarian homologs revealed a number of transcripts with conserved expression in cycling-cell populations from these distantly related flatworms ( Supplementary file 1c ) ( see Discussion ) . In summary , our analysis revealed a heterogeneous and complex mixture of cell types or states in the neck parenchyma as well as within the cycling-cell population . What role ( s ) do the newly identified cycling-cell genes play during regeneration ? We performed RNAi of target genes , confirmed knockdown by quantitative PCR ( Figure 4—figure supplement 1 ) , and assayed for defects in growth and regeneration ( Figure 4a ) . As a proof of principle , we knocked down h2b , which should compromise growth due to cycling cell loss , as observed in other flatworms ( Collins et al . , 2016; Solana et al . , 2012 ) . Knockdown of h2b , zmym3 , and pogzl each resulted in diminished growth and regeneration ( Figure 4b–c ) . The number of proglottids regenerated was also reduced , but could not be quantified as many RNAi worms were so thin and frail ( Figure 4b ) that proglottid definition was lost . Are these RNAi-induced failures in growth and regeneration due to defects in the cycling-cell population ? RNAi knockdown of h2b , zmym3 , and pogzl severely reduced the number of proliferative cells in the neck that could incorporate F-ara-EdU ( Figure 4d–e ) . We also observed fewer mcm2+ cells after RNAi ( Figure 4f ) . Taken together , these results indicate that the cycling-cell population is either lost or dysregulated . Therefore , h2b , zmym3 , and pogzl are necessary for the maintenance and/or proper function of cycling cells , likely including stem cells , in H . diminuta . Although we have identified heterogeneity within the cycling-cell population of the neck parenchyma and uncovered genes that are required for growth and regeneration , it remains unclear why regeneration competence is restricted to the neck . By WISH and FISH , all cycling-cell transcripts including zmym3 and pogzl were detected throughout the whole tapeworm body ( Figure 4g , Figure 3—figure supplement 1b–c ) . In the tapeworm posterior , zmym3 and pogzl were expressed in gonadal tissues ( which contain mitotic germ cells ) but also in somatic cells within the parenchyma ( Figure 4—figure supplement 2 ) . If zmym3 and pogzl mark stem cells , this suggests that stem cells reside even in posterior tissues that are not competent to regenerate . Since zmym3 and pogzl label all cycling cells , it is possible that stem cells of limited potential exist in the posterior , but an elusive subpopulation of pluripotent stem cells is confined to the neck . Since we observed an anterior bias in regenerative ability ( Figure 1h–i ) , we hypothesized that RNA-seq may reveal an anteriorly biased stem cell distribution that may point us to a pluripotent stem cell subpopulation . Thus , we performed RNA-seq of 1 mm anterior , middle , and posterior neck fragments ( Figure 1h ) , and identified 461 anterior-enriched and 241 anterior-depleted transcripts ( Supplementary file 1d ) . By WISH , anterior-enriched and anterior-depleted transcripts were often detected in corresponding gradients ( Figure 5a ) , but in patterns that were excluded from the neck parenchyma . When we overlaid the anterior-enriched and -depleted datasets with our irradiation-sensitive dataset , the majority of anterior-enriched transcripts ( 88% ) were not irradiation sensitive ( Figure 5b ) . Our results suggest that the A-P polarized signals across the neck region arise predominantly within the non-cycling compartments . Since our RNA-seq analysis identified 57 transcripts that were anterior enriched and irradiation sensitive , we examined expression patterns within this category . We found 15 transcripts expressed in a subset of cells within the neck parenchyma ( Figure 5c ) and initially hypothesized that these transcripts may represent subsets of stem cells . We were able to successfully test eight candidates by dFISH with cycling-cell markers and found that the majority ( 7/8 ) were not expressed in cycling cells ( Figure 5d , Supplementary file 1b ) . Only prox1 was co-expressed in cycling cells ( Figure 3—figure supplement 3b ) . At present , the identity and function of prox1+ cells is unknown . Furthermore , prox1 is expressed throughout the tapeworm body ( Figure 3—figure supplement 1 ) . Thus , our analyses have not revealed an anteriorly biased subpopulation of stem cells that confer regenerative ability . With no evidence for a unique neck-specific subpopulation of stem cells , we hypothesized that stem cells may be distributed throughout the tapeworm , but that extrinsic signals functioning in the neck are necessary to instruct stem cell behavior and/or proglottid regeneration . We designed a functional assay to test populations of cells for the ability to rescue regeneration , modelled after similar experiments performed on planarians ( Baguñà , 2012 ) . We exposed tapeworms to a lethal dose of x-irradiation ( 200 Gy ) , injected cells from wild-type donors into the neck region , amputated 5 mm anterior fragments , and assayed rescue of lethality and regeneration after 30 days in vitro ( Figure 6a ) . Remarkably , bulk-cell transplants were able to either partially or fully rescue irradiated worms that were destined to die ( Figure 6a , c ) . ‘Full’ rescue was ascribed to worms with normal adult appearance whereas ‘partial’ rescue was assigned to cases in which proglottids were regenerated but the worms displayed abnormalities , like contracted necks ( Figure 6—figure supplement 1a ) . We did not observe any proglottid regeneration in irradiated worms with or without buffer injection ( Figure 6a , c ) . Is the rescue ability described above dependent on tapeworm cycling cells ? We exposed donors to F-ara-EdU for 1 hr , to label cycling cells prior to transplantation into irradiated hosts ( Figure 6—figure supplement 1b ) . Though bulk-cell transplants were performed , injection sites contained 0 , 1 , or small groups of F-ara-EdU+ cells immediately after transplantations ( Figure 6—figure supplement 1c ) , likely due to technical challenges . Despite this issue , we were able to detect large colonies of F-ara-EdU+ cells 3 days post-transplantation ( Figure 6—figure supplement 1d ) . We also observed that some labeled cells were incorporated into terminally differentiated tissues at the animal edge ( Figure 6—figure supplement 1d: inset ) . Thus , cycling cells from donors are able to become established and differentiate inside the irradiated host . To test if the cycling-cell population is necessary to rescue lethally irradiated tapeworms , we depleted cycling cells from donor worms using 50 mM hydroxyurea ( HU ) , which resulted in 96 ± 3% loss of cycling cells after 6 days ( Figure 6—figure supplement 1e–f ) . Cycling cells are essential for rescue of regeneration as injected cells from HU-treated donors rescued only 1% of the time , compared to 26% rescue using cells sourced from sister donors that did not receive the drug ( Figure 6b–c ) . HU was used to deplete cycling cells instead of irradiation in order to avoid inducing DNA damage in the transplanted cells . Cells transplanted from HU-treated donors had otherwise comparable morphology to untreated cells ( Figure 6—figure supplement 1g ) . Our results suggest that tapeworm cycling cells contain bona fide stem cell activity . With this functional assay in hand , we examined the rescue ability of cells from anterior donor tissues ( including the regeneration-competent neck ) compared to donor tissues from the most posterior termini of 6-day-old tapeworms ( which are regeneration incompetent and exclusively comprised of proglottids ) . Cells from either region were able to rescue regeneration in lethally irradiated tapeworms ( Figure 6b–c ) . Thus , cells from posterior proglottids were competent to receive signals from the head and neck region that regulate regenerative ability . Interestingly , using pulse-chase experiments with F-ara-EdU , we find that the cycling cells of posterior proglottids can give rise to multiple differentiated cell types like muscle/tegument at the animal edge as well as flame cells of the protonephridial system marked by anti-acetylated α-tubulin antibodies ( Rozario and Newmark , 2015 ) ( Figure 6—figure supplement 2 ) . Thus , the cycling cells from tapeworm posteriors show hallmarks of stem cell activity , despite the fact that this tissue is not competent to regenerate . Taken together , the results of our study support the idea that the regeneration competence of the neck is due to extrinsic signals that regulate regeneration , rather than intrinsic properties of stem cells in the neck region ( see Discussion ) . It appears that in tapeworms , location matters enormously: the head and neck environment provide cues that regulate the ability of stem cells to regenerate proglottids , even though cycling cells ( and likely stem cells ) , are not anatomically confined . Our study shows that H . diminuta is a powerful developmental model for understanding intrinsic and extrinsic regulation of stem cells and regeneration . The regionally limited regenerative biology of H . diminuta and the technical advances put forth in this work show that we can exploit this tapeworm to understand the complexities of stem cell regulation in parasites . We defined heterogenous stem cells that are collectively pluripotent but that require extrinsic head-dependent signals to enable persistent proglottid regeneration . Understanding how the stem cell niche we describe is regulated may have broad implications for elucidating stem cell biology in parasitic flatworms , as well as other regenerative animals . Infective H . diminuta cysts were obtained from Carolina Biological ( 132232 ) . To obtain adult tapeworms , 100–400 cysts were fed to Sprague-Dawley rats by oral gavage in ~0 . 5 mL of 0 . 85% NaCl . Rats were euthanized in a CO2 chamber 6 days post-gavage , tapeworms were flushed out of the small intestine , and washed in 1X Hanks Balanced Salt Solution ( HBSS; Corning ) ( 140 mg/L CaCl2 , 100 mg/L MgCl2 . 6H2O , 100 mg/L MgSO4 . 7H2O , 400 mg/L KCl , 60 mg/L KH2PO4 , 350 mg/L NaHCO3 , 8 g/L NaCl , 48 mg/L Na2HPO4 , 1 g/L D-glucose , no phenol red ) . Rodent care was in accordance with protocols approved by the Institutional Animal Care and Use Committee ( IACUC ) of the University of Wisconsin-Madison ( M005573 ) . Biphasic parasite cultures were prepared based on the Schiller method ( Schiller , 1965 ) . Briefly , the solid phase was made in 50 mL Erlenmeyer flasks by mixing 30% heat-inactivated defibrinated sheep blood ( Hemostat ) with 70% agar base for 10 mL blood-agar mixture per flask . Fresh blood was heat-inactivated at 56°C for 30 min then kept at 4°C and used repeatedly for one week by first warming the blood to 37°C . The agar base was prepared from 8 g Difco nutrient agar and 1 . 75 g NaCl in 350 mL water , autoclaved , and stored at 4°C . Before use , the agar base was microwaved to liquify , and cooled to below 56°C before mixing with warmed blood . After the blood-agar mixture solidified , 10 mL of Working Hanks 4 ( WH4; 1X HBSS/4 g/L total glucose/1X antibiotic-antimycotic ( Sigma ) ) was added . Each flask was topped with a gas-permeable stopper ( Jaece Identi-plug ) and pre-incubated at 37°C in hypoxia ( 3% CO2/5% O2/92% N2 ) overnight before use . Before tapeworms were transferred into the flasks , the liquid phase was adjusted to pH7 . 5 with 200 μL 7 . 5% NaHCO3 ( Corning ) . Tapeworms were first washed in WH4 for 10 mins at 37°C in petri dishes pre-coated with 0 . 5% BSA to inhibit sticking . Transfers to pre-cultured flasks were performed by gently lifting the worms with a stainless-steel hook ( Moody Tools ) and immersing them in the liquid phase . Tapeworms were grown in hypoxia and transferred to fresh cultures every 3–4 days . Tapeworms were heat-killed by swirling in 75°C water for a few seconds until the worms relaxed and elongated , then fixative ( 4% formaldehyde in Phosphate Buffered Saline with 0 . 3% TritonX-100 ( PBSTx ) ) was added immediately for 30 min-2hr at room temperature or overnight at 4°C . For DAPI staining , samples were incubated in 1 μg/mL DAPI ( Sigma ) in PBSTx overnight at 4°C and cleared in 80% glycerol/10 mM Tris pH7 . 5/1 mM EDTA overnight at room temperature before mounting . For F-ara-EdU pulse , tapeworms were incubated in 0 . 1 μM F-ara-EdU ( Sigma ) in 1% DMSO at 37°C in WH4 . Tapeworms were heat-killed ( above ) and fixed in 4% formaldehyde/10% DMSO/1% NP40/PBSTx . Large tissues/worms were permeabilized by incubating in PBSTx at room temp for several days . Additional permeabilization was achieved by treatment with 10 μg/mL Proteinase-K/0 . 1% SDS/PBSTx for 10–30 min at room temperature , fixed in 4% formaldehyde/PBSTx for 10 min before samples were cut into small pieces or retained whole in PBSTx . Samples were further permeabilized in PBSTx/10% DMSO/1% NP40 for 20 min-1 hr ( depending on size ) before performing the click-it reaction ( Salic and Mitchison , 2008 ) with Oregon Green 488 azide ( Invitrogen ) . Signal was detected using anti-Oregon Green 488-HRP antibody ( 1:1000; Invitrogen ) in K-block ( 5% Horse serum/0 . 45% fish gelatin/0 . 3% Triton-X/0 . 05% Tween-20/PBS ) ( Collins et al . , 2011 ) followed by 10–20 min Tyramide Signal Amplification ( TSA ) reaction ( King and Newmark , 2013 ) . Tiled confocal z-stacks through the anterior of the worms were taken and cell numbers were counted using background subtraction on Imaris software . F-ara-EdU+ cells were normalized to worm area from maximum projections of the DAPI stain . Flame cells were stained using an anti-acetylated α-tubulin mouse antibody at 1:500 ( sc-23950 , Santa Cruz ) as described previously ( Rozario and Newmark , 2015 ) . Most irradiation was performed using a CellRad irradiator ( Faxitron Bioptics ) at 200 Gy ( 150 kV , 5 mA ) with two exceptions . Due to instrument failure , a cesium irradiator was used for one rescue experiment with donors + /- HU ( Figure 6b ) at 400 Gy ( 92 ± 5% cycling cell loss 3 days post-irradiation ) . The rescue experiment with + /- HU donors was performed a third time once we gained access to an x-irradiator ( Xstrahl RS225 Cell Irradiator ) , where the lethal dose was 200 Gy ( 63 ± 10% cycling cell loss 3 days post-irradiation ) . All three experiments gave similar results despite the use of different irradiators . In all cases , lethal irradiation was determined as the dosage at which tapeworms degenerated , had 0 proglottids , and were inviable after 30 days in culture . Irradiation was performed in WH4 in BSA-coated petri dishes . RNA was collected from five regions: 1 ) head and neck , 2 ) immature proglottids , 3 ) mature reproductive proglottids , 4 ) gravid proglottids , and 5 ) mixed larval stages isolated from beetles . The first three regions covered the entirety of 3 . 5-week-old adult tapeworms . Gravid proglottids were taken from posteriors of 10-week-old tapeworms . Paired-end libraries were constructed with 2 × 150 bp reads from a HiSeq2500 chip . 2 x ~ 30 million reads were obtained for each sample . The transcriptome was assembled from three components: 1 ) map-based assembly , 2 ) de novo assembly , and 3 ) Maker predictions from Wormbase Parasite . The map-based assembly was performed using TopHat2 with the 2014 H . diminuta draft genome courtesy of Matt Berriman ( Wellcome Sanger Institute , UK ) . 15 , 859 transcripts were assembled using TopHat . De novo assembly was performed using Velvet/Oases and resulted in 144 , 682 transcripts . There were 11 , 275 predicted Maker transcripts and 73 . 2% matched ( >95% along the length ) to the TopHat transcripts . The remaining predicted transcripts that were not represented in the TopHat dataset were added for a combined TopHat/predicted set of 17 , 651 transcripts . Most of the Oases transcripts matched to the TopHat/predicted set but 35 , 300 or 24 . 4% of the Oases transcripts did not ( >75% match cut-off ) . These transcripts could be transcripts missed in the genome , transcription noise , non-coding transcripts , or contamination . We found significant contamination from beetle tissue in the larval tapeworm sample ( more below ) . Initial filtering for contamination excluded 1388 transcripts ( from beetle , rat , bacterial , and viral sources ) . At this point 51 , 563 transcripts were retained from the three methodologies described above and were processed for further filtering . There was significant contamination from beetle tissues that had adhered to the tapeworm larvae , which produced transcripts with best hits to beetle proteins ( Ixodes scapularis , Harpegnathos saltator , Monodelphis domestica , Nasonia vitripennis , Pediculus humanus corporis , Solenopsis invicta , Tenebrio molitor , or Tribolium castaneum ) . Most of the transcripts were from the Oases de novo assembly and did not match the H . diminuta genome . Furthermore , they were strongly over-represented in the larval sample only . To filter out beetle contamination , we removed 11 , 918 transcripts from the Oases assembly without matches to the H . diminuta genome that showed >90% expression ( by RPKM ) in the larval sample . To the remaining 39 , 645 transcripts , we applied additional filters: 1 ) Remove transcripts if average RPKM <1 unless the transcript is long ( >1000 bp ) , has a long ORF ( >500 bp ) or is annotated . 11 , 615 transcripts were removed as they met none of these criteria . 2 ) A stringent expression cut-off was applied to the remaining Oases transcripts; transcripts were discarded if average RPKM <5 and maximum RPKM <10 unless the transcripts were long ( >1000 bp ) , had long ORFs ( >500 bp ) or were annotated . 8027 transcripts were removed . 3 ) 51 transcripts were removed because they are mitochondrial or rRNAs . 4 ) An ORF size filter was applied to remove all transcripts with ORF <300 bp unless they are annotated . 5331 transcripts were removed . 5 ) For the Maker predicted transcripts , expression and size filters were applied to remove transcripts with expression <1 RPKM and size <500 bp . 275 transcripts were removed . Our final transcriptome is comprised of 14 , 346 transcripts ( 84 . 9% TopHat , 8 . 4% Maker predictions , 6 . 1% Oases with match to genome , and 0 . 6% Oases without match to genome ) . The total transcriptome size is 34 Mb with average transcript length of 2 , 354 bp . This Transcriptome Shotgun Assembly project has been deposited at DDB/ENA/Genbank under the accession GHNR00000000 . The version described in this paper is the first version , GHNR01000000 . All sequence reads are available at GenBank Bioproject PRJNA546290 . Tissue was collected and immediately frozen on dry ice in 100 μL Trizol ( Life Technologies ) before RNA extraction . Tissue homogenization was performed as the mixture was in a semi-frozen state using RNase-free pestles and a pestle motor . RNA was purified using the Direct-zol RNA MiniPrep kit ( Zymo ) . RNA quality was assessed using Bioanalyzer , libraries were prepared with TruSeq Stranded mRNAseq Sample Prep kit ( Illumina ) , and sequenced on two lanes on a HiSeq2500 chip . We performed paired-end sequencing and obtained ~20 million reads per sample . Samples were obtained in triplicate . To identify irradiation-sensitive transcripts , 2 mm anterior tapeworm fragments were cut from 10 worms after 3 days in vitro . To identify differentially expressed transcripts across the neck A-P axis , 1 mm fragments were cut from 20 freshly obtained 6-day-old tapeworms . Paired-end reads were mapped to the transcriptome ( above ) using default settings on CLC Genomics Workbench 6 ( Qiagen ) except that read alignments were done with a relaxed length fraction of 0 . 5 . Differential gene expression analysis was done with the same software using estimate tagwise dispersions on total read counts and a total count filter cut-off of 5 reads . All sequence reads used for differential gene expression analyses are available at GenBank Bioproject PRJNA546293 . Target genes were amplified using PCR with Platinum Taq ( Life Technologies ) from cDNA generated from RNAs extracted from tapeworm anteriors to enrich for neck transcripts . PCR products were inserted via TA-mediated cloning into the previously described vector pJC53 . 2 ( Collins et al . , 2010 ) pre-digested with Eam11051 . Anti-sense riboprobes could be generated by in vitro transcription with SP6 or T3 RNA polymerases . For RNAi , dsRNA was generated using T7 RNA polymerase . For sequences and primers , refer to Supplementary file 1e . WISH and FISH protocols were modified from previously published methods for planarians ( King and Newmark , 2013 ) and the mouse bile-duct tapeworm Hymenolepis microstoma ( Olson et al . , 2018 ) . Tapeworms were heat killed and fixed in 4% formaldehyde/10% DMSO/1% NP40/PBSTx for 30 min at room temperature before washing and dehydration into methanol . Dehydrated samples were frozen at −30°C for at least 2 days . After rehydration , samples were permeabilized in 10 μg/mL Proteinase-K/0 . 1% SDS/PBSTx for 30 min , washed into 0 . 1 M Triethanolamine pH7-8 ( TEA ) , 2 . 5 μL/mL acetic anhydride was added for 5 min with vigorous swirling , acetic anhydride step was repeated , washed in PBSTx , and post-fixed in 4% formaldehyde/PBSTx for 10 min . Probe synthesis , hybridization , and staining were performed as previously described ( King and Newmark , 2013 ) using probe concentrations at ~50 ng/mL for 16–48 hr at 56°C . All probes were synthesized with either DIG or DNP haptens and detected using the following antibodies , all at 1:2000: anti-DIG-AP ( Sigma ) , anti-DIG-POD ( Sigma ) , anti-DNP-HRP ( Vector Labs ) . Colorimetric development was done using NBT ( Roche ) /BCIP ( Sigma ) or with Fast-Blue ( Sigma ) ( Currie et al . , 2016 ) . Fluorescent signal was visualized after 10–20 min TSA reaction ( King and Newmark , 2013 ) . DAPI staining and mounting were performed as described above . Confocal imaging was performed on a Zeiss LSM 880 with the following objectives: 20X/0 . 8 NA Plan-APOCHROMAT , 40X/1 . 3 NA Plan-APOCHROMAT , and 63X/1 . 4 NA Plan-APOCHROMAT . WISH samples and whole-mount DAPI-stained worms were imaged using Zeiss AxioZoom V16 macroscope . Image processing was performed using ImageJ for general brightness/contrast adjustments , maximum-intensity projections , and tile stitching ( Preibisch et al . , 2009 ) . dsRNA was synthesized as previously described ( Rouhana et al . , 2013 ) and resuspended at concentrations ~ 1 . 5–2 μg/μL . For control injections , 1 . 5 kb dsRNA derived from ccdB and camR-containing insert of the pJC53 . 2 vector was used ( Collins et al . , 2010 ) . 6-day-old tapeworms were obtained and microinjected with dsRNA using femtotips II via the Femtojet injection system ( Eppendorf ) to obtain spreading across the first ~3–4 mm anterior of the tapeworm . The spread of injected fluids could be detected by a temporary increase in opacity . 500 hPa injection pressure for 0 . 3–1 s was used per injection site . Whole tapeworms were cultured in vitro for 3 days , 2 mm anterior fragments were amputated , worms were re-injected with dsRNA on day 6 , and cultured in vitro for an additional 9 days before termination . Whole worms ( 6 days old ) were injected with dsRNA throughout and frozen in Trizol on dry ice after 6 days in vitro for RNA extraction according to manufacturer’s protocol and DNAse ( Promega ) treatment for 30 min at 37°C . cDNA synthesis was performed using SuperScriptIII First-Strand Synthesis System ( Invitrogen ) with Oligo ( dT ) 20 primers followed by iScript cDNA Synthesis Kit ( Bio-Rad ) . qPCR was performed using GoTaq Mastermix ( Promega ) on a StepOnePlus real-time PCR machine ( Applied Biosystems ) . 60S ribosomal protein L13 ( 60Srpl13 ) was used as an internal normalization control . For primers refer to Supplementary file 1e . Tapeworms were treated with HU ( Sigma ) or HBSS ( for controls ) every day for a total of 6 days . HU stock solution was made fresh every day at 2 M in HBSS . 250 μL was added to each flask of tapeworms for final concentration of 50 mM . HU is unstable at 37°C so worms were transferred into fresh HU-containing media every two days , and fresh HU was added every other day . For dissociated cell preparations , tapeworms were placed in a drop of calcium-magnesium free HBSS ( CMF HBSS , Gibco ) , minced into small pieces with a tungsten needle , incubated in 3X Trypsin-EDTA ( Sigma ) in CMF HBSS for 30 min at 37°C and dissociated using a dounce homogenizer ( Kontes ) . Cells were pelleted by centrifugation at 250 g for 5 min . The cell pellet was washed in CMF HBSS and passed through cell strainers at 100 μm , 40 μm , 20 μm , and 10 μm ( Falcon and Sysmex ) with one spin and wash in between . Cells were pelleted and resuspended in 200–400 μL WH4 with 0 . 05% BSA . Cell injections were performed using the Cell Tram Oil four injection system ( Eppendorf ) into the necks of irradiated worms . For + /- HU donors , cell concentrations were measured using a hemocytometer and normalized ( to ~108 cells/mL ) to ensure equal numbers of cells were injected . For all rescue experiments , cells were injected into irradiated hosts on the same day that the hosts were irradiated . After 3 days in vitro , 5 mm anterior fragments were amputated and grown for an additional 27 days . Statistical analyses were performed using Prism7 software ( GraphPad Prism ) . All experiments were repeated at least twice . All measurements were taken from distinct samples . Error bars , statistical tests , number of replicates ( N ) and sample sizes ( n ) are indicated in corresponding figure legends . Either Dunnett’s or Tukey’s multiple comparison tests were used for one-way ANOVAs . SD = standard deviation . P-values: ns = not significant , =p ≤ 0 . 5 , ***=p ≤ 0 . 0001 .	Many worms have remarkable abilities to regrow and repair their bodies . The parasitic tapeworms , for example , can reach lengths of several meters and grow much more quickly than tissues in humans and other complex animals . This growth allows tapeworms to counteract the continual loss of the segments that make up their bodies , known as proglottids – a process that happens throughout their lives . The capacity to regenerate thousands of lost body segments and maintain an overall body length suggests that tapeworms have groups of stem cells in their body which can grow and divide to produce the new body parts . Yet , regeneration in tapeworms has not been closely studied . Rozario et al . have now examined Hymenolepsis diminuta , the rat tapeworm , and identified the neck of the tapeworm as crucial for its ability to regrow lost body segments . Further analysis identified two genes , zmym3 and pogzl , that are essential for cell division during tapeworm growth . However , Rozario et al . showed that these genes are active elsewhere in the worm’s body and that it is the conditions found specifically in the tapeworm’s neck that create the right environment for stem cells to enable regeneration of new segments . Tapeworms provide a valuable example for studying the growth of stem cells and these findings highlight the important role that the cells’ surroundings play in driving stem cell activity . These findings could also lead to new insights into how stem cells behave in other animals and could potentially lead to new approaches to prevent or treat tapeworm infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "microbiology", "and", "infectious", "disease" ]	2019	Region-specific regulation of stem cell-driven regeneration in tapeworms
PIWI-interacting RNAs ( piRNAs ) protect the germ line by targeting transposable elements ( TEs ) through the base-pair complementarity . We do not know how piRNAs co-evolve with TEs in chickens . Here we reported that all active TEs in the chicken germ line are targeted by piRNAs , and as TEs lose their activity , the corresponding piRNAs erode away . We observed de novo piRNA birth as host responds to a recent retroviral invasion . Avian leukosis virus ( ALV ) has endogenized prior to chicken domestication , remains infectious , and threatens poultry industry . Domestic fowl produce piRNAs targeting ALV from one ALV provirus that was known to render its host ALV resistant . This proviral locus does not produce piRNAs in undomesticated wild chickens . Our findings uncover rapid piRNA evolution reflecting contemporary TE activity , identify a new piRNA acquisition modality by activating a pre-existing genomic locus , and extend piRNA defense roles to include the period when endogenous retroviruses are still infectious . A vertebrate germ-line genome faces repeated activation of transposable elements ( TEs ) as well as integration of new retroviruses that become endogenized . The genome has a marvelous ability to adapt to these challenges ( McClintock , 1984 ) . Among the adaptive responses , PIWI-interacting RNAs ( piRNAs ) are essential to protect the integrity of the germ-line genome by targeting ‘non-self’ sequences through base-pair complementarity . Disruption of piRNA pathways activates TEs in male and female fruit flies ( Wilson et al . , 1996; Lin and Spradling , 1997 ) , male and female zebrafish ( Houwing et al . , 2008 ) , and male mice ( Kuramochi-Miyagawa et al . , 2004; Carmell et al . , 2007 ) . piRNAs bind a specialized sub-family of Argonaute proteins , the PIWI proteins , which are expressed mainly in germ cells ( Kumar and Carmichael , 1998; Aravin and Hannon , 2008; Farazi et al . , 2008; Kim et al . , 2009; Thomson and Lin , 2009; Cenik and Zamore , 2011 ) . piRNAs guide PIWI proteins to their complementary RNA targets . PIWI proteins catalyze an endonucleolytic cleavage between the 10th and 11th positions of the RNA target relative to the piRNA 5´end . The cleaved product can then be loaded into another PIWI protein becoming a secondary piRNA . This results in a ‘Ping-Pong’ loop that amplifies antisense TE piRNAs ( Brennecke et al . , 2007; Gunawardane et al . , 2007 ) . The initial triggers of Ping-Pong amplification are produced from discrete genomic loci . The host needs to incorporate the foreign sequences into these piRNA-producing loci to recognize novel ‘nonself’ sequences , an RNA-based immune system similar to CRISPRs in prokaryotes ( Kumar and Chen , 2012 ) . New piRNA-producing loci can originate by duplication ( Assis and Kondrashov , 2009 ) , but duplication per se does not directly generate new piRNA sequences . The only known mechanism of new piRNA acquisition comes from studies of fruit flies , in which a TE inserts into an actively expressed piRNA cluster ( Khurana et al . , 2011 ) . However , considering that active piRNA-producing loci represent only a tiny fraction of the genome , and that no preference for insertions of TEs into piRNA-producing loci has been reported ( Kumar and Chen , 2012 ) , other piRNA acquisition mechanisms remain to be discovered . Endogenous retroviruses ( ERVs ) are distributed relative strictly in vertebrate genomes ( Gifford and Tristem , 2003; Eickbush and Jamburuthugoda , 2008 ) . Numerous distinct ERV families have invaded the chicken germ line ( Jurka et al . , 2005 ) , making chicken ( Gallus gallus ) an excellent model to study virus-host interplay . Compared to other TEs in the chicken genome , including the ancient CR1 superfamily ( Vandergon and Reitman , 1994 ) and DNA transposons , and largely absent short interspersed nuclear elements ( SINEs ) ( International Chicken Genome Sequencing Consortium , 2004 ) , chicken ERVs are more active and have led to phenotypic changes like blue egg-shells ( Wang et al . , 2013 ) and late feathering ( Boyce-Jacino et al . , 1989 ) . Chicken ERVs can also remain infectious , and may evolve into new viruses through recombination with host genes or exogenous viruses . Avian leukosis virus ( ALV ) was the first ERV to be discovered ( Temin , 1964; Weiss , 1969; Baluda , 1972 ) . Uninfected chickens sometimes spontaneously shed infectious ALV subtype E ( ALVE ) viruses ( Varmus et al . , 1972; Weiss , 2006 ) , and the infection can lead to cancer ( Weiss , 2006 ) . ALV acquisition of a host oncogene , SRC , generated a more acute transforming virus , Rous sarcoma virus ( RSV ) ( Stehelin et al . , 1976 ) . Recombination between ALVE and EAV-HP—a member of the endogenous avian retrovirus ( EAV ) family , has created ALV subtype J ( ALVJ ) , a new subgroup of ALVs that were associated with myeloid leukosis in meat-type chickens ( Smith et al . , 1999 ) . After spreading to China in 2002 , ALVJ mutated to infect egg-laying breeds with a wide spectrum of tumors ( Gao et al . , 2010 ) . Despite the threat of ERVs to the poultry industry , we lack a systematic investigation of ERV activity and piRNA-mediated suppression in the chicken germ line . Here , we identified the ERV activity and piRNA-producing loci in White Leghorn , the most popular egg-laying breed , dissected the interplay between ERVs and piRNAs , and traced the origination of recently acquired piRNAs in undomesticated wild chickens , Red Jungle Fowl ( Eriksson et al . , 2008 ) . We chose to focus on the White Leghorn because this domestic breed has suffered from ERV activation and thus its ERVs have been extensively studied ( Crittenden , 1991 ) . White Leghorn lays an average of 280 eggs per year ( Bao et al . , 2008 ) . Breeders have used the late-feathering trait as a convenient marker to select female layers at hatch ( Boyce-Jacino et al . , 1989 ) ; however , this trait is linked to a fully infectious ALVE provirus ( known as ev21 ) on the Z chromosome , which causes decreased performance ( Smith and Fadly , 1988; Fadly and Smith , 1997 ) . We found that 73 TE families , including ALVE , were active in White Leghorn testis , and all 503 TE insertions absent in Red Jungle Fowl were derived from these TE families . More than 60% of the active TEs belonged to ERV families , indicating that ERVs contribute to most TE activity in chickens . All active TEs are targeted by robust piRNA-mediated suppression . As TEs become inactivated , their targeting piRNAs erode away . We found that the ability of chickens to produce piRNAs targeting ALV is an evolutionarily recent acquisition—White Leghorn produced abundant ALV piRNAs while Red Jungle Fowl did not . The ALV piRNAs in White Leghorn were produced from a truncated ALV provirus that was known to render its host ALV resistant ( Robinson et al . , 1981 ) . The presence of this provirus predated domestication , indicating that the responsible genomic region exists in either an ‘on’ or ‘off’ state as a piRNA-producing locus . Active ERVs are transcribed and translated , and are able to transpose within the germ line . Comparison of RNA-seq data from 12 tissues ( Brawand et al . , 2011 ) indicated that chicken ERV families were ubiquitously expressed ( Figure 1—figure supplement 1A ) . Because many insertions are truncated , detection of ERV RNAs does not necessarily indicate that they are translated or competent for transposition . It has been shown that chicken ERVs are transcribed and translated in somatic cells ( Bolisetty et al . , 2012 ) . Because the highest expression was in testis and ovary—the only tissues where their expansion can become heritable ( Figure 1—figure supplement 1A ) , we decided to perform polysome profile analysis to determine whether ERVs were also being translated in testis . Adult White Leghorn testis lysates were separated in 10–50% sucrose-density gradients by ultracentrifugation ( Figure 1A ) . This fractionation separates non-translating ribonucleoproteins , small and large subunits of ribosomes , monosomes and polysomes , as shown by the distribution of rRNA . Actively translated β-ACTIN , CILI , and CIWI mRNAs co-sedimented with both monosome and polysome fractions , but the MALAT1 non-coding RNA did not co-sedimented with polysomes . CILI and CIWI are the two PIWI proteins in chickens ( Figure 1—figure supplement 2 ) . We tested the distribution of CR1-B and CR1-F families that belong to the CR1 superfamily , as well as the EAV-HP and ALVE that belong to ERV families . Although CR1 arose prior to the divergence of birds and reptiles and peaked ~45 million years ago ( Vandergon and Reitman , 1994 ) , CR1-F and CR1-B elements remain able to drive their own transcription ( Wicker et al . , 2005; Lee et al . , 2009 ) . CR1-B , CR1-F , EAV-HP , and ALVE transcripts co-sedimented with polysomes . These profiling results suggest that CR1-B , CR1-F , EAV and ALVE insertions are transcribed and translated in testis . 10 . 7554/eLife . 24695 . 003Figure 1 . Active ERVs in White Leghorn testis . ( A ) A254 absorbance profile of 10% to 50% sucrose density gradients of testis lysates from adult rooster . From top to bottom , plots show the relative abundance of 18S rRNA , β-ACTIN mRNA , CILI mRNA , CIWI mRNA , chicken Malat1 lncRNA , CR1B , CR1F , EAV-HP , and ALVE quantified by RT-qPCR . Data were normalized to a spike-in control RNA . ( B ) Scatter plots of transcript abundance versus ribosome density . Each black dot represents an mRNA expressed in testis . Each filled red circle represents an ERV family , and each open red circle represents any other TE family , including DNA transposons and CR1 superfamily; rpkm , reads per kilobase of transcript per million mapped reads; fpkm , fragments per kilobase of transcript per million mapped reads . ( C ) Normalized reads of White Leghorn RPFs ( Top ) , White Leghorn piRNAs ( Middle ) , and Red Jungle Fowl small RNA reads ( >23 nt ) ( Bottom ) . Blue represents sense mapping reads; Red represents anti-sense mapping reads . The gene organization of ALVE is also shown . Gag , group-specific antigen; Pol , polymerase; Env , envelope protein; ppm , parts per million . ( D ) Circos plot representing the locations , from periphery to center , of cytological position ( black lines represent centromeres ) , piRNA clusters in White Leghorn ( Black lines represent conserved piRNA clusters; White lines represent divergent piRNA clusters ) , putative new insertions discovered by TEMP ( tiles ) using genomic resequencing of White Leghorn , and 2 × 2 contingency table for Fisher’s exact test to assess the significance of the coincidence of transcription and translation of each TE family . The table data correspond to the number of TE families in each category and , in parentheses , the number of TE families in each category with recent transpositions . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 00310 . 7554/eLife . 24695 . 004Figure 1—figure supplement 1 . Tissue distribution of ERVs and piRNA pathway genes . ( A ) Box plots show the abundance of ERV families in different chicken tissues measured by RNA-seq data . ( B ) Expression of A-MYB , CILI , and CIWI in each tissue measured by RNA-seq data . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 00410 . 7554/eLife . 24695 . 005Figure 1—figure supplement 2 . PIWI proteins are conserved between mammals and birds . Phylogenetic tree of PIWI proteins , including four human PIWI proteins , three mouse PIWI proteins , and two chicken PIWI proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 00510 . 7554/eLife . 24695 . 006Figure 1—figure supplement 3 . Ribosome profiling in adult rooster testes . ( A ) Schematic of ribosome profiling library construction . ( B ) Length distributions of RPFs mapped to mRNA CDS ( black ) and ALVE ( purple ) . ( C ) Metagene plots of RNA-seq ( top ) and RPF ( bottom ) at 5´ leader , CDS , and 3´ trailer of mRNAs . The x-axis shows the median length of these regions , and the y-axis represents the mean of normalized abundance . ( D ) Discrete Fourier transformation of the distance spectrum of 5´ ends of RPFs across mRNA CDSs ( black ) and ALVE ( purple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 00610 . 7554/eLife . 24695 . 007Figure 1—figure supplement 4 . A recent ALVE insertion in the SOX5 gene detected by genome-resequencing of White Leghorn . From top to bottom , the genomic location of the insertion , the genome resequencing signals mapping to Crick strand and Watson strand , Ref-Seq track showing the first intron of SOX5 , RepeatMasker track showing no other TEs in these regions , and example reads that map to the first intron of SOX5 genes are shown ( the rest of reads that map to ALVE do not align to the reference genome ) . The 6 bp targeted site duplication is labeled on the example reads . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 007 To determine whether the observed co-sedimentation with ribosomes reflects the active translation , we performed ribosome profiling using testis lysates from adult White Leghorn . Ribosome profiling is based on the facts that the ribosome-bound fraction of mRNA is protected from RNase digestion in vitro ( Steitz , 1969 ) , and that the subsequent genome-wide sequencing of ribosome-protected fragments ( RPFs ) provides a snapshot of in vivo translation ( Ingolia et al . , 2009 ) . RNA fragments protected from RNase A and T1 digestion were isolated from 80S fractions and sequenced ( Figure 1—figure supplement 3A ) ( Ricci et al . , 2014; Cenik et al . , 2015 ) . Similar to reported RPF sizes in mammals , the RPF sizes in chicken from coding DNA sequences ( CDS ) ranged from 26–32 nt ( Figure 1—figure supplement 3B ) . While RNA-seq reads were distributed throughout the entire set of mRNA transcripts , RPF reads were enriched in CDS regions ( Figure 1—figure supplement 3C ) , and RPFs that mapped to CDS regions accounted for 96% of the RPFs that mapped to entire mRNA transcripts . The RPF reads mapping to open read frames displayed an obvious three-nt periodicity ( Figure 1—figure supplement 3D ) , reflecting the triplet nature of the genetic code during translation elongation . Based on the enrichment of RPFs at CDS regions and the observed codon periodicity of RPFs , we conclude that the ribosome profiling identified RNAs undergoing translation . We integrated the ribosome profiling and RNA-seq data in our analysis of transcription and translation of ERVs and other TE families . The ribosome density of each TE family correlated with their steady-state RNA levels ( ρ = 0 . 82 , p<2 . 2×10−16 ) ( Figure 1B ) . The median translational efficiency ( ratio of ribosome density to transcript abundance ) of ERVs was around 1/10 of the median translation efficiency of mRNAs that were expressed in testis . Consistent with our expectation that ALVE was active in White Leghorn , we detected RPFs mapping to ALVE ( Figure 1C ) . These ALVE RPF reads displayed a length distribution that was similar to that of RPFs from CDS regions ( Figure 1—figure supplement 3B ) , and they displayed codon periodicity ( Figure 1—figure supplement 3D ) . These RPFs were distributed throughout the entire ALVE transcripts but with higher abundance at gag and env than at pol ( Figure 1C ) . Most transcribed TEs were also translated ( the two events , transcription and translation , were significantly associated: Fisher's exact test , p<2 . 2×10−16; Figure 1D ) . We detected RPFs in 71 of 73 TE families that were transcribed ( 97 . 3% ) . Sometimes RPFs could not be unambiguously assigned to TEs due to their small sizes , resulting in false positive signals on transcriptionally silenced TEs . Based on RNA-seq data , polysome profiles , and ribosome profiling , we conclude that most transcribed TE families in the testis were also being translated . To detect new transposition events , we aligned the published resequencing data of the White Leghorn genome with ~100X coverage ( Oh et al . , 2016 ) to the Red Jungle Fowl reference genome . 503 putative TE insertions , absent in Red Jungle Fowl , were distributed throughout the chicken genome ( Figure 1D , Supplementary file 1 ) . Although Red Jungle Fowl are commonly called as the ‘ancestor’ of domestic chicken , they evolved thousands of years in parallel with domestic chicken after chicken domestication ( West and Zhou , 1988 ) , therefore our analysis cannot distinct lineage specific insertions in White Leghorn from lineage specific deletions in Red Jungle Fowl using structure variant alone . De novo ALVE insertions have been reported in domestic fowl ( Crittenden , 1991 ) , and we detected new ALVE insertions in SOX5 , as recently reported ( Rutherford et al . , 2016 ) . The identification of a 6 bp target site duplication typical for ALVE ( Figure 1—figure supplement 4 ) indicates a transposition event rather than a genome duplication event . No putative new insertion came from the 127 transcriptionally inactive TE families ( Figure 1D ) , confirming their inactive state . All putative new insertions came from the 73 actively transcribed TE families . Thus , combing multiple methods to detect active ERVs , we found that active TEs and their insertions in the White Leghorn genome ( Supplementary file 1 ) . Our data indicated that transposing activity of some TEs has been recent or may have been ongoing . Active TEs must be tightly controlled in germ cells . Using small RNA-seq data from the adult testis of White Leghorn ( Li et al . , 2013 ) , we detected abundant TE piRNAs , which accounted for 7 . 8% of total piRNAs , and exhibited a size range peaking at 24–25 nt . These small RNAs were resistant to oxidation ( Figure 2—figure supplement 1A , B ) . Oxidation by sodium periodate makes most small RNA species non-accessible for cloning into libraries , but 2´-O-methyl-modified 3´ termini protect piRNAs from oxidation ( Ghildiyal et al . , 2008 ) . Like piRNAs in other species , these TE piRNAs typically began with uracil ( 61 . 6% of species and 66 . 7% of reads , Figure 2—figure supplement 1C ) . Almost equal numbers of piRNAs mapped to sense versus antisense strands ( median ratio of sense to antisense piRNAs was 1 . 2 ) ( Figure 2—figure supplement 1D ) , and there was an adenine bias at the 10th position ( Figure 2—figure supplement 1C ) , indicating that secondary piRNAs are generated ( Brennecke et al . , 2007; Gunawardane et al . , 2007 ) . To test whether the anti-sense TE piRNAs were able to guide the PIWI proteins to cleave TE transcripts , we plotted the distance between the 5´ends of anti-sense piRNAs and the 5´ends of sense piRNAs from TE loci . We detected a significant Z score at a distance of 10 nt , a signature of robust Ping-Pong amplification ( Figure 2—figure supplement 1E ) ( Brennecke et al . , 2007; Gunawardane et al . , 2007 ) . These findings indicate that a piRNA mediated silencing pathway against TEs is active in the chicken germ line . The expression of piRNAs that target each TE family correlated with overall TE expression ( ρ = 0 . 81 , p<2 . 2 × 10−16 , Figure 2A; the two events were significantly associated: Fisher's exact test , p<2 . 2 × 10−16 , Figure 2B ) , although there were exceptions . All the 73 actively expressed TE families were targeted by piRNAs . The presence of this piRNA activity explains why expression of the TEs can be tolerated in White Leghorn . Most inactive TEs ( 108 families ) are not targeted by piRNAs; interestingly , 19 inactive TEs are targeted by piRNAs . Those piRNAs that target inactive TEs exhibit the authentic piRNA length distribution , resistance to oxidation , and first position ( 1st ) U bias ( Figure 2—figure supplement 2A , B ) , but they are less abundant than the piRNAs that target active TEs ( p<2 . 2×10−16 ) ( Figure 2C ) . piRNAs that target active TEs display a robust Ping-Pong amplification with a median Z score of 12 . 2; whereas , the piRNAs that target the inactive TEs do not show significant Ping-Pong amplification ( median Z score of 0 . 41 ) ( Figure 2C; Z-score >3 . 3 corresponds to p<0 . 01 ) , although both sense and antisense TE piRNAs are produced in equal abundance ( Figure 2—figure supplement 1B ) . The lack of a Ping-Pong signature for piRNAs targeting inactive TEs supports the function of Ping-Pong amplification as an adaptive response to TE activation rather than merely a consequence of piRNA production . 10 . 7554/eLife . 24695 . 008Figure 2 . Three groups of TEs based on TE expression and piRNA expression . ( A ) Scatter plots of TE transcript abundance versus TE piRNA abundance . Each filled circle represents a TE family . Here and in Figure 2—figure supplements 1 and 2 , young TE in purple , medium TE in yellow , and old TE in grey . ( B ) 2 × 2 contingency table for Fisher’s exact test to assess the significance of the coincidence of the TE transcript abundance and TE piRNA abundance . The table data correspond to the number of TE families in each category and , in parentheses , the number of ERV families in each category . ( C ) Top , box plots present piRNA abundance per TE family . Bottom , box plots present Ping-Pong amplification score per TE family . ( D ) Histograms of TE ages . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 00810 . 7554/eLife . 24695 . 009Figure 2—figure supplement 1 . piRNA-mediated TE suppression in rooster testes . ( A ) Length distributions of testis small RNAs that map to TE regions . Blue represents sense mapping piRNAs; Red represents anti-sense mapping piRNAs . ( B ) Scatter plots of piRNA abundance in total small RNA library and oxidized small RNA library . Each filled circle represents a TE family . Color identifies young , medium , or old TE as in Figure 2 . ( C ) Sequence logo showing the nucleotide composition of TE piRNA species; Top , sense mapping TE piRNAs; Bottom , anti-sense mapping TE piRNAs . ( D ) Scatter plots of sense piRNA abundance versus anti-sense piRNA abundance . Each filled circle represents a TE family . Color identifies young , medium , or old TE . ( E ) The 5´−5´ overlap between TE piRNAs from opposite strands was analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 00910 . 7554/eLife . 24695 . 010Figure 2—figure supplement 2 . Medium TE piRNAs are authentic piRNAs . ( A ) Length distributions of testis small RNAs that map to young TEs ( left ) and medium TEs ( right ) . ( B ) Sequence logo showing the nucleotide composition of Young TE piRNA species ( left ) and Medium TE piRNA species ( right ) ; Top , sense mapping TE piRNAs; Bottom , anti-sense mapping TE piRNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 010 Since inactive TEs are no longer a threat to the host genome , piRNAs that target them could either represent a remnant from suppression of past threats , or have acquired new functions beyond TE suppression . To distinguish these two possibilities , we grouped TEs based on their expression and based on TE piRNA expression ( Figure 2A , B ) , and compared TE age ( Figure 2D ) . If inactive TEs that were targeted by piRNAs have an intermediate age between active TEs and inactive TEs that were not targeted by piRNAs , the targeting more likely reflects a remnant of prior suppression function; if inactive TEs that were targeted by piRNAs are as old as other inactive TEs , it is more likely that these TEs and piRNAs represent possible new functions . We inferred TE age using organism information available in Repbase ( Jurka et al . , 2005 ) . We found that all active TEs that were targeted by piRNAs are recent invaders of the chicken genome . These young TEs are specific to Gallus gallus or other Gallus genera . ERVs comprised most of these young TEs ( 47 out of 73 families ) , which is consistent with the observations that non-ERV TEs lack recent activity in chickens ( International Chicken Genome Sequencing Consortium , 2004 ) . More than 90% of inactive TEs that were not targeted by piRNAs had invaded the chicken genome before birds and other amniotes diverged . These old TEs included 80 DNA transposons , 15 CR1s , and 4 ERVs . We found that the 19 inactive TEs that were targeted by piRNAs were of medium age , exhibiting invasion times that were distinct from both old and young TEs ( Figure 2D , χ2 , p≤2 . 5×10−9 ) . From these data , we conclude that piRNA expression reflects TE age—young TEs are targeted by abundant piRNAs , while TE inactivation leads to the erosion of piRNA production . Thus , we designate three TE groups: young , medium , and old based on the expression pattern of TEs and TE piRNAs ( Figure 2A , B ) . Our data imply that a rapid turnover of chicken piRNA sequences reflects contemporary TE activity . piRNAs have not previously been shown to suppress any infectious virus—exogenous or endogenous—in vertebrates; yet , we unexpectedly detected abundant piRNAs mapping to ALVE in adult rooster testis ( Figure 3A ) . These ALVE-mapping reads , which peaked at 25 nt were resistant to oxidation , indicating that they were authentic piRNAs . These piRNAs mapped to both strands of ALVE ( Figure 1C ) , spanning env and the 3´ half of pol . The sense piRNAs exhibited a typical 1st U bias , and the antisense piRNAs exhibited a 10th A bias ( Figure 2—figure supplement 2C ) , indicating production of secondary piRNAs . Indeed , robust Ping-Pong amplification signals were detected in ALVE piRNAs ( Figure 3B ) . ALVE piRNAs were produced to an abundance of 188 parts per million reads mapped to the genome ( ppm ) , which was roughly half the abundance of EAV-HP piRNAs ( 359 ppm ) . The EAV family underwent endogenization prior to Gallus speciation , and although its members no longer produce viral particles , they actively transpose and can cause new insertions ( Boyce-Jacino et al . , 1992 ) as reported in Supplementary file 1 . Because small RNAs recognize their targets without complete sequence complementarity ( Bartel , 2009 ) , the roughly 3000 ALVE piRNA species detected can recognize mutated ALVE , which thus might explain why other ALV family members failed to endogenize . The presence of these piRNAs likely improves fitness both by suppressing the mutagenic effects of germ-line activation , and by reducing the numbers of viral copies thereby reducing horizontal transmissions . 10 . 7554/eLife . 24695 . 011Figure 3 . ALVE piRNA acquisition . ( A ) Length distributions of testis small RNAs mapping to ALVE . Blue represents sense mapping piRNAs; Red represents anti-sense mapping piRNAs . ( B ) Sequence logo showing the nucleotide composition of ALVE piRNA species from White Leghorn ( left ) and ALVE species from Red Jungle Fowl ( right ) , top , sense ALVE mapping reads , bottom , anti-sense ALVE mapping reads . ( C ) Analysis of the 5´−5´ overlap between ALVE piRNAs from opposite strands was analyzed . Significance of ten-nucleotide overlap ( ‘Ping-Pong’ ) was determined from Z-score . Z-score >3 . 3 corresponds to p-value<0 . 01 . ( D ) Analysis of the 5´−5´ overlaps between EAV-HP piRNAs from opposite strands . ( E ) Scatter plots comparing mRNA abundance between White Leghorn and Red Jungle Fowl . Each black filled circle represents an mRNA expressed in testis , and each red filled circle represents an mRNA coding for a protein in the piRNA pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 011 ALVE was introduced into the chicken genome following speciation but prior to domestication ( Frisby et al . , 1979 ) . The Red Jungle Fowl genome carries one full length ALVE , ALVE-JFvB ( Weiss and Biggs , 1972 ) , and one truncated copy , ALVE6 ( known as ALVE-JFvA in Red Jungle Fowl and ev6 in White Leghorn ) ( Levin et al . , 1994; Benkel and Rutherford , 2014 ) . Using published RNA-seq data from the testis of Red Jungle Fowl ( Necsulea et al . , 2014 ) , we determined the expression of ALVE to be 17 . 1 fragments per kilobase of transcript per million mapped reads ( fpkm ) , which is roughly one-third of the abundance of EAV-HP ( 49 . 3 fpkm ) . Based on 41 single nucleotide polymorphisms ( SNPs ) that distinguish ALVE-JFevB and ALVE6 , we determined that the two copies were expressed at a 1:5 ratio . Given the level of testicular expression of ALVE , it was surprising that we did not detect robust ALVE piRNAs in Red Jungle Fowl ( Figure 1C ) . These ALVE-mapping reads in Red Jungle Fowl had neither a strong U bias ( Figure 3B ) , nor significant Ping-Pong amplification ( Figure 3C ) . The Ping-Pong analysis method based on the Z-score of piRNA pairs is not affected by sequencing depth ( Zhang et al . , 2011 ) . Therefore , the absence of robust expression of ALVE piRNAs in Red Jungle Fowl indicates that the germ-line endogenization of a new retrovirus is not sufficient to establish piRNA-mediated repression . To determine whether piRNAs are increased generally for all TEs or specifically for ALVE in White Leghorn , we tested the expression of EAV-HP piRNAs in Red Jungle Fowl . EAV-HP piRNAs exhibited robust Ping-Pong amplification in both White Leghorn and Red Jungle Fowl ( Figure 3D ) . Moreover , compared to the non-TE piRNAs , the overall percentage of TE piRNAs did not increase in White Leghorn ( χ2 , p=1 ) . We then compared expression of piRNA pathway genes in White Leghorn and Red Jungle Fowl ( Figure 3E ) . RNA silencing pathway genes , including CIWI , CILI , A-MYB ( Li et al . , 2013 ) , DDX4 ( Kuramochi-Miyagawa et al . , 2010 ) , MAEL ( Soper et al . , 2008 ) , L3MBTL4 ( Fagegaltier et al . , 2016; Sumiyoshi et al . , 2016 ) , MOV10l1 ( Frost et al . , 2010; Zheng et al . , 2010 ) , TDRD1 ( Chen et al . , 2009; Kojima et al . , 2009; Reuter et al . , 2009; Wang et al . , 2009 ) , TDRKH ( TDRD2 ) ( Saxe et al . , 2013 ) , TDRD3 , TDRD5 ( Yabuta et al . , 2011 ) , TDRD7 ( Tanaka et al . , 2011 ) , TDRD9 ( Aravin et al . , 2009; Shoji et al . , 2009 ) , and TDRD12 ( Aravin et al . , 2009; Shoji et al . , 2009 ) , exhibited a median abundance of 164 fpkm in White Leghorn testis , which was not significantly different from expression in Red Jungle Fowl ( median abundance of 167 fpkm , p=0 . 63 ) . This expression of TE piRNAs and piRNA processing genes at approximately the same levels in Red Jungle Fowl and White Leghorn suggests that the activation of piRNAs in White Leghorn is specific to ALVE . The presence of an active piRNA pathway in Red Jungle Fowl indicates that ALVE piRNA expression emerged or was selected subsequent to domestication . The ability of chickens to produce piRNAs against a new ERV provides a rare opportunity to study where new piRNAs are acquired . To identify the genomic source of the ALVE piRNAs , we defined all piRNA-producing loci , so-called piRNA clusters , in White Leghorn . Using our previously developed dynamic programming algorithm ( Li et al . , 2013 ) , in total , we identified 1633 piRNA clusters that accounted for 0 . 88% of the chicken genome , and explained 87 . 3% of total piRNA reads and 81 . 1% of uniquely mapping piRNAs ( Figure 4A ) . These piRNA clusters were distributed on most autosomes and the Z chromosome ( Figure 1D ) . Unlike divergently and uni-directionally transcribed mouse piRNA-producing loci , we observed that chicken piRNAs were produced from both strands of piRNA clusters ( Figure 4B , Figure 4—figure supplement 1A ) as reported previously ( Li et al . , 2013; Chirn et al . , 2015 ) , and were derived from convergently transcribed precursors detected by our RNA-seq data ( Figure 4B , Figure 4—figure supplement 1A ) . Over 70% of clusters ( 1173 out of 1633 ) included uniquely mapping piRNAs transcribed from either strand at a level of greater than 10% of total uniquely mapping piRNAs from that cluster . Based on these findings , we conclude that most chicken piRNA-producing loci are transcribed from both strands , and both transcripts are processed into piRNAs . 10 . 7554/eLife . 24695 . 012Figure 4 . ALVE6 is the primary piRNA-producing locus for viral piRNAs . ( A ) Cumulative distributions for all piRNAs ( Blue ) and for uniquely mapping piRNAs ( Red ) on the 1633 piRNA loci in White Leghorn . ( B ) An example of conserved piRNA-producing loci , Cluster33 , in chicken . Normalized RNA-seq reads of piRNA precursors , piRNAs in White Leghorn , and piRNAs in Red Jungle Fowl . Blue represents Watson-strand piRNAs; Red represents Crick-strand piRNAs . ( C ) Box plots showing abundance of piRNA precursors in 12 chicken tissues . ( D ) Venn diagram showing piRNA clusters defined in Red Jungle Fowl ( White ) and White Leghorn ( Black ) . ( E ) Normalized RNA-seq reads of piRNA precursors in White Leghorn , White Leghorn piRNAs , unique mapping piRNAs , and Red Jungle Fowl small RNA reads ( >23 nt ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 01210 . 7554/eLife . 24695 . 013Figure 4—figure supplement 1 . Divergent transcription of piRNA clusters . ( A ) Scatter plots of Watson-strand piRNA abundance versus Crick-strand piRNA abundance ( left ) ; scatter plots of Watson-strand piRNA precursor abundance versus Crick-strand piRNA precursor abundance ( right ) . Each filled circle represents a conserved piRNA cluster and each open circle represent a divergent piRNA cluster . ( B ) Length distributions of testis small RNAs from conserved piRNA clusters ( left ) and divergent piRNA clusters ( right ) . ( C ) Sequence logo showing the nucleotide composition of piRNA species from conserved piRNA clusters ( left ) and from divergent piRNA clusters ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 01310 . 7554/eLife . 24695 . 014Figure 4—figure supplement 2 . ALVE6 existed in chicken genome prior domestication . ( A ) From top to bottom , the genomic location of Cluster719 , White Leghorn genomic re-sequencing signals mapping to Crick strand and Watson strand , Ref-Seq track showing depletion of annotated gene , RepeatMasker track showing the annotated ALVE region , the position of primers used for genomic PCRs , and genomic PCR sequences . Separation of genomic PCR products is shown on the agarose gel , and the primers used for genomic PCRs are labeled in each lane . The sequences of these PCR products were blasted against Red Jungle Fowl with complete alignment . A red tick-mark represents a base substitution; an orange tick-mark represents an insertion . ( B ) Sequence logo showing the nucleotide composition of ALVE piRNA species from ALVE6 locus ( top ) and from new ALVE insertions ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24695 . 014 We tested tissue-specific expression of piRNA precursors using the RNA-seq data to measure the abundance of piRNA cluster transcripts in testis as well as in 11 other tissues ( both sexes were included ) ( Figure 4C ) . The median abundance of piRNA precursors in the testis was 2 . 6 fpkm , but we were unable to detect these transcripts in other tissues ( median abundance = 0 ) . The expression of two PIWI genes , CILI and CIWI , was also only detected in testis ( Figure 1—figure supplement 1B ) . The detection of piRNA precursors and PIWI mRNAs exclusively in testis is consistent with their role in protecting the germ-line genome . The lack of detection of PIWI mRNAs and piRNA precursors in ovary suggests that chicken piRNA pathways display sexual dimorphism . In mouse and chicken , the synthesis of most piRNA precursors and mRNAs of key piRNA pathway genes is driven by the transcription factor A-MYB ( Li et al . , 2013 ) . We found that A-MYB is also expressed exclusively in the testis , which may explain the transcriptional activation of piRNA pathways in chicken ( Figure 4C and Figure 1—figure supplement 1B ) . Thus , the chicken piRNA pathway is transcriptionally activated predominantly in the testis . Based on our finding that the expression of ALVE piRNAs was acquired recently , we reasoned that the ALVE piRNAs were either derived from new ALVE insertions or activated from pre-existing genomic elements . None of the new ALVE insertions in White Leghorn were found within or near the 1633 identified piRNA-producing loci ( Supplementary file 1 ) . To assess the second possibility , we systematically compared piRNA cluster locations in White Leghorn and Red Jungle Fowl . Using the same parameters in the dynamic programming algorithm , we defined the piRNA clusters in Red Jungle Fowl . Overall , the genomic location of 72% piRNA-producing loci ( 1168 of 1633 ) overlapped between the two breeds , but 468 piRNA clusters are specific to White Leghorn ( the right circle in Figure 4D ) . The piRNAs from divergent piRNA clusters exhibited authentic piRNA length distribution , resistance to oxidation , and 1st U bias ( Figure 4—figure supplement 1B , C ) . In White Leghorn , the conserved piRNA clusters accounted for 77 . 4% of uniquely mapping piRNAs , and the divergent piRNA clusters only accounted for only 3 . 7% of uniquely mapping piRNAs . In Red Jungle Fowl , 82 . 8% of total piRNAs and 74 . 9% of uniquely mapping piRNAs ( Meunier et al . , 2013 ) could be explained by the conserved piRNA clusters ( Figure 4A ) . Thus , piRNAs are predominantly produced from identical genomic locations but with notable divergence between the breeds . One divergent piRNA-producing locus ( cluster 719 ) contains the truncated ALVE provirus ( Figure 4E ) , ALVE6 . Cluster 719 produces abundant piRNAs ( 147 ppm ) in White Leghorn , but in Red Jungle Fowl produces few piRNAs ( Figure 4E ) . ALVE6 has lost its 5´LTR , gag , and half of pol , eliminating its transcriptional promoter ( Tereba , 1981 ) . The gene structure matches the distributions of piRNAs on ALVE , which starts in the middle of the pol gene ( Figure 1C ) . This defective provirus has been associated with ALVE resistance ( Robinson et al . , 1981 ) . The wide distribution of ALVE6 in commercial egg-laying breeds has been believed to reflect selection of nonshedders ( Hayward et al . , 1980; Kuhnlein et al . , 1989; Smith et al . , 1990b ) . ALVE6 is the only known ALVE provirus that is present in both White Leghorn and Red Jungle Fowl ( Levin et al . , 1994 ) . In addition to the resequencing analysis of White Leghorn , we used the longer sequencing reads of Sanger sequencing to confirm that although sequence polymorphisms exist , the genomic structure of the ALVE6 locus surrounding regions was remains the same as the reference locus in Red Jungle Fowl ( Figure 4—figure supplement 2A ) . These results indicate that ALVE6 existed in the chicken genome prior to domestication . Although the ALVE6 locus is defined as a piRNA cluster , it remains possible that ALVE piRNAs in White Leghorn are primarily derived from more recent insertions that occurred during domestication . Each ERV insertion event typically deposits a full-length provirus , as the ERVs are reverse transcribed to double-strand DNAs before their integration into the genome ( Lewinski and Bushman , 2005 ) . Additionally , an intact ALVE site cannot explain why uniquely mapping piRNAs come from the flanking genomic regions of ALVE6 ( Figure 4E ) . Moreover , because ALVE6 has accumulated SNPs that are observed as uniquely mapping reads differentiating ALVE6 from other ALVEs ( Figure 4E ) , among the piRNAs overlapped with the 33 SNPs that discriminate new ALVE insertions from ALVE6 , 73 . 3% were expressed from ALVE6 locus and exhibited a pronounced 1st U bias , indicating that ALVE6 was the primary source for ALVE piRNAs . The piRNAs expressed from the new ALVE insertions exhibited a pronounced 10th A bias ( Figure 4—figure supplement 2B ) , indicating that they represented secondary piRNAs generated during Ping-Pong amplification of the ALVE piRNAs . Finally , the presence of ALVE6 in chicken suppresses the spontaneous activation of intact ALVE copies , and enhanced fitness has been associated only with truncated ALVE and not with full length ALVE provirus ( Smith et al . , 1990a ) . Therefore , we conclude that the ALVE piRNAs are primarily produced from the pre-existing ALVE6 locus . In this study , we found that a truncated ALVE provirus gave rise to the piRNAs that target ALVE in White Leghorn . ALVE6 had been identified as a dominant gene that confers resistance to the horizontal spread of spontaneously expressed ALVE ( Robinson et al . , 1981 ) and to congenital transmission of ALVE ( Smith et al . , 1990a ) . Although ALVE6 is a defective provirus , and is not infectious , it is highly expressed in domestic fowl ( Hayward et al . , 1980 ) . The truncated provirus produces envelope glycoproteins , and it was proposed that products of ALVE6 compete for cellular receptors ( Robinson et al . , 1981 ) , thus , preventing ALVE replication . However , the expression of envelope proteins from ALVE6 only leads to a 3–4 fold reduction in virus penetration , and does not account for robust resistance to ALVE infection in chickens ( Robinson et al . , 1981 ) . More than 30 years ago , before the discovery of piRNAs , Robinson et al speculated that ‘the presence of endogenous virus…would protect the germ line from accumulation of provirus and provirus-associated mutations’ ( Robinson et al . , 1981 ) . This type of ‘immune response’ is also known as viral interference—most hosts are resistant to infection by viruses expressed by their germ-line provirus . Although chicken piRNA mutant is currently not available to estimate the magnitude of restriction rendered by the piRNA pathways , in mouse mutant with disrupted PIWI gene , the transposon expression increased up to 10 fold ( Aravin et al . , 2007 ) . Therefore , our discovery represents a new viral interference mechanism , and provides a critical missing piece to the puzzle: we attribute the ability to protect the chicken germ line , at least partially , to the function of piRNAs produced by truncated ALVEs . The example presented here of piRNAs targeting an infectious virus in vertebrates represents a previously unappreciated function and history of piRNAs . Generally , there is a clear boundary between TEs and viruses . Most TEs are in a long-term co-evolutionary relationship that minimizes deleterious impacts on the host , but most viruses adopt a more destructive lifestyle that often leads to intense conflict with their hosts ( Feschotte and Gilbert , 2012 ) . In fruit flies , as a response to different parasites , anti-viral responses are mediated by endogenous siRNAs , and TE silencing is mediated by piRNAs; the functional division of the two small RNA pathways is clear—piRNAs do not appear to play a role in anti-viral defense ( Goic et al . , 2013 ) . Vertebrate genomes , however , contain ERVs ( Malik et al . , 2000 ) , which blurs the boundary between the TEs and viruses . Most ERVs , due to loss-of-function mutations , have lost the ability to make infectious viral particles . This loss occurs evolutionarily , and their infectious capacity is likely maintained for some period of time after endogenization . In addition to ALV in chickens , infectious ERVs have been reported in mammals , including mouse mammary tumor virus ( MMTV ) ( Moore et al . , 1979 ) , Moloney Murine Leukemia Virus ( M-MuLV ) ( Stoye and Coffin , 1987 ) , Koala retrovirus ( KoRV ) ( Tarlinton et al . , 2008 ) , and Feline ERV-DC ( Anai et al . , 2012 ) . Acting as an essential immune response in germ-line defense , piRNAs would be expected to have evolved before ERVs lost their infectious capacity , in which case piRNAs would have contributed to host defense against the infectious viruses . Our results here expand the function of piRNAs to include resistance to infectious pathogens in vertebrates , and imply that the vertebrate piRNA pathway has evolved under selection pressures both from the mutagenic effects of TE propagation and from the deleterious effects of activation of infectious ERVs . The new piRNAs that we report were produced from an existing genomic element , rather than via ‘trapping’ a new ALVE insertion into a highly expressed piRNA cluster . The birth of new piRNA loci shown here is reminiscent of the origin of new genes from previously noncoding DNA ( Schlötterer , 2015 ) that acquire additional regulatory signals for transcription and acquired a functional ORF . The transcription of ALVE6 in White Leghorn could be activated by an adjacent transcriptional promoter as proposed previously ( Tereba , 1981 ) , or by de novo acquisition of a transcriptional binding motif derived via point mutations . Our detection of ALVE6 expression in Red Jungle Fowl by RNA-seq indicates that transcription alone is not sufficient to become a new piRNA-producing locus . Considering that the genomic structure of ALVE6 is similar between White Leghorn and Red Jungle Fowl , either point mutations or epigenetic changes mark the ALVE6 transcripts for piRNA production in White Leghorn . Although we do not understand the mechanisms that lead to conversion of quiescent genomic regions to emerge as active piRNA producing loci , our work identifies a new mechanism for piRNA acquisition for ERV defense through ‘twisting’ existing elements . piRNAs are the most recently discovered family of small silencing RNAs , and questions regarding the biogenesis and function of piRNAs remain . For example , a large proportion of non-TE piRNAs mysteriously enable mammalian sperm production ( Reuter et al . , 2011; Lim et al . , 2015 ) . Each of the available model organisms , including fruit fly , zebrafish and mouse , exhibits distinct piRNA features , and provides unique insights into piRNA pathway . Chicken piRNAs exhibit unique hybrid features of piRNAs found in other organisms . For example , the convergent transcription of piRNA-producing loci resembles the dual strands in fruit flies ( Brennecke et al . , 2007; Malone et al . , 2009 ) , but is distinct from that in frog ( Chirn et al . , 2015 ) , zebrafish ( Houwing et al . , 2007 ) , and mice ( Li et al . , 2013 ) . However , unlike piRNAs in fruit flies and zebrafish that derive mainly from TEs , fewer than 10% of chicken piRNAs come from TE regions . The majority of chicken piRNAs come from non-TE regions , similar to piRNAs in adult mouse testis . Thus , the 1633 piRNA-producing loci should provide a valuable resource for the study of chicken piRNAs that will enable us to unify distinct features in model organisms . Chicken breeding is based on quantitative traits . Putative new insertions , ERV activity , and the capacity to produce piRNAs can be potential contributors to the genetic changes that underlie phenotypic selection . We observed that 58 putative new ERVs inserted into protein coding genes in White Leghorn . Some of these were known to be associated with commercial traits; the others were mapped here for the first time . Each insertion is a mutational event , and has the potential for altering the phenotype . These insertions may contribute to the individual variations of chickens with respect to growth rate , egg production , woody meat , response to heat stress , and resistance to pathogens including newcastle disease virus , avian influenza virus , clostridium , campylobacter , and salmonell . All new insertions are derived from young TEs that are controlled by piRNAs encoded within piRNA-producing loci . Considering the intra-species diversity of piRNA producing loci , both repressive and non-repressive alleles may exist . During selective breeding , it is possible that a genomic region responsible for TE piRNA production is segregated from the active TEs , resulting in TE activation in germ line of F1 generation and increased TE insertions in the offspring of F1 . High ALVE levels and increased integrations have been associated with low body weight in chickens ( Ka et al . , 2009a , 2009b ) . Therefore , the identified active TEs and piRNA clusters may provide another angle for the discovery of functional polymorphisms underlying quantitative traits , and may also be used to guide breeding to modulate TE activity . Two observations in our studies indicated the presence of germ-line TE control mechanisms beyond piRNAs . To wit , intact ALVE is present and expressed in Red Jungle Fowl , and upon induction , it can produce infectious viral particles ( Weiss and Biggs , 1972 ) . Despite this , and despite the absence of ALVE piRNAs , variations of genomic copy numbers of ALVE in Red Jungle Fowl have not been reported . It could be that the somatic TE suppression mechanisms , such as histone modification and DNA methylation , are sufficient to protect Red Jungle Fowl from re-integration . Alternatively , activation , when it occurs , may be extremely deleterious , preventing spread to the general population ( Lu and Clark , 2010 ) . A second mechanism is suggested by the observation that no piRNA-producing loci or PIWI genes are expressed in the ovary , a site where ERVs are highly expressed . In mice , another small RNA silencing mechanism mediated by endogenous siRNAs is known to protect the murine oocytes from TE attacks ( Tam et al . , 2008; Watanabe et al . , 2008 ) . As a consequence , piRNA pathways are not essential for female mouse fertility ( Kuramochi-Miyagawa et al . , 2004; Carmell et al . , 2007 ) . Although the activation of siRNAs in oocytes is rodent-specific , suggesting that this piRNA-independent defense may not exist in other mammals ( Flemr et al . , 2013; Rosenkranz et al . , 2015 ) , our studies suggest that sexual dimorphism in piRNA pathway may be a conserved feature . In conclusion , chicken piRNAs have rapidly evolved to protect the germ-line genome from the contemporary threats . The robust Ping-Pong amplification in piRNAs targeting young TEs reflects an ongoing arms race . When TEs become inactive , the piRNAs gradually erode away , as shown by the low abundance of medium TE piRNAs and the death of piRNAs targeting old TEs . A mystery surrounds the means by which new piRNAs are acquired when a retrovirus is endogenized to a new host . The compact chicken genome , which includes a small fraction of TEs ( 10% ) ( International Chicken Genome Sequencing Consortium , 2004 ) , permitted pinpointing the ALVE piRNA-producing locus and tracing its evolutionary history . In chickens , ALVE6 , as a piRNA-producing locus , exhibits ‘on’ and ‘off’ states . Comparative studies among chicken breeds will delineate the molecular events that turn on piRNA production at the ALVE6 locus . The acquisition of piRNA that target a recently invaded ERV , as reported here , represents an opportunity to elucidate the mechanisms by why some transcripts produce piRNAs while some do not . Rooster testes from a 15 months-old White Leghorn of the Cornell Special C strain were used for polysome gradients , ribosomal profiling , RNA-seq , and genomic PCR . The same strain was used to construct the small RNA libraries in our previous studies ( Li et al . , 2013 ) . Testes were flash frozen in liquid nitrogen , and lysed in 1 ml lysis buffer ( 10 mM Tris-HCl , pH 7 . 5 , 5 mM MgCl2 , 100 mM KCl , 1% Triton X-100 , 2 mM DTT , 100 μg/ml cycloheximide , and 1× Protease-Inhibitor Cocktail ) . Lysates were homogenized with a pellet pestle for a total of ten strokes , and incubated at 4°C with inverted rotation for 10 min . The lysates were centrifuged at 1300 g for 10 min at 4°C , the supernatant was recovered , and the absorbance at 260 nm was measured . Five A260 absorbance units were used for polysome gradients and ribosome profiling . Samples were loaded on a 10–50% ( w/v ) linear sucrose gradient ( 20 mM HEPES-KOH , pH 7 . 4 , 5 mM MgCl2 , 100 mM KCl , 2 mM DTT , 100 μg/ml of cycloheximide ) and centrifuged in a SW-40ti rotor at 35 , 000 rpm for 2 hr 40 min at 4°C . Samples were then collected from the top of the gradient using the gradient Fractionation system ( BR-188 , Brandel , Boca Raton , FL , USA ) while monitoring absorbance at 254 nm was measured . Synthetic spike-in RNAs were added to each collected fraction before RNA purification . The collected fractions were incubated at 42°C in 1% SDS and proteinase K ( 200 μg/ml ) for 45 min . After proteinase K treatment , RNAs were extracted with one volume of Acid phenol ( pH 4 . 5 ) /chloroform/isoamyl alcohol ( 25:24:1 ) . The recovered aqueous phase was supplemented with 20 μg glycogen and precipitated with three volumes of 100% ethanol at 4°C for 1 hr . Pellets were washed with 70% ethanol , and RNAs were resuspended in water . Ribosome profiling was performed as described ( Guo et al . , 2010; Ingolia et al . , 2012; Ricci et al . , 2014; Cenik et al . , 2015 ) with the following modifications: Cleared testis lysates were incubated with 60 units of RNase T1 ( Fermentas , Waltham , MA , USA ) and 100 ng of RNase A ( Ambion , Waltham , MA , USA ) per A260 unit for 30 min at room temperature . Samples were loaded on a 10–50% ( w/v ) linear sucrose gradient , and after centrifuged , the fractions corresponding to 80S monosomes were recovered . Ribosome profiling Illumina-compatible sequencing libraries were prepared as follows ( Figure 1—figure supplement 3A ) : ( i ) the RPFs were resolved on a 15% acrylamide ( 19:1 ) 8 M urea denaturing gel for 1 hr 30 min at 35 W , and fragments ranging from 26 nt to 35 nt were size-selected from the gel; ( ii ) size-selected RNAs were extracted from the gel slice by electro elution using GeBAflex tubes ( Gerad Biotech , Oxford , OH , USA ) , and the rRNAs were removed by Ribo-Zero Gold ( Epicentre Biotechnologies , Madison , WI , USA ) ; ( iii ) the 3´ ends of recovered RNAs were dephosphorylated by T4 PNK ( New England BioLabs , Ipswich , MA , USA ) in MES buffer ( 100 mM MES-NaOH pH 5 . 5 , 10 mM MgCl2 , 10 mM β-mercaptoethanol and 300 mM NaCl ) at 37°C for 3 hr , followed by Alkaline Phosphatase ( New England BioLabs ) treatment at 37°C for 1 hr; ( iv ) dephosphorylated RNAs were used in our small RNA library construction protocol with an additional step of 5´ end phosphorylation by T4 PNK ( New England BioLabs ) using the PNK buffer with 1 mM ATP at 37°C for 1 hr before 5´ ligation . Strand-specific RNA-seq libraries were constructed following the TruSeq RNA sample preparation protocol as previously described ( Li et al . , 2013 ) . rRNAs were depleted from total RNAs by Ribo-Zero Gold ( Epicentre Biotechnologies , Madison , WI , USA ) . The library was sequenced using the paired-end 2 × 50 nt platform on a HiSeq 2000 . Extracted RNAs were treated with Turbo DNase ( Thermo Fisher , Waltham , MA , USA ) for 20 mins at 37°C and then size-selected to isolate RNA ≥200 nt ( DNA Clean and Concentrator−5 , ZYMO RESEARCH , USA ) before reverse transcription by SuperScript III ( Life Technologies , Carlsbad , CA , USA ) at 50°C . Quantitative PCR ( qPCR ) was performed using the ABI Real-Time PCR Detection System with SYBR Green qPCR Master Mix ( Bimake , Houston , TX , USA ) . Data were analyzed using DART-PCR ( Peirson et al . , 2003 ) . Spike-in RNA was used to normalize RNAs in different fractions . Supplementary file 1 lists the qPCR primers . PIWI protein sequences were obtained from the Ensembl genome browser ( SCR_013367 ) . Multiple sequence alignment and neighbour-joining clustering were performed with clustalw 2 . 0 . 12 ( Thompson et al . , 1994 ) . The R package ape ( Paradis et al . , 2004 ) was used to create the phylogenetic tree . We used 200 chicken TE families that are defined in both Repbase ( Jurka et al . , 2005 ) and RepeatMasker ( Smit et al . , 2016 , SCR_012954 ) . We downloaded the 233 Gallus gallus and ancestral ( shared ) repeats from Repbase , and first removed the 46 families containing tRNAs , rRNAs , and snRNAs . Because Repbase and RepeatMasker sometimes name TEs differently , we submitted the Repbase repeat sequences to CENSOR ( Kohany et al . , 2006 ) or to blast to identify the corresponding RepeatMasker name . In comparing the TE annotation between RepeatMasker and Repbase , we found that 9 Repbase repeats appeared to be truncations of existing repeats . For example , CAM1_GG appeared to be an incomplete sequence of CR1-C4 . Based on the latest chicken genome assembly ( Gallus_gallus-5 . 0 ) , we further removed 12 Repbase repeats did not have corresponding genomic copies . We also noticed that some TEs annotated in the genome by RepeatMasker were not included in the Gallus gallus repeats in Repbase . One example is EAV-HP , which is deposited in the archive Repbase21 . 08 , but is classified as being of virus origin rather than chicken origin . We extracted the 34 repeats that are annotated in the current chicken genome by RepeatMasker from the vertebrate archive Repbase . The final total set of 200 TE families and their corresponding names in Repbase and Repeatmasker are listed in Supplementary file 1 . Analyses were performed using piPipes v1 . 4 ( Han et al . , 2014 ) . All data from the small RNA-seq , ribosome profiling , RNA-seq , and genome sequencing were analyzed using the latest chicken genome release galGal5 ( GCA_000002315 . 3 ) . Generally , one mismatch is allowed for genome mapping and three mismatches are allowed for transcriptome mapping . For small RNA analysis , the transcriptome included the 200 TE families and 1633 piRNA clusters . For RNA-seq and ribosome profiling analysis , the transcriptome included mRNAs , lncRNAs , piRNA clusters , and TE families . Supplementary file 1 reports the statistics for the high-throughput sequencing libraries constructed in this study . In small RNA-seq analysis , reads were mapped to ALVE and EAV-HP sequences before being mapped to the genome , and three mismatches were allowed for alignment . The sequences of EAV-HP and ALVE came from NCBI with id: NC_005947 . 1 ( Sacco et al . , 2000 ) and AY013303 ( Johnson and Heneine , 2001 ) . We analyzed previously published testis small RNA libraries from White Leghorn ( GSM1096613 ) ( Li et al . , 2013 ) , and from Red Jungle fowl ( GSM995329 ) ( Meunier et al . , 2013 ) . Small RNA species with characteristic piRNA length ( >23 nt ) were defined as piRNAs ( Ghildiyal et al . , 2008 ) . The small RNA libraries from White Leghorn and Red Jungle Fowl were normalized to the sum of all piRNA reads . Oxidized samples were calibrated to the corresponding total small RNA library via the abundance of shared piRNA species . The piRNA abundance per TE or per piRNA cluster is reported either as parts per million reads mapped to the genome ( ppm ) or reads per kilobase pair per million reads mapped to the genome ( rpkm ) using a pseudo count of 0 . 01 . The pair-end total RNA-seq reads were aligned to the genome using TopHat 2 . 0 . 12 ( Trapnell et al . , 2009 , SCR_013035 ) . Reads were mapped using the ‘-g 100’ flag . The direct transcriptome mapping results were quantified using eXpress ( Roberts and Pachter , 2013 , SCR_006873 ) . The advantage of eXpress lies in the Expectation–Maximization algorithm to apportion multimapping reads , reporting the estimated numbers of fragments in each transcript ( Dempster et al . , 1977 ) . The eXpress results are normalized by the gene compatible reads calculated by Cufflinks per library; and the fpkm ( fragments per kilobase of transcript per million mapped reads ) value with a pseudo count of 0 . 01 was used for all analyses . We analyzed our RNA-seq library from testis of White Leghorn and the published RNA-seq libraries from different tissues of Red Jungle Fowl ( GSM752557 , GSM752558 , GSM752559 , GSM752560 , GSM752561 , GSM752562 , GSM752563 , GSM752564 , GSM752565 , GSM752566 , GSM752567 , GSM752568 , GSM1064853 , GSM1064854 , GSM1064855 , and GSM1196055 ) ( Brawand et al . , 2011; Necsulea et al . , 2014 ) . Ribosome profiling analysis was done according to the modified small RNA pipeline procedure , but including the junction mapping reads . Ribosome protected fragments ( RPFs ) 26–32 nt long were selected for further analysis . The RPF abundance per TE or per piRNA cluster was quantified by eXpress , and reported as reads per kilobase pair per million reads mapped to the genome ( rpkm ) using a pseudo count of 0 . 01 . The pair-end genome sequencing reads were aligned to the reference genome using BWA-aln ( -R 1000 ) ( Li and Durbin , 2009 , SCR_010910 ) . We analyzed the previously published genome resequencing libraries from White Leghorn ( SRX1121834 , SRX1121835 , SRX1121836 ) ( Oh et al . , 2016 ) and we combined the three replicates to increase detection sensitivity . The new transposition events were analyzed by TEMP ( Khurana et al . , 2011; Zhuang et al . , 2014 , SCR_001788 ) . The insertions that are supported by reads at both sides are listed in Supplementary file 1 . Statistical analyses were performed in R 3 . 0 . 2 ( Team , 2014 , SCR_001905 ) . The significance of the differences was calculated by Wilcoxon rank sum test except as indicated in the text . Ping-Pong amplification was analyzed by the 5′–5′ overlap between piRNA pairs from opposite genomic strands ( Li et al . , 2009 ) . Overlap scores for each overlapping pair were the product of the number of reads of each of the piRNAs from opposite strands . The overall score for each overlap extension ( 1–30 ) was the sum of all such products for all chromosomes . Heterogeneity at the 3′ ends of small RNAs was neglected . The Z-score for a 10 bp overlap was calculated using the scores of overlaps from 1–9 and 11–30 as background . Nucleotide periodicity was computed as described ( Pelechano et al . , 2015 ) with modifications . We first aligned the RPFs to each other using 5′–5′ overlap analysis from the same transcript , and reported the distance spectrum . An annotated ORF is not a prerequisite for this analysis . The distance spectrum of RPFs from mRNAs already showed a 3-nt periodicity pattern . We then transformed the distance spectrum using the ‘periodogram’ function of the GeneCycle package ( Wichert et al . , 2004 ) with the ‘clone’ method . The relative spectral density was calculated by normalizing to the value at the third position . We used the same dynamic programming algorithm that we developed previously ( Li et al . , 2013 ) to identify genomic regions with the highest piRNA density . The oxidized small RNA reads ( >23 nt ) ( SRR772069 ) were used to define the clusters in White Leghorn , and the small RNA reads ( >23 nt ) ( SRR553601 ) were used to define the clusters in Red Jungle Fowl . We assumed that piRNA clusters comprise at most 5% of the chicken genome . We first split the genome into one kbp non-overlapping windows , and computed piRNA abundance for each window . The mean of the top 5% of windows was used as the penalty score for the dynamic programming algorithm . The algorithm computes the cumulative piRNA abundance score as a function of the window index along each chromosome . The score at a window is the sum of: the score in the previous window , plus the piRNA abundance in the current window , minus the penalty score; negative scores were reset to 0 . The maximum score points to the largest piRNA cluster . We extracted the largest piRNA cluster , recomputed the scores at the corresponding windows , and searched for the next cluster . This process was continued iteratively until the scores for all windows were zero . The boundaries of each cluster were further refined by including those base pairs for which piRNA abundance exceeded the mean piRNA abundance of the top 5% windows . We required a piRNA cluster to have at least one unique mapping read . The coordinates of all 1633 piRNA-producing loci of White Leghorn and whether they are conserved in Red Jungle Fowl are reported in Supplementary file 1 . We used Cufflinks v2 . 2 . 1 ( Trapnell et al . , 2012 , SCR_014597 ) with parameters of ‘-u -j 0 . 2 --min-frags-per-transfrag 40 --overlap-radius 100’ to assemble transcripts using the strand specific pair-end RNA-seq data from adult testis of White Leghorn ( Supplementary file 1 ) . We assembled 59 , 614 transcripts . Using the TransDecoder/3 . 3 . 0 ( Haas et al . , 2013 ) with the BlastP ( retain ORFs with homology to known proteins ) and Pfam search ( identify common protein domains ) , we further identified the candidate coding regions of 49 , 962 mRNA transcripts . We performed our transcriptome analysis on 9505 mRNAs which had an abundance of at least 10 fpkm in testis: among these mRNAs , 9287 were reported in the latest release of RefSeq ( GCF_000002315 . 4 , SCR_003496 ) , and 218 were putative novel mRNAs . For lncRNAs , we first selected the 13 , 103 assembled transcripts that were reported to be lncRNA by RefSeq . Among the 724 lncRNAs with an abundance above 10 fpkm in testis , 218 of these transcripts had ORFs detected by TransDecoder . Following removal these putative false lncRNA , we performed transcriptome analysis on the remaining 502 lncRNAs . We created a covariance model ( CƒM ) using Infernal v1 . 1rc1 as previously described ( Nawrocki and Eddy , 2013 , SCR_011809 ) . Briefly , we built a CM based on the human mascRNA and menRNA alignment using the cmbuild program , which was calibrated for E-value reporting with the cmcalibrate program . We then searched the chicken genome for high scoring hits with the cmsearch program . Default Infernal v1 . 1rc1 parameters were used for all steps ( cmbuild , cmcalibrate , and cmsearch programs ) . Only one significant hit with an E value below 0 . 01 was identified by the CM model . This tRNA-like element is from the minus strand of chrUn_AADN03016580:1547–1491 with the E value of 3 . 8 × 10−9 . Manual inspection of its upstream sequence revealed a MALAT 3´ end like module with two T-rich motifs: TTTTCTTTT and TTTTGCTTTT , and one polyA-like moiety: AAAAAAAGCAAAA . This contig contains 6473 bp and does not harbor TEs . While ESTs mapped of this tRNA-line element , hundreds ESTs mapped to sites spanning the entire upstream region ( chrUn_AADN03016580:1492–6473 ) , suggesting that the promoter of this gene lies outside of this contig . Despite the lack of syntenic information , it has been shown to be a human MALAT1 homolog in chicken . The evolution of MALAT1 lncRNA and its 3´end module is reported in a manuscript under review ( Zhang et al . , 2017 ) . All sequence data reported here are available through the NCBI Gene Expression Omnibus under the accession number GSE93559 .	Viruses called retroviruses can infect animal cells and merge their genetic information with those of the animal causing damage to the animal’s genetic blueprints . Once retroviruses are integrated into a cell they can sometimes get passed down through the generations over the centuries . Almost half of the human genetic code , for example , is made from ancient retroviruses and other foreign sequences . Over time many of these ancient viruses lost the ability to infect other cells and became trapped within cells but they can still jump out and damage the animal’s genetic code under certain circumstances . These trapped foreign sequences are called transposable elements . Animal cells produce molecules called piRNAs to shut down transposable elements . Most piRNAs are produced from genetic information that originally came from integrated retroviruses and that has been hijacked to defend the cell , a similar strategy as Crisper system in bacteria . Domestic chickens produce piRNAs against a virus called avian leukosis virus ( or ALV for short ) – which commonly infects domestic fowl . The virus also infected the wild ancestors of chickens , known as red jungle fowl , but these birds do not produce piRNAs . This provides an ideal setting to study the evolution of piRNAs in an animal that is not too distantly related to humans ( chickens and humans both have backbones , and are therefore both warm-blooded vertebrates ) . Sun et al . examined cells from the testicles of domestic chickens and red jungle fowl as an example of the role of piRNAs in protecting genetic information in vertebrates . The investigation revealed that piRNAs against all previously trapped viruses in the chicken’s genetic code are produced in chickens to stop them from causing more damage . Sun et al . also observed the creation of piRNAs in chickens in response to ALV that had not yet become trapped in the chicken’s genetic code . Importantly , the piRNAs could control these retroviruses while they were still infectious . The experiments also revealed that piRNAs against ALV are produced from a single copy of ALV that is found in both domestic and wild chickens . The results showed that cells can produce new piRNAs using these pre-existing viral copies within their own genetics . This illustrates that production of piRNA from existing genetic material can be activated in response to certain cues . Further work will seek to discover how existing genetic information becomes a source of piRNAs . In the United States , 8 billion domestic chickens are consumed each year , and a better understanding of how these birds defend themselves against viral infections could increase the productivity of the poultry industry around the world . Moreover , because other viruses trapped in the chicken’s genetic code are related to similar viruses in humans , future discoveries made in this area could help to guide research that will benefit human health as well .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2017	Domestic chickens activate a piRNA defense against avian leukosis virus
Active zone proteins cluster synaptic vesicles at presynaptic terminals and coordinate their release . In forward genetic screens , we isolated a novel Caenorhabditis elegans active zone gene , clarinet ( cla-1 ) . cla-1 mutants exhibit defects in synaptic vesicle clustering , active zone structure and synapse number . As a result , they have reduced spontaneous vesicle release and increased synaptic depression . cla-1 mutants show defects in vesicle distribution near the presynaptic dense projection , with fewer undocked vesicles contacting the dense projection and more docked vesicles at the plasma membrane . cla-1 encodes three isoforms containing common C-terminal PDZ and C2 domains with homology to vertebrate active zone proteins Piccolo and RIM . The C-termini of all isoforms localize to the active zone . Specific loss of the ~9000 amino acid long isoform results in vesicle clustering defects and increased synaptic depression . Our data indicate that specific isoforms of clarinet serve distinct functions , regulating synapse development , vesicle clustering and release . The coordinated and precise release of synaptic vesicles from presynaptic compartments underlies neuronal communication and brain function . This is achieved through the concerted action of conserved proteins that make up the cytomatrix at the active zone , a protein dense region within the presynaptic bouton that is surrounded by synaptic vesicles . Active zone proteins regulate neurotransmission by recruiting synaptic vesicles to the plasma membrane , positioning calcium channels adjacent to the site of exocytosis , and priming synaptic vesicles for calcium-dependent release . In vertebrates , the main active zone proteins that coordinate synaptic vesicle release are Liprin-α , RIM , RIM-BP , Elks and Munc-13 ( Südhof , 2012; Ackermann et al . , 2015 ) . Two additional proteins , Bassoon and Piccolo , serve to cluster synaptic vesicles near the active zone ( Cases-Langhoff et al . , 1996; Langnaese et al . , 1996; Mukherjee et al . , 2010 ) . Although the core components of the active zone are conserved between vertebrates and invertebrates , Bassoon and Piccolo have long been considered exclusive to vertebrates . While the N-terminus of Drosophila Bruchpilot ( BRP ) contains significant sequence homology to vertebrate ELKS ( Wagh et al . , 2006; Kittel et al . , 2006 ) , like Bassoon and Piccolo it also has a large C-terminal domain rich in coiled-coil structures that is thought to function in tethering synaptic vesicles ( Matkovic et al . , 2013 ) . Recently , Drosophila Fife , a protein that contains ZnF , PDZ and C2 domains , was discovered based on sequence homology to the PDZ domain of vertebrate Piccolo , and shown to be an active zone protein ( Bruckner et al . , 2012 ) . Fife binds to and functionally interacts with Rim to dock synaptic vesicles and increase probability of release ( Bruckner et al . , 2017 ) . No clear homologs of Piccolo , Bassoon , Fife , or of the coiled-coil domain of BRP have been identified for C . elegans . We performed forward genetic screens in C . elegans for proteins required for synaptic vesicle clustering , and identified clarinet ( cla-1 ) . CLA-1 is required for normal synapse number and cla-1 null mutants exhibit reduced spontaneous synaptic vesicle release . Cla-1 mutants have a smaller dense projection and display defects in the clustering of the active zone protein SYD-2/Liprin-α . They exhibit a dramatic reduction in the number of synaptic vesicles contacting the dense projection , and increased synaptic depression . The cla-1 gene encodes three main isoforms ( CLA-1L , CLA-1M and CLA-1S ) containing PDZ and C2 domains with sequence homology to vertebrate Piccolo and RIM . While all three isoforms share a C-terminal region that localizes to the active zone , their genetic requirement in synapse function and development differ: the N-terminus of CLA-1L is specifically required for synaptic vesicle clustering and proper synaptic function during repeated stimulations , whereas the shorter isoforms or C-terminus are required for active zone assembly and proper synapse number . Together our findings indicate that cla-1 encodes novel active zone proteins that are required for proper synapse development , active zone structure and synaptic vesicle clustering , and thus play a role in synaptic function during prolonged activation . We performed unbiased forward genetic screens to identify molecules required for the localization of synaptic vesicle proteins in the serotonergic NSM neuron of the nematode C . elegans ( Figure 1A–C ) . From this screen , we identified allele ola104 , which displayed a diffuse distribution of the synaptic vesicle protein VMAT/CAT-1 as compared to wild type controls ( Figure 1—figure supplement 1A–D ) . In ola104 mutants , reduced intensity of synaptic puncta was accompanied by an increase in the extrasynaptic signal , suggestive of a defect in synaptic vesicle clustering at the synapse . Using single-nucleotide polymorphism mapping , we identified ola104 as a missense mutation in cla-1 ( Figure 1—figure supplement 1E ) . An independent allele , cla-1 ( ok560 ) , phenocopied and failed to complement ola104 ( Figure 1C–F and Figure 1—figure supplement 1F ) . cla-1 is predicted to encode six isoforms of different lengths ( Figure 1G ) . Based on the length of the proteins , we classified them into three categories: CLA-1L ( long ) including CLA-1a and b; CLA-1M ( medium ) including CLA-1c and d; CLA-1S ( short ) including CLA-1e and f ( Figure 1G ) . Distinct alleles affect different isoforms . cla-1 ( ok560 ) results in a deletion of the promoter and part of the coding region of cla-1L , and will be referred to henceforth as cla-1 ( L ) . cla-1 ( wy1048 ) , an allele we generated using CRISPR , eliminates most of cla-1S and M , including the PDZ and C2 domains . Because these domains are shared by all isoforms , this deletion is likely a null and the allele will henceforth be referred to as cla-1 ( S/M/L ) . Importantly , in cla-1 ( L ) deletion mutants , the shorter isoforms are still expressed , as evidenced by RT-PCR to the C-terminal PDZ domain ( Figure 1—figure supplement 1H ) . Synaptic vesicle clustering was examined in five alleles affecting different isoforms ( Figure 1G ) , and all alleles examined showed defects in synaptic vesicle clustering in NSM ( Figure 1H ) . Since the long-isoform-specific allele cla-1 ( L ) exhibited as dramatic a defect as the null allele , we hypothesize that CLA-1L may thus be specifically required for proper clustering of vesicles at the synapse . CLA-1L is composed of approximately 9000 amino acids and contains an extended repetitive region of about 4000 amino acids ( Figure 1G ) . The 12 kb cDNA sequence encoding the repetitive region is comprised of tandem repeats , with a 282 bp repeat unit ( Figure 1—figure supplement 1G ) . The secondary structure of the repetitive region is predicted to consist of random coils interlaced with alpha helices . CLA-1M is made up of ~3000 amino acids , whereas CLA-1S is ~1000 amino acids long . The common C-terminal domain for all three isoforms includes PDZ and C2 domains that are conserved with the mammalian active zone proteins Piccolo and RIM ( Figure 1I ) . Other than the PDZ and C2 domains , we did not identify other sequence similarities between the CLA-1 isoforms and vertebrate sequences . Based on a phylogenetic analysis using the PDZ domain sequences , we found that the cla-1 PDZ domain is most similar to that of RIM , but constitutes a distinct clade ( Figure 1I ) . This result , along with the lack of sequence homology between the rest of the CLA-1 protein ( other than the C2 domains ) and any known active zone proteins , suggests that cla-1 encodes a novel active zone protein . Its role in synaptic vesicle clustering suggested that it may be functionally homologous to Piccolo , Bassoon and Fife , and hence was given the name Clarinet ( CLA-1 ) to reflect its large size . To determine the expression pattern of CLA-1 isoforms , we created GFP reporters under the cla-1 promoters ( 2 kb fragments upstream of the L , M and S isoforms ) . We found that each isoform is expressed broadly within the nervous system , as evidenced by a high degree of colocalization with an mCherry reporter under the pan-neuronal rab-3 promoter ( Figure 1—figure supplement 2A–C ) . CLA-1S was expressed broadly throughout the nervous system , while CLA-1M and L were expressed in a subset of neurons . To probe the subcellular localization of CLA-1L , we inserted GFP at the N-terminus of the endogenous cla-1 locus via CRISPR ( Figure 2—figure supplement 1A; Dickinson et al . , 2015 ) . Using this strain , we determined that GFP::CLA-1L ( homozygous endogenous ) localizes to synapses at the developmental period in which the embryonic nervous system begins to form ( three-fold stage: Figure 1—figure supplement 2D , E ) . CLA-1L localized in a pattern reminiscent of synaptic vesicle marker RAB-3 . When we expressed mCherry::rab-3 cDNA under the NSM-specific promoter in the CRISPR strain , CLA-1L colocalized with RAB-3 in NSM ( Figure 2A–C ) , indicating that CLA1L localizes to synapses , at or near synaptic vesicle clusters . To determine whether CLA-1L regulates synaptic vesicle clustering cell-autonomously in NSM , we manipulated its expression in specific neurons using CRISPR-based strategies . Briefly , if CLA-1L acts cell-autonomously in NSM , cell-specific knockouts of CLA-1 should result in a cell-specific synaptic vesicle mutant phenotype , even in the context of all other cells expressing wild type CLA-1L . Conversely , in the context of all other cells lacking CLA-1L , cell-specific expression of wild type CLA-1L should result in cell-specific rescue of the synaptic vesicle phenotype . To achieve cell-specific knockouts of CLA-1L , we created transgenic strains with loxP sites inserted within the introns flanking exon 3 and exon 13 of cla-1L ( Figure 1H and Figure 2—figure supplement 1B ) . Insertion of loxP sites did not affect synaptic vesicle clustering in NSM , as predicted ( Figure 2E ) . However , cell-specific expression of Cre in NSM , which leads to NSM-specific deletion of CLA-1L , resulted in the cla-1L mutant phenotype in NSM . Namely , we observed a diffuse distribution of synaptic vesicle proteins in NSM ( Figure 2F and G ) . These findings indicate that CLA-1L is required in NSM for synaptic vesicle clustering and are consistent with it acting cell-autonomously in NSM . To examine whether cell-specific expression of CLA-1L is sufficient to mediate synaptic vesicle clustering in cla-1L null mutant animals , we created a conditional cla-1L-expressing strain . We inserted a GFP followed by a transcriptional terminator before the start codon of cla-1L ( Figure 2—figure supplement 1C ) . This construct drives GFP expression off the endogenous CLA-1L promoter , preventing the expression of the endogenous CLA-1L gene . In these animals , synaptic vesicle clustering was disrupted ( Figure 2I , arrow ) and GFP was observed throughout the nervous system , as predicated , and similar to transcriptional fusion transgenes previously examined ( Figure 2H and Figure 1—figure supplement 2A ) . Cell-specific expression of Cre in NSM removes the transcriptional terminator and transforms it into an in-frame , functional translational fusion of the CLA-1L gene product ( Figure 2—figure supplement 1C ) . In those animals , the resulting GFP::CLA-1L localized in a synaptic pattern in the NSM process ( Figure 2M , arrow ) , colocalized with the synaptic vesicle marker RAB-3 ( Figure 2K–M ) , and rescued the synaptic vesicle phenotype in NSM ( Figure 2L , arrow , and N ) . Our findings indicate that CLA-1L is required cell-autonomously in the NSM neuron , where it is both necessary and sufficient to mediate synaptic vesicle clustering . Given the broad expression pattern of cla-1 in the nervous system ( Figure 1—figure supplement 2A–C ) , we sought to determine whether CLA-1L functions to cluster synaptic vesicles in neurons other than NSM . We found that cla-1 ( L ) mutants exhibited diffuse synaptic vesicle patterns in the AIY interneuron ( Figure 3A–D ) and the PVD mechanosensory neuron ( Figure 3—figure supplement 1A–C ) , but not the GABAergic or cholinergic motor neurons that innervate body wall muscles ( Figure 3E–I and L; Figure 3—figure supplement 1D–F ) , ( although CLA-1L is expressed in at least a subset of these motor neurons; Figure 1—figure supplement 2A ) . These data demonstrate that CLA-1L is required for synaptic vesicle clustering at specific synapses in C . elegans , indicating that the molecular mechanisms for vesicle clustering may be cell ( or synapse ) specific . cla-1 ( S/M/L ) showed a similar phenotype to cla-1 ( L ) in NSM ( Figure 1H ) . Although cla-1 ( S/M/L ) did not induce a diffuse synaptic vesicle phenotype in motor neurons either ( Figure 3H and I ) , the number of synapses in these neurons was significantly reduced as compared to WT or to cla-1 ( L ) mutants ( Figure 3J ) . To more carefully quantify this effect , we examined the synaptic vesicle marker RAB-3 in a single cholinergic motor neuron , DA9 ( Figure 3K ) . Consistent with our previous observations , we observed that cla-1 ( S/M/L ) mutants have reduced numbers of RAB-3 puncta , suggesting a reduction in the number of synapses ( Figure 3L and M ) . This is in contrast to mutants for the most closely related synaptic gene unc-10/RIM , which did not show a decrease in synapse number or an enhancement of cla-1 ( Figure 3L and M ) . We note that while DA9 motor neurons also display a reduction in synapse number in cla-1 ( L ) mutants , the expressivity of this phenotype was more variable than that of the null allele ( Figure 3L and M ) . Taken together , given the distinct synaptic phenotypes observed in different neurons , our results suggest that cla-1 functions at specific synapses to regulate different aspects of synaptic development . To determine the subsynaptic localization of CLA-1 , we tagged the CLA-1S cDNA with either N- or C-terminal GFP and co-expressed it under a DA9 cell-specific promoter along with the synaptic vesicle protein RAB-3 ( Figure 4A and data not shown ) . N and C-terminal CLA-1S GFP fusion constructs were indistinguishable and showed specific punctate localization at the ventral tip of the presynaptic varicosity , where active zones ( or their ultrastructural correlates , dense projections ) are known to be located from electron microscopy studies ( Stigloher et al . , 2011 ) . Coexpression of CLA-1S with ELKS-1 ( Figure 4B ) or with the calcium channel UNC-2 ( Figure 4C ) led to near complete colocalization of CLA-1S with these active zone proteins , suggesting that CLA-1S specifically localizes to the active zone . To determine the spatial relationship between CLA-1S and CLA-1L , we simultaneously labeled CLA-1S ( tagged with N-terminal mRuby3 ) and CLA-1L ( endogenously tagged with N-terminal GFP ) and imaged their localization in DA9 neurons ( Figure 4D; see Materials and methods for labeling strategy of endogenous CLA-1L protein ) . As expected , both isoforms were enriched at the synapse . However , unexpectedly , they differed regarding their subcellular localization within the synaptic compartment . While the N-terminally tagged CLA-1S co-localized precisely with other active zone proteins , N-terminally tagged CLA-1L displayed a more diffuse pattern of localization in the synaptic region , away from the active zone . CLA-1L is a large ( ~9000 amino acid ) protein , and its N-terminal domain could lie far from its C-terminus . To examine this , we endogenously tagged the C-terminal region of the cla-1 genomic locus , which would label the C-termini of all CLA-1 isoforms , including CLA-1L . We observed that C-terminally tagged CLA-1 isoforms displayed a similar punctate pattern and precise colocalization with the CLA-1S isoform ( Figure 4E; note that additional puncta in the second panel correspond to synapses in neurons other than DA9 ) . Together our findings suggest that CLA-1S and CLA-1L are anchored at the active zone via their C-terminus , and that the N-terminus of CLA-1L may extend away from the active zone into other subcellular regions of the synaptic bouton . To understand the ultrastructural organization underlying our light-level observations , we conducted serial section electron microscopy ( EM; Figure 5A and Figure 5—figure supplement 1 ) . An average of 130 wild type and 166 cla-1 ( S/M/L ) mutant 40 nm sections were cut and reconstructed from three worms from each genotype ( encompassing 19 wild type and 12 mutant synapses ) . We found that cla-1 mutants had smaller terminal area size ( Figure 5B ) , and fewer total synaptic vesicles ( Figure 5—figure supplement 2A ) as compared to wild type animals . The vesicle density ( vesicle number normalized for terminal size ) was indistinguishable between mutant and wild type animals ( Figure 5—figure supplement 2B ) . The length of the dense projection was reduced in cla-1 mutants ( Figure 5C ) , suggesting a role for this protein in regulating the development of the dense projection . cla-1 mutants also exhibited a reduction in the number of undocked synaptic vesicles contacting the dense projection ( pseudocolored as pink vesicles in Figure 5A; Figure 5D ) , and a change in the distribution of docked vesicles ( Figure 5—figure supplement 2C ) , including an increase in the number of docked vesicles within 100 nm of the dense projection ( Figure 5E ) . Our findings suggest that CLA-1 is necessary for the development and clustering of synaptic vesicles at the dense projection , a region known to be crucial for vesicle release ( Stigloher et al . , 2011 ) . Defects in synaptic vesicle clustering or in the number of synaptic vesicle release sites frequently lead to changes in synaptic transmission ( Zhen and Jin , 1999; Hallam et al . , 2002 ) . Defects in synaptic transmission can be quantitatively measured by resistance to the acetylcholinesterase inhibitor aldicarb , which potentiates the action of secreted acetylcholine ( Ach ) ( Mahoney et al . , 2006 ) . Resistance to aldicarb is thus indicative of a reduction in secretion of ACh from cholinergic NMJs . Both cla-1L ( L ) and cla-1 ( S/M/L ) mutants exhibited resistance to aldicarb , suggesting compromised synaptic transmission ( Figure 6—figure supplement 1A ) . cla-1 ( S/M/L ) animals were more resistant to aldicarb than cla-1L ( L ) ( Figure 6—figure supplement 1A ) , suggesting that while the long isoform plays a role in synaptic transmission , the shorter isoforms and/or the C-terminus might execute additional functions that affect synaptic vesicle release . To determine more precisely how synaptic transmission was perturbed in the cla-1 mutants , we recorded spontaneous and evoked responses in postsynaptic muscle cells using patch clamp electrophysiology . In cla-1 ( S/M/L ) , but not in cla-1 ( L ) mutants , the frequency of spontaneous postsynaptic currents ( ‘minis’ ) was reduced by 46% ( Figure 6A , C ) . Since synapse number is also reduced in these mutants ( Figure 3M ) , the reduction in mini frequency could be partially or wholly attributable to the reduction in synapse number . Mini amplitude was unchanged ( Figure 6D ) , indicating that postsynaptic receptor function was not perturbed . While evoked response to a single presynaptic depolarization was unchanged in cla-1 mutants ( Figure 6B , E ) , subsequent release during a 20 Hz stimulation train was impaired ( Figure 6F and Figure 6—figure supplement 1B ) , mirroring the aldicarb results . An increase in depression upon repeated stimulation indicates a defect in the number of vesicles that can be readily recruited by depolarization , and might be a functional consequence of the reduced number of vesicles contacting the dense projection ( Figure 5D ) . Because cla-1 ( L ) and cla-1 ( S/M/L ) showed equally enhanced depression , we could not detect an additional role for the shorter CLA-1 isoforms . Our findings therefore suggest that the long isoform of CLA-1 might be solely responsible for synaptic vesicle recruitment to sustain release upon repetitive stimulation . Taken together our assays reveal functional consequences to the observed cell biological and ultrastructural phenotypes and suggest a specific role for CLA-1L in synaptic vesicle release in response to repeated depolarizations . In light of the reduction in synapse number and mini frequency , as well as the increase in synaptic depression , we were surprised to see no defect in the response to a single evoked stimulus ( Figure 6E ) . We hypothesized that this might reflect either a compensatory upregulation of vesicle release at the remaining synapses , or a redundancy between CLA-1 and another protein . Vesicle release is regulated by the related active zone protein UNC-10/RIM ( Wang et al . , 1997 ) . To test the genetic relationship of UNC-10/RIM in the context of CLA-1 function and the physiological output of the synapse , we recorded from double mutants of cla-1 and unc-10/rim . We found that cla-1 ( S/M/L ) ;unc-10/rim double mutants showed reduced evoked release in response to a single stimulus when compared to unc-10/rim mutants alone ( Figure 6E ) . Since we did not detect a change in the number of synapses between unc-10;cla-1 double mutants as compared to cla-1 mutants alone ( Figure 3M ) , we interpret the enhanced defect in evoked release in the double mutants to be the result of functional requirement for both proteins at the active zone rather than a synthetic effect due to changes in synapse number . Active zone proteins not only colocalize with each other but also interact genetically in synapse development ( Patel et al . , 2006; Van Vactor and Sigrist , 2017 ) . The scaffold molecule syd-2/Liprin-α and the rhoGAP syd-1/mSYD1A are among the first active zone proteins to arrive at the synapse ( Fouquet et al . , 2009 ) , but the precise mechanisms through which these and other active zone proteins are trafficked to and localized at synapses is still largely unknown . To better understand the genetic relationship of CLA-1 to other active zone proteins and the molecular program that localizes CLA-1 to synapses , we examined CLA-1S localization in other active zone protein mutants as well as in mutants for the synaptic vesicle motor unc-104/Kinesin-3 . We found that CLA-1S was greatly reduced , but not completely absent , in unc-104/Kinesin-3 mutants ( Figure 7A , C ) , consistent with previous studies showing down-regulation of active zone proteins in Drosophila kinesin-3 mutants ( Pack-Chung et al . , 2007; Li et al . , 2017 ) . Strikingly , CLA-1S was completely absent from the axon in mutants for the active zone scaffold protein SYD-2/Liprin-α , and greatly reduced in mutants for syd-1/mSYD1A ( Figure 7A , C ) . syd-2/Liprin-α mutants have a profound defect in synaptic vesicle accumulation ( Zhen and Jin , 1999; Patel et al . , 2006; Stigloher et al . , 2011 ) , which can be assessed by the distribution of RAB-3 puncta in DA9 ( Figure 7—figure supplement 1A ) . Consistent with CLA-1 being downstream of SYD-2 function in synaptic vesicle recruitment , double mutants of cla-1 and syd-2 did not show a detectable enhanced synaptic phenotype , or additional synthetic phenotypes ( Figure 7—figure supplement 1A ) . CLA-1S localization was also tested in several other synaptic mutants , including elks-1 , unc-10/RIM and rimb-1/RIM-BP ( and triple mutants for all three of these genes ) , but we could not detect a requirement for these genes in proper localization of CLA-1S ( data not shown ) . We also examined whether unc-104/Kinesin-3 , syd-1 and syd-2/Liprin-α mutants regulate the localization of endogenous CLA-1L . Since our CRISPR-tagged strain labels CLA-1L in many neurons , we were not able to examine CLA-1L distribution with single-cell resolution and assayed instead localization of these active zone proteins to the synapse-rich regions of the nerve ring . Consistent with our cell-specific analyses using CLA-1S , all three mutants resulted in reduced CLA-1L intensity at the nerve ring ( Figure 7—figure supplement 1B–E ) . Taken together , these results show that CLA-1 localization at synapses is dependent on SYD-2/Liprin-α and SYD-1 , but is independent of other active zone genes such as ELKS-1 and UNC-10/RIM . To determine whether loss of CLA-1 may itself affect active zone composition , we examined the synaptic distribution of endogenously tagged SYD-2/Liprin-α::GFP ( Figure 7D ) . Endogenous SYD-2 puncta in the dorsal nerve cord were dimmer in cla-1 mutants ( Figure 7E ) , and the overall fluorescence intensity was reduced ( Figure 7F ) . To gain cellular specificity , we examined the localization of GFP-tagged SYD-2/Liprin-α expressed in NSM in cla-1 mutants ( Figure 7—figure supplement 1F–I ) . We found that in cla-1 ( S/M/L ) but not cla-1 ( L ) mutants , SYD-2::GFP localization was more diffuse ( Figure 7—figure supplement 1F–I ) . Together these data demonstrate that loss of CLA-1 affects the recruitment or maintenance of SYD-2/Liprin-α at active zones . In this study , we used two deletion alleles to interrogate the function of cla-1 . The cla-1 ( L ) allele specifically deletes the start of the long isoform , but does not affect the short and medium isoforms . The cla-1 ( S/M/L ) allele deletes the PDZ and C2 domain-containing C-terminus shared by all three isoforms . By comparing phenotypes between these two alleles , we were able to assign distinct roles to the N-terminus of the long isoform , versus the common C-terminus , or the short/medium isoforms . Spontaneous synaptic vesicle release as well as inhibitory motor neuron synapse number were impaired in cla-1 ( S/M/L ) , but not in cla-1 ( L ) , suggesting that either the common C-terminus or only the CLA-1S/M isoforms are involved in these processes . However , the cla-1 ( L ) -specific mutant exhibited synaptic vesicle clustering defects in many sensory neurons , and synaptic transmission defects upon repetitive stimulation as well as increased aldicarb resistance . These data suggest that the long CLA-1 isoform is specifically required for synaptic vesicle clustering and functions during periods of sustained release . Importantly , our findings indicate that distinct CLA-1 isoforms might play specific roles to promote synaptic development and function . CLA-1S ( whether it is N- or C-terminally tagged ) colocalizes with active zone proteins . CLA-1L and CLA1S share the same C-terminal PDZ and C2 domains with sequence homology to vertebrate active zone proteins Piccolo and RIM . An endogenous C-terminal tag of all CLA-1 isoforms colocalizes with CLA-1S . These findings suggest that all CLA-1 isoforms may be anchored at the active zone by their C-terminus . The N-terminally tagged CLA-1L still localized to synaptic areas but was not confined to the synaptic subregions occupied by CLA-1S and known active zone proteins . CLA-1L is a large , ~9000 amino acid protein that , if anchored to the active zone area via its C-terminus , could possibly extend away to regions occupied by undocked synaptic vesicles . Because the sub-synaptic localization of the N-terminally tagged CLA-1L differed from that of the C-terminally tagged CLA-1 isoforms , our findings are consistent with a model in which the N- and C-termini of CLA-1L occupy distinct sub-synaptic areas ( Figure 8 ) . This model is analogous to the orientation of Drosophila BRP at the fly neuromuscular junction ( Fouquet et al . , 2009 ) and consistent with models of Piccolo as a protein oriented in a polarized manner and extending ~100 nm from the plasma membrane ( Dani et al . , 2010 ) . Since Clarinet is almost twice the size of Piccolo and exhibits more unstructured regions , it could potentially extend even farther . Our model is also consistent with the genetic and electrophysiological roles we identify for the long CLA-1L isoform in clustering vesicles at sensory synapses and possibly recruiting vesicles for release upon repeated stimulation . C . elegans Liprin-α , Drosophila BRP and Fife and mammalian Piccolo and Bassoon have all been implicated in clustering synaptic vesicles at the active zone ( Stigloher et al . , 2011; Kittelmann et al . , 2013; Hallermann et al . , 2010; Mukherjee et al . , 2010; Bruckner et al . , 2012; Bruckner et al . , 2017 ) . Deleting just the last 17 amino acids of BRP leads to the complete loss of synaptic vesicles adjacent to the T bar , as well as increased synaptic depression ( Hallermann et al . , 2010 ) , suggesting that the inability of BRP to tether synaptic vesicles to the T bar contributes directly to the sustained release defect upon repeated stimulation . Our model would indicate that while N-terminal protein sequence between active zone proteins Clarinet , BRP , Fife and Piccolo/Bassoon varies , they share analogous molecular architecture required to link the synaptic vesicle pool with the active zone to actuate their function at presynaptic sites . The smaller size of the dense projection in cla-1 mutants indicates that this protein is either a component of this presynaptic specialization , or is required for its development . The C . elegans dense projection is thought to organize synaptic vesicles and their release machinery , much like the Drosophila T bar and the ribbon structure in the mammalian visual system . We observe a co-dependency between CLA-1 and the active zone protein SYD-2 in their recruitment to the synapse , consistent with a requirement of these proteins for the assembly of the dense projection . Despite the presence of morphological and structural abnormalities , cla-1 mutants exhibit normal responses to single evoked stimuli . This may be due either to compensatory upregulation of vesicle release at the remaining synapses , or to redundancy with another active zone protein . Mutants for active zone proteins implicated in synaptic vesicle release , such as RIM/UNC-10 and UNC-13 , exhibit a reduction in the number of docked synaptic vesicles at the active zone ( Stigloher et al . , 2011; Weimer et al . , 2006; Gracheva et al . , 2008; Kaeser et al . , 2011; Han et al . , 2011; Wang et al . , 2016; Acuna et al . , 2016 ) . RIM/UNC-10 in particular localizes within 100 nm of the dense projection and is responsible for vesicle docking in this region ( Weimer et al . , 2006; Gracheva et al . , 2008 ) , precisely where cla-1 mutants exhibit an increase in docked vesicles . This increase in morphologically docked vesicles might be the structural correlates of a compensatory upregulation of primed vesicles . Consistent with this model , in the absence of UNC-10/RIM , loss of CLA-1 further reduces evoked responses after a single stimulus , suggesting that UNC-10/RIM could be responsible for a compensatory response in cla-1 mutants . Alternatively , it is also possible that docked synaptic vesicles accumulate in cla-1 mutants due to a cla-1-dependent release defect . BRP , Fife , Rim and Bassoon have all been shown to play a role in calcium channel localization ( Bruckner et al . , 2017; Kaeser et al . , 2011; Han et al . , 2011; Kittel et al . , 2006; Graf et al . , 2012; Frank et al . , 2010 ) , and both Drosophila Fife and mammalian rim mutant phenotypes are consistent with an impairment in the coupling of synaptic vesicles to calcium channels ( Bruckner et al . , 2017; Kaeser et al . , 2011; Han et al . , 2011 ) . Regardless of the cause of the additive phenotype in the unc-10;cla-1 double mutants , our findings indicate a genetic , and functionally significant interaction between CLA-1 and a protein known to function in synaptic vesicle release , UNC-10/RIM . Together these data underscore the functional consequences of loss of CLA-1 at the synapse . How synaptic vesicles are clustered at synapses remains poorly understood . Initial studies suggested that synapsin tethers synaptic vesicles to the actin cytoskeleton ( Bähler et al . , 1990 ) , but more recent evidence calls that model into question ( Pechstein and Shupliakov , 2010; Shupliakov et al . , 2011 ) and suggests that other as yet unidentified proteins may be involved in synaptic vesicle clustering ( Siksou et al . , 2007; Fernández-Busnadiego et al . , 2010; Stavoe and Colón-Ramos , 2012; Stavoe et al . , 2012 ) . Mammalian Piccolo has been shown to play a role in recruiting synaptic vesicles from the reserve pool through interactions with synapsin ( Leal-Ortiz et al . , 2008; Waites et al . , 2011 ) , and to maintain synaptic vesicle clustering at the active zone ( Mukherjee et al . , 2010 ) , while SYD-2 has been shown to cluster vesicles by influencing transport ( Edwards et al . , 2015 ) . Tomosyn has also been shown to regulate synaptic vesicle distribution between the reserve and recycling pools , possibly through interactions with synapsin ( Cazares et al . , 2016 ) . CLA-1L , which extends away from the active zone , may be an important link in understanding how synaptic vesicles are clustered and recruited . Our analyses of synaptic vesicle clustering at various synapses by confocal microscopy indicated that CLA-1L was required to cluster synaptic vesicles at synapses in several different classes of neurons , although not in excitatory or inhibitory motor neurons . Our ultrastructural analysis and functional assays revealed that at motor neuron synapses , CLA-1 is involved in tethering vesicles to the dense projection and CLA-1L itself is implicated in recruiting synaptic vesicles for release upon repeated stimulations . Our findings suggest that although CLA-1L might not display a change in synaptic vesicle clustering by fluorescence measurements in motor neurons , it could still play a role in synaptic vesicle recruitment to the active zone at these synapses . We speculate that CLA-1L may retain the recycling pool of vesicles ( i . e . vesicles recruited upon multiple stimulations ) at the dense projection ( Figure 8 ) . In certain neurons ( including NSM , AIY and PVD ) , this may lead in turn to the retention of the reserve pool of vesicles within the presynaptic bouton . Of all the isoforms , CLA-1L is the most enigmatic due to its large size and structure . Almost half of CLA-1L consists of a repetitive region , which is predicted to be disordered and has no sequence homology to vertebrate proteins . The structure , function , regulation and evolution of the repetitive region pose interesting questions . The distribution of this protein within the synaptic bouton and its function in synaptic vesicle release suggest a novel mechanism for clustering synaptic vesicles , with shared functional homology to vertebrate and Drosophila active zone proteins . The mechanisms uncovered in this study might therefore demonstrate how divergent strategies can be utilized for conserved purposes in organizing the development and function of synapses . Worms were raised on NGM plates at 20°C using OP50 Escherichia coli as a food source . N2 Bristol was used as the wild type reference strain . Hawaii CB4856 strain was used for SNP mapping . The following mutant strains were obtained through the Caenorhabditis Genetics Center: cla-1 ( ok560 ) IV , cla-1 ( gk352 ) IV , cla-1 ( ok937 ) IV , cla-1 ( ok2285 ) IV , unc-104 ( e1265 ) II , syd-2 ( ok217 ) X , syd-2 ( ju37 ) X , syd-1 ( ju82 ) II , unc-10 ( md1117 ) X and zxIs6 [unc-17p::ChR2 ( H134R ) ::YFP + lin-15 ( + ) ] V . nuIs168 [Pmyo-2::gfp + Punc-129::Venus::rab-3] was provided by Jihong Bai ( Fred Hutchinson Cancer Research Center , Seattle , Washington ) . juIs137 [Pflp-13::snb-1::gfp] was provided by Yishi Jin ( UCSD , San Diego , CA ) . kyIs445 [Pdes-2::mCherry::rab-3 + Pdes-2:sad-1::gfp] was provided by Cori Bargmann ( Rockefeller University , New York , NY ) . Other strains used in the study are as follows: olaIs1 [Ptph-1::mCherry + Ptph-1::cat-1::gfp] , olaEx3222 [Ptph-1::mCherry::rab-3]; cla-1 ( ola311 ) IV [GFP::CLA-1L] , olaEx3309 [Ptph-1::mCherry + Ptph-1::cat-1::gfp; Ptph-1::cre]; cla-1 ( ola324 ) IV [floxed cla-1L] , olaEx3289 [Ptph-1::mCherry::rab-3 + Ptph-1::cre]; cla-1 ( ola321 ) IV [GFP^CAS^cla-1L] , olaEx2897 [Pcla-1L::gfp + Prab-3::mCherry] , olaEx2898 [Pcla-1M::gfp + Prab-3::mCherry] , olaEx2924 [Pcla-1S::gfp + Prab-3::mCherry] , olaEx1106 [Ptph-1:: mCherry::rab-3 + Ptph-1::syd-2::gfp] , wyIs45 [Pttx-3::rab3::gfp] , wyIs85 [Pitr-1::GFP::RAB-3] , wyIs574 [Pmig-13::CLA1S::GFP] , wyIs226 [Pmig-13::mCherry::RAB-3] , wyEx8596 [Pmig-13::mRuby3::CLA-1S] , wyEx6368 [Pmig-13::CLA-1S::mCherry + Pmig-13::GFP::ELKS-1] , wyEx9404[Pmig13::FLPase + Pmig13::mRuby3::cla-1];cla-1 ( wy1186 ) IV [C-terminal FRT-stop-FRT GFP] , syd-2 ( wy1074 ) [endogenous N-term GFP] . Expression clones were made in the pSM vector ( Shen and Bargmann , 2003 ) . The plasmids and transgenic strains ( 0 . 5–50 ng/μl ) were generated using standard techniques and coinjected with markers Punc122::GFP ( 15–30 ng/μl ) , Punc122::dsRed ( 15–30 ng/μl ) , Podr-1::RFP ( 100 ng/μl ) or Podr-1::GFP ( 100 ng/μl ) . Worms expressing CAT-1::GFP and cytosolic mCherry in NSM neuron ( olaIs1 ) were mutagenized with ethyl methanesulfonate ( EMS ) as described previously ( Brenner , 1974 ) . The screen was performed as previously described ( Nelson and Colón-Ramos , 2013; Jang et al . , 2016 ) . CAT-1::GFP was diffusely distributed throughout neurites in six mutants , including cla-1 ( ola104 ) . The ola104 allele was mapped to a 2 . 1Mbp region on chromosome IV using SNP mapping coupled with whole-genome sequencing ( WGS ) ( Davis et al . , 2005; Doitsidou et al . , 2010 ) . WGS identified the genetic lesion in ola104 as a missense mutation in cla-1 . ola104/cla-1 ( ok560 ) trans-heterozygotes were examined for complementation . We generated a phylogenic tree to determine how related the CLA-1 PDZ domain was to the other family members ( Figure 1I ) . The PDZ domains of Piccolo/Fife-related proteins were identified by SMART ( Schultz et al . , 1998; Letunic et al . , 2012 ) . T-Coffee ( M-Coffee ) was used for multi-alignment of the sequences ( Notredame , 2010 ) . A rooted phylogenetic tree was determined from aligned sequences by neighbor joining with 100 bootstrap replicates using APE ( Paradis et al . , 2004 ) . PDZ domains of Dishevelled family proteins were used as an outgroup . A circle tree was built using ggtree ( Yu et al . , 2016 ) . RNA from wild type , cla-1 ( S/M/L ) and cla-1 ( L ) worms was prepared using Trizol ( Sigma Aldrich , St . Louis , MO ) . A cDNA library was created by reverse transcription using oligo dTs . PCR amplification was conducted using primers against the C-terminal PDZ domain of cla-1 , as well as against the housekeeping gene tba-1 . Images of fluorescently tagged fusion proteins were captured at room temperature in live C . elegans . Mid-L4 through young adult stage hermaphrodite animals were anesthetized using 10 mM levamisole ( Sigma-Aldrich ) or 50 mM muscimol ( Abcam ) in M9 buffer , mounted on 2–5% agar pads and imaged as follows: Images in Figures 1 , 2 and 3B-Hwere taken using a 60x CFI Plan Apochromat VC , NA 1 . 4 , oil objective ( Nikon ) on an UltraView VoX spinning-disc confocal microscope ( PerkinElmer ) . Images in Figures 4A–C and and 7A were taken using a Zeiss LSM710 confocal microscope ( Carl Zeiss ) with a Plan-Apochromat 63x/1 . 4 NA objective . Images in Figures 3L , 4D–E and and 7D were taken with a Zeiss Axio Observer Z1 microscope equipped with a Plan-Apochromat 63 × 1 . 4 objective and a Yokagawa spinning-disk unit . Maximum-intensity projections were generated using ImageJ ( NIH ) or ZEN 2009 software and used for all the confocal images . Quantification was performed on maximal projections of raw data . Quantification of synaptic vesicle clustering in Figures 1–3 and active zone protein clustering in Figure 7—figure supplement 1 was based on a previous protocol ( Jang et al . , 2016 ) . Briefly , fluorescence values for individual neurites ( ventral neurite for the NSM and PVD neurons , Zone3 for the AIY neuron , and dorsal neurite for DD GABAergic or cholinergic motor neurons ) were obtained through segmented line scans using ImageJ . A sliding window of 2 μm was used to identify all the local fluorescence peak values and trough values for an individual neuron . Synaptic enrichment was then calculated as % ΔF/F as previously described ( Dittman and Kaplan , 2006; Bai et al . , 2010 ) . To measure penetrance , animals were scored as displaying either ‘punctate’ or ‘diffuse’ phenotypes for synaptic vesicles proteins . Percentage of animals displaying diffuse distribution of synaptic vesicle proteins was calculated for each genotype . For each experiment , at least 30 animals were scored for each genotype and at least five independent experiments were performed . The number of synaptic vesicle puncta in DD GABAergic motor neurons was counted by ImageJ with the same settings for all images including threshold , size and circularity . DA9 synapse number in Figure 3 and SYD-2::GFP puncta fluorescence in Figure 7 was quantified using a Matlab ( Mathworks , Natick , MA ) script that counted and measured peaks above threshold from plot profiles of segmented line scans generated in ImageJ . To quantify synaptic fluorescence of CLA-1S or RAB-3 in Figure 7 , total integrated intensity of the line scans was analyzed using an ImageJ plugin . To create cla-1 ( wy1048 ) we chose sgRNAs ~13 kb apart designed to delete most of the M and almost all of the S isoform , including the shared PDZ and C2 domains . sgRNAs were injected at 30 ng/μl along with Cas9 plasmid at 50 ng/μl and F2 worms were screened by PCR . The resulting deletion is flanked by the following sequences: 5’ CCACAACAATCATTCCACCC , 3’ AGGTGTCGGCACACGTCATC . To endogenously tag CLA-1L at the N-terminus , a CRISPR protocol ( Dickinson et al . , 2015 ) was used to create cla-1 ( ola300[gfp:: SEC::cla-1L] ) , in which gfp::SEC ( Self-Excising Cassette ) was inserted before the start codon of cla-1L ( Figure 2—figure supplement 1A ) . SEC consists of a hygromycin resistance gene ( hygR ) , a visible marker [sqt-1 ( d ) ] ) and an inducible Cre recombinase ( Figure 2—figure supplement 1A ) . SEC is flanked by LoxP sites , and heat shock induced Cre expression removed the SEC , leaving GFP fused to CLA-1L in cla-1 ( ola311[gfp::cla-1L] ) ( Figure 2—figure supplement 1A ) . Two methods were used to demonstrate cell autonomy of CLA-1L . In the first method , a CRISPR protocol ( Paix et al . , 2014; Arribere et al . , 2014 ) was used to create cla-1 ( ola324 ) , in which two loxP sites were inserted into two introns of cla-1L ( Figure 1G and Figure 2—figure supplement 1B ) . We used three criteria to ensure that our insertion sites efficiently and specifically target CLA-1L . First , we avoided inserting loxP sites into small introns to prevent any effects on splicing . Second , to ensure that CLA-1M is unaffected after Cre-loxP recombination , the second loxP site was positioned about 4 kb away from the start codon of cla-1M . Third , the sequence flanked by loxP sites is about 16 kb and is close to the start codon of cla-1L . Thus removal of the sequence should result in a CLA-1L null mutation . Cell-specific removal of CLA-1L in NSM was achieved with a plasmid driving the expression of cre cDNA under the NSM-specific tph-1 promoter fragment as described previously ( Jang et al . , 2016; Nelson and Colón-Ramos , 2013 ) . In the second method , we modified a CRISPR protocol ( Dickinson et al . , 2015 ) to create cla-1 ( ola321[gfp:: CAS::cla-1L] ) , in which CAS consists of a hygromycin resistance gene ( hygR ) and a visible marker [sqt-1 ( d ) ] ) ( Figure 2—figure supplement 1C ) . Since CAS contains a transcriptional terminator , this strain is a cla-1L null allele . Since CAS is flanked by loxP sites , Cre-loxp recombination generates functional GFP fused to CLA-1L . Cell-specific rescue in NSM was achieved with a plasmid driving the expression of cre cDNA under the NSM-specific tph-1 promoter fragment . Detailed subcloning information will be provided upon request . A cell-specific CRISPR protocol ( Schwartz and Jorgensen , 2016 ) was used to insert a let-858 3’UTR flanked by FRT sites followed by GFP at the conserved C-terminus of cla-1 . Upon crossing to a strain containing cell-specific FLPase , the endogenous stop site and exogenous 3’UTR are excised , leaving the C-terminal GFP inserted in front of the endogenous 3’UTR . To achieve DA9-specific expression of CLA-1::GFP we used a FLPase driven by the Pmig-13 promoter , which has previously proven to be specific to DA9 within the posterior dorsal cord . However , Pmig-13 seems to express at very low levels in other neurons in this region ( enough to generate excision at the FRT sites ) , as evidenced by the fact that we see CLA-1::GFP puncta outside DA9 driven exogenously expressed CLA-1 ( Figure 4E ) . Animals were assayed for acute exposure to aldicarb ( Mahoney et al . , 2006 ) . Aldicarb ( ULTRA scientific ) was prepared as a stock solution of 200 mM stock in 50% ethanol . Aldicarb sensitivity was measured by transferring 25 animals to plates containing 1 mM aldicarb and then assaying the time course of paralysis . Animals were considered paralyzed once they no longer moved even when prodded with a platinum wire three times on the head and tail . The ratio of animals moving to the total number of animals on the plate was calculated for each time point . All strains used for this assay also contained zxIs6 in the background for consistency with electrophysiology assays . All assays were performed blinded to genotype . Electrophysiological recordings were obtained from the C . elegans neuromuscular junctions of immobilized and dissected adult worms as previously described ( Richmond , 2009 ) . Ventral body wall muscle recordings were acquired in whole-cell voltage-clamp mode ( holding potential , −60 mV ) using an EPC-10 amplifier , digitized at 1 kHz . Evoked responses were obtained using a 2 ms voltage pulse applied to a stimulating electrode positioned on the ventral nerve cord anterior to the recording site . For multiple stimulations , a five pulse train was delivered at 20 Hz . The 5 mM Ca2+ extracellular solution consisted of 150 mM NaCl , 5 mM KCl , 5 mM CaCl2 , 4 mM MgCl2 , 10 mM glucose , 5 mM sucrose , and 15 mM HEPES ( pH 7 . 3 , ~340 mOsm ) . The patch pipette was filled with 120 mM KCl , 20 mM KOH , 4 mM MgCl2 , 5 mM ( N-tris[Hydroxymethyl] methyl-2-aminoethane-sulfonic acid ) , 0 . 25 mM CaCl2 , 4 mM Na2ATP , 36 mM sucrose , and 5 mM EGTA ( pH 7 . 2 , ~315 mOsm ) . Data were obtained using Pulse software ( HEKA . Subsequent analysis and graphing was performed using mini analysis ( Synaptosoft ) , Igor Pro and Prism ( GraphPad ) . Worms underwent high-pressure freeze ( HPF ) fixation as described previously ( Weimer , 2006 ) . Young adult hermaphrodites were placed in specimen chambers filled with Escherichia coli and frozen at −180°C and high pressure ( Leica SPF HPM 100 ) . Samples then underwent freeze substitution ( Reichert AFS , Leica , Oberkochen , Germany ) . Samples were held at −90°C for 107 hr with 0 . 1% tannic acid and 2% OsO4 in anhydrous acetone . The temperature was then increased at 5 °C/h to −20°C , and kept at −20°C for 14 hr , and increased by 10 °C/h to 20°C . After fixation , samples were infiltrated with 50% Epon/acetone for 4 hr , 90% Epon/acetone for 18 hr , and 100% Epon for 5 hr . Finally , samples were embedded in Epon and incubated for 48 hr at 65°C . All specimens were prepared in the same fixation and subsequently blinded for genotype . Ultra thin ( 40 nm ) serial sections were cut using an Ultracut 6 ( Leica ) and collected on formvar-covered , carbon-coated copper grids ( EMS , FCF2010-Cu ) . Post-staining was performed using 2 . 5% aqueous uranyl acetate for 4 min , followed by Reynolds lead citrate for 2 min . Images were obtained on a Jeol JEM-1220 ( Tokyo , Japan ) transmission electron microscope operating at 80 kV . Micrographs were collected using a Gatan digital camera ( Pleasanton , CA ) at a magnification of 100 k . Images were quantified blinded to genotype using NIH ImageJ software and macros provided by the Jorgensen lab . Data were analyzed using MATLAB scripts written by the Jorgensen lab and Ricardo Fleury . Images of the dorsal cord were taken for three animals from each strain . Cholinergic synapses were identified by morphology ( White et al . , 1986 ) . A synapse was defined as a set of serial sections containing a dense projection and two flanking sections without dense projections from either side . Synaptic vesicles were identified as spherical , light gray structures with an average diameter of ~30 nm . To control for inherent variability in the size of synaptic terminals , we measured the density of synaptic vesicles in the terminal by dividing the number of synaptic vesicles by the area of the terminal in micrometers . Terminal area was defined as the average cross-sectional area of every profile containing a dense projection plus two flanking sections . A synaptic vesicle was considered docked if it contacted the plasma membrane . Vesicles that were within 1–4 nm of the plasma membrane that exhibited small tethers to the PM were not scored as docked . The total number of undocked vesicles contacting the dense projection were quantified per profile containing a dense projection . Statistics was determined using students t-test , one-way ANOVA or two-way ANOVA with Tukey’s post-hoc analysis . Error bars were calculated using standard errors of the mean . * signifies p<0 . 05 , p<0 . 01 , p<0 . 001 , ***p<0 . 0001 .	Nerve cells , or neurons , communicate with one another by sending messages across junctions called synapses . When the neuron on one side of a synapse becomes active , calcium ions flood into the cell . This causes the neuron to release signals called neurotransmitters , which activate the cell on the other side of the synapse . Neurons store their neurotransmitter molecules inside packages called vesicles , and keep them clustered close to their release site , a region called the active zone . The active zone contains a number of different proteins . Some of these hold vesicles in position . Others respond to calcium entry by fusing vesicles with the cell membrane . The identity of many of the proteins within the active zone remains unknown , especially those responsible for keeping vesicles clustered nearby . Xuan , Manning et al . therefore introduced mutations at random into the genome of the roundworm C . elegans , and searched for mutant worms with altered patterns of vesicles . Worms with mutations in a previously unknown gene , which they named clarinet , showed abnormal distribution of vesicles within the presynaptic compartment . They also released fewer vesicles compared to non-mutant worms . Further experiments revealed that the clarinet gene encodes three different proteins with varying sizes , all found at the active zone . Using microscopy and electrode recordings , as well as a genetic technique called CRISPR , Xuan , Manning et al . showed that the three forms of clarinet have different roles . The shorter ones contribute to the development of synapses . They help ensure that the active zone forms correctly and that neurons have an appropriate number of synaptic connections . The longest form of clarinet is responsible for clustering vesicles , which allows cells to continue releasing vesicles during bursts of repeated neuronal firing . Problems with synapses contribute to many brain disorders , including autism and intellectual disability . Xuan , Manning et al . hope that an increased understanding of how synapses form , and how they work , will provide insights into these and other conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2017	Clarinet (CLA-1), a novel active zone protein required for synaptic vesicle clustering and release
To maintain their speeds during navigation , insects rely on feedback from their visual and mechanosensory modalities . Although optic flow plays an essential role in speed determination , it is less reliable under conditions of low light or sparse landmarks . Under such conditions , insects rely on feedback from antennal mechanosensors but it is not clear how these inputs combine to elicit flight-related antennal behaviours . We here show that antennal movements of the honeybee , Apis mellifera , are governed by combined visual and antennal mechanosensory inputs . Frontal airflow , as experienced during forward flight , causes antennae to actively move forward as a sigmoidal function of absolute airspeed values . However , corresponding front-to-back optic flow causes antennae to move backward , as a linear function of relative optic flow , opposite the airspeed response . When combined , these inputs maintain antennal position in a state of dynamic equilibrium . When flying in unpredictable conditions , sensory cues from a single modality are often unreliable measures of the ambient environmental parameters . For instance , purely optic flow-based measurements of self-motion can be misleading for insects which experience sideslip while flying in a crosswind . Moreover , reliance on optic flow may be problematic under dimly lit or overcast conditions , or when flying over lakes or deserts which present sparse visual feedback . In such situations , sampling from multiple sensory cues reduces the ambiguity arising from variability in feedback from single modalities ( Wehner , 2003; Sherman and Dickinson , 2004; Wasserman et al . , 2015 ) . Hence , the integration of multimodal sensory cues is essential for most natural locomotory behaviours , including insect flight manoeuvres ( Willis and Arbas , 1991; Frye et al . , 2003; Verspui and Gray , 2009 ) . For flight control , the importance of optic flow cues detected by compound eyes is well-documented in diverse insects , including honeybees ( Srinivasan et al . , 1996; Baird et al . , 2005 ) , bumblebees ( Baird et al . , 2010; Dyhr and Higgins , 2010 ) , Megalopta ( Baird et al . , 2011 ) and Drosophila ( David , 1982; Duistermars et al . , 2009 ) . In recent years , mechanosensory feedback from antennae has also emerged as a key sensory input for insect flight ( Sane et al . , 2007; Yorozu et al . , 2009; Krishnan et al . , 2012; Fuller et al . , 2014 ) . This feedback is transduced primarily by two sets of mechanosensors - the chordotonal Johnston’s organ ( JO ) ( Gewecke , 1974 ) which senses a wide range of stimuli from high-frequency antennal vibrations to low-frequency ambient airflow or gravity ( Yorozu et al . , 2009; Dieudonné et al . , 2014 ) , and the antennal hair plates ( or Böhm’s bristles ) which are involved in the reflexive positioning of antennae during flight ( Krishnan et al . , 2012 ) . The characteristic positioning of the antennae at the onset of flight is ubiquitous in most , if not all , flying insects underscoring the evolutionary significance of its function ( Dorsett , 1962 ) . Disruption of antennal positioning due to Böhm’s bristle ablation or reduction of JO inputs severely impairs flight ( Willis et al . , 1995; Hinterwirth and Daniel , 2010; Sane et al . , 2007 ) . The control of antennal position is thought to be essential for the unambiguous sensing of inputs by the JO ( Hinterwirth et al . , 2012 ) . However , the mechanisms underlying this behaviour are not well-understood . Here , we show that the airflow cues sensed by Johnston’s organs and the optic flow cues sensed by eyes combine to maintain and control antennal position during flight in the honeybee , Apis mellifera . Each input influences antennal position in an opposite manner; frontal airflow causes antennae to move actively forward against the aerodynamic drag , whereas front-to-back optic flow causes them to move backward . The antennal positioning response thus offers a critical readout for understanding how honeybees integrate information about their own motion from airflow and optic flow cues . To characterize antennal response to ambient airflow , we provided flying bees with frontal airflow cues as would be experienced by them during forward flight . We then calculated the inter-antennal angle ( IAA ) , defined as the angle between the lines joining the base and tip of each antenna ( Sane et al . , 2007 ) , as a measure of antennal position ( Figure 1A ) . When the antennae move backwards , the IAA increases and when the antennae move forward , the IAA decreases . Of the three conventional variables used in such experiments ( David , 1982 ) , we experimentally set the windspeed ( i . e . velocity of ambient airflow relative to ground ) in a calibrated , laminar wind tunnel , whereas the bees controlled their own airspeed ( i . e . velocity of body relative to ambient air ) and therefore also their groundspeed ( i . e . velocity of the body relative to ground ) , which is the vector sum of airspeed and windspeed ( Figure 1B ) . As tethered bees are stationary relative to the ground , their airspeed equals windspeed . However , when freely flying in air currents , airspeed and windspeed are independent of each other . Analogous to swimmers swimming within water currents , insects control their airspeed relative to the air pocket drifting at some windspeed . 10 . 7554/eLife . 14449 . 003Figure 1 . Antennal responses to changing airflow . ( A ) Inter-Antennal Angle ( IAA ) is measured by digitizing 4 points ( red circles ) on the antennae in all frames . ( B ) Groundspeed is obtained by tracking Point#2 ( from A ) which is static relative to head . It is the vector sum of bee-controlled airspeed and experimenter-controlled windspeed . ( C ) Top panel: Response to ambient airflow in tethered bees . Tethered bees were positioned at the centre of the wind-tunnel test section , and facing upwind and the windspeed was linearly varied from 0 to 5 . 5 m/s . Two high-speed cameras positioned dorsally and laterally filmed the bees at 500 fps . Bottom panel: Normalised IAA response as a function of airspeed ( or windspeed ) . We normalized IAA values between 0 ( defined as the mean of values between 0 and 1 . 5 m/s ) and 1 ( defined as the mean of values between 3 and 5 . 5 m/s ) . Between 0–1 . 5 m/s and 3–5 . 5 m/s , normalized IAA did not significantly change ( p>0 . 05 , Moore’s test ) . Between 1 . 5 and 3 m/s , normalized IAA sigmoidally decreased with airspeed , changing with each step ( p<0 . 0001 , Moore’s test , N=10; each colour represents one individual ) relative to the preceding and succeeding values . Non-normalized data in Figure 1—figure supplement 1B . Here and everywhere we have plotted the means , and the error bars indicate the standard deviation of the mean . ( D ) Top panel: IAA response of freely-flying bees to ambient airflow . Bees were trained to enter the wind tunnel through a side-door and fly upwind past the test section to a feeder . High-speed cameras placed and operated as in ( C ) filmed their IAA response . Bottom panel: Normalised IAA response as a function of airspeed in free flight . Between airspeeds of 1 . 5 to 3 m/s , IAA changed significantly ( p<0 . 0001 , Moore’s test , N=10 ) relative to the preceding and succeeding values , but saturated at airspeeds less than 1 . 5 m/s and greater than 3 m/s . Non-normalized data in Figure 1—figure supplement 1D . ( E ) IAA responses to random sequence of ambient airflow in tethered bees . We presented the bees with airflow values between 0 and 5 . 5 m/s in a random sequence and plotted the normalised IAA response as a function of airspeed values reshuffled to lie in increasing order . As in 1C , IAA sigmoidally decreased with airspeed , significantly changing between 1 . 5 and 3 m/s ( p<0 . 0001 , Moore’s test , N=5; each colour represents one individual . ) From 0–1 m/s and 3 . 5–5 . 5 m/s , the normalized IAA did not significantly change . For non-normalized data , see Figure 1—figure supplement 1C . ( F and G insets ) Experiments with sham-treated and JO-restricted bees . Red crosses indicate the location of applied glue , and blue arrow the presence of airflow . ( F ) Normalised IAA vs . airspeed in sham-treated bees . Sham-treated bees ( N=5 ) show responses similar to untreated bees ( compare with Figure 1C ) . Each coloured line represents an individual bee ( Non-normalized data in Figure 1—figure supplement 1H ) . Change in IAA ( Figure 1—figure supplement 1H ) is in the same range as untreated bees ( compare with Figure 1—figure supplement 1B ) . ( G ) Bees with restricted JO do not respond to airspeed change . When the pedicel-flagellum joint is glued , IAA does not vary significantly with changing airspeed ( p>0 . 1 , Moore’s test , N=7 ) . Each colour represents an individual ( Non-normalized data in Figure 1—figure supplement 1I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14449 . 00310 . 7554/eLife . 14449 . 004Figure 1—figure supplement 1 . Antennal responses to changing airflow . ( A ) IAA values at 0 . 5 m/s airflow for a tethered ( blue plot ) and freely-flying honeybee ( red plot ) . IAA values are maintained for all 250 frames for tethered bees ( Mean=96° , S . D . = 0 . 8° ) and freely-flying bees ( Mean=85° , S . D . = 3° ) ( B ) Response of tethered flying bees to linearly changing airflow cues in the wind tunnel . The response is sigmoidal between 1 . 5 and 3 m/s . Each colour represents the same individual as in Figure 1C . Solid black lines represent the mean IAA at which saturation occurs . Normalized data shown in Figure 1C . ( C ) Response of tethered flying bees to randomly changing air flow cues in the wind tunnel . Each bee received a different sequence of random airflow value , and the x-axis was reshuffled in a linearly increasing fashion . The response is similar to that seen with linearly changing cues ( Figure 1—figure supplement 1B ) . ( D ) Response of freely flying bees to changing air flow cues in the wind tunnel . The response is sigmoidal between 1 . 5 m/s and 3 m/s . Each colour in the plot represents the same individual as in Figure 1D . ( E ) Comparison of antennal responses to airflow from Heran ( 1957 ) against data from Figure 1—figure supplement 1B . ( F ) Groundspeed of a freely flying bee as it flies against increasing windspeeds , shown as notched plots . Honeybees maintained a constant groundspeed at approximately an average of 0 . 43 m/s with increasing windspeeds , consistent with previously reported values ( Barron and Srinivasan , 2006 ) . ( G ) Interantennal angle as a function of the groundspeed of the freely flying bees . Different colours indicate different individuals . ( H and I ) Raw data for response of sham-treated tethered bees to changing airflow and JO-glued bees to changing airflow . DOI: http://dx . doi . org/10 . 7554/eLife . 14449 . 004 Honeybees modulated their antennal position as a function of their absolute airspeed . When presented with fixed airflow but no optic flow , both tethered and freely-flying bees held their antennae at constant IAA throughout a flight bout ( e . g . Figure 1—figure supplement 1A ) . As frontal airflow increased , both antennae moved forward and mean IAA at each airflow value decreased as a sigmoidal function of airspeed ( Figure 1C , D ) . The linear region of the sigmoid lay between 1 . 5 to 3 m/s ( grey bar , Figure 1C , D ) , changing by approximately 35o ( Figure 1—figure supplement 1B ) in tethered and 25° ( Figure 1—figure supplement 1D ) in freely-flying bees , significantly decreasing with each step change in airspeed ( step change = 0 . 5 m/s; p<0 . 0001; Moore’s paired test; N=10 ) . This behaviour was similar regardless of whether airspeed stimulus was presented in linearly increasing or decreasing steps ( Figure 1C , Figure 1—figure supplement 1B ) or in random order ( Figure 1E , Figure 1—figure supplement 1C ) . Thus , antennal position is calibrated against absolute airspeeds , independent of time history . At airspeeds less than 1 . 5 m/s or greater than 3 m/s , mean IAA at each airflow plateaued in both tethered ( Figure 1—figure supplement 1B , C ) and freely-flying bees ( Figure 1—figure supplement 1D ) . Previous studies on antennal responses of tethered bees to changing airflow did not report a zone of saturation ( Heran , 1957 ) perhaps due to coarser sampling of their data ( Figure 1—figure supplement 1E ) . We , however , consistently observed sigmoidal mean IAA responses , similar to locusts ( Gewecke , 1974 ) . As frontal airflow in the wind tunnel increased , freely-flying honeybees modulated their flight to maintain roughly constant groundspeed of ca . 0 . 4 m/s ( Figure 1—figure supplement 1F; also [Barron and Srinivasan , 2006] ) . In these bees , we observed no correlation between groundspeed and IAA ( Figure 1—figure supplement 1G ) . Previous researchers have implicated the antennal mechanosensory Johnston’s organs ( JO ) in sensing airflow cues ( Gewecke , 1974; Heran , 1957; Yorozu et al . , 2009 ) . JO spans the pedicel-flagellar joint in the antennae of all Neopteran insects , and tracks the motion of the flagella relative to pedicel . In most insects , the JO consists of several hundred scolopidial units that are range-fractionated . These enable both exquisite sensitivity and narrowly-tuned sensing over a large range of stimulus frequencies ( Yorozu et al . , 2009; Dieudonné et al . , 2014 ) . Does JO also mediate the observed antennal response to changes in airspeed ? To test this hypothesis , we attenuated JO feedback by gluing the pedicel-flagellar joint ( see Materials and methods ) in tethered honeybees , and measured their antennal responses to airflow . To control for the extraneous effects of glue on the antenna ( e . g . due to added weight ) , we glued the second annulus of the flagella from the tip ( sham-treatment ) in a separate group of the bees . Application of glue at the JO does not affect the movement of the antennae because the pedicellar-flagellar joints have no muscles , and hence motion around these joints are passive . Unlike sham-treated bees which showed the typical sigmoid response to frontal airflow ( Figure 1F , non-normalized data in Figure 1—figure supplement 1H , N=5 ) , bees with restricted JO positioned their antennae at the onset of flight ( Figure 1—figure supplement 1I; also [Krishnan et al . , 2012] ) but their mean IAA at each airflow was insensitive to changing airspeeds ( Figure 1G , N=7; p>0 . 1 , Moore’s paired test ) . These experiments established that JO input is required for antennal positioning during flight , but not at flight onset ( Krishnan et al . , 2012 ) . Thus , the antennal positioning at flight onset is a separate process from the inflight modulation of IAA . Flying insects rely heavily on optic flow for flight control , most notably during slower flight manoeuvres such as landing or hovering ( Baird et al . , 2013 ) . For example , visual feedback is critical for regulation of groundspeed and height , or in centring trajectories through narrow corridors in diverse flying insects ( bees: [Srinivasan et al . , 1996; Baird et al . , 2005] , Drosophila: [Straw et al . , 2010] , moths: [Kuenen and Baker , 1982] , review: [Collett et al . , 1993] ) . It also influences antennal position in insects that antennate during walking ( Ye et al . , 2003; Honegger , 1981 ) . Under natural flight conditions , front-to-back optic flow stimulus accompanies frontal airflow stimulus . Does the antennomotor system also respond to optic flow in addition to airflow ? To address this question , we simulated forward flight conditions by presenting tethered flying bees with changing temporal frequency of front-to-back optic flow cues on two LED screens , and measured their IAA ( Figure 2A ) . Note that , previous studies on optic flow dependent behaviours in honeybees have typically demonstrated that they extract angular velocity ( ratio of temporal and spatial frequency ) cues from the image motion , independent of spatial ( Srinivasan et al . , 1991 ) or temporal frequencies ( Baird et al . , 2013 ) . In the experiments reported here , we have kept spatial frequency constant and only varied temporal frequency as the main experimental variable which means that the angular velocity is the temporal frequency times some multiplication factor . 10 . 7554/eLife . 14449 . 005Figure 2 . Antennal responses to changing temporal frequency of optic flow . ( A ) IAA response to optic flow in tethered bees . Bees were positioned central and at 10 cm from the two screens separated by 2 cm at the apex . ( B ) Sample IAA response to linearly increasing ( red ) or decreasing ( green ) optic flow stimulus . ( Top panel ) Stimulus comprises of a visual grating moving from front to back at temporal frequencies ranging between 0–25 cycles/s ( cps ) , in steps of 1 cps . Each step lasts for 1 s ( blue ) . The IAA response to optic flow saturates beyond the threshold of 10 cps ( grey bar; also Materials and methods for threshold calculation ) . ( C–E ) IAA Response to randomized optic flow values . The above honeybee ( the individual shown in Figure 2B–E , bee #1 ) , was presented with randomized optic flow stimulus ( C ) . The precise temporal sequence of randomized stimulus varied between bees . Dotted line shows the 10 cps threshold . Each 1 cps step lasts for 1 s from 1–25 cps . IAA response is plotted in two ways: the IAA response to the randomized stimulus ( D , olive green bars ) , and the response reshuffled in increasing order of temporal frequencies ( E; blue bars ) . The peak at 20 cps is due to the sharp IAA readjustment at stimulus onset . Grey lines between C and D indicate step transitions in temporal frequency values . IAA is predicted to change when stimulus changes occur below or across threshold . Predicted changes in IAA are marked by black circle and no change by white circle . IAA is predicted to decrease ( down arrowhead ) when optic flow changes from high-to-low under or across threshold , and increase ( up arrowhead ) when values change from low-to-high under or across threshold . Changes in temporal frequency above threshold ( horizontal line ) yield no antennal response . In both tests , correct predictions are marked by green and wrong predictions by red circles , and fraction of correct predictions vs . total number indicated beside each test . In this instance , we correctly predicted when IAA would change with 83% accuracy ( 19/23=0 . 83 ) , and the direction of its change with 74% accuracy . ( F , G ) Summary figures showing the IAA response of three individuals ( shown in Figure 2B–E , Figure 2—figure supplement 1A–D and E–H ) ( F ) IAA responses for optic flow rates increasing linearly from 0 to 25 cps . The mean response from all three bees is shown in bold red and the spread of the data is shown in the background of the plot . ( G ) IAA responses for optic flow rates decreasing linearly from 25 to 0 cps . The time axis is shown in the opposite manner because the first optic flow rate that is presented to the tethered bee is 25 cps . The mean of the IAA responses from all three individuals is shown in bold green and the spread of the data shown in the background . ( H ) IAA response to sinusoidal moving visual gratings . Sinusoidal moving grating ( orange , amplitude=1 . 8 cps; period =10 s ) stimulus elicits correlated IAA responses ( blue ) . ( I ) Normalized IAA response vs . optic flow rate between 0 and 1 . 8 cps . Each step increase of 0 . 3 cps in the optic flow rate elicits significant changes in IAA ( p<0 . 05 , Moore’s test , N=9 ) . IAA values were normalized relative to maximum and minimum values for each individual ( raw data in Figure 2—figure supplement 2A ) . ( J , inset ) IAA Responses of the bees with restricted Johnston’s organs . Red crosses indicate glue location . To the set-up in Figure 2A , we added a ducted fan to provide collimated airflow . Red dot indicates absence of airflow , and blue arrows indicate presence of optic flow . ( J ) Normalized IAA response to changes in optic flow . IAA changes with change in optic flow ( p<0 . 0001 , Moore’s test , N=7 ) . In Figure 1G and 2J , the same individuals share color . Normalisation procedure is the same as in Figure 2G and the raw data has been shown in Figure 2—figure supplement 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 14449 . 00510 . 7554/eLife . 14449 . 006Figure 2—figure supplement 1 . Antennal responses to changing optic flow rates . Data for two additional individuals . Figure 2—figure supplement 1A–D show data from one individual and Figure 2—figure supplement 1E–H for a second individual . ( A and E ) IAA responses of linearly changing optic flow rates in tethered bees . Data from two bees showing the IAA response to linearly ascending ( red plot ) and linearly descending ( green plot ) temporal frequencies between 0 and 25 cps . The step changes in optic flow rates have been shown in blue at the top . In both cases , the IAA response to changing rates of optic flow saturates after 14 cps . ( B and F ) Optic flow stimulus provided to the tethered bees . Each optic flow stimulus lasted for 1 s . Black dotted line indicates the threshold after which the response to changing optic flow saturates ( threshold calculated from Figure 2—figure supplement 1A and E ) . Grey lines indicate the transition from one value of temporal frequency to the next . ( C and G ) IAA responses of the bees to randomly changing optic flow rates between 0 and 25 cps . Olive green bar graphs show the actual response of the bees to the stimulus pattern shown in Figure 2C and Figure 2—figure supplement 1H respectively . The threshold in both cases is 14 cps . Predictions and scores are described in Figure 2C , D . The score for predicting motion was 17/23=0 . 74 ( 74% accuracy ) and the score for prediction change in direction of antennal motion was 11/23=0 . 48 ( 48% accuracy , Figure 2—figure supplement 1C ) and 18/23 = 0 . 78 ( 78% accuracy ) and 17/23=0 . 74 ( 74% accuracy , Figure 2—figure supplement 1G ) . ( D and H ) IAA responses to randomized optic flows between 0 and 25 cps in two bees . IAA responses of the bee ( blue ) are reordered and represented against optic flow . Bar graphs show the mean and standard deviation of the IAA response to each optic flow rate value . The first optic flow experienced by the bee in Figure 2—figure supplement 1B was 8 cps and in Figure 2—figure supplement 1F was 5 cps . In both these cases , the IAA at the first optic flow rate stands out from the rest of the dataset . No significant trend emerges from the IAA response of the bee to randomised optic flow rates . DOI: http://dx . doi . org/10 . 7554/eLife . 14449 . 00610 . 7554/eLife . 14449 . 007Figure 2—figure supplement 2 . Antennal responses to changing optic flow rates . ( A ) Raw data for response of tethered flying bees to changes in optic flow . Each colour represents in the plot the same individual as the corresponding colour in Figure 2I . ( B ) Raw data for response of JO-glued bees to changing optic flow cues . Normalised data are shown in Figure 2J . DOI: http://dx . doi . org/10 . 7554/eLife . 14449 . 007 In a single flight bout ( defined as an uninterrupted session between initiation and cessation of flight ) , tethered bees first experienced a static screen of black-and-white gratings ( i . e . temporal frequency = 0 ) followed by a sequence of gratings moving from 0 to 25 cycle/s ( cps ) in discrete steps of 1 cps to simulate increasing flight speed ( red curve , Figure 2B; Figure 2—figure supplement 1A , E; summary figure showing data from all individuals , Figure 2F ) . In the subsequent flight bout , the temporal frequency decreased stepwise from 25 to 0 cps ( green curve , Figure 2B; Figure 2—figure supplement 1A , E; summary figure showing data from all individuals , Figure 2G ) to simulate decreasing flight speed . Each stimulus step lasted for 1 s and the entire protocol for 26 s . When temporal frequency increased from 0 to 25 cps , an initial sharp adjustment in IAA ( red curve , Figure 2B , summary figure Figure 2F ) was followed by a graded response between 0 to approx . 10 cps ( grey bar , Figure 2B ) , beyond which the value reached a plateau . The response curve for linearly decreasing optic flow ( 25 to 0 cps , green curve , Figure 2B , summary figure Figure 2G ) was similar to the increasing cues , except for the lack of initial sharp IAA adjustment from zero to non-zero optic flow ( also Figure 2—figure supplement 1A , E ) . In these experiments , the optic flow was patterned to monotonically increase or decrease . Graded IAA responses to such optic flow patterns suggested two possibilities . First , as in the case of airflow-based response , IAA response curve is innately calibrated against specific temporal frequency values , independent of its time history . If true , the response curve should remain invariant when the temporal frequency of the stimulus is presented in random order . This is an unlikely possibility because absolute values of optic flow are meaningless without prior knowledge of the spatial structure of the world . Second , that bees respond to changes in temporal frequency rather than their absolute value . If true , IAA would increase ( or decrease ) in response to positive ( or negative ) changes in temporal frequency within the operating range ( e . g . 0–10 cps for the bee in Figure 2B–E ) while remaining unchanged outside this range ( i . e . >10 cps ) . The threshold value was calculated separately for each bee ( see Materials and methods for details ) . The latter criteria lead to specific predictions of when to expect changes in IAA ( filled black circle; Figure 2C , D ) and in which direction ( arrow; Figure 2C , D ) . To test the above hypotheses , we presented tethered honeybees with grating speeds between 1 to 25 cps in randomized order ( Figure 2C–E and Figure 2—figure supplement 1B–D , Figure 2—figure supplement 1F–H ) and compared the measured vs . predicted IAA response ( Figure 2D; also Figure 2—figure supplement 1C , G ) . For each bee , we scored these as the fraction of correct predictions; a score of 1 corresponds to all correct , and 0 to all incorrect predictions . In all the cases , the total prediction score matched the actual observed score well above chance levels ( p<0 . 001; Student’s T test; see Materials and methods for details ) . When stimulus bins were rearranged in ascending order of temporal frequency , the reshuffled mean IAA response was not graded ( compare Figure 2B and E; Figure 2—figure supplement 1A and D; Figure 2—figure supplement 1E and H ) ruling out the possibility of innate calibration of IAA against optic flow . Thus , our data show that antennomotor activity in honey bees tracks changes in temporal frequency , and not their absolute value . Bees typically experience the steepest gradients in optic flow during landing or hovering over flowers when optic flow rates are low ( Baird et al . , 2013 ) . Hence , in the follow-up experiments , we focused on the lower range of stimuli from 0–1 . 8 cps ( Figure 2H–J ) . Even in this narrow range , IAA responses tracked the magnitude and direction of grating patterns on the screen . For example , bee antennae robustly tracked simple sinusoidal stimuli that were both , cycled between front-to-back ( 0 to 1 . 8 cps ) , and back-to-front grating movement ( 0 to -1 . 8 to 0 cps ) over the 10 s duration ( Figure 2H ) , thus verifying that the antennae move in both directions as a function of optic flow . To characterize their responses in this narrow range of temporal frequency ( 0–1 . 8 cps ) , the tethered bees initiated wing flapping in front of a blank screen , followed by a sequence of black-and-white gratings moving front-to-back from 1 . 8 to 0 . 3 cps . Each grating sequence lasted for 6 s , interspersed with a static screen ( zero optic flow ) for 3 s . Despite inter-animal variability in the set points for IAA at zero optic flow ( Figure 2—figure supplement 2A ) , all bees increased their IAA as a function of optic flow ( Figure 2I ) , with each step change of 0 . 3 cps eliciting significant increase in mean IAA ( p<0 . 0001; Moore’s paired test , N=10 ) at that optic flow rate . The IAA response of tethered bees to optic flow remained intact ( Figure 2J; Figure 2—figure supplement 2B ) even when their JO were restricted by gluing the pedicel-flagellar joint , establishing that it is independent of the JO pathway . Thus , front-to-back optic flow causes antenna to move backward , opposite to frontal airflow which causes antenna to move actively forward . The experiments outlined above established that inputs from two sensory modalities independently helped the bees to sense airspeed and optic flow . Whereas the JO-based antennal mechanosensory feedback was calibrated against absolute values of airspeed , visual feedback was calibrated relative to changes in optic flow under or across a threshold value . These experiments required flying bees to respond to single sensory cues at a time . Hence , we next measured IAA response of tethered bees to simultaneous visual and airflow cues . The paired stimuli combinations were arbitrarily drawn from values in previous experiments , and are not naturally correlated . How do multimodal cues affect IAA response when simultaneously presented ? Optic flow ( red , Figure 3A ) and airflow ( blue , Figure 3A ) cues were presented in three combinations , represented by three regimes A , B or C ( grey bars; Figure 3A ) . Responses of a single bee are shown in Figure 3B , C . In each trial , we started the experiment with an intermediate bimodal combination ( grey bars , middle Regime B; optic flow=0 . 9 cps , airspeed =2 . 5 m/s ) and in random order , either decreased ( to Regime A; optic flow =0 . 3 cps , airspeed =0 . 5 m/s; Figure 3B ) or increased these values ( to Regime C , optic flow =1 . 8 cps , airspeed =4 m/s; Figure 3C ) . Antennae maintained position when the cue combination changed from Regime B to Regime A ( green line; Figure 3B; Pp>0 . 1 , Moore’s test ) , but not when only airflow or optic flow was changed . When airspeed alone ( i . e . with static grating ) decreased from 2 . 5 to 0 . 5 m/s ( blue line; Figure 3B; p<0 . 001 , Moore’s test ) , both antennae moved backward . However , they moved forward when optic flow alone ( i . e . with fan off ) decreased from 0 . 9 to 0 . 3 cps ( red line; Figure 3B; p<0 . 001; Moore’s test ) . Similarly , when the cue combination changed from Regime B to Regime C , antennae again maintained position ( green line; Figure 3C; p>0 . 1 , Moore’s test ) , but moved forward when airflow alone increased from 2 . 5 to 4 m/s ( blue line; Figure 3C ) and backward when optic flow alone increased from 0 . 9 to 1 . 8 cps ( red line; Figure 3C ) . Thus , visual and airflow cues elicit opposite IAA responses which , when acting in concert , maintain antennal position . 10 . 7554/eLife . 14449 . 008Figure 3 . IAA responses to combinatorial stimuli . ( A ) IAA response curve of a tethered bee to changing airflow ( blue ) and optic flow ( red ) . Regime A represents the combination of low optic flow rate and airspeed ( temporal frequency=0 . 3 cps; airspeed=0 . 5 m/s ) ; Regime B represents combination of intermediate optic flow rate and airspeed ( temporal frequency =0 . 9 cps; airspeed=2 . 5 m/s ) , and Regime C represents high optic flow rate combined and airspeed ( temporal frequency =1 . 8 cps , airspeed=4 m/s ) . The dashed line ( grey ) at the top of the plot indicates the IAA response at 0 optic flow . ( B , C ) Representative data from a single individual when values transition from Regime B→ A ( B ) or from Regime B→C ( C ) . IAA responses to airspeed ( blue ) , and optic flow ( red ) , and combination of airspeed and optic flow ( green ) plot . ( D , E ) ΔIAA response to step changes in airspeed , optic flow or combined cues . We have represented the mean ΔIAA values as notched plots . The extent of box shows the inter-quartile range and the lower and upper bounds of the box represent the 25th and the 75th quartiles . The line in the box represents the median of the data and there is a ‘notch’ around this median for easy comparison of the notched boxes with each other . If the notches of two boxes do not overlap , their medians are statistically significantly different from each other . The red crosses indicate the outliers in the data . The whiskers extend to the most extreme data point that is not considered to be an outlier . represents statistically significant difference ( p<0 . 001 , Moore’s test , N=8 ) . Mean ΔIAA is significantly different from a hypothetical mean of zero ( p<0 . 05 , ANOVA; post hoc Tukey’s HSD test , N=8 ) when only airspeed or optic flow are varied , but not when a stimulus combination is co-varied ( p>0 . 5 , ANOVA; post hoc Tukey’s HSD test , N=8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14449 . 008 This was also borne out in the pooled data over multiple trials ( Figure 3D , E ) . Decrease of only airspeed resulted in a negative mean ΔIAA ( blue , Figure 3D; p<0 . 001; Moore’s test ) , whereas increase resulted in a positive mean ΔIAA ( blue , Figure 3E; p<0 . 001; Moore’s test ) . Similarly , a decrease in only optic flow resulted in positive mean ΔIAA values , but its increase led to negative mean ΔIAA ( red , compare Figure 3D , E; p<0 . 001; Moore’s test ) compared to the IAA values in Regime B . Here , mean ΔIAA values significantly differed from a hypothetical mean of zero ( one-way ANOVA with post-hoc Tukey’s Honest Significant Difference test , p<0 . 05 ) . However , when both cues were co-varied , ΔIAA did not significantly change ( green , Figure 3D , E; p>0 . 1 , Moore’s test ) and their means were statistically indistinguishable from zero . How does the presence of one cue alter the antennal response curve of the other cue ? We considered three possibilities . First , if the response curve of Cue A remains unaltered in presence of a constant Cue B , then it is likely that the system adapts to cue B . Second , if the response curve to Cue A is uniformly offset by the presence of constant Cue B , then the cross-modal influence is likely linear and summative . Third , if response curve to Cue A is non-uniformly offset in presence of constant Cue B , then crossmodal influences are likely to be non-linear . Testing for possibilities provides insights into how inputs from different modalities combine to determine the antennal position . We measured IAA against optic flow in presence of two constant values ( 0 and 1 m/s ) of airspeed . The two curves were uniformly offset over the range from 0 to 1 . 8 cps ( compare dotted and solid lines , Figure 4A , B; p<0 . 05 , Moore’s test , N=8 ) . Similarly , for airspeeds between 0 and 4 m/s , we observed a steady offset for greater optic flow , but in the opposite direction ( compare dotted and solid lines , Figure 4C , D; p<0 . 05 , Moore’s test , N=9 ) . In both experiments , the presence of constant optic flow ( or airflow ) altered the set point of the antenna to a new mean value that is greater ( or less ) than the original set point ( Figure 4B , D ) and the antenna responded to variation in airflow ( or optic flow ) relative to the new set point . Thus , the antennomotor system linearly combines the multimodal cues and recalibrates accordingly . 10 . 7554/eLife . 14449 . 009Figure 4 . IAA responses to changes in optic flow in presence of constant airflow ( A , B ) , and to changes in airspeed in presence of constant optic flow cues ( C , D ) . ( A ) IAA responses to optic flow rates varying from 0 to 1 . 8 cps in presence of still air ( IAAOp; solid green line vs . steady 1 m/s airflow ( IAAW+Op; dotted green line ) . Mean difference between IAAOp and IAAW+Op is significant at each optic flow rate value ( p<0 . 0001 , Moore’s test , N=8 ) . ( B ) We have represented the mean ΔIAA values as notched plots . The range of ΔIAA ( =IAAW+Op - IAAOp ) is not significantly different over various optic flow values . Each mean ΔIAA is significantly different from a hypothetical mean of 0 ( p<0 . 05 , ANOVA and post hoc Tukey’s HSD test , N=8 ) . ( C ) IAA responses to varying airspeed at 1 , 2 . 5 and 4 m/s in presence of no optic flow ( IAAW; solid red line; top left panel ) vs . steady optic flow of 1 . 8 cps ( IAAW+Op; dotted red line; top right panel ) . Again , mean IAA values at each rate of optic flow are significantly different ( p<0 . 0001 , Moore’s test , N=9 ) for the two cases . ( D ) The range of ΔIAA plotted as notched plots are again not significantly different across the various airspeeds , but each mean ΔIAA value is significantly different from a hypothetical mean of 0 ( *p<0 . 05 , N=9 , ANOVA and post hoc Tukey’s HSD test ) . ( E ) A general model of the antennal positioning response to airspeed and optic flow cues , including the role of mechanosensory hair plates ( Böhm’s bristles ) in antennal positioning response . ( F ) The crossmodal calibration hypothesis proposes that insects simultaneously sample airflow and optic flow , and use response characteristics of airflow sensing to calibrate optic flow . Determining how the sampled optic flow varies in the dynamic range of airspeeds enables insects to linearly extrapolate the optic flow response curve over a greater range . The grey bar represents the stimulus range in which such simultaneous airflow ( blue ) and optic flow ( black on red ) measurements are made . According to this hypothesis , once a specific airflow value is correlated against the observed optic flow , which could be slow or fast , it can then be used to make measurements over much greater range of airspeeds . DOI: http://dx . doi . org/10 . 7554/eLife . 14449 . 009 The above study provides us with insights into how the insect antennomotor system integrates the multiple sensory cues that it encounters during flight . This can be summarized in a general schematic of the multi-modal integration of antennal positioning behaviour in flying honeybees ( Figure 4E ) . For the sake of completeness , this model must also include the role of the mechanosensory hair plates ( or Böhm’s bristles ) in reflexive antennal positioning , which appears conserved across insects according to our studies in moths ( Krishnan et al . , 2012 ) , bees and crickets ( Sant and Sane , 2016 ) . To summarize these studies: antennal hair plates are stimulated when the antenna undergoes a substantial movement in the scape-head capsule or pedicel-scape joints . The mechanosensory neurons underlying these bristles directly project into the Antennal Motor and Mechanosensory Center ( AMMC ) where they arborize on the dendritic fields of antennal motor neurons and activate them . Any gross change in antennal position thus elicits a rapid reflexive correction of the antennal position with latencies under 10 ms in hawk moths ( Krishnan et al . , 2012 ) and probably on the same order in honeybees and other insects . Based on these data , we proposed a model to describe the antennal positioning reflex loop mediated via antennal mechanosensors ( Krishnan et al . , 2012 ) . Although arrangement of the hair plates in different insects varies , the underlying neural circuitry and hair plate function is conserved in moths , bees , and most other insect orders ( Krishnan and Sane , 2015 ) . Previous studies in hawk moths have described only how the antenna , once positioned , reflexively maintains this position ( Krishnan et al . , 2012 ) , and also their response to visual cues ( Krishnan and Sane , 2014 ) . However , the combinatorial role of these cues in antennal positioning behaviour during flight remained unclear . Here , we show that the reflexive maintenance of in-flight antennal positioning is simultaneously modulated by both visual ( from eyes ) and airflow ( from JO ) feedback during flight . This study also shows that their action is via multi-modal pathways which combine in a mutually antagonistic fashion; whereas frontal airflow detected by JO reduces IAA , the front-to-back optic flow detected by compound eyes increases IAA . The combination of these inputs modifies IAA to a new set point , which is then maintained by the antennal hair plates via a negative feedback loop that ensures rapid maintenance and correction of the antennal position . The block diagram in Figure 4E thus provides a model for multimodal sensory control of antennal positioning response . It is possible that other inputs , such as olfaction , also additionally modulate the antennomotor responses , which is the subject of future studies . Our study shows that the initial positioning of antennae at the flight onset ( antennal deployment ) is separate from its later inflight maintenance by the multimodal inputs ( inflight antennal positioning ) . Both behaviours require the hair plate-mediated reflex pathway but the set-point of this reflex system is under multimodal control in the latter case . In particular , restriction of the JO feedback has no effect on the antennal deployment behaviour , whereas it completely disrupts inflight antennal positioning . It has been suggested that inflight modulation of IAA maintains the scolopidial units of the JO in their operating range ( Hinterwirth et al . , 2012 ) . Although this hypothesis remains unaddressed and is beyond the scope of this paper , our data clearly show that inflight antennal movements are precisely modulated , and product of both visual ( from compound eyes ) and mechanosensory ( from the hair plates and JO ) input . Visual feedback also induces directionally-sensitive antennal movements in other insects such as hawk moths ( Krishnan and Sane , 2014 ) , Drosophila ( Mamiya et al . , 2011 ) , and many orthopteran insects ( Honegger , 1981; Ye et al . , 2003 ) . Unlike the hair plate reflexes which are strictly unilateral ( Krishnan et al . , 2012 ) , visual feedback drives the activity of both ipsi- and contralateral antennal motor neurons in moths ( Krishnan and Sane , 2014 ) and may therefore serve to coordinate the movements of both antennae . The mechanosensory pathways mediated by the hair plates then provide rapid reflexive , local correction of the intended position of each antenna . Recent studies have also described how the integration of antennal mechanosensory and visual inputs mediates abdominal flexion in moths ( Hinterwirth and Daniel , 2010 ) which in turn is relevant for flight control and balance . In honeybees too , their combined effect on the abdominal streamlining behaviour enables drag reduction during flight ( Luu et al . , 2011; Taylor et al . , 2013 ) . In freely-flying moths , the stimulation of antennal muscles elicits abdominal flexion accompanied by a change in flight trajectory ( Hinterwirth et al . , 2012 ) . Similarly , integration of mechanosensory and visual information mediates flight control in other insects such as Drosophila ( Sherman and Dickinson , 2003; Fuller et al . , 2014 ) . Whereas the relative influence of mechanosensory input is greater during rapid turns , visual input is more important during slower rotations ( Sherman and Dickinson , 2003; 2004 ) . The integration of these multimodal cues quite likely occurs in the Antennal Motor and Mechanosensory Centre ( AMMC ) region of the insect brain , which houses the soma of the antennal motor neurons . This region also receives the arbors of motion-sensitive visual interneurons ( Hertel and Maronde , 1987 ) , and likely also the inputs from cephalic hair mechano-afferents . These different modalities are known to influence diverse flight behaviours . It is thus likely that descending interneurons transduce multimodal sensory information that elicits these behaviours . In locusts , the tritocerebral commissure giant ( TCG ) interneurone integrates visual and airflow information ( Bacon and Tyrer , 1978 ) , while other descending interneurons in crickets integrate antennal mechanosensory and visual inputs ( Gebhardt and Honegger , 2001 ) . The importance of AMMC as the site of integration of multimodal sensory feedback is an important topic of future study . The data on antennal position are also relevant to studies on speedometry in honeybees . A key finding of this paper is that IAA responds to absolute airspeed values from 1 . 5 to 3 m/s , which is narrow compared to the typical range of airspeeds ( between 0 and 7 m/s; [Wenner , 1963] ) of honeybees in free flight . IAA is also calibrated against optic flow cues , but this feedback depends on the spatial structure of the environment . As shown in studies over the past two decades , the odometer in honeybees is quite clearly visual ( Srinivasan , 2014 ) . The visual system can extract the overall image motion , which is perceived by and integrated over the motion detectors in the retina ( Zanker et al . , 1999 ) . Honeybees trained to find food at a specific distance in tunnels lined with visual stripes can correctly judge distances even when the spatial frequency of the stripes is altered ( Baird et al . , 2005; Srinivasan et al . , 1996; 1997; 2000; 2014 ) . Moreover , they estimate distances correctly despite head or tailwinds , suggesting that parameters such as windspeed , flight duration and number of wing beats etc . play no role in the odometry . Indeed , image motion cues can also be used to estimate depth and distinguish between near and far objects independent of their size ( Kirchner and Srinivasan , 1989; Srinivasan et al . , 1989; Zhang et al . , 1992 ) . Thus , bees appear to be able to separately process size and velocity cues ( Srinivasan et al . , 1993 ) . Complement to the visual odometer , we show here that honeybees are also able to extract information about both airspeed and groundspeed . Importantly , the airspeed calibration mediated by JO is absolute , but works over a narrow speed range , whereas the groundspeed calibration mediated by optic flow can be tuned over a broader range . We propose the hypothesis that honeybees calibrate their optic flow speedometer using their JO-based system that measures absolute airspeeds ( Figure 4F ) . At the onset of a regular flight bout , bees simultaneously sample the optic flow ( red lines ) and airspeed ( blue line ) . The three red lines correspond to temporal frequency of optic flow that is slow ( i . e . if objects are far away ) to fast ( i . e . if objects are close ) . Sampling the optic flow ( Figure 4F; black segments on the red lines ) corresponding to the dynamic range of the JO-based airspeed sensor ( e . g . between ~1 . 5 to 3 m/s ) provides bees with the means to cross-calibrate average optic flow against absolute airspeeds . As the airspeed-based estimation only works in a narrow range from 1 . 5 to 3 m/s , the optic flow-based speedometer requires extrapolation well beyond the range of the antennal system . As shown in Figure 4A–D , the presence of one cue linearly offsets the response curve for the other , analogous to using a standardized albeit narrow-ranging measure ( e . g . airflow-based speedometer ) to calibrate another arbitrarily scalable measure ( e . g . the optic flow based speedometer ) . A disadvantage of such a system is that it must be recalibrated at the beginning of each bout , or within bouts in case of sudden gusts of wind which may confound the calibration . Under such circumstances , we predict that honeybees would have to slow down to recalibrate their optic flow based speedometer against the airspeed-based speedometer , before resuming flight . Our data throw some light on puzzling observations from previous studies . In their study on visual regulation of groundspeeds in honeybees , Barron and Srinivasan ( Barron and Srinivasan , 2006 ) observed that honeybees freely flying within a tunnel lined with checkerboard patterns flew at speeds of approximately 0 . 4 m/s . However , when the checkerboard pattern was replaced by an axial-stripe pattern which offered sparse optic flow cues , its groundspeed was maintained with low variance at a value of 1 . 4 m/s . Maintenance of ground speeds ( which equals airspeed in still air ) requires that bees be able to ‘sense’ their speed even when optic flow cues are sparse . How is this possible if speed is maintained only via optic flow ? To explain this result , we propose that , in absence of optic flow , the bees rely on their JO-based airflow sensors to both set and maintain groundspeed . It is worth noting that the value of 1 . 4 m/s lies at the cusp of the airflow response curve ( Figure 1C ) , and is maintained as the new reference at different wind speeds in sparse optic flow ( see Figure 3 in Barron and Srinivasan [2006] ) . A JO-based antennal airflow sensing , can in principle offer a ‘true’ airspeed measure for odometry . It is not presently clear if the visual odometer would need such information as the current model for visual odometry is able to quite robustly explain most observations . One would expect such a speedometer system to accumulate errors as the range increases , as has also been observed with the honeybee odometer , which follows the Weber’s law ( Cheng et al . , 1999 ) . Future studies will be required to determine if the several predictions that emerge from this model about honeybee speed control hold true . The antennal positioning responses thus provide unique insights into how honeybees and perhaps other insects sense and combine information from multiple modalities to help reduce ambiguities arising from drift in any one modality . After capturing individual forager bees at the hive , we cold-anaesthetized them on ice until they were inactive . The anaesthetized bees were then dorsally tethered to a bent metal rod ( 30 mm in length and 0 . 2 mm diameter ) using a synthetic rubber based adhesive ( Fevibond , Pidilite , Mumbai , India ) . The bee was then provided with sucrose solution and left to recover for 45 min . When performing experiments involving visual stimuli , we dark-adapted the bees in a darkened box during this recovery period . To elicit flight , we provided the bee with a piece of tissue paper to hold , and then suddenly withdrew it to elicit flight due to tarsal reflex . We constructed an open-circuit wind tunnel ( 30 cm x 30 cm cross-section and 120 cm in length , with a 40 cm x 30 cm x 30 cm test section ) with a fan , driven by a motor drive at one end that drew air through the wind tunnel . The wind tunnel was calibrated using a constant temperature mini-anemometer ( Kurz 490S , Kurz Instruments , Inc . , Monterey , CA ) , which was modified to take direct voltage readings ( Sane and Jacobson , 2006 ) . The voltage readings were consistently reproducible between 0–6 m/s at a step size of 0 . 5 m/s . To minimise confounding visual cues , the floor and walls of the wind tunnel were covered with white paper . To determine how antennal position varies with ambient airflow , we provided tethered and freely flying bees with frontal airflow between 0 and 5 . 5 m/s and between 0 and 3 . 5 m/s in steps of 0 . 5 m/s respectively . At their fastest , freely flying honey bees have been estimated to fly at ~7 . 5 m/s in natural outdoor environments . Presence of headwinds or tailwinds of about 3 m/s does not affect the speed of the bees ( Wenner , 1963 ) in their natural flight . In both cases , we filmed bees from the top and side with two synchronised high-speed cameras and calculated their IAA . We performed both tethered and free flight assays in the wind tunnel to look at how the antennae respond to changes in airflow . To investigate how visual motion influences the antennal positioning response , we modified a MATLAB demo routine from the Psychophysics toolbox ( Mathworks Inc , Natick , MA ) ( Brainard , 1997; Kleiner and Pelli , 2007 ) to generate a moving visual grating pattern . The patterns were displayed on two LED monitors ( Beetel 8”x14” , 640x480 resolution , 60 Hz refresh rate ) and controlled simultaneously by a master computer equipped with a dual-VGA graphics card ( nVIDIA GeForce 9800GT ) . These screens did not show noticeable flicker at 1000 fps , as determined by filming them with a high-speed camera . The monitors were placed in a V configuration , with a 2 cm gap between them . The tethered bee was positioned at the centre of the two monitors , facing the apex , at a linear distance of 10 cm from each monitor ( Figure 2A ) . The floor below the screens was layered with white paper to increase the contrast of the antennae during filming . We provided open-loop forward translational stimuli ( simulating front-to-back movement of the visual field ) to the tethered bees , at a spatial frequency of 0 . 44 cm-1 and various temporal frequencies ( 0 to 25 cps for the first set of the experiments and 1 . 8 , 1 . 5 , 1 . 2 , 0 . 9 and 0 . 3 cps for the second set of experiments ) , corresponding approximately to angular speeds of 0 to 300 deg/s and 18 , 15 , 12 , 9 , 6 and 3 deg/s respectively . Using two synchronized high-speed video cameras ( Phantom v7 . 3 , Vision Research , Inc . Wayne , NJ ) , we filmed tethered flying bees at 250 fps , from both a side view and an overhead view . A third camera , synchronized to the first two , filmed the visual stimulus being displayed to the bee . We provided each bee with a range of temporal frequencies over a single flight bout , and digitized and analysed the whole video , the entire flight bout of the bee in that trial for the first set of experiments . In the second set of experiments in which we gave the bees optic flow from 0 to 1 . 8 cps , we digitised and analysed 50 frames of video per temporal frequency for 50 wing beats . In our analysis , we discarded bees that did not fly continuously through the entire protocol . To provide both visual and airflow stimuli to the tethered bees , we placed a 4 wire DC fan in front of the bee in a 2 cm gap between the two screens in the set up described above , such that the tethered bee received frontal airflow ( Figure 2A ) . The airflow was collimated and fairly laminar , and we ensured that there were no other sources of airflow . The speed of the fan was adjustable and we could measure the speed of the frontal airflow near the head of the bee using a hot wire anemometer . The fan was placed inside a rectangular box with a plexiglass tunnel in front of it that opened between the two screens so that the tethered bee received collimated frontal airflow .	Insects combine information from different senses to help them navigate during flight . Flying insects see moving images , which the brain can use to measure their speeds . Insect antennae also help to judge speed , as they signal to the brain about the physical forces that result from the insect moving through the air . To accurately detect these forces , and also to detect odors from the surrounding environment , insects must precisely position their antennae as they fly . To investigate how honeybees use different types of sensory information to position their antennae during flight , Roy Khurana and Sane first placed freely-flying and tethered bees in a wind tunnel . Flying forward causes air to flow from the front to the back of the bee . The experiments revealed that a bee brings its antennae forward and holds them in a specific position that depends on the rate of airflow . As the bee flies forward more quickly ( or airflow increases ) , the antennae are positioned further forward . Roy Khurana and Sane then investigated how the movement of images across the insect’s eyes causes their antennae to change position . This unexpectedly revealed that moving images across the eye from front to back , which simulates what bees see when flying forward , causes the bees to move their antennae backward . However , exposing the bees to both the frontal airflow and front-to-back image motion as normally experienced during forward flight caused the bees to maintain their antennae in a fixed position . This behaviour results from the opposing responses of the antennae to the two stimuli . Future challenges will be to determine how the brain of a honeybee combines the information from different senses to position the antennae , and to discover what this behaviour implies for insect flight in general .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "neuroscience" ]	2016	Airflow and optic flow mediate antennal positioning in flying honeybees
Non-alcoholic fatty liver disease ( NAFLD ) is the most common liver disease in industrialized countries and is increasing in prevalence . The pathomechanisms , however , are poorly understood . This study assessed the unexpected role of the Hedgehog pathway in adult liver lipid metabolism . Using transgenic mice with conditional hepatocyte-specific deletion of Smoothened in adult mice , we showed that hepatocellular inhibition of Hedgehog signaling leads to steatosis by altering the abundance of the transcription factors GLI1 and GLI3 . This steatotic 'Gli-code' caused the modulation of a complex network of lipogenic transcription factors and enzymes , including SREBP1 and PNPLA3 , as demonstrated by microarray analysis and siRNA experiments and could be confirmed in other steatotic mouse models as well as in steatotic human livers . Conversely , activation of the Hedgehog pathway reversed the "Gli-code" and mitigated hepatic steatosis . Collectively , our results reveal that dysfunctions in the Hedgehog pathway play an important role in hepatic steatosis and beyond . As the most common liver disease in the western countries and a disease with an increasing prevalence , non-alcoholic fatty liver disease ( NAFLD ) is the subject of many investigations and studies ( Ahmed et al . , 2015; Dongiovanni et al . , 2015; Zhang and Lu , 2015 ) to identify unknown risk factors and new treatment strategies . Hepatic steatosis is the hallmark feature of NAFLD ( Enomoto et al . , 2015 ) and has the potential to develop into more severe steatohepatitis ( NASH ) , which can progress to fibrosis , cirrhosis and cancer . Steatosis occurs when the rate of fatty acid delivery exceeds the rate of fatty acid removal ( oxidation and export ) . A comprehensive knowledge of the hepatic lipid metabolism and its control mechanisms is crucial for preventing and treating liver steatosis . To date , these control mechanisms , e . g . the factors governing hepatocyte heterogeneity in lipid metabolism , remain largely unknown and have not yet received adequate attention in the discussion of NAFLD ( Postic and Girard , 2008 ) . Metabolic zonation of the liver is of considerable importance for the optimal integration and regulation of the plethora of different hepatic functions and metabolic homeostasis ( Gebhardt , 1992; Gebhardt and Matz-Soja , 2014 ) . Recently , Wnt/beta-catenin signaling was recognized as a master regulator of the zonal distribution of nitrogen metabolism in the adult liver ( Benhamouche et al . , 2006; Gebhardt and Hovhannisyan , 2010; Monga , 2015 ) and it influences the balance between the anabolic and catabolic functions of glucose metabolism ( Chafey et al . , 2009 ) . The Hedgehog ( Hh ) signaling cascade is another important pathway that determines embryonic patterning , cell growth and tissue repair ( Omenetti et al . , 2011; Gu and Xie , 2015 ) , and often acts in close crosstalk with Wnt/beta-catenin signaling ( Toku et al . , 2011 ) . Similar to Wnt/beta-catenin signaling , the Hh pathway can act in canonical and non-canonical manners , as explicitly described by Teperino et al . ( Teperino et al . , 2014 ) . To activate canonical Hh signaling in mammals , the ligands Sonic , Indian and Desert Hedgehog ( SHH , IHH and DHH ) interact with the surface receptors Patched 1 and 2 ( PTCH1 and PTCH2 ) which remove their inhibition of the co-receptor Smoothened ( SMO ) . Active SMO triggers the activation of the GLI ( Glioma-associated oncogene ) transcription factors ( TFs ) ( GLI1 , 2 and 3 ) by preventing the conversion of GLI2 and GLI3 into transcriptional repressors ( Ruiz i Altaba , 1999; Infante et al . , 2015 ) . According to Ruiz I Altaba et al . , the 'Gli-code' , i . e . the combinatorial and cooperative function of the repressing and activating forms of all GLI factors , forms the basis of the integration of Hh signals ( as well as of multiple other morphogens and cytokines ) in embryogenesis and carcinogenesis ( Ruiz i Altaba , 1999; Ruiz i Altaba et al . , 2003; 2007 ) . To date , Hh signaling has been found to play a considerable role in various scenarios of adult liver regeneration from progenitor cells ( Sicklick et al . , 2006 ) and in the regulation of the compensatory outgrowth of progenitors and myofibroblasts ( Jung et al . , 2010 ) . Regarding liver metabolism , we recently showed that the regulation of the IGF-axis ( insulin-like growth factor ) in mature hepatocytes is controlled by Hh signaling ( Matz-Soja et al . , 2014 ) . Moreover , we were also able to show that the TFs GLI1 , GLI2 and GLI3 form a unique self-stabilizing network in hepatocytes and regulate several metabolic genes , including lipid associated factors ( Schmidt-Heck et al . , 2015 ) . Therefore , in this study , we have focused on the alterations in lipid metabolism as a consequence of hepatic Hh signaling modulations in vivo and in vitro . To address this aim , a mouse model with a conditional hepatocyte-specific deletion of Smo in adult mice was established , in order to avoid interference with developmental effects of Hh signaling . For the in vitro investigations , we used RNA interference ( RNAi ) technology to knock down several genes of the Hh pathway in cultured hepatocytes ( Böttger et al . , 2015; Schmidt-Heck et al . , 2015 ) . The results of our investigations clearly show that the Hh pathway is a strong regulator of lipid metabolism in the adult liver . Furthermore , we show that impaired Hh signaling leads to increased expression of lipogenic TFs and enzymes with different zonal preference which finally results in steatosis . Conversely , we demonstrate that slight activation of the pathway by Sufu knockdown , small molecule agonists or GLI overexpression can mitigate lipid accumulation in steatotic livers . As depicted in Figure 1A–B , triple transgenic mice which have a conditional hepatocyte-specific ablation of Smo in response to doxycycline ( abbreviated SLC mice ) were generated . Thus , the Smotm2Amc/J mice ( Jackson Laboratories ) , which possess loxP sites flanking exon 1 of the Smo gene ( Long et al . , 2001 ) , were crossed with the LC-1/rTALAP-1 mice which are working with a tetracycline-controlled transcriptional activation of the Cre recombinase protein ( provided by Hermann Bujard ) ( Schonig et al . , 2002 ) . In the LC-1/rTALAP-1 mice the synthetic transactivator variant ( rtetR ) of the tet-repressor present in rTALAP-1 mice is driven by the LAP-promoter ( PLAP ) . In the presence of doxyxycline , rtetR binds to an array of seven tet operator sequences ( tetO7 ) activating transcription of the Cre recombinase gene ( tet-on-system ) ( Figure 1A ) . The offspring were genotyped by PCR for the Smo wildtype ( Smo WT ) and Smo floxed ( Smo flx ) alleles , the doxycycline responsible element ( rtetR ) and the Cre recombinase ( Primer: Supplementary file 1A ) . 10 . 7554/eLife . 13308 . 003Figure 1 . Strategy for conditional and hepatocyte-specific deletion of Smo . ( A ) Scheme of the tet-on system in the LC-1/rTALAP-1 mice . ( B ) Structure of Smo with loxP sites . ( C ) PCR , for Cre recombinase , yielded a 400 bp fragment in SLC-KO mice only . ( D ) PCR from liver tissue during treatment with doxycycline , yielding a 600 bp amplicon for SLC-WT alleles and a 350 bp amplicon for the recombinant Smo alleles in the SLC-KO mice . ( E ) Immunohistochemical staining of Cre recombinase in liver sections of the SLC-WT and the SLC-KO mice . Bar: 100 µm . ( F ) qRT-PCR of Smo in different tissues and isolated hepatocytes of the SLC-WT ( n=6–10 ) and the SLC-KO ( n=6–10 ) mice . Source files of all data used for the quantitative analysis are available in the Figure 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 00310 . 7554/eLife . 13308 . 004Figure 1—source data 1 . Source data of qRT-PCR of Smo in different tissues and isolated hepatocytes of the SLC-WT and the SLC-KO mice ( Figure 1F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 00410 . 7554/eLife . 13308 . 005Figure 1—figure supplement 1 . Influence of doxycycline on lipogenic gene expression qRT-PCR of Ppara , Pparb/d , Pparg , Elovl3 , Elovl6 and Fasn in isolated hepatocytes of the male SLC-WT mice treated with doxycycline ( n=3–4 ) for 10 days compared to male SLC-WT mice without doxycycline ( n=3–4 ) at the same age . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 005 At the age of 8 weeks ( to avoid the hormonal complications of adolescence ) , hepatocyte-specific ablation of Smo was induced by adding 2 mg/ml doxycycline hydrochloride ( Sigma , Germany ) to the drinking water for 10 days to promote the expression of the Cre recombinase ( Figure 1C ) . During this period , the Smo rec . ( recombinant ) primer yielded a 350 bp fragment , indicating the deletion of the floxed domain ( Figure 1D ) ( Primer: Supplementary file 1A ) . After 10 days , nearly all hepatocytes were positive for Cre recombinase protein ( Figure 1E ) . After this treatment , the mice were maintained under normal conditions until 12 weeks of age . After sacrifice , the specificity of the Smo deletion was measured via qRT-PCR , indicating that there was a significant decrease in Smo expression in the liver material and isolated hepatocytes ( Figure 1F ) . As already shown in our previous article , no adverse side effect of the doxycycline treatment could be observed on body weight ( Matz-Soja et al . , 2014 ) . Likewise , the expression of important genes of lipid metabolism in hepatocytes was not affected by doxycycline ( Figure 1—figure supplement 1 ) . When maintained without doxycycline , the transgenic SLC mice developed without a phenotype . The deletion of Smo at 8 weeks of age resulted in pronounced liver steatosis within 5 weeks ( Figure 2A ) . The H&E and fat red staining clearly showed lipid droplet accumulations , which are most prominent in the midzone to periportal zone , but occasionally encompassed the entire parenchyma ( Figure 2A ) . Quantification of the fat red staining revealed a 7-fold increase in the SLC-KO mice compared with the WT controls ( Figure 2B ) . Because the insulin and glucose levels changed only marginally ( Table 1 ) , we conclude that insulin resistance does not contribute to the steatosis in SLC-KO mice at least during the first five weeks after doxycycline administration . This assumption is confirmed by the fact that we could not detect any significant changes in hepatic expression for insulin receptor ( Insr ) , insulin receptor substrate 1 and 2 ( Irs1/2 ) ( Figure 2—figure supplement 1 ) . With the exception of the steatosis and lower liver/body weight ratio compared with the SLC-WT mice ( Figure 2C ) , there were no other indications of overt liver damage in the SLC-KO mice . Accordingly , the serum activities of ASAT ( aspartate aminotransferase ) , ALAT ( alanine aminotransferase ) and GLDH ( glutamate dehydrogenase ) were not different from those of the SLC-WT mice ( Table 1 ) . The plasma triglyceride levels were significantly increased in the VLDL ( very low density lipoprotein ) fraction , while no changes were found in LDL ( low density lipoprotein ) and HDL ( high density lipoprotein ) fractions ( Figure 2—figure supplement 2 ) . These findings indicate that steatosis is not caused by an impairment of triglyceride secretion from the hepatocytes . 10 . 7554/eLife . 13308 . 006Figure 2 . Liver phenotype of the SLC mice . ( A ) The H&E and fat red staining of liver sections and hepatocytes showed strong steatosis in the male SLC-KO mice compared to the SLC-WT mice ( bars: 200 µm , 100 µm and 50 µm ) ( pc: pericentral , pc: periportal ) . ( B ) Quantification of the fat red-stained liver sections from the male SLC-WT ( n=10 ) and SLC-KO ( n=7 ) mice . ( C ) Comparison of the liver/body ratio . Source files of all data used for the quantitative analysis are available in the Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 00610 . 7554/eLife . 13308 . 007Figure 2—source data 1 . Source data of quantification of the fat red-stained liver sections from the male SLC-WT and SLC-KO mice ( Figure 2B ) and comparison of the liver/body ratio ( Figure 2C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 00710 . 7554/eLife . 13308 . 008Figure 2—figure supplement 1 . Gene expression of insulin signaling in isolated hepatocytes from SLC mice . Relative gene expression determined by qRT-PCR in isolated hepatocytes from the male SLC-WT ( n=14 ) and the male SLC-KO ( n=14 ) mice of insulin receptor ( Inrs ) and insulin receptor substrate 1 and 2 ( Irs1 , Irs2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 00810 . 7554/eLife . 13308 . 009Figure 2—figure supplement 2 . Serum lipoprotein levels of the SLC mice . ( A–D ) Total cholesterol , HDL , LDL and VLDL in the serum of the male SLC-WT ( n=5 ) and male SLC-KO mice ( n=6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 00910 . 7554/eLife . 13308 . 010Figure 2—figure supplement 3 . Influence of cyclopamine on accumulation of neutral lipids in cultured mouse and human hepatocytes . ( A ) Mouse and ( B ) human hepatocytes were cultured in the absence ( left; vehicle only ) or presence ( right ) of 10 µM cyclopamine ( in 0 . 1% DMSO ) . After 72 h , cultures were fixed and stained with Nile red . Bar: 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 01010 . 7554/eLife . 13308 . 011Table 1 . Serum concentrations of glucose , insulin and enzyme activities of ALAT , ASAT and GLDH in SLC mice . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 011parameterSLC-WTnSLC-KOninsulin ( pmol/l ) 80 . 06 ± 9 . 9813110 . 88 ± 31 . 317glucose ( mmol/dl ) 10 . 16 ± 1 . 8057 . 68 ± 1 . 026ALAT ( µkat/l ) 1 . 02 ± 0 . 3250 . 81 ± 0 . 035ASAT ( µkat/l ) 2 . 95 ± 0 . 9052 . 50 ± 0 . 415GLDH ( µkat/l ) 0 . 42 ± 0 . 1350 . 28 ± 0 . 055 In addition to these results in vivo , the inhibition of Hh signaling in cultured mouse and human hepatocytes using the SMO inhibitor cyclopamine ( Hovhannisyan et al . , 2009 ) also resulted in marked steatosis within 48 to 72 hr ( Figure 2—figure supplement 3 ) . To study the signaling cascade after the deletion of Smo , we analyzed the alterations of Hh-related genes in isolated hepatocytes . Regarding Hh ligands , we could observe a down-regulation of Ihh and Shh , consistent with a decrease in pathway activity due to the deletion of Smo ( Figure 3A ) . It is worth noting that Ihh is the most abundantly expressed ligand in hepatocytes , whereas Shh is hard to detect and Dhh was not measureable . Regarding the receptors , we did not observe significant changes in the Ptch1 , Ptch2 and Hhip ( Hedgehog-interacting protein ) transcripts , which was also true for Fu ( Fused ) and Sufu ( Suppressor of fused ) ( Figure 3B , C ) . However , the Gli1 and Gli3 mRNAs were significantly decreased in the SLC-KO mice , while the Gli2 mRNA remained unchanged ( Figure 3D ) . To visualize the amount and the distribution of the GLI1 and GLI3 protein in the liver parenchyma of the SLC mice , we performed immunohistological stainings of GLI1 and GLI3 as well . ( Figure 3E , F ) . The results clearly demonstrate that GLI1 and GLI3 , are well detectable in hepatocyte nuclei in SLC-WT mice , but are absent in nuclei of SLC-KO hepatocytes ( white arrows ) . In SLC-KO livers , these TFs are only present in non-parenchymal cells , e . g . bile duct epithelial cells ( Figure 3E , F , yellow arrowheads ) . These results were confirmed by analyses of the GLI3 protein content by western blotting in isolated hepatocytes from SLC-WT and SLC-KO mice . The results clearly show that the amount of GLI3/A ( full length activator protein ) was significantly reduced in SLC-KO hepatocytes ( Figure 3—figure supplement 1A , B ) . In order to find out whether these distinct alterations of the GLI factors are characteristic for steatotic livers , we measured the expression signature of the Gli TFs in isolated hepatocytes from melanocortin-4-receptor-deficient mice ( MC4R ) and Lepob/ob mice , which are characterized by massive steatosis as a result of over-nutrition ( Sandrock et al . , 2009; Itoh et al . , 2011; Trak-Smayra et al . , 2011 ) . In both mouse models , the expression of Gli1 and Gli3 was obviously reduced , whereas Gli2 expression either did not change ( MC4R-KO mice ) or was increased ( ob/ob mice ) ( Figure 3—figure supplement 2A , B ) . Furthermore we could detect the same transcriptional changes of the Gli factors in human patients with clinical relevant steatosis compared to non-steatotic patients ( Figure 3—figure supplement 2C ) . 10 . 7554/eLife . 13308 . 012Figure 3 . Expression of genes and proteins related to Hh signaling in SLC mice . ( A–D ) qRT-PCR data from isolated hepatocytes from the male SLC-WT ( n=6–10 ) and the SLC-KO ( n=6–10 ) mice illustrating the expression of ( A ) the ligands Ihh , Shh and Dhh ( n . d . : not detectable ) ; ( B ) the ligand binding molecules Ptch1 , Ptch2 and Hhip1; ( C ) the downstream genes Fu and Sufu and ( D ) the TFs Gli1 , Gli2 and Gli3 of the Hh signaling pathway . ( E–F ) Immunohistochemistry of liver sections from male SLC-WT and SLC-KO mice of ( E ) GLI1 and ( F ) GLI3 . Labeled hepatocyte nuclei for both Gli factors are seen only in WT , but not in KO animals ( white arrows ) . Staining in non-parenchymal cells , e . g . bile ductular cells ( yellow arrowheads ) is not affected by the knockout . Scale bars: 200 μm; 100 μm and 50 μm . Source files of all data used for the quantitative analysis are available in the Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 01210 . 7554/eLife . 13308 . 013Figure 3—source data 1 . Source data of the expression of genes related to Hh signaling in SLC mice ( Figure 3A–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 01310 . 7554/eLife . 13308 . 014Figure 3—figure supplement 1 . Western Blot of GLI3 in the SLC mice . ( A ) Representative Western blot from isolated hepatocytes from male SLC-WT ( n=4 ) and SLC-KO ( n=4 ) mice . The blot shows the expected molecular weight from the two isoforms of GLI3 , the full length activator form at 170–190 kDa ( GLI3/A ) and the truncated repressor form at ~80 kDa ( GLI3/R ) . ( B ) Densitometric quantification of GLI3/A , normalized to loading control ACTB , was done from three Western blots from SLC-WT ( n=9 ) and SLC-KO ( n=9 ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 01410 . 7554/eLife . 13308 . 015Figure 3—figure supplement 2 . The “steatotic Gli-code” extents to several mouse models and humans with steatosis . ( A ) qRT-PCR data from isolated hepatocytes from the male MC4R-WT ( n=4 ) and the MC4R-KO ( n=4 ) mice illustrating the expression of Gli1 , Gli2 and Gli3 at the age of 6 month; ( B ) qRT-PCR data from isolated hepatocytes from the male WT ( n=4 ) and the ob/ob ( n=4 ) mice illustrating the expression of Gli1 , Gli2 and Gli3 at the age of 6 month; ( C ) qRT-PCR data from healthy female human liver tissue ( control , n=29 ) and steatotic samples ( n=33 ) illustrating the expression of Gli1 , Gli2 and Gli3 . ( D ) Quantitative fat red staining in isolated hepatocytes from the male MC4R-KO mice after incubation with SAG ( 300 nM ) and control vehicle for 48 hr . ( E ) Quantitative fat red staining in isolated hepatocytes from the male ob/ob mice after incubation with recombinant SHH ( 0 . 25 µg ) and control vehicle for 72 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 015 To get an impression of the global changes in gene expression in response to the deletion of Smo , microarray studies were performed . In order to account for possible inter-individual variations of gene expression , total RNA was prepared from freshly isolated hepatocytes of four pairs of SLC-KO and SLC-WT mice . At a cut-off level of 1 . 5-fold , 179 genes were up-regulated in Chip Arrays and 106 genes were down-regulated . Gene set enrichment analysis ( GSEA ) using ClueGO revealed highly significant changes in a number of metabolic functions , particularly those involved in lipid metabolism ( Figure 4A ) ( Figure 4—source data 1—data ) . For instance , the GO term ‘lipid metabolic process’ showed pronounced up-regulation of 30 genes ( p-value of 7 . 70E-11 ) many of which ( e . g . Ppara , Pparg , Srebf1 , Aacs , Elovl6 and Fas ) were verified by qRT-PCR . In addition , the GSEA revealed that many regulated genes in the SLC-KO mice ( e . g . Cd36 , Avpr1a , Ttc23 , Lifr andFabp2 ) belong to the top 50 ones described to be mostly correlated with elevated hepatic triglyceride level among 100 unique inbred mouse strains ( Hui et al . , 2015 ) . The GO term ‘metabolic process’ showed pronounced up-regulation of 113 genes reaching a p-value of 1 . 24E-9 indicating that genes from other metabolic processes also responded to the Smo knockout . The GO terms ‘organic acid metabolic process’ and ‘steroid biosynthetic process’ showed up-regulation of several genes with p-values of 1 . 61E-6 and 6 . 09E-5 , respectively . Furthermore , genes belonging to the GO term ‘response to hormone stimulus’ such as Lipin1 and Ramp2 were found to be up-regulated . Among down-regulated genes those involved in ‘activation of protein kinase activity’ and ‘microtubule-based processes’ prevailed with p-values of 1 . 59E-3 and 1 . 62E-3 , respectively . 10 . 7554/eLife . 13308 . 016Figure 4 . Gen and protein expression of hepatic TFs involved in lipid metabolism in SLC mice . ( A ) Volcano blot visualizing differentially expressed genes in male SLC-KO mice detected by Affymetrix microarray analysis ( n=4 ) . All colored dots ( blue and magenta ) indicate an expression fold change equal or higher than 1 . 5; magenta: central genes of lipid metabolism; blue: other regulated genes . ( B ) qRT-PCR of Chrebp , Srebf1 , Srebf2 , Ppara , Pparb/d and Pparg from hepatocytes of male SLC-WT ( n=6–13 ) and SLC-KO ( n=6–13 ) mice . ( C–D ) Immunohistochemistry in liver sections from male SLC-WT and SLC-KO mice . ( C ) SREBP1 is strongly induced and shows a higher incidence of nuclear staining in pericentral hepatocytes of SLC-KO mice . ( D ) PPARG shows a much higher incidence in hepatocyte nuclei and a slight cytoplasmic increase in pericentral hepatocytes of SLC-KO mice . Scale bars: 200 μm and 100 μm . Source files of all data used for the quantitative analysis are available in the Figure 4—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 01610 . 7554/eLife . 13308 . 017Figure 4—source data 1 . Gene set enrichment analysis of isolated hepatocytes from SLC-WT and SLC-KO mice . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 01710 . 7554/eLife . 13308 . 018Figure 4—source data 2 . Source data of gene expression of hepatic TFs involved in lipid metabolism in SLC mice ( Figure 4B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 018 QRT-PCR was used to confirm major results of the microarray analysis and to detect additional regulated genes with higher sensitivity and accuracy . As shown in Figure 4B , the significant up-regulation of several TFs involved in the regulation of lipid and carbohydrate metabolism was detected including Chrebp ( Carbohydrate-responsive element-binding protein ) , Srebf1 , Srebf2 ( Sterol regulatory element binding transcription factor 1/2 ) , Ppara and Pparg ( Peroxisome proliferator activated receptor alpha/gamma ) in SLC-KO mice . To confirm this data on protein level , immunohistochemical staining was performed for the expression of SREBP1 and PPARG in liver sections of SLC-WT and SLC-KO mice . The SREBP1 protein showed a slightly pericentral and rare nuclear preference in liver parenchyma from SLC-WT mice and was very strongly enhanced in the pericentral zone of livers from SLC-KO mice . In particular , the frequency of nuclear staining in hepatocytes was much higher in these mice compared to control mice ( Figure 4C ) . Likewise , PPARG protein was much more frequent in hepatocyte nuclei of SLC-KO mice than in SLC-WT mice . Also with PPARG heterogeneous distribution of the protein was obvious ( Figure 4D ) . Additionally , the gene expression of the anti-adipogenic TF Gata4 ( Patankar et al . , 2012 ) was significantly reduced , whereas Gata6 remained unchanged ( Figure 5A ) . The Nfyb ( Nuclear transcription factor-Y beta ) mRNA was also markedly reduced , while the Nfyg ( Nuclear transcription factor-Y gamma ) and Lxra ( Liver X receptor alpha ) mRNAs remained unaffected . The expression of Foxa1 and Foxa2 ( Forkhead box A1/2 ) , which are known to mediate the effects of insulin on lipid metabolism ( Wolfrum et al . , 2004 ) , did not change ( Figure 5B ) . Interestingly , the expression of Nr1d2 ( Rev-ErbA beta ) was strongly down-regulated , while that of Nr1d1 ( Rev-ErbA alpha ) remained unchanged ( Figure 5B ) . 10 . 7554/eLife . 13308 . 019Figure 5 . Expression of the hepatic TFs and enzymes involved in lipid metabolism in SLC mice . ( A–D ) qRT-PCR data from hepatocytes from the male SLC-WT ( n=6–13 ) and SLC-KO ( n=6–13 ) mice . ( A ) Gata4 , Gata6 , Nfyb , Nfyg and Lxra . ( B ) Foxa1 , Foxa2 , Nr1d1 and Nr1d2 . ( C ) Acaca , Acacb , Fasn , Gpam , Elovl6 and Elovl3 . ( D ) Aacs , Hmgcr , Lss and Pnpla3 . ( E ) The PPI ( protein-protein interaction ) network obtained from the STRINGv10 database using steatosis- and Hh-signaling-related genes as the query . The colored lines indicate co-expression ( black ) , experimental data ( purple ) , database scan ( blue ) and published scientific abstracts ( green ) . Source files of all data used for the quantitative analysis are available in the Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 01910 . 7554/eLife . 13308 . 020Figure 5—source data 1 . Source data of expression of the hepatic TFs and enzymes involved in lipid metabolism in SLC mice ( Figure 5A–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 020 Of the key lipogenic enzymes , we found increased expression of those involved in the biosynthesis of fatty acids and triglycerides , including Acaca ( Acetyl-Coenzyme A carboxylase alpha ) , Fasn ( fatty acid synthase ) , Elovl6 ( Elongation of long chain fatty acids ) and Gpam ( Glycerol-3-phosphate acyltransferase ) ( Figure 5C ) . In contrast , the expression of Acacb was not changed , and Elovl3 expression was significantly decreased ( Figure 5C ) . The transcripts for enzymes involved in cholesterol biosynthesis , such as Aacs ( acetoacetyl-CoA synthetase ) , Hmgcr ( 3-hydroxy-3-methylglutaryl-Coenzyme A reductase ) , and Lss ( Lanosterol synthase ) , were also increased ( Figure 5D ) . These findings suggest a coordinated response of genes favoring fatty acid , triglyceride and cholesterol biosynthesis . Intriguingly , the newly discovered NAFLD-associated gene Pnpla3 ( Adiponutrin ) ( Chow et al . , 2014; Smagris et al . , 2015 ) was also dramatically increased ( Figure 5D ) . Using this data , the results of our microarray analysis , and published databases , we created a hypothetical protein-protein interaction network ( PPI ) of the studied steatosis- and Hh-signaling-related genes with the STRINGv10 software ( Figure 5E ) . This network provides an overview of the studied proteins and confirms the observed diverse and highly complex connections of the TFs and enzymes related to lipid metabolism and Hh signaling . PNPLA3 expression is connected to SREBF1 activity , and the PPAR family is linked to the FOXA and GLI TFs . The lipogenic enzymes ( e . g . FASN , ACACA/B , ELOVL6 , and GPAM ) are closely connected to the TF SREBF1 and to each other as part of the fatty acid and triglyceride biosynthetic pathways . Immunohistochemical analyses of FASN were performed to confirm the differences in expression at the protein level . The stronger staining not only confirmed the up-regulation of FASN in the SLC-KO mice , but , surprisingly , revealed a shift from the known pericentral ( Gebhardt , 1992; Postic and Girard , 2008 ) to the periportal localization in the SLC-WT mice , matching the preferential site of lipid accumulation ( Figure 6A ) . These results suggest that in addition to Wnt/beta-catenin signaling ( Benhamouche et al . , 2006; Gebhardt and Hovhannisyan , 2010 ) , Hh signaling is required to maintain the normal zonation of the liver parenchyma , at least with respect to lipid metabolism . Furthermore , we measured the activity of several lipid metabolism pathways in cultured hepatocytes . Fatty acid biosynthesis , as determined by the incorporation of [14C]-labelled acetate , was almost doubled in the hepatocytes from the SLC-KO mice ( Figure 6B ) . However , cholesterol biosynthesis from labelled acetate was not changed in vitro ( Figure 6B ) , corresponding to the unchanged serum cholesterol levels in vivo ( Figure 1—figure supplement 1A ) . Interestingly , the hepatocytes from the SLC-KO mice showed a significantly higher rate of fatty acid synthesis from [14C ( U ) ]-glucose in the presence of 10 mM glucose and 0 . 1 µM insulin ( Figure 6C ) , suggesting increased channeling of the high concentrations of glucose into fatty acid biosynthesis . The fact that neither the glycogen content ( Figure 6D ) , nor basal or stimulated glycolysis as determined by the conversion of [14C ( U ) ]-glucose to lactate ( Figure 6E ) were altered in the hepatocytes from the re-fed SLC-KO mice , suggests that there is no shortage of potential fuel for glycolytic acetyl-CoA formation . 10 . 7554/eLife . 13308 . 021Figure 6 . Expression of genes and proteins involved in lipid and mitochondrial energy metabolism in SLC mice . ( A ) Immunohistochemistry of FASN in liver sections of male SLC-WT and SLC-KO mice ( pp: periportal , pc: pericentral ) . ( B–H ) Measurements in freshly isolated hepatocytes from male SLC-WT and SLC-KO mice ( B ) Determination of fatty acid and cholesterol biosynthesis . ( C ) Conversion of [14C ( U ) ]-glucose to fatty acids . ( D ) Glycogen content . ( E ) Determination of glycolysis . ( F ) Electron microscopy of liver tissue . ( G ) ATP content . ( H–I ) qRT-PCR data from the male SLC-WT ( n=6 ) and the SLC-KO ( n=6 ) mice: ( H ) Acox1 , Cpt1a , Cpt2 and Acadvl , ( I ) Slc25a1 , Slc25a2 and Slc25a20 . ( J ) Serum concentrations of ketone bodies . Eight to ten SLC mice were used in each of the experiments depicted in ( B , C , D , E , G , H , I , J ) . Source files of all data used for the quantitative analysis are available in the Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 02110 . 7554/eLife . 13308 . 022Figure 6—source data 1 . Source data of expression of genes and proteins involved in lipid and mitochondrial energy metabolism in SLC mice ( Figure 6B–J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 02210 . 7554/eLife . 13308 . 023Figure 6—figure supplement 1 . Expression of respiratory chain complexes in SLC mice . Enzyme activity of citrate synthase ( CS ) and respiratory chain complexes I-IV in liver tissue of the male SLC-WT ( n=6 ) and the male SLC-KO ( n=6 ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 023 There is growing evidence that mitochondrial dysfunction plays a central role in the pathogenesis of NAFLD ( Petrosillo et al . , 2007; Wei et al . , 2008 ) . Therefore , we investigated the impact of the Hh pathway on the mitochondria in hepatocytes . Electron microscopy revealed larger and swollen mitochondria in the SLC-KO mice ( Figure 6F ) . Further analyses showed significantly reduced ATP levels ( Figure 6G ) . However , the activity of the respiratory complexes was not changed ( Figure 6—figure supplement 1 ) . The expression of carrier proteins ( e . g . Cpt1a , Cpt2; Carnitine palmitoyltransferases 1a/2 ) and enzymes ( Acox1 , acyl-Coenzyme A oxidase 1 , and Acadvl , acyl-Coenzyme A dehydrogenase , very long chain ) of beta oxidation were also not changed ( Figure 6H ) . Instead , we found the significant decrease of Slc25a5 ( mitochondrial carrier ) expression , which transports ADP and ATP through the mitochondrial inner membrane ( Figure 6I ) . The fact that no significant changes in the ketone body concentrations could be measured ( Figure 6J ) , lead us to speculate that the decrease of SLC25a5 could be one explanation of the low ATP production . Furthermore , we wanted to know whether the down-regulation of Gli1 and Gli3 might be sufficient to explain the changes in expression of the TFs and their metabolic consequences; thus , we performed a siRNA-mediated knockdown of the Gli factors . Gli3 knockdown led to a significant increase in the expression of lipogenic TFs including Ppara , Pparg , and Srebf1 ( Figure 7A ) . Likewise , only the hepatocytes treated with Gli3 siRNA responded with an increase in the expression of Fasn and Elovl6 and a reduced expression of Elovl3 ( Figure 7—figure supplement 1 ) . To validate the transcriptional data we focused on the lipid content of siRNA-treated cultures . Mainly the hepatocytes transfected with the Gli3 siRNA showed pronounced accumulation of fat droplets ( Figure 7B ) and elevated lipid staining after 72 hr ( Figure 7C ) . To a minor extent Gli1 knockdown may aid in inducing steatosis , because Srebf1 was also induced under this condition ( Figure 7A ) . Chrebp expression did not change in these experiments ( Figure 7—figure supplement 1 ) , suggesting that the up-regulation observed in the SLC-KO mice ( Figure 4B ) is either due to a GLI-independent mechanism or reflects another adaptive response in vivo . 10 . 7554/eLife . 13308 . 024Figure 7 . Influence of siRNA-mediated knockdown of Gli1 , Gli2 and Gli3 on expression of genes of lipid metabolism . ( A–C ) Isolated hepatocytes from male C57BL6/N mice ( n=4–9 ) treated with the control , Gli1 , Gli2 and Gli3 siRNAs . ( A ) Relative expression of Ppara , Pparg , Srebf1 and Srebf2 , determined by qRT-PCR . Data for Ppara , Pparg , Srebf1 and Srebf2 were taken from our publication ( Schmidt-Heck et al . , 2015 ) with the same mouse model for simplifying comparison . ( B ) Qualitative and ( C ) quantitative fat red staining . Source files of all data used for the quantitative analysis are available in the Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 02410 . 7554/eLife . 13308 . 025Figure 7—source data 1 . Source data of the influence of siRNA-mediated knockdown of Gli1 , Gli2 and Gli3 on expression of genes of lipid metabolism ( Figure 7A , C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 02510 . 7554/eLife . 13308 . 026Figure 7—figure supplement 1 . Influence of siRNA-mediated knockdown of Gli1 , Gli2 and Gli3 on gene expression of lipogenic enzymes . ( A–D ) Isolated hepatocytes from male C57BL6/N mice ( n=4–7 ) treated with the control , Gli1 , Gli2 and Gli3 siRNAs . Relative expression of ( A ) Chrebp , ( B ) Fasn , ( C ) Elovl6 and ( D ) Elovl3 was determined by qRT-PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 02610 . 7554/eLife . 13308 . 027Figure 7—figure supplement 2 . Chromatin immunoprecipitation experiments for GLI3 binding sites . ( A ) Consensus sequence of the GLI3 binding site at -1767 bp of the Fasn promoter , at -2674 bp of the Ppara promoter and at -1156 bp of the Srebf1 promoter according to MotivMap . ( B ) qRT-PCR analyses of precipitated DNA using primers for the promotor region of Fasn , Ppara and Srebf1 ( supplementary material table 3 ) . DNA immunoprecipitated with Histone H3 Antibody as positive control ( black ) , GLI3 antibody ( grey ) , and IgG as a negative control ( white ) was calculated relative to 2% input as described in Materials and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 027 In order to evaluate the role of GLI3 in mediating the influence of the Hh pathway activity on the expression of downstream lipogenic TFs ( Ppara and Srebf1 ) and enzymes ( Fasn ) several chromatin immunoprecipitation experiments ( ChIP ) were performed . Using data mining by MotivMap analyses ( Daily et al . , 2011 ) several possible binding sites for the GLI3 protein within the range of − 10000 bp upstream to 10000 bp downstream of the transcription start site of the Fasn , Ppara and Srebf1 genes were identified . As consensus sequence we selected TGTGTGGTC for the ChIP analysis ( Figure 7—figure supplement 2A ) . The data provided in Figure 7—figure supplement 2B give a first hint that direct binding of GLI3 to the predicted binding site in the promoter region of Srebf1 ( at -1156 bp ) might be involved in transmitting GLI factor activity to the expression of SREBP1 . No such hint was found , however , in the case of Fasn ( at the predicted binding site at -1767 bp ) and Ppara ( at the predicted binding site at -2674 bp ) making it more likely that these proteins are not primary , but secondary targets of GLI3 ( mediated among else by SREBP1 , see below ) . Finally , we investigated whether it is possible to prevent steatosis via activation of the Hh pathway . Therefore , we used several activating strategies including ( i ) the knockdown of the suppressor of fused ( Sufu ) ( a well-known negative regulator of the signaling cascade; Law et al . , 2012 ) , ( ii ) the incubation with SMO agonists ( SAG ) or recombinant SHH , and ( iii ) GLI overexpression in hepatocytes from different mouse models . Ad ( i ) , Sufu siRNA significantly reduced Sufu mRNA levels ( Figure 8A ) and strongly up-regulatedexpression of all Gli TFs ( Figure 8B ) in hepatocytes from C57BL6/N mice . Regarding the lipogenic TFs , a significant decrease of Srebf1 and a slight down-regulation of Srebf2 and Elovl6 were detected in the treated cells ( Figure 8C ) indicating that up-regulation of the Gli TFs , i . e . reversal of the steatotic GLI code , causes anti-steatotic modulation of lipogenic genes . To go a step further , we determined whether Sufu knockdown could reduce hepatic steatosis . Using isolated hepatocytes from MC4R-KO mice , which retain their steatotic phenotype in vitro , we demonstrated a loss of staining intensity in the cultures treated with the Sufu siRNA compared with the cultures transfected with the control ( Figure 8D , E ) . 10 . 7554/eLife . 13308 . 028Figure 8 . Influence of siRNA-mediated knockdown of Sufu on lipid metabolism in vitro . ( A-C ) Isolated hepatocytes from male C57BL6/N mice ( n=8 ) treated with the control and Sufu siRNA . Relative expression of the following genes was determined by qRT-PCR 48 hr post-transfection ( A ) Sufu and Fu , ( B ) Gli1 , Gli2 and Gli3 , ( C ) Srebf1 , Srebf2 , Pparg , Elovl3 and Elovl6 . ( D ) Qualitative and ( E ) quantitative fat red staining in hepatocytes isolated from the MC4R-KO mice in response to treatment with the control and Sufu siRNA . Source files of all data used for the quantitative analysis are available in the Figure 8—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 02810 . 7554/eLife . 13308 . 029Figure 8—source data 1 . Data source of the influence of siRNA-mediated knockdown of Sufu on lipid metabolism in vitro ( Figure 8 A-C , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 029 Ad ( ii ) , We also could observed a significant reduction in fat red staining when the steatotic hepatocytes from the MC4R-KO and ob/ob mice were incubated with the SMO agonist SAG and recombinant SHH ( Figure 3—figure supplement 2D , E ) . These experiments indicate that activation of Hh signaling upstream of Sufu on the Gli TF level by an Hh agonist or an Hh ligand has similar anti-steatotic effects as the Sufu knockdown . Ad ( iii ) , to proof the assumption that reversal of the level of the Gli TFs ( GLI1 and GLI3 ) is able to overcome steatosis we treated isolated hepatocytes from steatotic ob/ob mice with overexpression plasmids for GLI1 , GLI2 , GLI3 and a combination of GLI1 and GLI3 . The results clearly show that the transfection of the GLI plasmids leads to a strong increase in the corresponding Gli mRNA expression , while the combined transfection of GLI1 and GLI3 expression plasmids resulted in a slightly lower expression of the corresponding Gli mRNAs ( Figure 9A–C ) . The results from quantitative and qualitative analysis of fat red staining indicate that overexpression of one Gli factor alone is not able to reduce the lipid content . However , the combination of GLI1 and GLI3 significantly mitigated the steatosis in the hepatocytes from ob/ob mice ( Figure 9D , E ) indicating that parallel changes in both GLI factors are required to reverse steatosis . 10 . 7554/eLife . 13308 . 030Figure 9 . Influence of GLI1 , GLI2 and GLI3 overexpression on lipid content in hepatocytes of ob/ob mice . Isolated hepatocytes from male ob/ob mice ( n=3 ) transfected with the control plasmid ( MOCK ) and human overexpression plasmids of GLI1 , GLI2 , and GLI3 or the combination of GLI1 and GLI3 , 72 hr post-transfection as described in Materials and Methods . ( A-C ) Relative expression of the following genes was determined by qRT-PCR . ( A ) Gli1 , ( B ) Gli2 ( C ) Gli3 . ( D ) Quantitative and ( E ) qualitative fat red staining . Lipid content was reduced only in the presence of the combined overexpression of GLI1 and GLI3 , but not after expression of each Gli factor alone . Source files of all data used for the quantitative analysis are available in the Figure 9—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 03010 . 7554/eLife . 13308 . 031Figure 9—source data 1 . Source data of the influence of GLI1 , GLI2 and GLI3 overexpression on lipid content in hepatocytes of ob/ob mice ( Figure A-D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 031 Interestingly , quantitative and qualitative determination of fat red staining demonstrates that only the combined overexpression of GLI1 and GLI3 ( but not of each Gli factor alone ) is able to significantly reduce the lipid content in the hepatocytes from ob/ob mice ( Figure 9D , E ) . These results emphasize the strong influence of the GLI-code and , thus , the mediating function of Hh signaling in an animal model of steatosis of completely different origin ( i . e leptin deficiency ) . In this study , we demonstrate that the hepatocyte-specific impairment of canonical Hh signaling by the conditional ablation of Smo results in considerable changes in liver lipid metabolism that ultimately lead to hepatic steatosis . Hence , the undisturbed Hh pathway is necessary to maintain the proper balance between the synthesis/uptake and degradation/export of fatty acids and triglycerides in adult hepatocytes ( Figure 10A ) . These results were confirmed on the metabolic and the transcriptional level . Metabolically , we demonstrated the increased accumulation of triglycerides and the significant increase of VLDL particles , which are a hallmark of hepatic steatosis and NAFLD ( Fon Tacer and Rozman , 2011 ) . On the transcriptional level , we showed that a large set of lipogenic TFs and enzymes were significantly up-regulated following the deletion of Smo . Among the responding TFs were Srebf1 , Ppara and Pparg strongly increased which are characteristic for liver steatosis ( Fabbrini et al . , 2010; Ables , 2012 ) . Interestingly , combinatorial interactions between Srepf1 , Nfyb and Nfyg ( Nuclear transcription factor Y ) , and Sp1 ( Transacting transcription factor 1 ) have been shown to determine the expression of specific sets of target genes , including hundreds of genes with distinct roles in lipid metabolism and other functions ( Reed et al . , 2008 ) . Especially the role of SREBP1 is further discussed below . Some studies with certain hepatic cells already revealed the contributions of Nfyb/g to the overexpression of lipid metabolizing enzymes and several other metabolic events ( Woo et al . , 2005; Damiano et al . , 2009 ) . Likewise , the differential response of Nr1d1 and Nr1d2 , which relate the circadian rhythm to metabolism and disease ( Duez and Staels , 2008; Ramanathan et al . , 2014 ) , may indicate that Hh signaling in hepatocytes could also influence the circadian fluctuations of lipid metabolism . 10 . 7554/eLife . 13308 . 032Figure 10 . Regulation of hepatic lipid metabolism by Hh signaling . ( A ) Intact Hh signaling allows a self-supporting network of the GLI TFs ( Schmidt-Heck et al . , 2015 ) . Gli1 and Gli3 exert attenuating effects on lipogenic TFs resulting in balanced liver lipid metabolism . ( B ) The impairment of Hh signaling down-regulates ( green ) Gli1 and Gli3 , while Gli2 is not affected . These selective changes , in turn , lead to the up-regulated expression ( red ) of Srebf1 and members of the Ppar family and eventually other TFs . ( C ) These secondary TFs contribute to a complex regulatory network leading to the up-regulation of lipogenic enzymes , which ultimately cause steatosis . The dashed lines indicate weaker effects than the solid lines . DOI: http://dx . doi . org/10 . 7554/eLife . 13308 . 032 Further support for a highly complex but coordinated response to down-regulation of Hh pathway activity in hepatocytes that shifts the normal balance of lipid metabolism in favour of liponeogenesis is provided by the observed response of the genes encoding the central lipogenic enzymes . For instance , increased Acaca but not Acacb expression ( Chow et al . , 2014 ) and increased Elovl6 expression , particularly when coupled to the down-regulation of Elovl3 , should trigger lipogenesis ( Jakobsson et al . , 2005 ) . The genome-wide dimension of the role of Hh signaling in controlling central transcriptional events of NAFLD formation and progression was revealed by our microarray studies . In particular , the GSEA revealed that many regulated genes in SLC-KO mice belong to the top 50 abundant changed genes associated with NAFLD tested in 100 unique inbred mouse strains ( Hui et al . , 2015 ) . Intriguingly , one of these changes concerned the increased expression of Pnpla3 which has frequently been found to be associated with steatosis where it acts as a downstream gene of SREBP-1c to promote lipid accumulation ( Smagris et al . , 2015 ) . Our observation that liponeogenesis is further increased in the presence of high glucose concentrations adds another piece of the puzzle . Thus , Hh signaling may also influence carbohydrate metabolism in the presence of excess glucose , possibly by increasing Chrebp expression . Another important result of our investigation is that the Hh pathway appears to play an important role in liver zonation , which is shown by the completely changed zonation of FASN in response to the loss of Smo . This result points to a deeper impact of this pathway not only on lipid metabolism but also on Metabolic Zonation as a whole ( Schleicher et al . , 2015 ) . Most importantly , the strong enhancement of SREBP1 und PPRAG protein expression in the SLC-KO mice occurred in the pericentral zone which is in line with published data from mice fed with a high fat diet ( Inoue et al . , 2005; Liu , 2012 ) . For SREBP1 it was shown that this protein has binding sites at -7000 bp and -500 bp of the Fasn gene ( Amemiya-Kudo et al . , 2002; Morishita et al . , 2014 ) . Furthermore it was also shown that overexpression of SREBP1c , an isoform of SREBP1 , lead to significant upregulation of Fasn gene expression in hepatocytes ( Dentin et al . , 2004 ) . Obviously , SREBP1 does not act alone but in combination with several other TFs such as the NFYs and SP1 ( Reed et al . , 2008 ) . To identify the downstream signaling pathways that lead to the observed lipid accumulations in hepatocytes , we extended our previous work on the Gli TFs , which form a self-stabilizing network in parenchymal cells ( Schmidt-Heck et al . , 2015 ) . This network mediates the prominent up-regulation of lipogenic TFs ( e . g . Ppara , Pparg and Srebf1 ) when hepatocytes are treated with Gli3 siRNA ( Schmidt-Heck et al . , 2015 ) . In our current study , we revealed additional Gli-dependent transcriptional effects . Even more importantly , we found that the expression of Gli1 and Gli3 were significantly reduced in the SLC-KO mice . This indicates that a specific transcriptional signature within the Gli TF network which can be termed the 'steatotic Gli-code' lead to the observed steatosis in vivo and in vitro . This signature is characterized by reduced expression of Gli1 and Gli3 , while Gli2 is unchanged ( or slightly increased ) , which can be abbreviated as Gli1/Gli2/Gli3 . Using different mouse models with strong steatosis by hyperphagy or leptin deficiency ( e . g . MC4R and ob/ob mice ) , we observed the same direction of the regulation of the Gli TF’s . This was also supported by measurements of Gli expression in liver samples from human patients with steatosis . These results indicate that the postulated “steatotic Gli-code” is universal across several strains and species promoting lipid accumulation in hepatocytes regardless of the manner of the etiology of steatosis . Based on the different functional possibilities of the GLI proteins , Ruiz I Altaba has originally postulated the concept of the 'Gli-code' , which describes the combinatorial and cooperative function of the GLI transcription factors ( Ruiz i Altaba , 1999; Ruiz i Altaba et al . , 2003; 2007 ) . This concept has proven useful in connecting different levels of Hh signaling with the regulation of cell fates and cancer ( Stecca and Ruiz i Altaba , 2010 ) . Because the GLI proteins are also regulated by other signaling pathways , e . g . EGFR-MEK/ERK , RAS/AKT , TGF-beta and Wnt/beta-catenin signaling ( for review see [Fernandez-Zapico , 2008] ) , the 'Gli-code' may reflect the outcome of integrating different signaling pathways . Regarding lipid metabolic processes , Suh and co-workers found similar changes in the expression of the Gli factors during adipogenic induction of 3T3-L1 fibroblasts ( Suh et al . , 2006 ) . These similarities lead us to hypothesize that the transition from the normal GLI-code ( GLI1/GLI2/GLI3 ) to the postulated 'steatotic Gli-code' ( GLI1/GLI2/GLI3 ) provides a novel paradigm for the regulation of lipid metabolism in the liver ( Figure 10B ) . In case of down-regulated Gli1 and Gli3 , prominent lipogenic transcription factors ( e . g . Ppara/g , Srebf1 ) are up-regulated and influence an entire network of several genes associated with lipid metabolism which finally ends up with hepatic steatosis ( Figure 10B ) . This concept is considerably supported by the finding that the combined overexpression of GLI1 and GLI3 ( but not of each GLI factor alone ) is able to significantly lower the lipid content of hepatocytes from steatotic ob/ob mice . In order to illustrate the complexity of the downstream gene sets affected by the GLI factor network the 55 overrepresented GO categories identified in the ClueGO analysis were reduced to 13 groups by grouping similar GO categories . Figure 10C shows the 7 selected groups ( out of the 13 ) that are related to lipid metabolism only . Besides direct GLI-dependent regulation also possible indirect mechanisms may be included . Using ChIP experiments we could provide first hints for a direct regulation of Srebf1 gene expression via binding sites for the GLI3 protein in its promotor region . However , given the complex interactions of the GLI factors with regulatory binding sites of target genes and the lack of reliable antibodies for full length , truncated and/or phosphorylated GLI factors the results of the CHIP analysis deserve confirmation by detailed analyses using different molecular techniques . Importantly , work by Gurdziel and coworkers ( Gurdziel et al . , 2016 ) published during revision of our manuscript supports our findings by identifying a comprehensive library of enriched GLI binding motifs in which among else the promotor regions of Srebf1 and Pnpla3 are predicted to harbor Hh activity via GLI binding sites ( Gurdziel et al . , 2016 ) . Even the down-regulation of IGF1 which we observed in the SLC-KO mice and the closely related SAC mice ( Matz-Soja et al . , 2014 ) may cause indirect alterations in the transcriptome in the liver and in other organs . Whether these effects lead to hepatic changes in the expression of Chrebp which does not seem to be a direct target of the GLI factors remains to be established . The relevance of our findings for the human liver is emphasized by our observation that increased lipid droplets occur in human hepatocytes when the Hh pathway was inhibited by cyclopamine . This fits to similar results reported for HepG2 cells . Furthermore , the obvious down-regulation of Gli1 and Gli3 in human livers with steatosis supports the universality of the postulated 'steatotic Gli-code' and reveals new insights in the regulation of hepatic lipid metabolism . In addition , the strong up-regulation of Pnpla3 also correlates with the expression changes observed in human NAFLD ( Romeo et al . , 2008 ) . Finally , we would like to point out that the already published changes in the IGF-Axis due to impairment of Hedgehog signaling ( Matz-Soja et al . , 2014 ) are even a hallmark for human steatosis . Both , the dramatic down-regulation of hepatic Igf1 expression and the upregulation of Igfbp1 , which were reflected in changes of the serum levels of these proteins , were observed in several clinical studies and are strongly associated with NAFLD and the metabolic syndrome ( Völzke et al . , 2009; Alderete et al . , 2011; Mallea-Gil et al . , 2012 ) . The central role of the Hh signaling pathway in liver steatosis is emphasized by our findings in vitro that activation of Hh signaling via the Sufu siRNA , recombinant SHH , the Hh agonist SAG and the combined overexpression of GLI1 and GLI3 was able to significantly reduce the expression of Srebf1 and mitigate the accumulation of lipids in steatotic hepatocytes . These findings suggest possible new therapeutic strategies for NAFLD . However , a limiting factor may be that high over-activation of Hh signaling may enhance the risk for carcinogenesis . Collectively , our study reveals an important role for Hh signaling in regulating hepatic lipid metabolism and its zonation . These findings suggest a new paradigm for the development of liver steatosis . The potential of impaired Hh signaling to trigger steatosis independent of nutritional changes suggests that malfunctions in this pathway may pave the way for the development of NAFLD long before other cues may lead to further aggravation . When the study was being designed the appropriate sample size was computed using the Sigma Plot software with a power of 0 . 8 and alpha = 0 . 05 . In addition we used also our experience from previous studies dealing with liver related investigations with the transgenic and even the non-transgenic mice ( Matz-Soja et al . , 2014; Schmidt-Heck et al . , 2015 ) . For isolated hepatocytes from transgenic SLC-WT , SLC-KO and non-transgenic C57BL6/N mice the same procedure was used . For assays with cultured hepatocytes ( siRNA experiments , use of agonists and inhibitors ) we used the paired t-Test with a power of 0 . 8 and alpha = 0 . 05 to calculate the required sample size . Most experiments were performed 2–3 times with different numbers of biological replicates ( animals ) of n = 4–17 as indicated in each figure . The number of technical replicates depended on the specific type of measurement and was duplicate in most cases and triplicate in some few experiments . Outliers were analyzed using the ROUT method on GraphPad Prism 6 software . The aggressiveness of the test was set to 0 . 2 % . The cleaned data was used for the subsequent statistical analyzes and data depiction in the figures . By using material from transgenic mice with Smo deletion , the amount of Smo mRNA was quantified via qRT-PCR . When the remaining expression in liver and hepatocytes from SLC-KO mice was more than 50 % of the mean of the SLC-WT mice , we excluded the sample . Only samples where the expression of Smo was lower than 50 % were included . The mice were maintained in a pathogen-free facility on a 12:12 hr LD cycle , according to the German guidelines and the world medical association declaration of Helsinki for the care and safe use of experimental animals . The animals had free access to regular chow ( sniff M-Z V1124-0 composed of 22 . 0 % protein , 50 . 1 % carbohydrate , 4 . 5 % fat; usable energy: 13 . 7 kJ/g; ssniff Spezialdiäten GmbH , Germany ) and tap water throughout life . Before sacrifice ( between 9 and 11 am ) , the mice were starved for 24 hr and re-fed with regular chow for 12 hr to obtain a synchronized feeding state . Primary hepatocytes were isolated from male transgenic and C57BL/6N mice using collagenase perfusion of the liver , as previously described ( Gebhardt et al . , 2010; Matz-Soja et al . , 2014 ) . The cell suspension was cleared of the non-parenchymal cells by differential centrifugation . Finally , the hepatocytes were suspended in Williams Medium E containing 10 % fetal calf serum and the indicated additions and were plated in 6-well or 12-well plates that had been pre-coated with collagen type I ( Klingmuller et al . , 2006 ) . After 4 hr , the cells were switched to serum-free medium , which was used throughout cultivation . Cryopreserved human hepatocytes were purchased from TebuBio ( Germany ) . They were thawed according to existing protocols ( Klingmuller et al . , 2006 ) and cultured in 6-well plates at the same cell density and culture conditions as the mouse hepatocytes , except for the omission of dexamethasone after 4 hr . The hepatocytes were incubated in the presence of 10 µM cyclopamine or vehicle ( 0 . 1 % DMSO ) control for 72 hr . In some experiments , 300 nM SAG ( N-Methyl-Nʹ- ( 3-pyridinylbenzyl ) -Nʹ- ( 3-chlorobenzo[b]thiophene-2-carbonyl ) -1 , 4-diaminocyclohexane ) ( Sigma-Aldrich , Germany ) was used and DMSO was used as the vehicle . In other experiments , 0 . 25 µg of recombinant SHH ( R&D Systems , USA ) or an equivalent amount of vehicle ( PBS , containing 0 . 1 % bovine serum albumin ) were used . Human liver tissues for mRNA expression analysis were obtained from patients without and patients with simple steatosis . Steatosis was histologically examined by a pathologist . Surgery was done because of hepatic metastases of extrahepatic tumors and only healthy ( non-tumorous ) tissue was used . Experimental procedures were performed according to the guidelines of the charitable state controlled foundation Human Tissue and Cell Research ( HTCR ) , with the written informed patient's consent approved by the local ethical committee of the University of Regensburg ( 12-101-0048 ) . Blood samples were taken from the beating heart of anesthetized mice . Insulin was detected in the serum using the ELISA kits from DRG Instruments ( EIA 3439; Mediagnost , Germany ) . The serum enzyme activities of ASAT ( aspartate aminotransferase ) , ALAT ( alanine aminotransferase ) and GLDH ( glutamate dehydrogenase ) and the serum concentrations of ketone bodies were measured with an automated analyzer ( Roche modular ) using standardized assays ( Roche , Germany ) . Lipoproteins were isolated by sequential ultracentrifugation from 60 μl of plasma at densities ( d ) of <1 . 006 g/ml ( very low density lipoprotein , VLDL ) , d≤1 . 063 g/ml ( intermediate density lipoprotein and low density lipoprotein , LDL ) and d>1 . 063 g/ml ( high density lipoprotein , HDL ) in a LE-80K ultracentrifuge ( Beckman , Germany ) as described . Cholesterol in the lipoprotein fractions was determined enzymatically using a colorimetric method ( Roche , Germany ) . The blood glucose levels were determined using a blood glucose meter ( Freestyle Mini , Abbott , Germany ) . Paraffin sections ( 3 µm ) were stained with H&E to visualize the tissue histology . For electron microscopy , separate small pieces of liver tissue ( approx . 1 mm3 ) were fixed in 2 . 5% glutaraldehyde in PBS ( pH 7 . 6 ) . Further processing was performed as previously described ( Gebhardt , 1992 ) . Frozen sections were cut at 6 µm for the quantitative and qualitative lipid analysis . A Leica DM5000B microscope ( Germany ) was used to examine the stained liver sections using a DCF 320 color camera and bright field settings . The fluorescent images were digitally captured using a DFC 350FX fluorescence camera . The fat red staining of the cryostat sections was assessed by bright-field microscopy . Digital images were captured from three contiguous microscopic fields per section , covering the entire parenchyma between large vessels . Using the UTHSCA Image Tool 3 . 0 software ( University of Texas Health Science Centre , USA ) , the images were transformed to a binary format after appropriate thresholding . The same threshold was applied to all images from all sections . The fat red staining was quantified in the cultured hepatocytes as previously described ( Nunnari et al . , 1989 ) . The values were normalized to amount of cellular protein quantified by Bradford assay ( Bradford , 1976 ) . For Nile red staining , a 200 nM solution in PBS was prepared from a 1 mM stock solution in DMSO and added directly to the fixed cells . After 20 min incubation at room temperature , the cells were washed in PBS . The nuclei were counterstained with DAPI . Immunohistochemistry on paraffin sections ( 3 µm ) was performed as previously described ( Zellmer et al . , 2009 ) . The sections were boiled ( 3 × 5 min ) in buffer ( 0 . 01 M sodium citrate , pH 6 . 0 ) and the slides were incubated for 30 min in 5% goat serum ( Sigma-Aldrich , Germany ) to block nonspecific binding . The following antibodies were used: rabbit anti-FAS ( 1:400 , Cell Signaling Technology , USA ) , rabbit anti-Cre ( 1:4000; Abcam , Cambridge ) , anti-GLI3 ( 1:1000; GeneTex , USA ) , anti-GLI1 ( 1:1000; GeneTex , USA ) , anti-SREBP1 ( 1:200; Abcam , Cambridge ) anti-PPARG ( 1:250 , ThermoFisher Scientific , Germany ) , biotinylated goat anti-rabbit IgG ( Millipore , Germany ) and Extravidine ( Sigma-Aldrich , Germany ) . POD and counterstaining with hematoxylin were performed as previously described ( Zellmer et al . , 2009 ) . In total the proteins from isolated hepatocytes from 9 SLC-WT and 9 SLC-KO mice were separated by 8% SDS-PAGE and transferred onto PVDF membranes ( Roth , Germany ) , and incubated overnight at 4°C in blocking buffer ( 50 mM Tris HCl , pH 7 . 4 , 150 mM NaCl , 0 , 1% Tween-20 , 5% milkpowder ) . GLI3 antibody ( GeneTex , USA ) was diluted 1:20000 in Solution 1 from SignalBoost™ Immunoreaction Enhancer Kit ( Calbiochem , Germany ) ; ACTB antibody ( Abcam , Cambridge ) was diluted 1:5000 in 1% blocking buffer and incubated over night at 4°C . Blots were subsequently incubated with secondary anti rabbit POD antibody ( Sigma , Germany ) . For the GLI3 Blots the secondary antibody was diluted 1:25000 in Solution 2 from the SignalBoost™ Immunoreaction Enhancer Kit . Chemiluminescent was used for detection . Densitometry quantification was performed with the Phoretix 1D Quantifier ( Nonlinear dynamics , USA ) . The biosynthesis of non-saponifiable lipids ( sterols ) and free fatty acids was performed as described ( Gebhardt et al . , 2010 ) , with minor modifications . Briefly , the hepatocyte cultures were incubated in 1 ml of culture medium supplemented with 9 µM 1[14C]acetate ( 2 . 04 GBq/mmol; Amersham International ) for 2 hr . Then , the cells were washed with 0 . 9% NaCl and lysed by incubation in 1 ml KOH ( 15% ) overnight at 37°C . The samples were saponified at 95°C for 90 min and the neutral lipids were extracted 3 times with a total volume of 8 ml of petroleum ether . The residual aqueous phase was acidified with 500 µl HCl ( conc . ) , and the protonated fatty acids were extracted 3 times with a total volume of 8 ml of petroleum ether . Glycolysis was determined as described by Probst et al . ( Probst et al . , 1982 ) with minor modifications . Briefly , the hepatocyte cultures were incubated in 1 ml of Hanks buffered solution supplemented with 20 mM NaHCO3 , and 1 . 9 µM [14C ( U ) ]glucose ( 9 . 7 GBq/mmol , Hartmann Analytic GmbH , Germany ) for 2 hr . Then , 200 µl of the culture supernatant were applied to a Dowex 1X8 ( formiate form ) column and eluted with 7 ml sodium formiate ( 0 . 4 M ) . After adding Ultima-Gold™ AB solution ( PerkinElmer GmbH , Germany ) , the radioactivity in each sample was counted in a liquid scintillation counter ( Tri-carb 2500TR ) . The glycogen determination was based on the microassay described by Gomez-Lechon et al . ( Gomez-Lechon et al . , 1996 ) . Glycogen from bovine liver was used for calibration . The DNA of the cultured cells was measured by a fluorimetric method adapted to the 96-well format as previously described ( Gomez-Lechon et al . , 1996 ) . The Hoechst 33 , 258 dye was replaced by DAPI ( 4’ , 6-diamidino-2-phenylindole ) at a concentration of 2 . 14 µg/ml in 40 mM Tris buffer ( pH 7 . 0 ) containing 2 M NaCl and 1 mM EDTA . Fluorescence was measured using a microplate reader ( Lumat LB 9501 Luminometer , Berthold , Germany ) . The protein concentrations were determined with the Bradford assay according to the standard microplate protocol ( Bradford , 1976 ) . The hepatocytes were broken down using a TOS-UCD-200-EX Bioruptor to measure their ATP content . After centrifugation at 10 , 000 rpm at 4°C , the ATP content was determined using the CellTiter-Glo® Luminescent Cell Viability Assay according to the manufacturer’s instructions . For normalization , the protein concentrations were determined using the Bradford assay according to the standard microplate protocol ( Bradford , 1976 ) . The mitochondria were isolated from 10–25 mg of liver tissue to obtain a functional , purified , and intact mitochondrial fraction . The tissue was homogenized 3–4 times with a Teflon-on-glass Potter Elvehjem in 10 mM Tris-HCl ( pH 7 . 8 ) buffer with 0 . 2 mM EDTA and 0 . 25 M sucrose . The activity of the respiratory chain complexes was assessed as described by Claus et al . ( Claus et al . , 2013 ) . After isolation of hepatocytes or tissue , the material was immediately frozen in liquid nitrogen and stored at -80°C up to 4–8 weeks . The total RNAs from the hepatocytes and tissues were extracted using TRIzol ( VWR , Germany ) , and the RNeasy Lipid Tissue mini Kit ( Qiagen , Germany ) was used for the adipose tissue . The RNA was quantified by using a NanoTrop ( VWR , Germany ) . 20 µl cDNA was prepared using 1 µg of RNA and oligo ( dt ) primers in combination with the IM Prom II reverse transcriptase ( Promega , Germany ) for each sample according to the manufacturer’s instructions . For each gene , specific intron-spanning primers ( Supplementary file 1B ) were designed using the online tools Universal ProbeLibrary Probe-Finder software , Perl Primer and Primer 3 . Therefore the RNA treatment with DNase was renounced . The levels of all mRNA transcripts were determined in duplicate by qRT-PCR using the LightCycler 2 . 0 Instrument and the LightCycler FastStart DNA Master PLUS SYBR Green I ( Roche , Germany ) according to the manufacturer’s instructions . Using the standard curve method , the absolute amount of the specific PCR products for each primer set was quantified . Actb ( beta Actin ) was amplified from each sample for normalization as reference gene . The knockdown of Gli1 , Gli2 and Gli3 in isolated hepatocytes from non-transgenic C57BL/6N mice , including the specific siRNA primers for the Gli factors , was performed as previously described ( Schmidt-Heck et al . , 2015 ) . For Sufu knockdown , a specific siRNA and the respective nonsense oligo control siRNA were purchased from Invitrogen , Germany ( Supplementary file 1C ) . The hepatocytes were seeded at a density of 100 , 000 cells per well on 12-well plates . After 4 h , the cells were transfected with the Sufu siRNA ( 10 nM ) with INTERFERin ( VWR , Germany ) according to the manufacturer’s instructions . 24 hr after transfection , the medium was changed , and fresh medium without the siRNA was added . The changes in gene expression were analysed by qRT-PCR at different time points . After isolation , hepatocytes from three male ob/ob mice at the age of 12 weeks were cultivated at 0 . 25 Mio . cells per well in 6-well plates in 1 . 5 ml medium ( William’s Medium E enriched with 10% fetal calf serum , 2 mM L-glutamine , 100 nM dexamethasone and Pen/Strep ) . After 2 h medium was changed Twenty four hours after seeding , transfection was performed using the jetPEI transfection reagent ( VWR , Germany ) according to the manufacturer's instruction with 1 . 0 μg DNA per well for each single plasmid . In the chase of the transfection of GLI1 and GLI3 together , the DNA amount of each plasmid was 0 . 5 µg DNA per well . For stable gene expression of Gli1 , Gli2 and Gli3 the pFN1A HaloTag®CMV Flexi® Vector was used which was designed from Promega ( Germany ) in cooperation with the Kazusa DNA Research Institute ( KDRI ) ( Japan ) ( Nagase et al . , 2008 ) . For control , only the DNA of the vector was transfected into the hepatocytes ( MOCK transfection ) . RNA isolation and fat red quantification was performed 72 hr post transfection . Primers for human Gli expression are listed in Supplementary file 1D . To determine the binding of the transcription factor GLI3 to the promoter region of Srebf1 , Ppara and Fasn , the Simple- ChIP Plus Enzymatic Chromatin IP Kit ( Magnetic Beads ) ( Cell Signaling , Germany ) was used according to the manufacturer’ s instructions . Freshly isolated hepatocytes from three male C57BL/6N mice were pooled , washed and cross-linked with 37% formaldehyde . As positive experimental control the Histone H3 ( D2B12 ) XP Rabbit mAb ( #2729 ) was used . For negative control the normal Rabbit IgG was used . The cross-linked GLI3 DNA complex was precipitated with the goat anti mouse GLI3 antibody ( 20 μg ) ( R&D , Germany ) . The quantification analysis was performed using qRT-PCR with 2 μl of each DNA sample and specific primers listed in Supplementary file 1E . Primer pairs for the putative GLI3 binding site in the Srebf1 , Ppara and Fasn promoter region were designed using Primer-BLAST software . The range of interest in the promoter regions ( Fasn: ~ − 1767 bp; Ppara: -2674 bp; Srebf1: -1156 bp ) was obtained by the MotifMap analysis ( Daily et al . , 2011 ) . The primers for Actb were used to analyse the unspecific DNA-antibody binding . Rpl30 primers ( provided in the SimpleChIP Plus Enzymatic Chromatin IP Kit ) were used as a positive control for the histone H3 antibody precipitation . Microarray analysis was conducted using isolated hepatocytes from four independent SLC-WT and four SLC-KO mice , at the microarray core facility of the Interdisziplinäres Zentrum für klinische Forschung ( IZKF ) Leipzig ( Faculty of Medicine , Leipzig University ) described by Zellmer et al . ( Zellmer et al . , 2009 ) . Briefly , GeneChip Mouse Genome 430 2 . 0 Arrays ( Affymetrix ) were used for hybridization . The gene expression data were pre-processed by Probe-level Linear Models using the ‘affyPLM’ packages of the Bioconductor Software . To explore the biological function of differentially expressed genes ( DEGs ) the ClueGO ( Bindea et al . , 2009 ) Cytoscape plugin was applied to identify overrepresented GO categories and possible correlation between these categories . In addition , the ClueGO plugin was used to reduce redundancy by combining the significantly overrepresented categories to new groups . The raw data for the micro array analysis is available at the temporary access link: http://seek . virtual-liver . de/data_files/3403 ? code=FC%2FJdE%2Ba94vTyN%2FcnSjG0uUL9elSQP4huKDrpFbV The STRING database version 10 ( http://string-db . org ) was used to explore the interactions between the studied genes/proteins related to steatosis and Hh signaling ( Szklarczyk et al . , 2015 ) . A hypothetical network was constructed with four different data sources: known co-expression of the inputted proteins , experimental data , the results from various databases and a scan of published scientific abstracts . A high confidence score ( 0 . 7 ) was chosen . Values are expressed as means ± standard error of the mean ( SEM ) . Statistical evaluations of data from different mice were made by unpaired t-test . For the RNAi experiments the paired student’s t-test was used . The null hypothesis was rejected at the p<0 . 05 ( * ) ; p<0 . 01 ( ) and p<0 . 001 ( * ) levels .	The liver is one of the main organs responsible for processing everything that mammals eat and drink . Nutrients absorbed by the gut like sugars and lipids ( fats ) are processed by the liver and are stored or distributed to provide energy to other organs . Sometimes these metabolic processes become unbalanced . This can lead to lipids accumulating in the liver – a process known as steatosis , which is a feature of human non-alcoholic fatty liver disease . In organs like the liver , cells are instructed how to behave via signaling pathways . A protein outside the cell signals to specific proteins inside , which switch on a set of target genes . One such pathway is the Hedgehog pathway , which primarily regulates tissue regeneration and the development of embryos . A component of this pathway is the Smoothened gene , which indirectly switches on proteins called GLI factors that regulate metabolic genes , including those involved in lipid metabolism . The Hedgehog pathway has been found to control the metabolism of lipids in fat tissue but it is not known whether it is important for lipid metabolism in the liver . Matz-Soja et al . investigated this possible role of the Hedgehog pathway in the liver using mice with a Smoothened gene that could be deleted specifically in that organ . This deletion disrupted Hedgehog signaling and led to lipids accumulating in the liver and eventually to steatosis . These changes were associated with an increase in the amounts and activityof several enzymes ( and the proteins that regulate these enzymes ) that help to synthesize lipids . Steatosis was also associated with low amounts of two of the three GLI factors; indeed , this seems to be key for triggering problems with lipid metabolism . Human livers with steatosis showed the same changes in levels of the GLI factors . Increasing the amount of GLI factors in liver cells taken from mice with steatosis reduced the accumulation of lipids and brought lipid metabolism back to its normal balance . A focus of future studies will be to understand how the Hedgehog signaling pathway interacts with other signaling pathways known to regulate liver lipid metabolism , such as insulin signaling . This knowledge will help clinicians to design new treatments for lipid-associated diseases like non-alcoholic fatty liver disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2016	Hedgehog signaling is a potent regulator of liver lipid metabolism and reveals a GLI-code associated with steatosis
The budding yeast Saccharomyces cerevisiae is a long-standing model for the three-dimensional organization of eukaryotic genomes . However , even in this well-studied model , it is unclear how homolog pairing in diploids or environmental conditions influence overall genome organization . Here , we performed high-throughput chromosome conformation capture on diverged Saccharomyces hybrid diploids to obtain the first global view of chromosome conformation in diploid yeasts . After controlling for the Rabl-like orientation using a polymer model , we observe significant homolog proximity that increases in saturated culture conditions . Surprisingly , we observe a localized increase in homologous interactions between the HAS1-TDA1 alleles specifically under galactose induction and saturated growth . This pairing is accompanied by relocalization to the nuclear periphery and requires Nup2 , suggesting a role for nuclear pore complexes . Together , these results reveal that the diploid yeast genome has a dynamic and complex 3D organization . The genome is actively organized in the nucleus in both space and time , and this organization impacts fundamental biological processes like transcription , DNA repair , and recombination ( Taddei et al . , 2010 ) . The budding yeast S . cerevisiae has been a useful model for studying eukaryotic genome conformation and its functional implications ( Taddei et al . , 2010 ) . The predominant feature of yeast 3D genome organization is its Rabl-like orientation ( Jin et al . , 1998 ) ( Figure 1A ) : during interphase , the centromeres cluster at one end of the nucleus , attached to the spindle pole body , and chromosome arms extend outward toward the nuclear periphery where the telomeres associate ( Schober et al . , 2008; Therizols et al . , 2010 ) , like in anaphase . In addition , the ribosomal DNA array forms the nucleolus , opposite the spindle pole ( Yang et al . , 1989 ) , splitting chromosome XII into two separate domains that behave as if they were separate chromosomes . This organization largely persists through the cell cycle ( Jin et al . , 1998 ) and even in stationary phase , albeit with increased telomere clustering and decreased centromere clustering ( Guidi et al . , 2015; Rutledge et al . , 2015 ) . 10 . 7554/eLife . 23623 . 003Figure 1 . Diverged hybrids provide a genome-wide view of diploid chromosome conformation . ( A ) Schematic of the Rabl-like orientation . CEN , centromere; SPB , spindle pole body; TEL , telomere; NUC , nucleolus . ( B ) Hi-C contact map for saturated S . cerevisiae and S . uvarum mixture control , at 32 kb resolution . Each axis represents the S . cerevisiae genome followed by the S . uvarum genome in syntenic order , separated by a black line . Ticks indicate centromeres . Odd-numbered centromeres are labeled . Rows and columns with insufficient data are colored grey . ( C ) Hi-C contact map for saturated S . cerevisiae x S . uvarum hybrid , as in ( B ) . ( D ) The portion of the map outlined in black in ( C ) , is enlarged with annotated features of the Rabl-like orientation . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 00310 . 7554/eLife . 23623 . 004Figure 1—figure supplement 1 . Mappability of hybrid yeast genomes . ( A ) Proportion of simulated reads from each hybrid ( 80 bp for interspecific and 150 bp for intraspecific hybrids , in 10 bp windows across the reference genome ) that can be remapped correctly to the reference with a mapping quality score ( MAPQ ) of at least 30 . ( B ) Number of reads ( in thousands ) from separate S . cerevisiae ( left ) or S . uvarum ( right ) Hi-C libraries mapping to each 32 kb genomic bin in the S . cerevisiae x S . uvarum hybrid reference genome . x ticks indicate centromeres; odd-numbered centromeres are labeled . ( C ) Same as ( B ) for S . cerevisiae ( left ) and S . paradoxus ( right ) mapping to S . cerevisiae x S . paradoxus . ( D ) Same as ( B ) for S . cerevisiae Y12 ( left ) and S . cerevisiae DBVPG6044 ( right ) mapping to Y12 x DBVPG6044 . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 00410 . 7554/eLife . 23623 . 005Figure 1—figure supplement 2 . Mixture control experiments for S . cerevisiae x S . paradoxus and S . cerevisiae x S . cerevisiae hybrids . ( A ) Hi-C contact map for a mixture of S . cerevisiae and S . paradoxus in exponential growth , at 32 kb resolution . Each axis represents the S . cerevisiae genome followed by the S . paradoxus genome , in syntenic order , separated by a black line; tick marks indicate centromere positions , and odd-numbered chromosome centromeres are labeled . Intrachromosomal interactions are outlined by black squares along the diagonal . Rows and columns with an average of less than one read pair per bin were filtered out , and are colored grey . ( B ) Hi-C contact map for a mixture of S . cerevisiae Y12 and S . cerevisiae DBVPG6044 haploids in exponential growth , as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 00510 . 7554/eLife . 23623 . 006Figure 1—figure supplement 3 . Reproducibility of Hi-C across replicates and restriction enzymes . ( A and B ) , Hi-C contact maps at 32 kb resolution for the S . cerevisiae x S . uvarum hybrid in saturated cultures , using the restriction enzyme Sau3AI ( A ) or HindIII ( B ) . Each axis represents the S . cerevisiae genome followed by the S . uvarum genome , in syntenic order , separated by a black line; ticks indicate centromeres , and odd-numbered centromeres are labeled . Rows and columns with insufficient data are colored grey . ( C and D ) Same as ( A ) and ( B ) for two biological replicates of S . cerevisiae x S . uvarum hybrid in exponential growth , using the restriction enzyme Sau3AI . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 00610 . 7554/eLife . 23623 . 007Figure 1—figure supplement 4 . Revisions to S . paradoxus and S . uvarum reference genomes . ( A ) Start and end positions of homologous genes on the original S . cerevisiae and S . paradoxus reference genomes . Each point represents the start or end of a homologous gene pair . Horizontal and vertical lines indicate starts of chromosomes , and ticks indicate centromere positions . Odd numbered chromosomes are numbered . Red arrows indicate unexpected rearrangements . ( B ) Raw contact map for S . paradoxus chromosome IV using the original reference genome . Arrows above the heat map represent segments of the chromosome , labeled A-D in order and orientation of synteny . ( C ) Same as ( A ) for S . cerevisiae and S . uvarum . Red arrows indicate an unexpected rearrangement in chromosome III , and S . uvarum chromosomes X and XII , which are homologous to S . cerevisiae chromosome XII and X , respectively . ( D ) Same as ( B ) for S . uvarum chromosome III . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 007 Genome-wide chromosome conformation capture methods like Hi-C have both confirmed these microscopy observations and permitted systematic analyses of the functional clustering of genomic elements like tRNA genes and origins of replication ( Duan et al . , 2010 ) . However , multiple studies have argued that a simple volume-exclusion polymer model of chromosomes in a Rabl-like orientation is sufficient to explain microscopy and Hi-C data of the budding yeast genome ( Tjong et al . , 2012; Wong et al . , 2012 ) , at least in haploids grown under standard lab conditions . These studies have argued that even the functional clustering that is observed may simply be a consequence of the Rabl-like orientation coupled with biases in the chromosomal positions of genomic elements , rather than active molecular interactions ( Rutledge et al . , 2015; Tjong et al . , 2012 ) . Although this simplicity is attractive , diploidy and variable environmental conditions may add complexity to yeast genome conformation . In diploid yeast , homologous chromosomes can pair not only in meiosis , but also during mitotic growth ( Burgess and Kleckner , 1999; Burgess et al . , 1999; Dekker et al . , 2002; Weiner and Kleckner , 1994 ) , as they do in Drosophila ( Metz , 1916 ) . However , the extent of mitotic homolog pairing has been debated due to discrepancies between studies ( Barzel and Kupiec , 2008 ) . One explanation for these discrepancies is potential artifacts in the microscopy methods used to detect pairing . In fluorescence in situ hybridization ( FISH ) , signal loss can lead to false inference of colocalization ( Lorenz et al . , 2003 ) . It has also been suggested that tagging of genomic loci with repetitive arrays of GFP for live-cell imaging can directly cause pairing via GFP dimerization ( Mirkin et al . , 2014 ) . Furthermore , the Rabl-like orientation alone can create the appearance of homolog pairing if not controlled for , by juxtaposing chromosomal loci at the same distance from centromeres , including homologous loci ( Lorenz et al . , 2003 ) . Discrepancies in the extent of pairing between studies might be attributable to variation in pairing strength across the genome; however , mitotic homolog pairing has only been examined at a few loci . In both haploid and diploid yeasts , genome conformation can also change in response to environmental conditions . Genes that respond to signals like galactose induction ( GAL1 , HXK1 ) , inositol starvation ( INO1 ) , oxidative stress , and heat shock ( HSP104 ) have been observed by microscopy to relocate to the nuclear periphery upon activation via interactions with nuclear pore complexes ( Ahmed et al . , 2010; Brickner and Walter , 2004; Brickner et al . , 2016; Casolari et al . , 2004; Dieppois et al . , 2006; Dultz et al . , 2016; Taddei et al . , 2006 ) . Nuclear pore interactions can mediate clustering of genes that share Gene Recruitment Sequences , including homologous alleles ( Brickner et al . , 2015 , 2012 , 2016; Randise-Hinchliff et al . , 2016 ) , and even impact the conformation of chromosomes well beyond the induced gene ( Dultz et al . , 2016 ) . However , because such changes in conformation are primarily detected by microscopy , systematic studies of how inducible gene relocalization impacts global genome conformation have been lacking . Here , we present a genome-wide analysis of diploid chromosome conformation in budding yeasts in multiple environmental conditions . We utilize hybrid yeasts resulting from mating diverged yeast species to perform homolog-resolved Hi-C . Our genomic approach allows us to more fully account for the Rabl-like orientation in assessing the extent of homolog pairing , and to detect whether some regions of the genome pair more strongly than others . We find that the strength of pairing varies across both growth conditions and the genome . Notably , the homologous HAS1-TDA1 alleles on chromosome XIII pair specifically in galactose induction and saturated growth , but not during exponential growth in glucose . The condition-specific pairing is accompanied by relocalization to the nuclear periphery and in galactose requires Nup2 , a component of the nuclear pore , suggesting a role for the nuclear pore complex . However , the genetic requirements of HAS1-TDA1 relocalization and pairing differ from that of previously known relocalized genes , suggesting a potentially novel mechanism . Together , our results demonstrate the underappreciated complexity of the 3D organization of the yeast genome . We performed Hi-C on interspecific hybrids between diverged Saccharomyces species to obtain the first genome-wide view of chromosome conformation in diploid yeasts . The sequence identity of homologous chromosomes in diploid S . cerevisiae precludes observation of interactions between them using sequencing-based methods . However , divergent Saccharomyces species can form stable hybrids ( González et al . , 2006; Mertens et al . , 2015 ) , e . g . between S . cerevisiae and S . paradoxus ( 90% nucleotide identity in coding regions [Kellis et al . , 2003] ) or its more distant relative S . uvarum ( also known as S . bayanus var . uvarum; 80% nucleotide identity in coding regions [Kellis et al . , 2003] ) . These interspecific hybrids are sufficiently diverged to allow straightforward sequence-level discrimination of homologs ( Figure 1—figure supplement 1A ) but have maintained nearly complete synteny ( Fischer et al . , 2000 ) . For comparison , we also analyzed hybrids between S . cerevisiae strains Y12 and DBVPG6044 , which are much less diverged ( ~99% nucleotide identity ) ( Liti et al . , 2009 ) . We confirmed the minimal impact of mapping and experimental artifacts by mapping Hi-C data from each individual species or strain ( Figure 1—figure supplement 1B–D ) and mixtures thereof ( Figure 1B , Figure 1—figure supplement 2 ) to the hybrid reference genomes . The most prominent features of Hi-C data from diploid yeast are the signatures of a Rabl-like orientation ( Figure 1C , D ) . As in all Hi-C datasets , the contact map exhibits a strong diagonal signal indicating frequent intrachromosomal interactions between adjacent loci . In addition , pericentromeric regions interact frequently with one another , but infrequently with regions far from centromeres , as expected from the clustering of centromeres at the spindle pole body . Telomeric regions also preferentially interact , consistent with their clustering at the nuclear periphery . Finally , the rDNA-carrying chromosomes each behave as two separate chromosomes divided by the nucleolus , with frequent interactions on either side of the rDNA array but not across it . In addition to these previously described phenomena , we observed an off-diagonal line of increased interaction suggestive of homolog pairing ( Figure 1C ) . Homologous loci tend to be closer together than nonhomologous loci in multiple assays , including microscopy ( Burgess et al . , 1999; Weiner and Kleckner , 1994 ) , recombination efficiency ( Burgess and Kleckner , 1999 ) , and chromatin conformation capture ( Dekker et al . , 2002 ) . Mitotic homolog pairing could be the result of transient pairing between homologous nucleosome-free DNA ( Danilowicz et al . , 2009; Gladyshev and Kleckner , 2014 ) or interactions among proteins bound to DNA ( Mirkin et al . , 2014 ) . However , it has also been suggested that the observation of homolog proximity is an artifact of the Rabl-like orientation or microscopy methods ( Lorenz et al . , 2003; Mirkin et al . , 2014 ) . This debate remains unresolved in part due to the targeted nature of previous studies , wherein each pair of homologous loci is only compared to a limited number of nonhomologous loci . To systematically investigate whether homolog proximity can be explained by the Rabl-like orientation , we compared our experimental data from S . cerevisiae x S . uvarum hybrids to simulated data from a volume-exclusion polymer model of the Rabl-like orientation . This model did not explicitly encode homolog pairing ( Tjong et al . , 2012 ) and served as a negative control . We quantified homolog proximity by comparing the frequency of each interaction between a pair of homologous loci to the set of nonhomologous interactions involving either locus ( Figure 2—figure supplement 1 ) . This naive comparison appears to suggest strong homolog proximity in both experiments ( Figure 2A , left panel ) , but in fact , the equally strong signal from the polymer model suggests that the apparent signal is a consequence of the Rabl-like orientation . We therefore controlled for the Rabl-like orientation by restricting comparisons to interactions with loci at a similar distance from the centromere ( at a resolution of 32 kb ) , as previous studies have done ( Burgess et al . , 1999; Lorenz et al . , 2003 ) . Using this approach , we find that the polymer simulation still predicts strong homolog proximity ( Figure 2A , middle panel ) , suggesting that the long-used approach of comparing homologous interactions to nonhomologous interactions at the same centromeric distance may not fully account for the Rabl-like orientation . Polymer models of the Rabl-like orientation suggest that short chromosomes interact preferentially , due to their dual telomeric tethering at the nuclear periphery and centromeric tethering at the spindle pole ( Tjong et al . , 2012 ) . Therefore , we further restricted comparisons to loci on chromosome arms of similar length ( within 25% ) . This additional restriction dramatically reduced the signal of homolog proximity for the polymer model , but not for the experimental data ( Figure 2A , right panel ) . 10 . 7554/eLife . 23623 . 008Figure 2 . Homolog proximity exceeds predicted effects of Rabl-like orientation . ( A ) Violin plot of the distribution of 10 , 000 sampled estimates of genomic homolog proximity ( ratio of homologous to nonhomologous interaction frequencies ) in the S . cerevisiae x S . uvarum hybrid , as a function of increasing comparison stringency ( left to right ) to account for Rabl-like orientation ( Figure 2—figure supplement 1 ) . Saturated culture data are shown in red , exponential growth in blue , and simulated data from a homology-agnostic polymer model in grey . dCEN , distance from centromere . Boxplot indicates median and interquartile range . Whiskers correspond to the highest and lowest points within 1 . 5× interquartile range . ( B ) Variation in homolog proximity across the S . cerevisiae x S . uvarum hybrid genome at 32 kb resolution , in saturated culture ( red ) , exponential growth ( blue ) , and the polymer model ( grey ) . Nonhomologous interactions were restricted to similar centromere distance and chromosome arm length . Bins with insufficient data ( <2 comparisons ) are left blank . Data are plotted by S . cerevisiae genome position . x ticks indicate ends of chromosomes . ( C and D ) Schematics and Hi-C contact maps ( at 32 kb resolution ) of interactions between S . uvarum chromosome XII ( SuXII ) and either S . cerevisiae chromosome XII ( ScXII ) or S . cerevisiae chromosome V ( ScV ) , in wild-type S . cerevisiae x S . uvarum hybrids ( C ) and a strain with a translocation between ScXII and ScV ( D ) , both in saturated cultures . Exponential growth data are shown in Figure 2—figure supplement 2 . Ovals indicate centromeres and slanted lines indicate the locations of rDNA arrays . Double-headed arrows indicate enhanced interactions . Dashed lines indicate translocation breakpoints . ( E ) Violin plot of homolog proximity across conditions , polymer models , and hybrids . S . cerevisiae x S . cerevisiae indicates hybrid between Y12 and DBVPG6044 strains . Calculated as in ( A ) , but excluding chromosome XII and all 32 kb bins that include centromeres . Saturated culture data are shown in red , exponential growth in blue , nocodazole-arrested in green , and polymer models in grey . Boxplot indicates median and interquartile range . Whiskers correspond to the highest and lowest points within the 1 . 5× interquartile range . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 00810 . 7554/eLife . 23623 . 009Figure 2—figure supplement 1 . Schematic of homolog proximity analysis . Representation of how homologous interactions ( black squares ) were compared to various subsets of nonhomologous interactions ( grey squares ) , either including all interactions with either homologous locus ( red squares ) ( A ) , restricted to interactions with loci at a similar centromeric distance , or dCEN ( same number of 32 kb bins ) ( B ) , or restricted to interactions with loci at a similar centromeric distance and on a chromosome arm of similar length ( within 25% ) ( C ) . Left panels show all interactions between the S . cerevisiae and S . uvarum genomes; middle panels show enlarged view of the area outlined in the left panels . Tick marks indicate centromere positions . Right panels represent the nonhomologous interactions being used for comparison; different colors represent nonhomologous chromosomes , and double-headed arrows represent interactions with the locus of interest ( circled ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 00910 . 7554/eLife . 23623 . 010Figure 2—figure supplement 2 . rDNA-carrying chromosomes interact preferentially due to shared tethering . ( A and B ) Schematics and contact maps of rDNA-carrying chromosomes S . uvarum chromosome XII ( SuXII ) and S . cerevisiae chromosome XII ( ScXII ) , and S . cerevisiae chromosome V ( ScV ) , in wild-type S . cerevisiae x S . uvarum hybrids ( A ) and a strain with a translocation between ScXII and ScV ( B ) , both in exponential growth . Ovals indicate centromeres and slanted lines indicate the rDNA arrays . Double-headed arrows indicate enhanced interactions . Dashed lines in the contact maps indicate the translocation breakpoints . ( C ) Schematic of how the rDNA-carrying chromosomes S . cerevisiae ( Sc ) chrXII and S . uvarum ( Su ) chrXII preferentially interact due to shared tethering . The proximal halves ( left diagram ) of the chromosomes , which contain the centromeres ( CEN ) , are tethered at the spindle pole body ( SPB ) at their centromeres , at the periphery at their telomeres ( TEL ) , and at the nucleolus ( NUC ) at their rDNA arrays ( rDNA ) . The distal halves ( right diagram ) are tethered at their telomeres and rDNA , but not their centromeres . These combinations of tethering points are not found in other chromosomes ( shown in grey ) . Su , S . uvarum; Sc , S . cerevisiae . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 010 Comparing homolog proximity across the genome , we noticed extensive interactions between the homologous chromosomes carrying the rDNA arrays ( Figure 2B ) . To test whether this enrichment for interactions is due to sequence-dependent homolog pairing , we generated a translocation that swapped most of the centromeric half of S . cerevisiae chromosome XII with an equivalently sized portion of S . cerevisiae chromosome V , thereby moving the rDNA array to S . cerevisiae chromosome V . In this translocation-bearing strain , interactions between S . uvarum chromosome XII and S . cerevisiae chromosome V are enriched instead of S . cerevisiae chromosome XII ( Figure 2C , D and Figure 2—figure supplement 2A , B ) , suggesting that homolog proximity of chromosomes carrying the rDNA arrays is due to the presence of the rDNA rather than the particular sequence of the chromosome that carries it . We propose that the rDNA-carrying chromosomes are uniquely positioned within the nucleus due to their tethering at the nucleolus ( Duan et al . , 2010 ) ( Figure 2—figure supplement 2C ) . This shared tethering would then cause enhanced interactions between the homologous proximal and distal segments of these chromosomes and inflate the signal for apparent homology-dependent pairing . Based on these findings , we excluded the rDNA-carrying chromosomes from estimates of homolog proximity . Even with these stringent constraints , we find that the observed interaction between homologous alleles exceeds that predicted based on the Rabl-like orientation ( Figure 2E ) . Of note , the left arm of chromosome III and the right arm of chromosome IX exhibit particularly strong homolog proximity ( Figure 2B ) ; proximity at chromosome III is possibly due to pairing of the silenced mating-type loci ( Miele et al . , 2009 ) . In all hybrids , homolog proximity is substantially greater in saturated cultures approaching stationary phase than in exponential growth ( Figure 2E ) , consistent with previous observations ( Burgess et al . , 1999 ) . One explanation for this result is differences in the strength of sequence-dependent homolog pairing between growth conditions , perhaps mediated by differences in nucleosome positioning and DNA-bound proteins . However , this difference could also be a consequence of the reduced cell cycling coupled with loss of homolog proximity during S-phase ( Burgess et al . , 1999 ) or smaller nuclear size in cells approaching stationary phase ( Guidi et al . , 2015 ) . To test whether we also observe cell cycle dependence of homolog proximity , we performed Hi-C on nocodazole-arrested cells , which were previously reported to exhibit reduced homolog proximity ( Burgess et al . , 1999 ) . We find that nocodazole arrest does not substantially reduce homolog proximity in the diverged hybrid S . cerevisiae x S . uvarum ( Figure 2E ) . However , it remains possible that the lack of S-phase cells in saturated cultures contributes to the difference in homolog proximity between exponentially growing and saturated cultures . We next sought to evaluate whether changes in nuclear size across growth conditions could explain the observed variation in homolog proximity . The nucleus is known to decrease in size in saturated cultures ( Guidi et al . , 2015 ) , so we created alternate versions of the polymer model of the Rabl-like orientation with proportionally smaller nuclei , at 80% and 64% of the original size . In these models , smaller nuclear size led to decreased , rather than increased , homolog proximity ( Figure 2E ) . These models suggest that the difference in homolog proximity between saturated and exponentially growing cultures cannot be explained by the effect of differences in nuclear size , and provide additional support for homolog pairing beyond the Rabl-like orientation . We also searched our dataset for evidence of highly specific changes in genome conformation at the scale of individual genes . Microscopy studies have revealed inducible genes that relocate to the nuclear periphery upon activation due to association with nuclear pores , for example GAL1 ( Brickner et al . , 2016; Casolari et al . , 2004; Dultz et al . , 2016 ) , INO1 ( Brickner and Walter , 2004 ) , HXK1 ( Taddei et al . , 2006 ) , TSA2 ( Ahmed et al . , 2010 ) , and HSP104 ( Dieppois et al . , 2006 ) , which can increase gene expression ( Ahmed et al . , 2010; Brickner and Walter , 2004; Brickner et al . , 2016; Taddei et al . , 2006 ) . Although DNA interactions with components of the nuclear pore complex have been identified genome-wide by chromatin immunoprecipitation ( Casolari et al . , 2004 ) , it remains unclear whether relocalization of specific genes impacts global genome conformation . We first focused on the galactose metabolism gene GAL1 . This gene and its neighbors GAL7 and GAL10 move upon galactose induction from their location near the spindle pole body to a nuclear pore complex at the nuclear periphery ( Casolari et al . , 2004; Dultz et al . , 2016 ) ( Figure 3A ) . Consistent with this expectation , we found using Hi-C that both GAL1 loci interacted less with pericentromeric regions upon galactose induction ( Figure 3B–D ) . Despite previous reports that the homologous GAL1 loci preferentially interact with each other during galactose induction ( Brickner et al . , 2016; Zhang and Bai , 2016 ) , we do not see a clear signal for increased pairing ( Figure 3—figure supplement 1 ) , perhaps because of the high basal interaction frequency between pericentromeric loci or the divergence between S . cerevisiae and S . uvarum . 10 . 7554/eLife . 23623 . 011Figure 3 . GAL1 shifts away from centromeres upon galactose induction . ( A ) Schematic of GAL1 positioning ( dark blue ) in glucose ( left ) and galactose ( right ) . NPC , nuclear pore complex; CEN , centromere; chr , chromosome; SPB , spindle pole body . ( B ) Example region of differential Hi-C map of S . cerevisiae x S . uvarum hybrids in galactose vs . glucose , at 32 kb resolution . Interactions that strengthen in galactose are in red , while those that weaken are in blue . Ticks indicate centromeres; black lines indicate chromosomes . Arrows indicate location of S . uvarum GAL1 . ( C ) Boxplot of the difference in S . uvarum GAL1 interaction frequency in galactose vs . glucose across the S . cerevisiae x S . uvarum genome , excluding intrachromosomal interactions and binned by distance from the centromere ( in 32 kb bins ) . Whiskers correspond to the highest and lowest points within the 1 . 5× interquartile range . p<0 . 05 after Bonferroni correction ( n = 9 ) ; Mann-Whitney test . Note: some outliers are beyond the plot range and are not shown . ( D ) Same as ( C ) for S . cerevisiae GAL1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 01110 . 7554/eLife . 23623 . 012Figure 3—figure supplement 1 . GAL1 homologs do not detectably pair during galactose induction . ( A ) Scatter plot of normalized interaction frequencies in galactose ( x-axis ) and in glucose ( y-axis ) , between the GAL1 homologs ( in red ) and other interactions between loci at the same centromeric distance ( in grey ) . All interactions are between 32 kb bins . ( B ) Normalized interaction frequencies between the S . cerevisiae GAL1 and the S . uvarum chromosome IV in glucose ( blue ) and galactose ( red ) . Arrow points toward interaction between GAL1 homologs , and the dashed line indicates the location of the centromere . ( C ) Same as ( B ) for the S . uvarum GAL1 and the S . cerevisiae chromosome II . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 012 Having established that we could detect the known inducible relocalization of the GAL1 gene , we looked for other specific changes in genome conformation in the well-studied environmental conditions of galactose induction and growth saturation ( approaching stationary phase ) . Surprisingly , we observed markedly increased interactions between homologous loci surrounding the genes HAS1 and TDA1 ( subsequently abbreviated as ‘HAS1-TDA1 loci’ ) on chromosome XIII under both growth saturation and galactose induction , compared to standard exponential growth in glucose ( Figure 4A ) . In fact , under inducing conditions , this interaction is among the strongest genome-wide , excluding pericentromeric and subtelomeric regions ( top interaction out of over 83 , 000; Figure 4—figure supplement 1 ) . No canonical galactose-induced genes are in or near this region . Nevertheless , this inducible homolog proximity appears to be evolutionarily conserved , as it occurs in all three tested interspecific hybrids , at least in saturated culture ( Figure 4A–C; galactose not tested in all hybrids ) . 10 . 7554/eLife . 23623 . 013Figure 4 . Inducible pairing of HAS1-TDA1 homologs is evolutionarily conserved and sequence-specific . ( A ) , ( B ) , and ( C ) Hi-C contact maps of chromosome XIII interactions at 32 kb resolution in S . cerevisiae x S . uvarum ( A ) , S . cerevisiae x S . paradoxus ( B ) , and S . paradoxus x S . uvarum ( C ) hybrids in exponential growth ( left column ) , saturated cultures ( middle column ) , and in S . cerevisiae x S . uvarum hybrids ( A ) , galactose ( right column ) . White arrows indicate the interaction between the homologous HAS1-TDA1 loci . ( D ) Genome browser shot of open-reading frames ( ORFs; blue boxes ) and tested deletions ( brackets ) in the S . cerevisiae region surrounding the genes HAS1 and TDA1 , from positions 840 , 000–860 , 000 ( Figure 4—figure supplement 1A ) . Arrows indicate ends and directionalities of ORFs . ( E ) Strength of HAS1-TDA1 homolog pairing at 32 kb resolution ( red lines ) compared to similar interactions ( grey violin plots; i . e . interactions between an S . cerevisiae locus and an S . uvarum locus , where both loci are ≥15 bins from a centromere and ≥1 bin from a telomere , and not both on chromosome XII ) in wild-type and deletion strains of S . cerevisiae x S . uvarum . See Figure 4—figure supplement 2 for Hi-C contact maps of deletion strains . ( F ) Two example z-stacks of images used to measure distances between HAS1-TDA1 alleles tagged with LacO arrays targeted by LacI-GFP ( shown in green ) , with membranes labeled by ER04 mCherry ( shown in red ) . The yellow outline indicates the images chosen for analysis . White brackets indicate the measured distance . Scale bar = 1 µm . ( G ) Distributions of the distance between the HAS1-TDA1 alleles measured by microscopy in S . cerevisiae diploids in glucose ( blue ) , galactose ( purple ) , and saturated cultures ( red ) . n = 100 for each condition . p-Values were calculated using the Wilcoxon rank sum test . ( H ) Frequency of HAS1-TDA1 alleles less than 0 . 55 µm apart , measured as in ( G ) , in glucose ( blue ) , galactose ( purple ) , or saturated cultures ( red ) . p-Values were calculated using Fisher’s exact test . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 01310 . 7554/eLife . 23623 . 014Figure 4—source data 1 . Raw cell counts for Figure 4G and H . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 01410 . 7554/eLife . 23623 . 015Figure 4—figure supplement 1 . Exceptional inducible homolog pairing at HAS1-TDA1 locus . ( A ) Raw Hi-C contact maps at 20 kb resolution for S . cerevisiae and S . uvarum chromosome XIII . Arrow points to strongest interaction in saturated culture excluding regions near centromeres , at positions 840 , 000–860 , 000 on the S . cerevisiae chromosome XIII , the target of deletion studies . ( B ) Histogram comparing HAS1-TDA1 homolog pairing interaction frequency ( red line ) to all other interactions ( grey ) between an S . cerevisiae locus and an S . uvarum locus ( 32 kb bin ) , with both loci ≥3 bins from a centromere , >1 bin from a telomere , and not both on an rDNA-carrying chromosome . Inset shows enlarged view of right end of plot . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 01510 . 7554/eLife . 23623 . 016Figure 4—figure supplement 2 . HAS1-TDA1 homolog pairing does not shift nearby upon deletion . Hi-C contact maps of chromosome XIII interactions at 32 kb resolution in S . cerevisiae x S . uvarum hybrids in saturated cultures , in wild-type ( WT ) and deletion strains . See Figure 4D for deletion boundaries . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 01610 . 7554/eLife . 23623 . 017Figure 4—figure supplement 3 . HAS1-TDA1 homolog pairing is recapitulated ectopically by the HAS1 and TDA1 promoters . Genomic relocation of the S . cerevisiae HAS1pr-TDA1pr 1 kb region causes inducible ectopic pairing with the S . uvarum HAS1-TDA1 allele . ( Left ) Hi-C contact maps of interactions between S . uvarum chromosome XIII and S . cerevisiae chromosomes XIII and XIV in the S . cerevisiae x S . uvarum hybrid with the S . cerevisiae HAS1pr-TDA1pr region moved to S . cerevisiae chromosome XIV ( location indicated by vertical black arrow ) , in saturated culture ( A ) and in exponential growth ( B ) , compared to the wild-type hybrid ( C ) in saturated culture . Interactions between the S . uvarum HAS1-TDA1 locus and the original S . cerevisiae HAS1-TDA1 locus are indicated by a blue arrow , whereas interactions with the new HAS1pr-TDA1pr locus are indicated by a red arrow . ( Right ) Histograms comparing the pairing frequency of the S . uvarum HAS1-TDA1 locus and either the normal S . cerevisiae HAS1-TDA1 locus ( blue line ) or the ectopic HAS1pr-TDA1pr locus ( red line ) to the frequency of all similar interactions , that is , those between an S . cerevisiae locus and an S . uvarum locus ( 32 kb bin ) , with both loci ≥15 bins from a centromere ( or dCEN >480 kb ) , >1 bin from a telomere , and not both on an rDNA-carrying chromosome . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 017 To explore whether this pairing depends on the presence of specific sequences , we created various deletions of the S . cerevisiae copy of the region , ranging from a 20 kb region from NGL2 through YMR295C ( Figure 4—figure supplement 1A ) to a single 1 kb intergenic region containing the promoters for HAS1 and TDA1 ( HAS1pr-TDA1pr; Figure 4D ) . Every deletion that included this intergenic region reduced the interaction frequency between HAS1-TDA1 homologs in saturated growth conditions back to uninduced levels , indicating that this inducible pairing is sequence-dependent ( Figure 4E and Figure 4—figure supplement 2 ) . In contrast , deletion of the HAS1 coding sequence had minimal impact , which shows that the deletion construct itself did not impede inducible pairing ( Figure 4E and Figure 4—figure supplement 2 ) . To test whether the HAS1pr-TDA1pr region is sufficient to produce inducible pairing , we moved the S . cerevisiae copy of this region to the left arm of S . cerevisiae chromosome XIV . The ectopic HAS1pr-TDA1pr allele exhibited inducible interactions with the S . uvarum HAS1pr-TDA1pr , although not to the same extent as the endogenous allele ( Figure 4—figure supplement 3 ) . The diminished extent of inducible pairing may reflect the contribution of chromosomal homolog pairing , which would be disrupted in the ectopic location , or of additional regions that are not sufficient to produce pairing on their own . To verify whether this pairing occurs in homozygous S . cerevisiae diploids in addition to diverged hybrids , we labeled both HAS1-TDA1 loci with integrated LacO arrays targeted by LacI-GFP and measured the distance between them in a population of cells by confocal microscopy ( Figure 4F ) . Consistent with our Hi-C data , the HAS1-TDA1 homologs were closer together in galactose-induced and saturated cultures than in glucose ( Figure 4G , H and Figure 4—source data 1 ) . Based on previous studies of relocalized genes ( Brickner et al . , 2012 , 2016 ) , we hypothesized that pairing between the homologous HAS1-TDA1 loci might be mediated by interactions of both alleles with nuclear pores ( Figure 5—figure supplement 1 ) . Therefore , we tested whether the HAS1-TDA1 loci are relocalized to the nuclear periphery in a condition-dependent manner . We tagged the HAS1-TDA1 locus with a LacO array as before and counted the proportion of cells in which HAS1-TDA1 colocalized with the mCherry-labeled nuclear membrane , in haploid S . cerevisiae . Indeed , the HAS1-TDA1 locus shifted to the nuclear periphery upon galactose induction and in saturated culture conditions ( Figure 5A and Figure 5—source data 1 ) . To confirm whether this inducible reorganization was dependent on association with nuclear pores , we repeated our analysis in strains with deletions of nuclear pore components NUP2 or NUP100 , or pore-associated protein MLP2 ( Figure 5A and Figure 5—source data 1 ) . As in other cases of gene relocalization , Nup2 but not Nup100 was required for peripheral localization of the HAS1-TDA1 locus . However , unlike other relocalized genes ( Ahmed et al . , 2010; Brickner et al . , 2016; Luthra et al . , 2007 ) , HAS1-TDA1 locus relocalization did not require Mlp2 , suggesting that the HAS1-TDA1 locus may interact with nuclear pores via a distinct mechanism . We performed these initial analyses in haploids , to facilitate deletion of nuclear pore components , but pairing cannot occur in haploids . Thus , we confirmed that the HAS1-TDA1 loci are peripherally relocalized in diploids , by reanalyzing the images we used to measure distances between HAS1-TDA1 alleles in diploid cells ( Figure 5B and Figure 5—source data 2 ) . We then asked whether pairing of the HAS1-TDA1 alleles only occurs at the periphery , by determining the proportion of cells in each category of peripheral localization that have paired HAS1-TDA1 alleles ( Figure 5C and Figure 5—source data 2 ) . In fact , HAS1-TDA1 alleles can remain paired in galactose in the nucleoplasm . However , this need not imply that the nuclear pores do not play a role in pairing . Previous studies of gene relocalization to nuclear pores have reported that the cell cycle affects when genes relocalize to the nuclear periphery; namely , genes tend to move to the nucleoplasm during S-phase ( Brickner and Brickner , 2010 ) . Alleles can remain paired in the nucleoplasm , but cannot actively pair during S-phase ( Brickner et al . , 2012 ) . In agreement with these studies , the peripheral localization of HAS1-TDA1 loci that we observe also exhibits cell cycle dependence ( Figure 5D , E and Figure 5—source data 2 ) , which may explain the presence of pairing in the nucleoplasm . 10 . 7554/eLife . 23623 . 018Figure 5 . Inducible peripheral localization and pairing of HAS1-TDA1 alleles involve nuclear pore interactions . ( A ) Proportions of haploid S . cerevisiae cells exhibiting peripheral HAS1-TDA1 localization in strains with and without deletions of nuclear pore components , in glucose ( in blue ) , galactose ( in purple ) , or saturated culture ( in red ) . Experiments were performed in biological triplicate , with n ≥ 30 per experiment . p<0 . 05 , Student’s t-test . Center values and error bars represent mean ± s . e . m . ( B ) Proportion of diploid S . cerevisiae cells exhibiting two ( ON-ON ) , one ( ON-OFF ) , or zero HAS1-TDA1 alleles with peripheral localization , in glucose ( blue ) and galactose ( purple ) . p-Value calculated using chi-squared test . ( C ) Proportion of diploid S . cerevisiae cells with HAS1-TDA1 alleles clustered ( <0 . 55 µm apart ) , in glucose ( blue ) and galactose ( purple ) as a function of the peripheral localization of HAS1-TDA1 alleles . Same images used as in ( B ) and Figure 4G , H . ( D ) Example images of cells in G1 , S , and G2/M phases of cell cycle . ( E ) Proportions of haploid S . cerevisiae cells exhibiting peripheral HAS1-TDA1 localization in different phases of the cell cycle , in glucose ( in blue ) and galactose ( in purple ) . p<0 . 05 , Student’s t-test . Center values and error bars represent mean ± s . e . m . ( F ) Strength of HAS1-TDA1 homolog pairing at 32 kb resolution ( red lines ) compared to similar interactions ( grey violin plots; that is , interactions between an S . cerevisiae locus and an S . uvarum locus , where both loci are ≥15 bins from a centromere and ≥1 bin from a telomere , and not both on chromosome XII ) in wild-type and homozygous nup2△ strains of S . cerevisiae x S . uvarum . ( G and H ) Nup60-TAP ChIP-seq read coverage tracks for IP and input in galactose ( purple ) and glucose ( blue ) , zoomed into a 35 kb region surrounding GAL1-GAL10-GAL7 on chromosome II ( G ) , and a 35 kb region surrounding HAS1 and TDA1 on chromosome XIII ( H ) . ( I ) Nup60-TAP ChIP qPCR as IP/input normalized to the negative control PRM1 , for the positive control GAL1pr and five sets of primers in TDA1 . Primer sequences are provided in Supplementary file 2 . Center values and error bars represent mean ± s . e . m . of three biological replicates . p<0 . 05 , Student’s t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 01810 . 7554/eLife . 23623 . 019Figure 5—source data 1 . Raw cell counts for Figure 5A . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 01910 . 7554/eLife . 23623 . 020Figure 5—source data 2 . Raw cell counts for Figure 5B and C . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 02010 . 7554/eLife . 23623 . 021Figure 5—source data 3 . Raw cell counts for Figure 5E . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 02110 . 7554/eLife . 23623 . 022Figure 5—figure supplement 1 . Schematic of how nuclear pore association mediates homologous HAS1 pairing . NPC , nuclear pore complex; CEN , centromere; chr , chromosome; SPB , spindle pole body . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 02210 . 7554/eLife . 23623 . 023Figure 5—figure supplement 2 . Mock-IP on Nup60-TAP . Nup60-TAP ChIP qPCR using BSA instead of anti-TAP antibody , plotted as IP/input normalized to the negative control PRM1 , for the positive control GAL1pr and five sets of primers in TDA1 . Primer sequences are provided in Supplementary file 2 . Center values and error bars represent mean ± s . e . m . of three biological replicates . No values are significant ( p<0 . 05 , Student’s t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 023 To test whether the nuclear pore complex is required for pairing as well as relocalization of the HAS1-TDA1 loci , we performed Hi-C on an S . cerevisiae x S . uvarum hybrid strain with a homozygous deletion of NUP2 . In this strain , HAS1-TDA1 pairing was not observed in galactose , as expected , but still occurred at full strength in saturated growth ( Figure 5F ) . These data indicate that Nup2 is required for HAS1-TDA1 homolog pairing in galactose but not in saturated culture , suggesting distinct and/or additional mechanisms of pairing . To test biochemically whether the HAS1-TDA1 loci interact with nuclear pores , we performed chromatin immunoprecipitation ( ChIP ) sequencing on the nuclear basket protein Nup60 , which unlike the dynamic Nup2 cannot dissociate from the nuclear pore complex , tagged with the tandem affinity purification ( TAP ) tag ( Ghaemmaghami et al . , 2003 ) , in haploid S . cerevisiae grown in either glucose or galactose . As expected , we observed a clear enrichment of the galactose metabolism gene cluster GAL1-GAL10-GAL7 in the immunoprecipitated DNA from cells grown in galactose , compared to those grown in glucose ( Figure 5G ) . In contrast , we observed little if any such enrichment of the HAS1-TDA1 locus ( Figure 5H ) . We also performed qPCR on the same ChIP DNA , which gave the same results ( Figure 5I and Figure 5—figure supplement 2 ) . Together , these data suggest that although HAS1-TDA1 locus relocalization and pairing requires Nup2 in galactose , HAS1-TDA1 differs from GAL1 in how it interacts with the nuclear pore complex and may pair via alternative mechanisms as well , particularly in saturated growth . Nuclear pore association is thought to play a role in transcriptional regulation , generally but not always leading to greater or faster activation ( Akhtar and Gasser , 2007; Taddei , 2007; Taddei et al . , 2010 ) . To explore how nuclear pore association might affect transcription at the HAS1-TDA1 locus , we performed RNA sequencing on haploid S . cerevisiae grown in glucose , galactose , and saturated growth conditions . HAS1 is downregulated in both pairing conditions ( Figure 6A ) , whereas TDA1 is weakly upregulated in both conditions ( Figure 6B ) . However , transcriptional changes at HAS1 and TDA1 are relatively unremarkable; in both galactose and saturated growth , dozens to hundreds of genes are more strongly up- or downregulated than HAS1 or TDA1 ( Figure 6C , D ) . For example , GAL1 is upregulated nearly 1000-fold in galactose ( Figure 6C ) . This suggests that nuclear pore association is not solely a function of strong transcriptional activation . 10 . 7554/eLife . 23623 . 024Figure 6 . Transcriptional changes in galactose and saturated culture . ( A and B ) Bar plots of gene expression in haploid S . cerevisiae grown in glucose , galactose , or to saturation , for HAS1 ( A ) and TDA1 ( B ) . Asterisks indicate p-values<0 . 05 ( * ) , 0 . 01 ( ) , 0 . 001 ( * ) , or 0 . 0001 ( **** ) , Student’s t-test . Center values and error bars represent mean ± s . e . m . of three biological replicates ( C and D ) Histogram of log2 fold change in gene expression in galactose ( C ) or saturated growth ( D ) compared to glucose . Vertical lines indicate values for HAS1 , TDA1 , and GAL1 ( C only ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23623 . 024 Homologous chromosomes pair along their lengths leading up to and during meiosis , but may recognize each other and preferentially interact even in normal mitotic growth , perhaps to facilitate homology-directed repair or prepare for meiosis under stressful conditions . However , whether and to what extent this mitotic homolog pairing occurs has remained controversial , in part due to the lack of genome-wide data and the apparent homolog pairing caused by the Rabl-like orientation . We performed Hi-C in diverged hybrid diploid yeast , which allowed us to resolve homologous chromosomes ( Figure 1 ) and thus infer homolog pairing strength on a genome-wide basis ( Figure 2B ) . After using a polymer model of the Rabl-like orientation to calibrate our estimates , we find that even in hybrid diploids with homologs diverged to less than 80% nucleotide identity , homologous chromosomes do interact preferentially during mitotic growth , albeit subtly ( Figure 2E ) . It would be interesting to compare the strength of pairing across hybrids with varying levels of divergence; however , our homolog pairing analysis requires filtering and stratifying genomic regions and thus may not be directly comparable across different reference genomes . Our data do not necessarily imply end-to-end chromosome alignment as occurs in meiosis . Instead , our data indicate an increased frequency of contact between homologous loci above the expectation based on random collisions , with substantial variation across the genome . Nevertheless , that such distant homology is sufficient for at least some homolog interactions is perhaps remarkable , and may hint at the role of DNA-bound proteins , which are more conserved than DNA , in mediating the interactions . Homolog pairing strength also depends on both growth conditions and genomic location , sometimes jointly: the homologous HAS1-TDA1 loci on chromosome XIII pairs during saturated growth and galactose induction , but not exponential growth in glucose ( Figure 4 ) . This region is not remarkably conserved , suggesting that homolog pairing is at least partly due to specific interactions mediated by proteins , rather than direct DNA-DNA homology interactions ( Danilowicz et al . , 2009; Gladyshev and Kleckner , 2014 ) . In all tested hybrids , the HAS1-TDA1 locus exhibits surprisingly strong homolog proximity ( Figure 4A–C; Figure 4—figure supplement 1B ) . How does HAS1-TDA1 pairing occur , and why ? The nuclear pore component Nup2 seems to play a role , although not exclusively , in mediating pairing . The HAS1-TDA1 locus moves to the nuclear periphery under pairing conditions , and both this relocalization and pairing are Nup2-dependent in galactose ( Figure 5A , F and Figure 5—source data 1 ) . However , Nup2 is not required for HAS1-TDA1 pairing in saturated growth ( Figure 5F ) . Together , the Nup2-independence of HAS1-TDA1 pairing in saturated growth , the Mlp2-independence of HAS1-TDA1 peripheral localization , and the lack of HAS1-TDA1 enrichment in Nup60 ChIP suggest that HAS1-TDA1 may interact with nuclear pores by a mechanism distinct from the previously studied GAL1 and INO1 , and possibly by different mechanisms in galactose and in saturated culture . However , more experiments are needed to fully elucidate the role and mechanism of the nuclear pore complex in HAS1-TDA1 homolog pairing . Regardless of the molecular mechanism , nuclear pore interactions may confine the HAS1-TDA1 alleles to the relatively small space near the nuclear periphery , thus speeding up the rate at which they randomly contact each other ( Figure 5—figure supplement 1 ) . Once in physical proximity , additional mechanisms such as protein-protein interactions between DNA-binding proteins could prolong the duration of contact , even after the alleles are no longer at a nuclear pore . Indeed , the presence of paired HAS1-TDA1 alleles in the nucleoplasm suggests that such nuclear pore-independent pairing mechanisms may act at HAS1-TDA1 ( Figure 5C and Figure 5—source data 2 ) . Interestingly , a recent study showed that Nup2 is involved in meiotic homolog pairing ( Chu et al . , 2017 ) , suggesting that Nup2 may more generally play a role in homolog pairing . Why do the homologous HAS1-TDA1 alleles pair and relocalize to the periphery , and why does this interaction appear to be unique ? Many genes associate with nuclear pores and relocalize to the nuclear periphery upon activation , including GAL1 , but we do not observe strong pairing of GAL1 homologs . It is possible that Hi-C may be failing to capture pairing at GAL1 due to its pericentromeric location , but these data may also reflect particularly strong pairing at the HAS1-TDA1 loci . Divergence between S . cerevisiae and S . uvarum , particularly in their galactose metabolism pathways ( Roop et al . , 2016 ) , may also contribute to the lack of pairing at the GAL1 locus . Lack of GAL1 pairing need not correspond to lack of peripheral localization ( Brickner et al . , 2016 ) ; in our hybrid , the S . cerevisiae and S . uvarum GAL1-binding proteins may be able to each interact with the nuclear pores but not with each other . Which gene is driving HAS1-TDA1 homolog pairing , HAS1 or TDA1 , or both ? Given the association between nuclear pore interactions and transcriptional activation , transcriptional changes in growth conditions that induce pairing may provide a clue . The genes at the HAS1-TDA1 locus , HAS1 and TDA1 , demonstrate opposing changes in gene expression in galactose and saturated culture: HAS1 is downregulated , whereas TDA1 is upregulated . The upregulation of TDA1 , athough not particularly strong in magnitude ( Figure 6C , D ) may be of functional importance . Tda1 is a kinase required for phosphorylation of Hxk2 , the primary hexokinase in yeast ( Kaps et al . , 2015; Kettner et al . , 2012 ) . Unphosphorylated Hxk2 can interact with Mig1 to repress various alternative carbon source metabolism genes in the presence of glucose . Phosphorylation of Hxk2 by Tda1 in low-glucose conditions prevents its interaction with Mig1 and thus leads to release from glucose repression . While we have not yet tested whether disrupting peripheral localization would affect TDA1 or HAS1 transcription , we hypothesize that nuclear pore interaction may aid the upregulation of TDA1 in response to low-glucose concentrations , perhaps by facilitating efficient transcription or mRNA export from the nucleus . Other questions remain about the mechanism and functional implications of HAS1-TDA1 pairing and peripheral relocalization . Increased transcription may itself contribute to localization at nuclear pores via interactions between nascent mRNA and mRNA processing and export factors at nuclear pores ( Akhtar and Gasser , 2007 ) , and may be involved in establishment of HAS1-TDA1 pairing as it is for GAL1 ( Brickner et al . , 2016 ) . However , given the abundance of other genes with similar or greater changes in transcription that do not pair ( Figure 6 ) , transcription alone likely cannot explain our data . For some genes , nuclear pore interactions mediate rapid reactivation in a phenomenon termed epigenetic transcriptional memory ( D'Urso and Brickner , 2017; D'Urso et al . , 2016; Light et al . , 2010 ) ; it is also possible that the nuclear pore interactions with HAS1-TDA1 may be involved in epigenetic transcriptional memory . The pairing of the HAS1-TDA1 alleles may also serve a distinct function , potentially including trans gene regulation like at GAL1 ( Zhang and Bai , 2016 ) , but further experiments are needed to test this possibility . The principles and functional implications of genome conformation remain open questions . Although the budding yeast S . cerevisiae is thought to have a simple genome organization , it serves as a versatile and relevant model system amenable to integrating multiple approaches to studying genome conformation , including Hi-C , polymer simulations , live-cell imaging , and genetic perturbations . While yeast nuclear organization may differ from that of other eukaryotes in some ways , our findings may , nevertheless , be applicable to other organisms: recent studies in the fruit fly Drosophila have provided evidence for the generality of the role of nuclear pores in transcriptional regulation first observed in yeast ( Pascual-Garcia et al . , 2017 ) . Our study illustrates both the utility of combining orthogonal methodologies and that we have much more to learn about genome organization , even in the simple budding yeast . All yeast strains used in this study are listed in Supplementary file 1 . All primers used in this study , including those used for generation and validation of strains , are listed in Supplementary file 2 . Hybrid strains were created by mating haploid strains and then performing auxotrophic selection . The ScV-ScXII translocation , S . cerevisiae x S . uvarum strain was generated by first creating the translocation in the haploid S . cerevisiae strain BY4742 , followed by mating with haploid S . uvarum . A cassette containing hphMX followed by the first half of URA3 , an artificial intron , and a lox71 site was amplified from pBAR3 ( Levy et al . , 2015 ) and integrated into the intergenic region between YLR150W and YLR151C . A second cassette containing a lox66 site , an artificial intron , the second half of URA3 , and natMX was amplified from pBAR2-natMX ( pBAR2 [Levy et al . , 2015] with natMX in place of kanMX ) and integrated into the intergenic region between YER151C and YER152C . The translocation was induced by transforming the resulting strain with the galactose-inducible Cre plasmid pSH47-kanMX ( pSH47 ( Güldener et al . , 1996 ) with kanMX in place of URA3 ) , and then inducing Cre recombination by plating on YP + galactose medium . Successful translocation strains were selected by growing in medium lacking uracil , and verified by PCR across the translocation junctions . This S . cerevisiae strain was then mated with S . uvarum strain ILY376 . Heterozygous deletion strains were made in S . cerevisiae x S . uvarum hybrids , by replacing regions of interest with the hphMX cassette . Homozygous deletion strains were made by making deletions in haploids and then mating the haploid strains . Strains were verified by PCR across each deletion junction . The knock-in strain was made by integrating the HAS1pr-TDA1pr region followed by the natMX cassette into the region between YNL266W and YNL267W ( PIK1 ) on S . cerevisiae chromosome XIV in the S . cerevisiae x S . uvarum hybrid YMD3269 ( HAS1pr-TDA1pr deletion ) . Plasmids pAFS144 ( Straight et al . , 1996 ) , p5LacIGFP ( Randise-Hinchliff et al . , 2016 ) , pER04 ( Randise-Hinchliff et al . , 2016 ) , and pFA6a-kanMX6 ( Longtine et al . , 1998 ) have been described . To tag HAS1-TDA1 with the LacO array , 1 kb downstream of the HAS1 ORF was PCR amplified and TOPO cloned to create pCR2 . 1-HAS1_3’UTR . Plasmid p6LacO128-HAS1 was made by inserting HAS1 from pCR2 . 1- HAS1_3’UTR into p6LacO128 ( Brickner and Walter , 2004 ) . Cells were grown overnight , shaking at 30°C ( room temperature for S . uvarum ) in YPD medium ( 1% yeast extract , 2% peptone , 2% dextrose ) , YP + raffinose ( 2% ) , or YP + galactose ( 2% ) . For saturated culture samples , they were crosslinked at this point by resuspension and incubation in 1% formaldehyde in PBS for 20 min at room temperature . Crosslinking was quenched by addition of 1% w/v solid glycine , followed by incubation for 20 min and a PBS wash . For all other experiments , fresh cultures were inoculated to OD600 = 0 . 1 in appropriate medium and grown to OD600 = 0 . 6–0 . 8 . Exponential growth samples were crosslinked at this point , while for nocodazole-arrested samples , cultures were supplemented with 15 µg/mL nocodazole and grown at 30°C for 2 hr following addition of drug prior to crosslinking . Arrested cultures were checked by flow cytometry . For mixture controls , samples were mixed prior to crosslinking . Hi-C libraries were created as described ( Burton et al . , 2014 ) with the exceptions that the restriction endonuclease Sau3AI or HindIII was used to digest the chromatin and the Kapa Hyper Prep kit ( Kapa Biosystems , Wilmington , MA ) was used to create the Illumina library instead of the Illumina TruSeq kit . Libraries were pooled and sequenced on an Illumina NextSeq 500 ( Illumina , San Diego , CA ) , with 2 × 80 bp reads for interspecific hybrids and 2 × 150 bp reads for intraspecific S . cerevisiae hybrids . Hi-C libraries were similar across the two restriction enzymes and biological replicates ( Figure 1—figure supplement 3 ) . All Hi-C libraries are listed in Supplementary file 3 . The S . cerevisiae references and annotations were downloaded from the Saccharomyces Genome Database ( version R64 . 2 . 1 ) . The S . paradoxus and S . uvarum references and annotations were downloaded from saccharomycessensustricto . org ( Scannell et al . , 2011 ) but modified to correct misassemblies evident based on synteny and Hi-C data ( Figure 1—figure supplement 4 ) . S . paradoxus chromosome IV was rearranged so bases 1–943 , 469 were followed by 1 , 029 , 253–1 , 193 , 028 , then 1 , 027 , 718–1 , 029 , 252 , then 943 , 470–1 , 027 , 717 in reverse order , followed by the remainder of the chromosome . S . uvarum chromosome III was rearranged so bases 219 , 500 onward were placed at the beginning ( left end ) of the chromosome , followed by the first 219 , 399 bases , and then new sequence determined by Sanger sequencing with primers CATTCCCATTTGTTGATTCCTG and GGATTCTATTGTTGCTAAAGGC: TAATAAGGAAGAACTGCTTATTCTTAATTATTTCTACCTACTAAACTAACTAATTATCAACAAATATCATCTATTTAATAGTATATCATCACATGCGGTGTAAGAGGATGACATAAAGATTGAGAAACAGTCATCCAGTCTAATGGAAGCTCAAATGCAAGGGCTGATAATGTAATAGGATAATGAATGACAACGTATAAAAGGAAAGAAGATAAAGCAATATTATTTTGTAGAATTATCGATTCCCTTTTGTGGATCCCTATATCCTCGAGGAGAA . S . uvarum chromosomes X and XII were also swapped , based on homology to S . cerevisiae . The S . cerevisiae Y12 and DBVPG6044 strain references were sequenced to 145- and 315-fold coverage using the PacBio ( Pacific Biosciences , Menlo Park , CA ) single-molecule , real-time ( SMRT ) sequencing platform with P6-C2 chemistry . Each genome was assembled with FALCON ( Chin et al . , 2016 ) , version June 30 , 2015 hash: cee6a58 , and polished with Quiver ( Chin et al . , 2013 ) version 1 . 1 . 0 to generate chromosome-length contigs ( with the exclusion of chromosome XII , which was split at the rDNA array , and Y12 chromosome XIV , which was split into one large and one small contig ) . To call centromeres in S . paradoxus , we searched the region on each chromosome between the genes homologous to those nearest the centromeres in S . cerevisiae ( e . g . YEL001C and YER001W on chromosome V ) for the sequence motif N2TCAC ( A/G ) TGN95-100CCGAAN6 ( based on an alignment of S . uvarum , S . mikatae , and S . kudriavzevii centromeres [Scannell et al . , 2011] ) or its reverse complement . When this motif was absent ( chromosomes VII and VIII ) , we called the centromere as the middle 120 bp of the region . To call centromeres in Y12 and DBVPG6044 , we mapped the S . cerevisiae S288C centromere sequences to the new references . Simulated reads for each hybrid genome ( as in experimental data , 80 bp for interspecific hybrids and 150 bp for the intraspecific S . cerevisiae hybrid ) were generated by taking sequences of the read length at 10 bp intervals . These reads were then remapped to the hybrid genome using bowtie2 ( Langmead and Salzberg , 2012 ) with the --very-sensitive parameter set . The proportion of reads that mapped with mapping quality ≥30 to the correct location was then calculated . Sequencing reads were first pre-processed using cutadapt ( Martin , 2011 ) : reads were quality-trimmed ( option -q 20 ) , trimmed of adapter sequences , and then trimmed up to the ligation junction ( if present ) , excluding any read pairs in which either read was shorter than 20 bp after trimming ( option -m 20 ) . The two reads in each read pair were then mapped separately using bowtie2 ( Langmead and Salzberg , 2012 ) with the --very-sensitive parameter set . For interspecific hybrids , reads were mapped to a combined reference containing both species references , where if secondary mappings were present the best alignment must have a score ≥10 greater than the next best alignment . For intraspecific S . cerevisiae hybrids , reads were mapped separately to both strain references , keeping only read pairs in which both reads mapped to both references—perfectly to one reference and with ≥2 mutations including ≥1 substitution to the other . PCR duplicates ( with identical fragment start and end positions ) were removed , as were read pairs mapping within 1 kb of each other or in the same restriction fragment , which represent either unligated or invalid ligation products . The genome was then binned into 32 kb fragments ( except the last fragment of each chromosome ) , and the number of read pairs mapping to each 32 kb genomic bin was counted based on the position of the restriction sites that were ligated together . Due to gaps in the reference genomes of S . uvarum , some repetitive sequences were only represented once and therefore artifactually mapped uniquely; therefore , reads mapping to annotated repetitive sequences were masked from further analysis . Similarly , gaps in the S . paradoxus reference led to mismapping of reads to the corresponding S . cerevisiae sequence; therefore , for S . cerevisiae x S . paradoxus libraries we masked regions in the S . cerevisiae genome where >1 read from a S . paradoxus Hi-C library mapped , and vice versa . We took a similar approach to mask regions of the Y12 and DBVPG6044 references that were prone to mismapping , as estimated by haploid Y12 and DBVPG6044 Hi-C libraries . For knock-in experiments , the HAS1pr-TDA1pr region was masked to account for its altered genomic location . The resulting matrices were then normalized by excluding the diagonal ( interactions within the same genomic bin ) , filtering out rows/columns with an average of less than one read per bin , and then multiplying each entry by the total number of read pairs divided by the column and row sums . The volume-exclusion polymer model of the Rabl-like orientation was a modified version of the Tjong et al . tethering model ( Tjong et al . , 2012 ) . Briefly , beads representing segments of the genome are randomly positioned and then adjusted until constraints ( e . g . consecutive beads must be adjacent , and no two beads can occupy the same space ) are met . The model was extended from 16 chromosomes to 32 , with the lengths of the S . cerevisiae and S . uvarum chromosomes . The parameters for nuclear size , centromeric constraint position and size , telomeric constraint at the nuclear periphery , and nucleolar position and size were scaled by a factor of 1 . 25 to reflect the roughly doubled volume of diploid nuclei ( cell volume correlates with ploidy ( Mortimer , 1958 ) , and nuclear volume correlates with cell volume [Jorgensen et al . , 2007] ) . To test the effect of smaller nuclei , all parameters were scaled by a factor of 0 . 8 or 0 . 64 from this initial diploid model . For each model , the modeling procedure was repeated 20 , 000 times to create a population of structures . From this population , we simulated Hi-C data by calling all beads within 45 nm of each other as contacting each other , and then counting the number of contacts between each pair of 32 kb bins . The resulting matrix of counts was normalized using the same pipeline as the experimental Hi-C data . In order to assess homolog proximity genome-wide , we first determined which bins represented interactions between homologous sequences , and then compared the normalized interaction frequencies in those bins compared to a set of ‘comparable’ nonhomologous bins . In the interspecific hybrids , we determined homology by counting the number of starts or ends of one-to-one homologous gene annotations falling into each bin . Genes whose ‘SGD’ and ‘BLAST’ gene annotations differed were ignored . To find homologous interaction bins for genomes 1 and 2 , for each bin of genome 1 we considered the bin in genome 2 where the most homologous gene ends fell to be homologous . In the intraspecific hybrids where inter-strain mapping was much more reliable , we simulated 150 bp reads from the Y12 genome at 10 bp intervals , then mapped them to the DBVPG6044 reference . Here , for each bin of genome 1 we considered the bin in genome 2 where the most reads mapped with MAPQ ≥30 to be homologous . To eliminate minor ‘homology’ arising from repetitive sequences ( e . g . telomeres ) , we excluded isolated homologous interaction bins lacking any other homologous interaction bins within two bins . To fully exclude homologous interactions from our estimates of nonhomologous interactions , any interaction bins within 2 bins of homologous interaction bins were excluded from analyses . After determining homologous bins , we compared each homologous bin to other intergenome interactions ( i . e . between chromosomes from different species/strains ) involving one of the two genomic bins involved in the homologous interaction . To control for the effects of the Rabl-like orientation , we further filtered the nonhomologous interaction bins for those in which the centromeric distance ( in units of 32 kb bins ) was equivalent , and then for those in which the chromosome arm lengths of the two loci were within 25% of each other ( in units of 32 kb bins ) . We also considered exclusion of the rDNA carrying chromosome XIIs as well as the centromeric bins , for which we could not fully control chromosome arm lengths . In all cases , we only considered homologous bins with at least two comparable nonhomologous bins . To estimate genome-wide homolog proximity , we compared the sum of normalized interaction frequencies across the homologous bins to those of an equal number of randomly chosen nonhomologous bins , one comparable to each homologous bin , with replacement . We repeated this 10 , 000 times to obtain a distribution of genomic homolog proximity . To obtain a view of homolog proximity strength across the genome , we compared the normalized interaction frequency in each homologous bin to the median of that in the similar nonhomologous bins , and then plotted the ratio of homologous/nonhomologous across the S . cerevisiae genome . Gene positioning at the nuclear periphery and inter-allelic clustering were determined as described previously ( Brickner and Walter , 2004; Brickner et al . , 2016; Egecioglu et al . , 2014 ) . Briefly , cells bearing an array of 128 Lac operators integrated downstream of the HAS1 coding sequence and expressing both the ER04 mCherry membrane marker ( Egecioglu et al . , 2014 ) and the GFP-LacI ( Robinett et al . , 1996 ) were imaged on a Leica SP5 line-scanning confocal microscope ( Leica Microsystems , Wetzlar , Germany ) . Cultures were grown in synthetic minimal media with 2% glucose or 2% galactose overnight at 30°C with constant shaking and harvested in log phase ( OD600 <0 . 5 ) or late log/stationary phase ( OD600 >1 . 0 ) . Unless noted , cultures were grown in the designated media overnight prior to imaging . Cultures were concentrated by brief centrifugation , and then 1 µl was spotted onto a microscope slide for visualization . For all experiments , cells were illuminated at 10–15% power with 488 nm and 561 nm using argon and diode pumped solid state lasers , respectively . Stacked images of ~150 µm x 150 µm fields were collected; ~ 15–20 z-slices of 0 . 34 µm thickness each . The optical thickness of the slices is ~0 . 73 µm . The z-slice in which the green dot ( s ) is most focused and bright is selected for analysis ( Figure 4F ) . For peripheral localization experiments , cells in which the center of the dot colocalizes with the nuclear envelope , as measured by mCherry fluorescence , are scored as peripheral . All other cells are scored as nucleoplasmic . Cells in which the dot was at the top or bottom of the nucleus were excluded . Each experiment was performed three times , counting ~30 cells per replicate . The percent of cells scored as peripheral was averaged and the standard error of the mean was calculated . Student’s t-test was used to compare these distributions . To monitor inter-allelic clustering of the HAS1-TDA1 locus , haploid strains bearing the LacO array integrated downstream of HAS1 were mated to create a diploid strain . These strains were imaged as above and , in cells in which the two alleles were either in the same z-slice or adjacent z-slices , the distance between the centers of the dots was measured using LAS AF software . Cells in which the two dots were not in the same or adjacent z-slices , or cells in which the two dots were unresolvable , were excluded . For each experiment , 100 cells were measured and both the distribution of distances among 0 . 15 µm bins and the fraction of cells in which the two alleles were <0 . 55 µm was calculated . To compare distributions , the Wilcoxon Rank Sum test was used . To compare the fraction of the cells in which the two alleles were <0 . 55 µm , Fisher’s exact test was used . Chromatin immunoprecipitation was performed as in ( Egecioglu et al . , 2014 ) , with modifications . Nup60-TAP yeast ( Ghaemmaghami et al . , 2003 ) were grown overnight , diluted to OD600 = 0 . 125–0 . 15 in 50 ml medium , then grown to OD600 = 0 . 75–0 . 85 at 30°C in either YPD ( for glucose samples ) or YP +2% galactose ( for galactose samples ) , then crosslinked with 1% formaldehyde ( v/v ) for 5 min at room temperature . Crosslinked cells were quenched in 150 mM glycine , washed twice in Tris-buffered saline , and then stored at −80°C . Crosslinked cell pellets were resuspended in 500 µl lysis buffer with 1x cOmplete Protease Inhibitor tablet ( Roche , Basel , Switzerland ) , and then 700 µl of 500 µm acid-washed glass beads were added . The cells were vortexed for 12 min total , in cycles of 2 min shaking and 2 min resting on ice . The lysate was pelleted and resuspended in fresh lysis buffer , and then sonicated for 3 × 10 min runs on a Diagenode Bioruptor ( Diagenode , Liège , Belgium ) , on high power with cycles of 30 s on and 30 s off , to an average of ~300 bp . The sonicate was cleared by centrifugation , and the resulting supernatant was split into an input aliquot ( 1/20 of IP volume ) and two halves for the IP and mock-IP ( BSA instead of antibody ) . For each sample , 5 µg of anti-TAP antibody ( Thermo Fisher Scientific , Waltham , MA; #CAB1001 , Lot #RL240352 ) was used with 10 µl Dynabeads Protein A ( Thermo ) . The antibody or BSA was incubated with pre-washed beads for 2 hr and then washed twice in fresh lysis buffer before being added to the lysate and then incubated overnight . After washing the beads four times in lysis buffer and eluting , the eluate was reverse crosslinked and then treated with RNase A and Proteinase K . DNA was purified using Zymo ChIP DNA Clean & Concentrator ( Zymo Research , Irvine , CA ) and eluted in 30 µl water . IP/mock-IP samples were diluted 1:8 and inputs were diluted 1:320 , and then 4 µl were used in each 10 µl qPCR reaction . qPCRs were performed in triplicate in 384-well plates on a ViiA7 ( Applied Biosystems , Foster City , CA ) , with Kapa Robust 2G Hot Start 2x master mix ( thermocycling as recommended , with 20 s extension for 40 cycles ) and 0 . 2x SYBR Green I dye . CT values were calculated and normalized to a genomic DNA standard curve using the ViiA7 software . IP/input ratios were normalized to those for the negative control PRM1 . ChIP-seq libraries were prepared using the Accel-NGS 2S Plus DNA Library Kit ( Swift Biosciences , Ann Arbor , MI ) , from equal volume pools of the biological replicates , either 1 ng total from input samples or 21 µl of total IP sample . Input and IP libraries were amplified for six and seven cycles , respectively . Libraries were sequenced to ~5–6 million read pairs using 2 × 75 bp reads on an Illumina MiSeq . Sequencing reads were first pre-processed using cutadapt ( Martin , 2011 ) : reads were quality-trimmed ( option -q 20 ) , trimmed of adapter sequences , excluding any read pairs in which either read was shorter than 28 bp after trimming ( option -m 28 ) . The two reads in each read pair were then mapped jointly to the sacCer3 S . cerevisiae reference using bowtie2 ( Langmead and Salzberg , 2012 ) with the --very-sensitive parameter set , requiring the two reads to be within 2000 bp of each other ( option -X 2000 ) . Fragments ( read pairs ) in which both reads had a mapping quality score of at least 30 were then deduplicated by fragment start and end positions and then aggregated into a coverage track using bedtools ( Quinlan and Hall , 2010 ) . Genome browser tracks were generated using the UCSC Genome Browser ( Kent et al . , 2002 ) . BY4741 yeast were grown overnight in YPD ( for exponential growth and saturated samples ) or YP +2% galactose ( for galactose samples ) at 30°C , then diluted to OD600 = 0 . 1–0 . 125 in 50 ml medium , and then grown to OD600 = 0 . 5–0 . 6 , pelleted and stored at −80°C . RNA was purified using acid phenol extraction , and then treated with the DNA-free DNase kit ( Thermo ) . Illumina libraries were then prepared using the TruSeq RNA Library Prep Kit v2 ( Illumina ) , with six cycles of amplification . Libraries were sequenced to ~15–20 million read pairs using 2 × 75 bp reads on an Illumina NextSeq 500 . Sequencing reads were first pre-processed using cutadapt ( Martin , 2011 ) : reads were quality-trimmed ( option -q 20 ) , trimmed of adapter sequences , excluding any read pairs in which either read was shorter than 28 bp after trimming ( option -m 28 ) . The two reads in each read pair were then mapped jointly to the sacCer3 S . cerevisiae reference using bowtie2 ( Langmead and Salzberg , 2012 ) with the --very-sensitive parameter set , requiring the two reads to be within 500 bp of each other ( option -X 500 ) . Fragments ( read pairs ) in which both reads had a mapping quality score of at least 30 were overlapped with annotated genes using HTSeq ( Anders et al . , 2015 ) . Global fold-change analyses were performed using DESeq2 ( Love et al . , 2014 ) . Code for all bioinformatic analyses is available at https://github . com/shendurelab/HybridYeastHiC ( Kim , 2017 ) . A copy is archived at https://github . com/elifesciences-publications/HybridYeastHiC . GEO accession number: GSE88952	Most of the DNA in human , yeast and other eukaryotic cells is packaged into long thread-like structures called chromosomes within a compartment of the cell called the nucleus . The chromosomes are folded to fit inside the nucleus and this organization influences how the DNA is read , copied , and repaired . The folding of chromosomes must be robust in order to protect the organism’s genetic material and yet be flexible enough to allow different parts of the DNA to be accessed in response to different signals . A biochemical technique called Hi-C can be used to detect the points of contact between different regions of a chromosome and between different chromosomes , thereby providing information on how the chromosomes are folded and arranged inside the nucleus . However , most animal cells contain two copies of each chromosome , and the Hi-C method is not able to distinguish between identical copies of chromosomes . As such , it remains unclear how much the chromosomes that can form pairs actually stick together in a cell’s nucleus . Unlike humans and most organisms , two distantly related budding yeast species can mate to produce a “hybrid” in which the chromosome copies can easily be distinguished from each other . Kim et al . now use Hi-C to analyze how chromosomes are organized in hybrid budding yeast cells . The experiments reveal that the copies of a chromosome contact each other more frequently than would be expected by chance . This is especially true for certain chromosomal regions and in hybrid yeast cells that are running out of their preferred nutrient , glucose . In these cells , the regions of both copies of chromosome 13 near a gene called TDA1 are pulled to the edge of the nucleus , which helps the copies to pair up and the gene to become active . The protein encoded by TDA1 then helps turn on other genes that allow the yeast to use nutrients other than glucose . Many questions remain about how and why DNA is organized the way it is , both in yeast and in other organisms . These findings will help guide future experiments testing how the two copies of each chromosome pair , as well as what purpose , if any , this pairing might serve for the cell . A better understanding of the fundamental process of DNA organization and its implications may ultimately lead to improved treatments for genetic diseases including developmental disorders and cancers .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2017	The dynamic three-dimensional organization of the diploid yeast genome
The straight-tusked elephants Palaeoloxodon spp . were widespread across Eurasia during the Pleistocene . Phylogenetic reconstructions using morphological traits have grouped them with Asian elephants ( Elephas maximus ) , and many paleontologists place Palaeoloxodon within Elephas . Here , we report the recovery of full mitochondrial genomes from four and partial nuclear genomes from two P . antiquus fossils . These fossils were collected at two sites in Germany , Neumark-Nord and Weimar-Ehringsdorf , and likely date to interglacial periods ~120 and ~244 thousand years ago , respectively . Unexpectedly , nuclear and mitochondrial DNA analyses suggest that P . antiquus was a close relative of extant African forest elephants ( Loxodonta cyclotis ) . Species previously referred to Palaeoloxodon are thus most parsimoniously explained as having diverged from the lineage of Loxodonta , indicating that Loxodonta has not been constrained to Africa . Our results demonstrate that the current picture of elephant evolution is in need of substantial revision . In the late Miocene in Africa , the last of several major radiations within Proboscidea gave rise to the family Elephantidae , which comprises living elephants and their extinct relatives including mammoths ( genus Mammuthus ) and various dwarf elephant species from Mediterranean islands . The three living elephant species ( the African savanna elephant , Loxodonta africana , the African forest elephant , L . cyclotis and the Asian elephant , Elephas maximus ) , represent the last remnants of this family and of the formerly much more widely distributed and species-rich order Proboscidea . Apart from mammoths , the elephant genus with the most abundant fossil record in Eurasia is Palaeoloxodon ( straight-tusked elephants; Figure 1 ) , which appears in Eurasia around 0 . 75 million years ago ( Ma ) ( Lister , 2016 ) . Based on morphological analyses , Palaeoloxodon is widely accepted as being more closely related to the extant Asian elephant than to mammoths or extant African elephants ( Shoshani et al . , 2007; Todd , 2010 ) and is often subsumed into the genus Elephas ( Maglio , 1973; Sanders et al . , 2010 ) . Across its range from Western Europe to Japan , Palaeoloxodon probably comprised several species ( Shoshani et al . , 2007 ) , and , based on morphological comparisons , all of them are considered to be derived from the African Palaeoloxodon ( or Elephas ) recki ( Maglio , 1973; Saegusa and Gilbert , 2008 ) , which was the predominant proboscidean lineage in Africa during the Pliocene and Pleistocene but went extinct around 100 thousand years ago ( ka ) ( Owen-Smith , 2013 ) . Straight-tusked elephants may have survived in mainland Eurasia until around 35 ka , although the youngest reliably dated remains are from the last interglacial , 115–130 ka ( Stuart , 2005 ) . 10 . 7554/eLife . 25413 . 003Figure 1 . Palaeoloxodon antiquus , geographic range based on fossil finds ( after Pushkina , 2007 ) . White dots indicate the locations of Weimar-Ehringsdorf and Neumark-Nord . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 003 Recent technological progress has pushed back the temporal limit of ancient DNA research , enabling , for example , recovery of a low coverage genome of a ~700 , 000 year-old horse preserved in permafrost ( Orlando et al . , 2013 ) . For more temperate regions , however , evidence of DNA preservation reaching far beyond the last glacial period is still limited to a single locality , Sima de los Huesos in Spain , where DNA has been recovered from ~430 ka old hominin and bear remains ( Dabney et al . , 2013; Meyer et al . , 2016 ) . While genetic analyses of the extinct interglacial fauna remain a challenging undertaking , recent advances in ancient DNA extraction ( Dabney et al . , 2013 ) and sequencing library construction ( Meyer et al . , 2012 ) have improved access to highly degraded DNA . To better understand the evolutionary relationships between the extinct straight-tusked elephants and other elephant species , we attempted DNA extraction and sequencing from several P . antiquus fossils , four of which we investigated in depth . Three of these , which were all unambiguously assigned to P . antiquus based on their morphology , were from Neumark-Nord ( NN ) 1 in Germany , a fossil-rich site that has been proposed to date to MIS 5e ( ~120 ka ) or MIS 7 ( ~244 ka ) or both ( Mania , 2010; Schüler , 2010; Penkman , 2010 ) . This site has yielded one of the largest collections of P . antiquus remains known to date . The fourth fossil was recovered during recent active mining in the travertine deposits of Weimar-Ehringsdorf ( WE ) , Germany , a quarry that has for more than a century yielded a rich collection of fossils representing a typical European interglacial fauna ( Kahlke , 1975 ) . Weimar-Ehringsdorf is best known for the discovery of Neanderthal remains in the early 20th century , and the assemblage is dated to MIS 7 ( Mallick and Frank , 2002 ) . The Paleoloxodon bone fragment from Weimar-Ehringsdorf is morphologically undiagnostic with respect to species . However , it was found in the Lower Travertine , which was dated to ~233 ka ( Schüler , 2003 ) and where P . antiquus is the only elephantid found so far . We performed DNA extraction , library preparation , hybridization capture and high-throughput sequencing on all four fossils ( Figure 2—source data 1 ) and obtained full mitochondrial genome sequences for all of them ( Figure 2—figure supplement 1 ) . All sequences show short fragment lengths ( Figure 2—figure supplement 2 ) and signals of cytosine deamination compatible with the old age of the specimens ( Figure 2—figure supplement 3 ) . We inferred a phylogeny using the four Paleoloxodon mitochondrial genomes and mitochondrial genomes from 16 M . primigenius , 2 E . maximus and 13 Loxodonta individuals . The latter were chosen for a diversity of haplotypes , including forest elephant derived ( ‘F-clade’ ) haplotypes as well as ‘S-clade’ haplotypes found only among savanna elephants ( Debruyne , 2005 ) . For calibration , we used an estimated divergence of the African elephant lineage from that of Asian elephants and mammoths of 6 . 6–8 . 6 Ma ( Rohland et al . , 2007 ) . Surprisingly , P . antiquus did not cluster with E . maximus , as hypothesized from morphological analyses . Instead , it fell within the mito-genetic diversity of extant L . cyclotis , with very high statistical support ( Figure 2 ) . The four straight-tusked elephants did not cluster together within this mitochondrial clade , but formed two separate lineages that share a common ancestor with an extant L . cyclotis lineage 0 . 7–1 . 6 Ma ( NN ) and 1 . 5–3 . 0 Ma ( WE ) ago , respectively . However , mitochondrial DNA represents a single , maternally inherited locus and does not reflect the full evolutionary history of populations or species . Furthermore , the transfer of mitochondrial DNA between hybridizing species is not unusual when gene flow is strongly male-mediated ( Petit and Excoffier , 2009; Li et al . , 2016; Cahill et al . , 2013 ) , as is the case with elephants . For example , mitochondrial sequences of the F-clade have also been found in some L . africana individuals ( Debruyne , 2005 ) despite the very substantial divergence of their nuclear genomes ( Roca et al . , 2005; Rohland et al . , 2010 ) , a pattern that has been attributed to mitochondrial gene flow from forest to savanna elephants ( Roca et al . , 2005 ) . 10 . 7554/eLife . 25413 . 004Figure 2 . Phylogenetic trees relating the mitochondrial and nuclear sequences of P . antiquus ( NN and WE ) to other elephantids . ( A ) Maximum clade credibility ( MCC ) tree resulting from a BEAST ( Drummond et al . , 2012 ) analysis of 35 complete mitochondrial genomes using 15 , 447 sites . Node bars and numbers show the 95% highest posterior density estimates for node ages and clade support , respectively . Mitochondrial partitioning scheme and molecular and coalescent models are described in ‘Materials and methods’ . ( B ) Pairwise-distance Neighbor-joining tree from between 210 million and 2 . 5 billion base pairs of nuclear shotgun sequence data . Bootstrap support values from 100 replicates are shown inside nodes . Summary statistics of the underlying sequence data are available in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 00410 . 7554/eLife . 25413 . 005Figure 2—source data 1 . This spreadsheet contains summary statistics of all sequence data generated in this study , the sequences of PCR primers used for reconstructing mtDNA sequences of extant elephants , as well as amino acid racemization data on opercula of Bithynia tentaculata from Amersfoort . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 00510 . 7554/eLife . 25413 . 006Figure 2—figure supplement 1 . Sequence coverage of the NN and WE mitochondrial genomes . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 00610 . 7554/eLife . 25413 . 007Figure 2—figure supplement 2 . DNA fragment size distribution inferred from full-length mtDNA sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 00710 . 7554/eLife . 25413 . 008Figure 2—figure supplement 3 . Frequency of C to T substitutions for each position in the sequence alignments . ( A ) Substitution frequencies in mitochondrial alignments . Substitution frequencies are depressed in the Neumark-Nord libraries due treatment with uracil-DNA-glycosylase ( UDG ) . ( B ) In nuclear sequence alignments , the deamination signal could be partly restored by limiting analysis to cytosines in CpG content . Since the majority of cytosines in CpG dinucleotides are methylated in mammalian genomes , deamination leaves thymines , which are not excised by UDG . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 00810 . 7554/eLife . 25413 . 009Figure 2—figure supplement 4 . Maximum likelihood tree from concatenated nuclear protein-coding sequences with bootstrap support values shown inside nodes . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 00910 . 7554/eLife . 25413 . 010Figure 2—figure supplement 5 . Amino acid racemization data . D/L values of Asx , Glx , Ala and Val for the free amino acid ( FAA , panels on the left ) and total hydrolysable amino acid ( THAA , panels on the right ) fraction of bleached Bithynia tentaculata opercula from Amersfoort , Neumark-Nord 1 and 2 . Ranges for samples from UK sites correlated with MIS 5e and MIS seven are indicative only , as effective diagenetic temperatures are likely to have differed significantly between Britain and continental Europe . The boundary of the box closest to zero indicates the 25th percentile , the dashed line within the box marks the mean and the boundary of the box farthest from zero indicates the 75th percentile . The 10th and 90th percentiles are represented by lines above and below the boxes . The results of each duplicate analysis are included in order to provide a statistically significant sample size . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 010 We therefore performed shotgun sequencing of DNA libraries prepared from the two best-preserved NN individuals ( a petrous bone and a molar ) to recover nuclear DNA sequences . When mapped to the L . africana reference genome , 39% and 28% of the sequence reads generated from these specimens were identified as elephant , respectively ( Figure 2—source data 1 ) . A neighbor-joining phylogenetic tree based on ~770M ( petrous ) and ~210M ( molar ) base pairs of P . antiquus nuclear DNA placed P . antiquus and L . cyclotis as sister taxa to the exclusion of L . africana ( Figure 2 ) . A tree with identical topology was obtained using coding sequences only and a maximum likelihood approach ( Figure 2—figure supplement 4 ) . Despite the high sequence error rates associated with the low-coverage genomes generated from the P . antiquus specimens , all nodes in the nuclear trees show maximal bootstrap support . The mitochondrial and nuclear phylogenies thus support a sister group relationship between P . antiquus and L . cyclotis . Despite their geographical proximity , the WE and NN specimens are found in different positions in the mitochondrial tree . Given that the three NN specimens show highly similar mitogenome sequences , we considered whether the sites date to different interglacials . Electron Spin Resonance ( ESR ) dating of tooth enamel has been applied to both sites , suggesting an age of ~117 ka ( range 97–142 ka [Schüler , 2010] ) for the NN1 layers from which our samples originate and of 233 ka ( range 216–250 ka [Schüler , 2003] ) for the WE specimen . In order to better estimate the age of the NN1 site , we used amino acid racemization of snail opercula ( Penkman et al . , 2011 ) , including samples from the continental Eemian type-site of Amersfoort ( Zagwijn , 1961 ) ( Figure 2—source data 1 ) , which is correlated with MIS 5e ( Cleveringa et al . , 2000 ) . The NN1 opercula show similar ( perhaps slightly lower ) levels of biomolecular degradation compared to Amersfoort , suggesting an Eemian age for NN1 ( Figure 2—figure supplement 5 ) . Importantly , NN1 shows very similar levels of amino acid racemization in intra-crystalline protein decomposition as a second site at Neumark-Nord ( NN2 ) , indicating that both sites are of the same age . Considerable evidence supports an Eemian age for NN2 , including palaeomagnetic data that shows a correlation with the MIS 5e Blake event and thermoluminescence dating of flint to ~126 ± 6 ka ( Sier et al . , 2011 ) . These results therefore indicate an Eemian age also for NN1 . Since the WE specimen likely dates to the previous interglacial , this suggests that the very different mitogenomes between WE and NN1 may reflect the contraction and re-expansion of the range of P . antiquus across glacial cycles . Our results have implications both for understanding elephant evolutionary history and for the use of morphological data to decipher phylogenetic relationships among elephants . The strongly supported mitochondrial and nuclear DNA phylogenies clearly demonstrate that Palaeoloxodon antiquus is more closely related to Loxodonta than to Elephas ( Figure 3 ) , suggesting that Elephas antiquus should not be used synonymously for Paleoloxodon antiquus when referring to the taxon . The new phylogeny suggests a remarkable degree of evolutionary transformation , from an ancestor that possessed the features of the cranium and dentition of Loxodonta , shared by both L . africana and L . cyclotis ( Maglio , 1973; Sanders et al . , 2010 ) , to a descendant that is highly similar to Elephas ( sensu stricto , the lineage of the Asian elephant ) in many morphological features . However , it should be noted that currently available genomic data from elephantids only allow for reconstructing the broad picture of elephant evolution . More complex evolutionary scenarios are conceivable , which might explain the presence of some Elephas-like traits in P . antiquus . These could for example involve gene flow , as has been shown for L . africana and L . cyclotis based on mitochondrial evidence ( Roca et al . , 2005 ) . In addition , the very large effective population size of the forest elephants ( Rohland et al . , 2010 ) could have allowed the retention of ancestral traits by incomplete lineage sorting . 10 . 7554/eLife . 25413 . 011Figure 3 . A revised tree of phylogenetic relationships among elephantids , color-coded by their presumed geographical range . DOI: http://dx . doi . org/10 . 7554/eLife . 25413 . 011 In summary , the molecular results presented here urge for a re-examination of morphology across the Elephantidae . This is especially important as the fossil record for elephants dates back several million years , well beyond the survival of ancient DNA . If , for example , P . recki , which was the most abundant Pleistocene elephant species in Africa , is indeed ancestral to P . antiquus and thus also represents a member of the Loxodonta lineage , the interpretation of the fossil record of elephantids in Africa is in strong need of revision . Furthermore , in contrast to the genera Mammuthus and Elephas , which also had their origin in Africa , the lineage of Loxodonta is generally assumed never to have left Africa . Although Osborn , 1942 placed Palaeoloxodon in the Loxodontinae on the basis of several cranial characters , later authors ( Shoshani et al . , 2007; Todd , 2010 ) have rejected this placement in favor of a placement in Elephantinae , restricting Loxodonta ( and Loxodontinae ) geographically to Africa . However , our data reveal that the Loxodonta lineage ( as Paleoloxodon ) also colonized the Eurasian continent . Last , the finding that L . africana is genetically more distant from L . cyclotis than is P . antiquus strongly supports previous evidence that urged recognition of L . cyclotis and L . africana as distinct species and underlines the importance of conservation efforts directed toward African forest elephants . In January 2014 , a fragment of an elephant long bone was discovered during work at the Ehringsdorf quarries . The specimen was removed from the lower travertine ( ~3m above the base and 2 . 5 m below the Pariser horizon ) , which has been dated by micro probe U/Th-series dating of primary travertine ( Mallick and Frank , 2002 ) and ESR dating of tooth enamel ( Schüler , 2003 ) to ~233 ka . A piece of the bone ( inventory number 14/18–1 ) was transferred to the ancient DNA laboratory at the MPI-EVA in Leipzig . Ten extracts were prepared using between 36 and 53 mg of bone ( totaling 425 mg ) material following the method of Dabney et al . 2013 ( Dabney et al . , 2013 ) . From these extracts , 30 libraries were prepared using single-stranded library preparation ( Gansauge and Meyer , 2013 ) with input volumes of 4 , 8 and 12 µl DNA extract ( of 25 µl extract volume ) , respectively . The number of library molecules was determined by digital droplet PCR using Bio-Rad's ( Hercules , CA ) QX200 system with EvaGreen chemistry ( QX200 ddPCR EvaGreen Supermix , Bio-Rad ) and primers IS7 and IS8 ( Meyer and Kircher , 2010 ) following the manufacturer’s instructions ( Figure 2—source data 1 ) . Libraries were then amplified using AccuPrime Pfx DNA polymerase ( Thermo Fisher Scientific , Waltham , MA ) ( Dabney and Meyer , 2012 ) and labeled with two sample-specific indices ( Kircher et al . , 2012 ) . Ancient DNA work on the Neumark-Nord specimens was carried out in the ancient DNA laboratory at the University of Potsdam . Initially , eight specimens were screened for the presence of elephant DNA of which the two best preserved ones ( individual 23 , Landesmuseum Halle museum inventory number HK 2007:25 . 285 , 117; a molar fragment [NEU2A] and a fragmentary upper jaw [NEU8B]; inventory number HK 92:990 ) were selected for further analyses . In addition , in 2013 , the petrous bone of individual 30 ( NEPEC; inventory number HK 2007:25:280 = E15 . 1 . 96 ) was sampled and also used in the analysis . Fourteen DNA extracts were prepared from individual 30 , and six from each of the two other specimens , using approximately 50 mg of bone powder in each extraction . DNA extraction and library preparation were performed as described above but including Archaeoglobus fulgidus uracil-DNA glycosylase in library preparation ( Gansauge and Meyer , 2013 ) , which removes the majority of uracils that are typically present in ancient DNA fragments . In addition , to maximize yields in library preparation , two extracts ( 25 µl each ) from each specimen were combined and 40 µl were used as input for library preparation . Reaction volumes in steps 1–3 of the protocol were doubled to accommodate larger input volumes of extract . Optimal amplification cycle numbers were established using qPCR ( PikoReal Real-Time PCR system , Thermo Fisher Scientific ) with primers IS7 and IS8 ( Gansauge and Meyer , 2013 ) . Libraries were then amplified and labeled with one sample-specific index . After purification ( MinElute PCR purification kit , Qiagen , Germany ) , the different libraries for each sample were pooled . 52-mer capture probes for the enrichment of mtDNA sequences from elephants were designed using the published mtDNA genome sequences of African forest elephant ( NC_020759 ) , Asian elephant ( NC_005129 ) , African savanna elephant ( NC_000934 ) and the mastodon ( NC_009574 ) , with one probe starting at each position in these genomes . Probes containing simple repeats longer than 24 bp ( repetition of the same 1–8 bp sequence motif ) were removed . Single-stranded biotinylated DNA probes were generated as described elsewhere ( Fu et al . , 2013 ) and used for two successive rounds of hybridization capture following a bead-based protocol ( Maricic et al . , 2010 ) . Enriched libraries were pooled and sequenced on one lane of a HiSeq2500 ( Illumina , San Diego , CA ) in paired-end mode ( 2 × 76 cycles plus 2 × 7 cycles index reads; Weimar-Ehringsdorf libraries ) or on an Illumina NextSeq 500 ( 2 × 76 cycles plus 1 × 8 cycles index read; Neumark-Nord libraries ) . Sequences were assigned to their source library requiring perfect matches to one of the expected indices or index pairs and overlap-merged to reconstruct full-length molecule sequences ( Renaud et al . , 2014 ) . Due to the different properties of the data obtained from Weimar-Ehringsdorf and Neumark-Nord with regard to sequence length distribution and damage patterns ( Figure 2—figure supplements 2 and 3 ) , two different strategies were used for mapping and consensus calling . To minimize the loss of alignments due to the high frequencies of damaged-induced substitutions in the Weimar-Ehringsdorf data , mapping to the L . cyclotis mtDNA genome ( JN673264 ) was performed as previously described for the Sima de los Huesos mtDNA assemblies ( Dabney et al . , 2013 ) , using BWA and allowing up to five C to T substitutions but not more than three of other types . The sequences from Neumark-Nord were mapped with ‘ancient’ parameters as described elsewhere ( Meyer et al . , 2012 ) . PCR duplicates were removed with bam-rmdup ( Stenzel , 2014 Biohazard , available from https://bitbucket . org/ustenzel/biohazard ) by calling a consensus from sequences with identical alignment start and end coordinates . Sequences shorter than 30 bp were discarded . An overview of the DNA extracts , libraries and sequences generated in this study is provided in Figure 2—source data 1 . When visually inspecting the Weimar-Ehringsdorf sequence alignments , we identified several regions in the mitochondrial genome where more than one sequence variant was present . Based on BLAST searches on a subset of these sequences , we found that they derived from present-day human or microbial contamination . We thus aligned all sequences to the identified contaminant genomes ( GenBank accession nos . NC_012920 , AF365635 and CP008889 ) and removed sequences that showed a greater similarity to one of the contaminants than to the African forest elephant mtDNA . No removal of contaminant sequences was necessary for the Neumark-Nord samples . To minimize the impact of damage-derived C to T substitutions on consensus calling , all T occurring in the first and last three positions of the Weimar-Ehringsdorf sequences were substituted by N . Next , a position-based tabular output was generated from the alignment files using the ‘mpileup’ function of SAMtools ( Li et al . , 2009 ) . This file was used to call the consensus at positions with minimum sequence coverage of 3 if the sequences were in at least 67 . 0% agreement . At three positions in the mtDNA genome ( positions 384 , 8467 , 8469 ) with low consensus support , we spotted obvious alignment errors in one or all specimens and determined the consensus base manually . Apart from a ~500 bp stretch of repetitive sequence in the D-loop , which cannot be reconstructed with short DNA fragments , only four positions remain undetermined in the Weimar-Ehringsdorf sequence and even fewer ( between none and three ) in the Neumark-Nord sequences . We estimated mitochondrial phylogenies using the software BEAST ( Drummond et al . , 2012 ) v 1 . 8 . 2 and a data set including 31 complete mitochondrial genomes ( GenBank accession nos . ; L . cyclotis: JN673264 , JN673263 , KJ557424 , KJ557423 , KY616976 , KY616979 , KY616978; L . africana: WA4020 , KR0014 , KR0138 , NC000934 , DQ316069 , AB443879; E . maximus: NC005129 , DQ316068; M . primigenius: DQ316067 , NC007596 , EU155210 , EU153449 , EU153455 , EU153456 , EU153458 , EU153445 , EU153446 , EU153447 , EU153448 , EU153452 , EU153453 , EU153454 , JF912200; M . columbi: JF912199 ) . For three of the L . cyclotis individuals ( LO3505 , LO3508 and DS1511 ) and three L . africana individuals ( WA4020 , KR0014 , KR0138 ) , only partial mitochondrial sequences were previously published . Full genome sequences were obtained using previously collected samples ( Ishida et al . , 2013 ) and the amplification and sequencing strategy detailed by Brandt et al . ( 2012 ) , except that additional primers were used in sequencing ( Figure 2—source data 1 ) . The complete mitochondrial genome sequences were partitioned prior to analysis into four partitions , representing concatenated genes ( with ND6 reversed ) , tRNAs , rRNAs , and the control region , and analyses were performed with and without the control region fragment . All BEAST analyses were performed assuming the flexible skygrid coalescent model ( Gill et al . , 2013 ) and the uncorrelated lognormal relaxed molecular clock ( Drummond et al . , 2006 ) . We calibrated the molecular clock using the ages of ancient tips and a lognormal prior with a mean of 7 . 6 million years and standard deviation of 500 , 000 years for the divergence of the Loxodonta and Elephas/Mammuthus lineages ( Rohland et al . , 2007 ) . Ages of the ancient samples were sampled from normal distributions derived from stratigraphic and previously estimated radiometric dates: Neumark-Nord: 142–92 ka ( Schüler , 2010 ) ; Ehringsdorf 250–216 ka ( Mallick and Frank , 2002 ) . Separate evolutionary rates and models of nucleotide substitution , as estimated using jModelTest ( Posada , 2008 ) , were estimated for each partition in the alignment . We ran two MCMC chains for 60 million iterations each , with trees and model parameters sampled every 6000 iterations . Chain convergence and parameter sampling were examined by eye using Tracer v 1 . 6 ( Rambaut A , Suchard MA , Xie D & Drummond AJ ( 2014 ) Tracer v1 . 6 , available from http://beast . bio . ed . ac . uk/Tracer ) . The first 10% of samples were discarded from each run after which the two runs were combined . Trees were summarized and maximum clade credibility ( MCC ) trees identified using TreeAnnotator v 1 . 8 . 2 , which is distributed as part of the BEAST package . MCC trees were edited and annotated using FigTree v1 . 4 . 2 ( http://tree . bio . ed . ac . uk/software/figtree/ ) . Libraries from the three Neumark-Nord samples and another sample ( not included in this study ) were pooled in equimolar concentrations and shotgun-sequenced on an Illumina NextSeq 500 ( 2 × 76 bp cycles ) at Harvard Medical School . Following determination of endogenous content and complexity in each library , two of them ( NEPEC from the petrous bone and NEU2A from the molar fragment ) were chosen for additional sequencing and were pooled together with another sample ( not included in this study ) for one NextSeq 500 run . Sequences were assigned to their source library according to their index allowing for one mismatch . Adapters were trimmed and paired-end sequences were merged with SeqPrep 1 . 1 ( https://github . com/jstjohn/SeqPrep ) using default parameters but with a modification in the source code to retain the best base quality scores in the merged region . Merged sequences shorter than 30 bp were discarded . Alignment to the African savanna elephant reference genome ( loxAfr4; downloaded from ftp://ftp . broadinstitute . org/distribution/assemblies/mammals/elephant/loxAfr4/ ) was performed with BWA’s version 0 . 7 . 8 ( Li and Durbin , 2009 ) using ‘ancient’ parameters and SAMtools’ v . 0 . 1 . 19 ‘samse’ command ( Li et al . , 2009 ) . A custom script was used to remove duplicates , which takes into account the alignment coordinates of both ends of the sorted sequences and their orientation . From the first sequencing run , 47% and 35% of the sequences from the libraries from the petrous bone and the molar fragment ( NEPEC and NEU2A , respectively ) aligned to the reference genome while only 0 . 5% of the sequences from the third library ( NEU8B ) aligned to the reference genome . The high percentage of mapped sequences in the petrous sample is consistent with previous reports on the superior DNA preservation in this part of the skeleton ( Gamba et al . , 2014 ) . Following the second sequencing run , the total endogenous content of the first two libraries was estimated to 39% and 28% , with an average sequence length of 39 bp and 38 bp , respectively ( Figure 2—source data 1 ) . The average depth of coverage was 0 . 65-fold for NEPEC and 0 . 14-fold for NEU2A . Both of them showed low frequencies of C to T substitutions at the 5’ and 3’ end , which are characteristic for Afu UDG-treated single-stranded libraries ( Gansauge and Meyer , 2013 ) , except for in CpG context , where deamination of 5-methylcytosine leaves thymine and not uracil ( Figure 2—figure supplement 3 ) . We also processed sequencing data from an African forest elephant ( SL0001 ) that was sequenced to high-coverage at the Broad Institute and re-processed sequencing data of an Asian elephant from ( Lynch et al . , 2015 ) . We trimmed adapters with SeqPrep 1 . 1 using default parameters and aligned paired-end reads to loxAfr4 using BWA’s ‘aln’ algorithm and SAMtools’ ‘sampe’ command . Duplicate reads were removed with SAMtools’s ‘rmdup’ . Moreover , we used the high-coverage genome of a woolly mammoth ( Wrangel ) from ( Palkopoulou et al . , 2015 ) . The woolly mammoth alignments were re-processed for removal of duplicate reads with the custom script mentioned above . To determine the phylogenetic relationships between the two P . antiquus specimens and other members of the Elephantidae family , we called pseudo-haploid consensus sequences for all autosomes of the two P . antiquus samples ( ~770 and~210 million sites , respectively ) . Sites with base quality below 30 and reads with mapping quality below 30 were filtered out . To exclude post-mortem damage-derived C to T substitutions , we trimmed 2 bp from the ends of all reads . We included regions of the loxAfr4 genome for which at least 90% of all possible 35-mers do not find a match at another position allowing for up to one mismatch , similar to the mappability filter described in ( Prüfer et al . , 2014 ) . We used a majority-allele calling rule that required at least one read aligned at each position of the genome . Using the same approach , we called sequences for an Asian elephant ( Uno [Lynch et al . , 2015] ) , a woolly mammoth ( Wrangel [Palkopoulou et al . , 2015] ) and an African forest elephant ( SL0001; Broad Institute ) . We also used the reference sequence loxAfr4 , as an African savanna elephant . We estimated the number of differences per base-pair for pairwise comparisons of all sequences and constructed a distance matrix , from which we built a Neighbor-joining ( NJ ) tree using PHYLIP version 3 . 696 ( Felsenstein , 2005 ) . To obtain support values for the nodes of the tree , we performed a bootstrap analysis ( 100 replicates ) by splitting all autosomes in blocks of 5 Mb and randomly sampling blocks with replacement and built a majority-rule consensus tree . We also extracted coding DNA sites ( CDS ) of protein-coding genes using the Ensembl 87 release for the loxAfr3 genome ( downloaded from http://www . ensembl . org/ ) from each elephant genome sequence . CDS mapping to unknown chromosomes as well as CDS containing partial codons were excluded , resulting in a total of 86 , 212 CDS . Multiple sequence alignments were generated for each gene using MAFFT –ginsi ( Katoh et al . , 2002; Katoh and Standley , 2013 ) with 1000 iterations , which were concatenated into a single fasta file . Maximum likelihood phylogenetic analysis was performed with RAxML v8 . 2 ( Stamatakis , 2014 ) with the GTRGAMMA model of nucleotide substitutions and 100 bootstrap trees . The resulting phylogeny is identical in its topology to that of the NJ tree with 100% bootstrap support ( Figure 2—figure supplement 4 ) . Amino acid racemization ( AAR ) analyses were undertaken on the intra-crystalline protein from four individual Bithynia tentaculata opercula from the Eemian type-site , Amersfoort ( Cleveringa et al . , 2000 ) : Amersfoort-1 , upper depth 27 . 71 , lower depth 28 . 50 ( NEaar 2982–3 , 3972 and 4681 ) and compared with previously published data from a single horizon at Neumark-Nord 1 ( 15 . 5 . 87/2 , Schluffmudde , 25 cm under Anmoor = surface of the lower shore area; NEaar 5698–5703 [Penkman , 2010] ) and several horizons from Neumark-Nord 2 ( Sier et al . , 2011 ) . All samples were prepared using procedures of isolating the intra-crystalline protein by bleaching ( Penkman et al . , 2008 ) . Two subsamples were then taken from each shell; one fraction was directly demineralized and the free amino acids analyzed ( referred to as the 'free' amino acids , FAA , F ) , and the second was treated to release the peptide-bound amino acids , thus yielding the 'total' amino acid concentration , referred to as the ‘total hydrolysable amino acid fraction ( THAA , H* ) . Samples were analyzed in duplicate by RP-HPLC . During preparative hydrolysis , both asparagine and glutamine undergo rapid irreversible deamination to aspartic acid and glutamic acid , respectively ( Hill , 1965 ) . It is therefore not possible to distinguish between the acidic amino acids and their derivatives and they are reported together as Asx and Glx , respectively . The D/L values of aspartic acid/asparagine , glutamic acid/glutamine , alanine and valine ( D/L Asx , Glx , Ala , Val ) are then assessed to provide an overall estimate of intra-crystalline protein decomposition ( Penkman et al . , 2011 ) .	Understanding how extinct species are related to each other or to their living relatives is often a difficult task . Many extinct species have been identified only from incomplete fragments of some of their bones . However , even if complete skeletons have been found , determining the relationships between species can be tricky because researchers often have to rely solely on the shapes of the bones . It is sometimes possible to retrieve DNA sequences from fossil bones . This is easier with younger fossils and those that have been recovered from cold environments . Ancient DNA sequences have been retrieved from only a few fossils older than 100 , 000 years , but such DNA sequences can be tremendously useful in determining how different species are related to each other . Today there are three living elephant species: the African forest elephant , the African savanna elephant and the Asian elephant . However , there are many extinct elephant species . For example , the European straight-tusked elephant went extinct at least 30 , 000 years ago , although most of the fossils that have been discovered are at least 100 , 000 years old . Straight-tusked elephants are generally assumed to be closely related to the Asian elephant , but this conclusion had been based solely on reconstructing skeletons . Meyer et al . have now obtained DNA sequences from fossils of four straight-tusked elephants ranging from around 120 , 000 to 240 , 000 years in age . These sequences were analysed to determine how straight-tusked elephants are related to the three living elephant species and the extinct mammoth , the DNA sequences for which can be found in public databases . The analyses revealed that straight-tusked elephants are in fact most closely related to the African forest elephant , not the Asian elephant as previously thought . This result completely changes our picture of elephant evolution and suggests that it is extremely difficult to determine elephant relationships based on the shape of their skeleton alone . It also shows that the African elephant lineage was not restricted to the African continent ( the place where all elephant lineages originated ) , but that it also left Africa . Overall , the results presented by Meyer et al . confirm that DNA sequences are of critical importance for understanding the evolution of animals . Future research should include obtaining DNA sequences from additional extinct elephant species as well as careful re-evaluation of skeletal measurements for reconstructing elephant evolution .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Material", "and", "methods" ]	[ "short", "report", "genetics", "and", "genomics" ]	2017	Palaeogenomes of Eurasian straight-tusked elephants challenge the current view of elephant evolution
Caveolae-associated protein 3 ( cavin3 ) is inactivated in most cancers . We characterized how cavin3 affects the cellular proteome using genome-edited cells together with label-free quantitative proteomics . These studies revealed a prominent role for cavin3 in DNA repair , with BRCA1 and BRCA1 A-complex components being downregulated on cavin3 deletion . Cellular and cell-free expression assays revealed a direct interaction between BRCA1 and cavin3 that occurs when cavin3 is released from caveolae that are disassembled in response to UV and mechanical stress . Overexpression and RNAi-depletion revealed that cavin3 sensitized various cancer cells to UV-induced apoptosis . Supporting a role in DNA repair , cavin3-deficient cells were sensitive to PARP inhibition , where concomitant depletion of 53BP1 restored BRCA1-dependent sensitivity to PARP inhibition . We conclude that cavin3 functions together with BRCA1 in multiple cancer-related pathways . The loss of cavin3 function may provide tumor cell survival by attenuating apoptotic sensitivity and hindering DNA repair under chronic stress conditions . Caveolae are an abundant surface feature of most vertebrate cells . Morphologically , caveolae are 50–100 nm bulb-shaped structures attached to the plasma membrane ( Parton and del Pozo , 2013 ) . One of the defining features of this domain is the integral membrane protein caveolin-1 ( CAV1 ) . CAV1 is a structural component of caveolae regulating diverse cellular processes , including endocytosis , vesicular transport , cell migration , and signal transduction ( Parton and del Pozo , 2013 ) . Recently , we and others have characterized a caveolar adaptor molecule , caveolae-associated protein 3 ( cavin3 ) ( McMahon et al . , 2009 ) . Cavin3 belongs to a family of proteins that includes caveolae-associated protein 1 ( cavin1 ) , caveolae-associated protein 2 ( cavin2 ) , and the muscle-specific member caveolae-associated protein 4 ( cavin4 ) ( Ariotti and Parton , 2013; Bastiani et al . , 2009; Hansen et al . , 2009; Kovtun et al . , 2015; Lo et al . , 2015; McMahon et al . , 2009 ) . Cavin3 is epigenetically silenced in a range of human malignancies ( Xu et al . , 2001 ) , principally due to hypermethylation of its promoter region ( Carén et al . , 2011; Kim et al . , 2014; Lee et al . , 2008; Lee et al . , 2011; Martinez et al . , 2009; Tong et al . , 2010; Zöchbauer-Müller et al . , 2005 ) . Furthermore , cavin3 has been previously suggested to interact with BRCA1 , although no data has been formally published to support this interaction ( Xu et al . , 2001 ) . Several studies have implicated cavin3 in a broad range of cancer-related processes including proliferation , apoptosis , Warburg metabolism , as well as in cell migration and matrix metalloproteinase regulation; however , the molecular basis of its actions is poorly understood ( Hernandez et al . , 2013; Toufaily et al . , 2014 ) . BReast CAncer gene 1 ( BRCA1 ) is a significant breast cancer suppressor gene . It is one of the most frequently mutated genes in hereditary breast cancer ( King and Marks , 2003; Miki et al . , 1994; Venkitaraman , 2002 ) . Also , BRCA1 levels are reduced or absent in many sporadic breast cancers due to gene silencing by promoter methylation or downregulation of the gene by other tumor suppressors or oncogenes ( Mueller and Roskelley , 2003; Turner et al . , 2004 ) . BRCA1 has been implicated in a remarkable number of processes , including cell cycle checkpoint control , DNA damage repair , and transcriptional regulation ( reviewed by Lord and Ashworth , 2016; Savage and Harkin , 2015 ) . At the molecular level , accumulated evidence suggests that BRCA1 plays an integral role in the formation of several macromolecular complexes ( BRCA1 A , BRCA1 B , and BRCA C , with different associated proteins ) that participate in distinct processes to repair DNA damage ( Deng and Brodie , 2000; Huen et al . , 2010; Roy et al . , 2012; Scully et al . , 1997; Scully et al . , 1999; Scully and Livingston , 2000; Wang et al . , 2007 ) . Specifically , the BRCA1 A-complex consists of BRCA1 in association with RAP80 , the deubiquitinating ( DUB ) enzymes BRCC36 and BRCC45 , MERIT-40 , and the adaptor protein ABRAXAS1 ( Harris and Khanna , 2011; Her et al . , 2016; Savage and Harkin , 2015; Wang et al . , 2007 ) . The BRCA1 A-complex participates in DNA repair by targeting BRCA1 to ionizing radiation ( IR ) -inducible foci; this occurs when RAP80 interacts with K63 poly-ubiquitin chains at sites of double strand breaks ( DSBs ) where the DNA damage marker γH2AX is phosphorylated ( Yan and Jetten , 2008 ) . BRCA1-A complex is thought to target BRCA1 to sites of DSB through interaction with ubiquitin-interacting motifs of RAP80 , which recognize the Lys63 poly-ubiquitin chains of H2AX ( Sobhian et al . , 2007; Wang et al . , 2007; Yan and Jetten , 2008 ) . BRCA1 is also bound to BRCA1-associated Ring Domain 1 ( BARD1 ) , an interaction that is necessary for BRCA1 protein stability , nuclear localization , and E3 ubiquitin ligase activity ( Irminger-Finger et al . , 2016 ) . In addition , BRCA1 is also a nuclear-cytoplasmic shuttling protein , and increasing evidence suggests that BRCA1 function can be controlled via active shuttling between subcellular compartments ( Fabbro et al . , 2002; Feng et al . , 2004 ) . We identify a novel function for cavin3 mediated through its interaction with BRCA1 , leading to regulation of BRCA1 levels , subcellular location , and function . We show that cavin3 controls BRCA1 functions in UV-induced apoptosis and cell protection against DNA damage through downregulated recruitment of the BRCA1 A-complex to DNA lesions in response to UV damage . As a first step to investigate the cell biology of cavin3 , we undertook an unbiased approach to characterize its cellular proteome , using label-free quantitative ( LFQ ) proteomics . We deleted cavin3 by genome editing in HeLa cells , a well-characterized model system that has been used extensively to study caveolae ( Bohmer and Jordan , 2015; Boucrot et al . , 2011; Hao et al . , 2012; Hirama et al . , 2017; Pang et al . , 2004; Rejman et al . , 2005; Sinha et al . , 2011; Figure 1A , Figure 1—figure supplement 1A ) . Global proteome analyses were carried out with three replicates from matched WT and cavin3 KO HeLa cells . Cells were SILAC-labeled and subjected to mass spectrometric analysis after lysis . Relative protein expression differences were then determined using label-free quantitation ( Figure 1A ) . A total of 4206 proteins were robustly quantified with >2 unique peptides and an FDR < 1 . 0% in at least two out of three replicates ( Figure 1A , details in Supplementary file 1 ) . To validate these results , we immunoblotted for several proteins involved in diverse cellular processes . Levels of these proteins were consistent with the proteomic analysis ( Figure 1—figure supplement 1B ) . Their levels were restored by the expression of exogenous cavin3 , confirming the specificity of the KO effect ( Figure 1—figure supplement 1C ) . Our analysis revealed distinct cavin3-dependent protein networks that might yield new insights into its cellular function . Initial inspection of differentially expressed protein by Gene Ontology analysis revealed that many proteins involved in DNA repair were altered in cavin3 KO cells ( Figure 1B , C and Supplementary file 2 ) ; see Supplementary file 3 for further analysis of cavin3-dependent pathways . Strikingly , BRCA1 ( ~1 . 5-fold decrease ) and many components of the BRCA1 A-complex , BRCC36 ( ~1 . 5-fold decrease ) , MDC1 ( ~1 . 7-fold decrease ) , and the newly described UBE4A ( ~2 . 2-fold decrease , Baranes-Bachar et al . , 2018 ) , were reduced in cavin3 KO cells that were confirmed by western analysis ( Figure 1D , Figure 1—figure supplement 1D ) . In contrast , 53BP1 protein levels were increased in cavin3 KO cells ( Figure 1D , Figure 1—figure supplement 1D ) . Accordingly , we elected to pursue the relationship between cavin3 and BRCA1 in greater detail . First , we asked whether cavin3 and BRCA1 might interact in the cytosol . Recent studies suggest that the release of cavin proteins into the cytosol can allow interaction with intracellular targets ( Gambin et al . , 2014; McMahon et al . , 2019; Sinha et al . , 2011 ) . To test whether non-caveolar cavin interacts with BRCA1 , we used MCF7 cells as a model system . These cells lack endogenous CAV1 , cavins , and caveolae ( Gambin et al . , 2014; McMahon et al . , 2019 ) , and so expressed cavin proteins are predominantly cytosolic . BRCA1-GFP was coexpressed in MCF7 cells with exogenous mCherry-tagged cavins-1 , 2 , 3 , and mCherry-CAV1 , and interactions between these proteins were measured in cytoplasmic extracts using two-color single-molecule coincidence ( SMC ) detection . The numbers of photons detected in green and red channels were plotted as a function of time where each fluorescent burst was analyzed for the coincidence between the GFP and cherry fluorescence that reflects co-diffusion of at least two proteins with different tags , the total brightness of the burst , indicating the number of proteins present in the oligomer and the burst profile that is determined by the rate of diffusion and reflects the apparent size of the complex ( Gambin et al . , 2014 ) . This revealed a specific association between BRCA1 and cavin3-mCherry , but not with the other cavin proteins ( Figure 2A–E ) . Quantitatively , 60% of BRCA1-GFP associated with cavin3-mCherry ( Figure 2D ) . The distribution of bursts revealed the behavior of monomeric GFP . This data was used to calibrate the brightness profile and estimate the number of BRCA1-GFP molecules . We concluded that overexpressed BRCA1 primarily exists in a dimeric state when expressed in MCF7 cells and that a dimer of overexpressed BRCA1 interacts with a monomer of exogenous cavin3 ( Figure 2F ) . Similar results were obtained when BRCA1-GFP and cavin3-mCherry were coexpressed in MDA-MB231 cells , a cell line with endogenous caveolar proteins and abundant caveolae at the plasma membrane ( Figure 2—figure supplement 1A–E ) . These findings implied that BRCA1 and cavin3 can interact in the cytosol , irrespective of the cells' caveolar state . We then used a Leishmania cell-free system ( Gambin et al . , 2014; Sierecki et al . , 2013 ) to test whether these proteins can interact directly . Indeed , a construct bearing the first 300 amino acids of BRCA1 ( 1–300 , tr-BRCA1 ) , which contains the nuclear export signal ( NES ) and BARD1 binding sites ( Figure 2G ) , was associated with cavin3 ( Figure 2J ) , but not with the other cavin proteins ( Figure 2H , I ) . These data suggest that cavin3 directly binds to the N-terminus of BRCA1 . Finally , we used in situ proximity ligation assay ( PLA ) technology ( Söderberg et al . , 2007 ) to probe for the protein-protein association within intact cells . GFP-tagged cavins or CAV1-GFP were expressed in MCF7 cells , and potential associations between transgenes and endogenous BRCA1 were analyzed using anti-BRCA1 and anti-GFP antibodies . Positive interactions in PLA analyses are revealed by fluorescent puncta ( Figure 3A–E ) . Puncta were evident throughout the cytosol of cells expressing cavin3-GFP , but not with the other cavins , CAV1-GFP or GFP alone ( Figure 3A–E , quantitation in Figure 3F ) . Additional experiments using different combinations of antibodies ( e . g . , rabbit antibodies against endogenous BRCA1 together with mouse anti-GFP antibodies; Figure 3—figure supplement 1 ) yielded similar results . Control experiments ( GFP alone , BRCA1 alone , absence of PLA probes , and no antibody ) yielded few puncta ( Figure 3—figure supplement 2A–E ) . Collectively , these studies suggest that BRCA1 can interact with cavin3 directly in vitro and that expressed cavin3 can associate with endogenous BRCA1 in cells . We next examined the relationship between cavin3 and the subcellular localization of BRCA1 . Immunofluorescence revealed a typical nuclear staining pattern for endogenous BRCA1 with little cytoplasmic staining in control MCF7 cells and cells expressing cavin1-GFP ( Figure 4A ) . In contrast , the expression of cavin3-GFP increased cytosolic staining for endogenous BRCA1 ( Figure 4A ) , and this was confirmed by quantitative analysis of the protein distribution ( Figure 4B ) . Western blotting revealed that cavin3-GFP selective increased total cellular levels of BRCA1 ( Figure 4C , quantitation in Figure 4—figure supplement 1A ) . This represents a post-transcriptional effect of cavin3 as BRCA1 mRNA levels were not significantly increased ( Figure 4—figure supplement 1B ) . Interestingly , the proteasome inhibitor , MG132 , increased BRCA1 levels in control cells , consistent with evidence for proteasomal degradation of BRCA1 ( Choudhury et al . , 2004 ) . However , it did not increase the already-elevated levels of BRCA1 found in cavin3-GFP cells ( Figure 4D , quantitation in Figure 4—figure supplement 1C ) . Dependence of BRCA1 on cavin3 was also evident when cavin3 was depleted in either A431 and MDA-MB231 cells , using two different siRNAs ( Figure 4E , quantitation in Figure 4—figure supplement 1D , Figure 4—figure supplement 2A , C ) . These cell lines express cavin3 , CAV1 , and BRCA1 proteins and present caveolae at the plasma membrane ( Figure 4—figure supplement 1E ) . In both cases , cavin3 depletion caused a significant decrease in BRCA1 ( Figure 4E , quantitation in Figure 4—figure supplement 1D , Figure 4—figure supplement 2A , C ) , and this was abrogated by proteasome inhibition ( Figure 4G ) . Immunofluorescence staining revealed that BRCA1 was reduced in the cytosol and nuclei of cavin3 siRNA cells ( Figure 4—figure supplement 3 ) . Interestingly , depletion of BRCA1 with two independent siRNAs significantly decreased endogenous cavin3 protein levels in these cells ( Figure 4F , quantitation in Figure 4—figure supplement 1F , Figure 4—figure supplement 2B , D ) . Taken with our earlier work on HeLa cells , these results collectively show that cavin3 can support BRCA1 protein levels in a variety of cancer cell systems . What might induce cavin3 to interact with BRCA1 ? A variety of stresses cause caveolae to flatten and disassemble , releasing cavins into the cytosol . We , therefore , hypothesized that stimuli that induce caveola disassembly might induce the association of cavin3 with BRCA1 . First , we tested a role for mechanical stress by swelling cells with hypo-osmotic medium . We used A431 cells for these experiments as they have abundant caveolae . The total association between endogenous cavin3 and endogenous BRCA1 , and their association in the nucleus , was significantly increased by hypo-osmotic stimulation , as measured by PLA ( Figure 5A ) . No interaction was seen with a range of control proteins , including the nuclear proteins PCNA , flottilin1 and Aurora kinase ( Figure 5B–E ) . These findings suggested that mechanical disassembly of caveolae could promote the association of cavin3 with BRCA1 both in the cytosol and the nucleus . Nest , we tested the effect of non-mechanical stimuli by exposing cells to either UV ( 2 min pulse , 30 min chase ) or oxidative stress with hydrogen peroxide ( H2O2 , 200 μM , 30 min ) . PLA showed that the interaction between endogenous BRCA1 and cavin3 was increased by both these stimuli ( Figure 6A–D , top panel , quantitation in Figure 6E ) . A more extended time course further demonstrated that association between these proteins was evident at 30 min and maintained at low levels for up to 4 hr ( Figure 6F , G ) . Interestingly , this coincided with a decrease in the interaction between cavin3 and cavin1 , which occurs in caveolae ( Figure 6A–D , bottom panel , quantitation in Figure 6G ) . Similar effects were seen in MDA-MB231 cells ( Figure 6—figure supplement 1 ) . Control experiments ( knockdown of cavin3 or BRCA1 in untreated and UV-treated A431 cells ) yielded few puncta ( Figure 6—figure supplement 2 ) , consistent with the notion that cavin3 was moving from caveolae into the cytosol to interact with BRCA1 . Our findings indicate that cavin3 can be released to interact with BRCA1 when caveolae disassemble in response to various mechanical and non-mechanical stimuli . Next , we sought to evaluate the potential functional consequences of this stress-inducible association of cavin3 with BRCA1 . As cytoplasmic BRCA1 has been implicated in cell death pathways ( Dizin et al . , 2008; Thangaraju et al . , 2000; Wang et al . , 2010 ) , we asked if cavin3 affects the sensitivity of cells to apoptosis induced by UV exposure . We found that LDH release , used as an index of membrane damage , was consistently increased after 2 min UV exposure in MCF7 cells that overexpressed cavin3-GFP , but not with cavin1-GFP ( Figure 7A ) . This cell damage reflected apoptosis induction confirmed by staining for annexin V ( which marks early apoptosis , Figure 7B ) and the DNA dye 7-amino-actinomycin 7 ( 7-AAD , late apoptosis , Figure 7C ) . Both apoptotic markers were enhanced by cavin3-GFP overexpression . Thus , cavin3 could sensitize MCF7 cells to UV-induced apoptosis . We then asked whether this effect also operated in cancer cells with endogenous expression of cavin3 . Indeed , overexpression of cavin3-GFP significantly increased LDH release from UV-treated A431 and MDA-MB231 cells ( Figure 7D , F ) . Furthermore , depletion of endogenous cavin3 reduced LDH release from these cells after UV stimulation ( Figure 7E , G , controls in Figure 7—figure supplement 1A , B ) . Together , these findings indicate that cavin3 sensitizes cells to apoptosis induced by UV . BRCA1 also sensitized A431 and MDA-MB231 cells to apoptosis , as evident when exogenous BRCA1 was overexpressed or the endogenous protein was depleted ( Figure 7E , G , controls in Figure 7—figure supplement 1C , D ) . Therefore , we further examined the relationship between BRCA1 and cavin3 . Overexpression of BRCA1 in cavin3-depleted A431 or MDA-MB231 cells or overexpression of cavin3 in BRCA1-depleted cells restored UV-induced apoptosis to control levels . This indicated that these two proteins have a similar sensitizing effect on UV-induced apoptosis ( Figure 7E , G ) . These results suggest a pro-apoptotic role for both cavin3 and BRCA1 in stress-induced cancer cells . Similarly , in MCF7 cells expression of cavin3 alone or in combination with BRCA1 restored the sensitivity of BRCA1 KD cells to UV-induced apoptosis ( Figure 7—figure supplement 1E ) . We further exposed WT and cavin3 KO HeLa cells to a range of stresses that allow interaction with BRCA1 , including hypo-osmotic medium , UV , and oxidative stress ( Figure 7—figure supplement 2A–D ) . Cavin3 KO cells exhibited enhanced resistance to all stressors , and apart from oxidative stress , this was time-dependent ( Figure 7—figure supplement 2A–D ) . Overall , these findings suggest that BRCA1 and cavin3 participate together in the cellular stress response . In addition to promoting apoptosis , BRCA1 , notably via its BRCA1 A-complex , has also been implicated in DNA repair to limit the mutational risk in stressed cells that evade apoptosis . As noted earlier , we found that BRCA1 A-complex components were reduced at steady state in cavin3 KO HeLa cells ( Figure 1A ) . Next , we examined UV treatment on the level of these components in WT and cavin3 KO HeLa cells . As shown in Figure 8A–C , UV treatment of WT cells upregulated the expression of cavin3 , BRCA1 , the DNA damage marker , RAD51 , and the A-complex proteins MDC1 , Rap80 , RNF168 , and Merit40 . Strikingly , the upregulation of BRCA1 , RAD51 , and the BRCA1 A-complex proteins was dramatically reduced in cavin3 KO cells ( Figure 8A , C , quantitation in Figure 8—figure supplement 1 ) . This suggested that cavin3 can influence the ability of BRCA1 to repair damaged DNA . To test this , we first examined the response of BRCA1 to DNA damage . BRCA1 relocates to form foci at sites of DNA DSBs . Indeed , we found that BRCA1 foci increased within 30 min of UV irradiation ( Figure 8D , E ) ; however , this was significantly reduced in cavin3 KO cells ( Figure 8E ) . Similarly , the recruitment of RAP80 and γH2AX was reduced in cavin3 KO cells , suggesting that DNA repair might be fundamentally compromised in these cells ( Figure 8E ) . Previous studies have shown that loss of functional BRCA1 protein leads to defects in DSB repair by homologous recombination and renders cells hypersensitive to PARP inhibitors through the mechanism of synthetic lethality ( Ashworth , 2008; Bryant et al . , 2005; Farmer et al . , 2005; Helleday et al . , 2005 ) . Therefore , we asked whether cavin3 KO cells that are BRCA1 deficient are also sensitive to the PARP inhibitor , AZD2461 . Clonogenic survival assays and cell viability studies revealed that cavin3-deficient HeLa cells ( red dots ) were more sensitive to the PARP inhibitor AZD2461 at nM concentrations than control WT HeLa cells ( black dots , Figure 8—figure supplement 2 ) . As another means to look at PARP loss , WT and cavin3 HeLa KO cells were also depleted of PARP1 using CRISPR/Cas9 genome editing . Cavin3 and PARP1 KO cells failed to produce colonies in clonogenic survival assays with reduced cell viability ( pink dots , Figure 8—figure supplement 2 ) . These findings suggest that cavin3-deficient HeLa cells are sensitive to PARP inhibition , suggesting that cavin3 and BRCA1 are involved in homologous recombination repair . Furthermore , these findings suggest that PARP1 is a potential synthetic lethal partner for cavin3 . We evaluated DNA strand breaks in control and PARP-treated WT HeLa and cavin3 KO cells using a comet assay , which revealed increased DNA damage only in PARP-treated cavin3 KO cells following a 6-day treatment ( Figure 8F ) . Recent reports have linked 53BP1 loss to PARP inhibitor resistance , presumably , as loss of 53BP1 partially restores homologous recombination repair in BRCA1-deficient cells ( Bouwman et al . , 2010; Bunting et al . , 2009; Cao et al . , 2009; Turner et al . , 2007; Yang et al . , 2017 ) . This restoration is made possible because homologous recombination and non-homologous end-joining repair pathways compete to repair DNA breaks during DNA replication . Therefore , we determined the dependence of the physiological outcomes on BRCA1 in cavin3 KO cells by rescue experiments with concomitant knockout of 53BP1 . Loss of 53BP1 in cavin3 HeLa KO cells could revert the PARP sensitivity of these cells to WT cell levels as demonstrated in clonogenic survival and cell viability assays ( orange dots , Figure 8—figure supplement 2A–C ) . These findings agree with several studies demonstrating that homologous recombination DNA repair is partially restored in BRCA1-deficient cells following 53BP1 loss ( Bouwman et al . , 2010; Bunting et al . , 2009; Cao et al . , 2009; Turner et al . , 2007; Yang et al . , 2017 ) . We further evaluated several other proteins: chromodomain helicase DNA-containing protein 3 ( CHD3 , an epigenetic modulator ) and Fanconi anemia ( FA ) complementation Group 2 ( FANCD2 , a DNA damage sensor protein ) that were specifically upregulated in cavin3 KO cells and that are involved in different aspects of DNA repair . These proteins represent potential targets and mediators of synthetic lethality in cancers ( Burdak-Rothkamm and Rothkamm , 2021 ) . Deficiencies in homologous recombination have been ascribed to cells with defects in several members of the FA pathway , including FANCD2 ( Ceccaldi et al . , 2016; Jenkins et al . , 2012; McCabe et al . , 2006; Ridpath et al . , 2007 ) ; hence , we examined whether FANCD2-depleted cavin3 KO cells were sensitive to PARP inhibition . CHD3 is a chromatin remodeler related to CHD4 , which is implicated as a tumor suppressor in several female malignancies ( Li and Mills , 2014 ) . It has been demonstrated that CHD3 can function like CHD4 in the nucleosome-remodeling ( NuRD ) complex and acts in the DNA damage response in active recruit of DNA repair factors to sites of lesions to promotion DNA repair ( Hoffmeister et al . , 2017; Smith et al . , 2018 ) . Both CHD3 and FANCD2 were depleted in HeLa WT and cavin3 KO cells . Depletion of CHD3 and FANCD2 specifically in cavin3 KO cells induced profound cellular sensitivity to PARP inhibition in clonogenic survival and cell viability assays ( Figure 8—figure supplement 2A–C ) . These findings suggest that cavin3 KO cells represent a novel cellular system to begin to dissect the interactions that occur in the DNA damage response , compensated that may occur by other components in a similar or different pathway for cell survival , and how this information can be used to identify new drug agents and treatment strategies in cancer . Here we describe a novel role for caveolae and the cavin3 protein in regulating the critical tumor suppressor , BRCA1 . Our studies raise the intriguing possibility that by releasing cavins , which can be triggered by mechanical and non-mechanical stimuli such as UV and oxidative stress ( McMahon et al . , 2019 , and this study ) , caveolae can act as general sensors and transducers of cellular stress . Our findings suggest that defining the role of the cavin proteins may provide new insights into the functions of caveolae in pathological conditions such as cancer . Cavin3 may represent a promising therapeutic target in breast cancer through its ability to act both inside and outside of caveolae , by modulating specific signaling pathways ( Hernandez et al . , 2013 ) and by interacting with and modulating the expression of many proteins such as BRCA1 , as shown here , and PP1alpha as previously described ( McMahon et al . , 2019 ) . The possibility of an interaction between BRCA1 and cavin3 was first suggested some 20 years ago , yet , no experimental evidence to support this interaction has been published to date ( Xu et al . , 2001 ) . Our results provide the first clear evidence that cavin3 directly interacts with BRCA1 and that this occurs when cavin3 is released from caveolae in response to cellular stressors . We established this using multiple techniques , including PLA in MCF7 , MDA-MB231 and A431 cells , SMC detection in multiple cancer cell lines ( MCF7 and MDA-MB231 cells ) , and in vitro synthesized BRCA1 and cavin3 . We were not able to reproducibly coimmunoprecipitate BRCA1 and cavin3 . However , this technique can fail to detect weak or transient interactions ( Berggård et al . , 2007 ) . Instead , the combination of cell-based methods ( PLA and single-molecule approaches ) and a cell-free direct interaction approach , as used here , provides unequivocal evidence for the proposed interaction between the N-terminus of BRCA1 and cavin3 . We propose that cavin3 can modulate BRCA1 function via multiple mechanisms: direct interaction with the RING domain of BRCA1 ( Figure 2J ) , increased localization of BRCA1 to the cytosol ( Figure 4A , B ) , regulation of BRCA1 protein levels ( Figure 4C , F , Figure 4—figure supplement 2 ) , modulation of proteasome-mediated protein degradation ( Figure 4G ) , by facilitating the localization of components of the BRCA1-A-complex in response to UV-induced DNA damage ( Figure 8E ) and in DNA repair , as cavin3-deficient cells were sensitive to PARP inhibition , suggesting that these cells are deficient in homologous recombination DNA repair ( Figure 8F ) . We show that the ubiquitin-proteasomal degradation pathway plays a role in the coordinated protein stability of BRCA1 and cavin3 ( Figure 4G ) . Previous studies have identified the RING domain region of BRCA1 as the degron sequence necessary for polyubiquitination and proteasome-mediated protein degradation , which coincides with the interaction domain of BRCA1 identified here for cavin3 ( Lu et al . , 2007 ) . Our data further supports studies that the ubiquitin-proteasome plays an important role in regulating BRCA1 during genotoxic stress ( Lu et al . , 2007 ) . Interaction of BRCA1 with BARD1 protein reduces proteasome-sensitive ubiquitination and stabilization of BRCA1 expression ( Choudhury et al . , 2004 ) . BARD1 levels were downregulated in cavin3 KO cells ( Figure 1—figure supplement 1D ) . Downregulation of BARD1 would be expected to impair BRCA1 function further in cavin3 KO cells as this interaction stabilizes both proteins , which then has a significant role in homologous recombination DNA repair ( Xia et al . , 2003 ) . Further experiments are required to determine if cavin3 disrupts the interaction between BRCA1 and BARD1 and the contribution of BARD1 to the loss of BRCA1 stability and function in these cells . In addition to its expression , BRCA1 subcellular localization is a significant contributor to its cellular functions ( Henderson , 2012 ) . Our findings imply that cavin3 may play a role in the cytosolic translocation of BRCA1 ( Figure 4A , B ) . It is intriguing to hypothesize that BRCA1 , together with cavin3 , executes its tumor suppressor function by its critical role in DNA repair in the nucleus and through signaling pathways and interactions that induce the apoptotic machinery in the cytoplasm . This implies that failed repair of DNA damage in the nucleus is linked to the induction of cell death processes . The elimination of damaged cells occurs in the cytosol and that BRCA1-cavin3 may contribute to this pathway . Interestingly , cells expressing tr-BRCA1 , which was identified here as the BRCA1 domain interacting with cavin3 ( Figure 2J ) , have been shown to cause BRCA1 translocation to the cytosol and enhance sensitivity to UV ( Wang et al . , 2010 ) . Ongoing investigations to test this idea may provide further insight into the role of BRCA1 nuclear-cytoplasmic shuttling and determination of cell fate ( survival vs . death ) . Furthermore , these data also point to the potential use of BRCA1 shuttling as a novel therapeutic strategy by which manipulation of BRCA1 localization can control cellular function and sensitivity to therapy . Cavin3 KO cells exhibited a reduction in recruitment of the BRCA1 A-complex to UV-induced DNA damage foci ( Figure 8E ) . This was further correlated with a decrease in the protein levels of the components of the BRCA1 A-complex , specifically in these cells ( Figure 8D ) . This is consistent with the observation that the loss of any member of the RAP80-BRCA1 complex eliminates observable BRCA1 foci formation as the BRCA1 A-complex requires all its protein components to be stable to optimally recruit BRCA1 to DSBs ( Jiang and Greenberg , 2015 ) . Recent studies from our laboratory have shown that γH2AX phosphorylation is compromised in cavin3 KD cells and that γH2AX forms a complex with the protein phosphatase PP1alpha , whose activity was regulated by cavin3 ( McMahon et al . , 2019 ) . γH2AX is one of the initial factors that recruit checkpoint and DNA repair proteins to DSBs . Failure of cavin3 KO cells to phosphorylate H2AX may further compromise DNA repair mechanisms in these cells . In addition , LFQ proteomics revealed that cavin3 KO cells upregulate many proteins involved in the protection and maintenance of the replication fork and postreplication repair , suggesting involvement of cavin3 in alternative DNA repair pathways that ultimately leads to cell survival ( Figure 1—figure supplement 1 ) . These pathways collectively may account for many of the characteristic features of genomic instability in familial breast and ovarian cancers , and cavin3 KO cells provide an alternative model cell line for further investigation ( see Supplementary file 3 for further analysis of cavin3-dependent pathways ) . Recent clinical evidence has shown that mutations in BRCA1 do not entirely account for the treatment benefits seen with PARP inhibitors ( O'Shaughnessy et al . , 2011; Javle and Curtin , 2011; Pilié et al . , 2019 ) . Loss of cavin3 expression has been observed in many human malignancies ( Carén et al . , 2011; Kim et al . , 2014; Lee et al . , 2008; Lee et al . , 2011; Martinez et al . , 2009; Tong et al . , 2010; Xu et al . , 2001; Zöchbauer-Müller et al . , 2005 ) . Several studies have shown that low expression of cavin3 promotes cisplatin resistance and oxaliplatin resistance in lung and colorectal cancers , respectively ( Fu et al . , 2020; Moutinho et al . , 2014 ) . This is in contrast to BRCA1-deficient cells that are sensitive to these platinum drugs ( Mylavarapu et al . , 2018 ) . These findings suggest that knowing the status of cavin3 in tumors in addition to BRCA1 may be used to better stratify patients in predicting drug sensitivity , that is , PARP inhibitors versus platinum drugs in the clinic . These findings also suggest that cavin3 KO cells may provide a unique platform to understand platinum drug resistance in the absence of BRCA1 expression . This may involve alterations in non-homologous end-joining repair , replication fork protection , upregulation of cellular drug efflux pumps , and alterations to the tumor microenvironment that can now be explored in these cells . Previous studies have shown that cavin3 knockout mice are not cancer-prone ( Hernandez et al . , 2013 ) . This raises the question as to how cavin3 may act as a tumor suppressor . Cavin3 inactivation may contribute to tumor progression by reducing cellular sensitivity to stressors as shown here as well as in previous published studies contributing to overall cell survival ( Lee et al . , 2011 ) . Cavin3 mRNA is increased in response to numerous stresses , suggesting regulation by stress signaling and cellular damage ( Lee et al . , 2011 ) . This may involve p53 as cavin3 increases the stability of p53 and its target gene expression and loss or reduction in tumor cells lessen p53 response to stresses , which contribute to malignant tumor progression ( Lee et al . , 2011 ) . Here , we have shown that cavin3 also interacts with BRCA1 where the two proteins work together to regulate DNA repair or , in extreme conditions , trigger apoptosis . Collectively our studies suggest that loss of cavin3 function might provide tumor cells' survival and growth advantages by attenuating the apoptotic sensitivity to various stresses and hindering DNA repair under chronic stress conditions . Loss of cavin3 expression is more prevalent in late-stage/high-grade cancers than in early-stage/low-grade cancers ( An et al . , 2020; Carén et al . , 2011; Lee et al . , 2008; Wikman et al . , 2012 ) . Cavin3 expression is lost due to promoter methylation in numerous cancer types ( Lee et al . , 2008; Lee et al . , 2011; Martinez et al . , 2009; Tong et al . , 2010; Xu et al . , 2001; Zöchbauer-Müller et al . , 2005 ) . Silencing of a DNA repair gene such as cavin3 by hypermethylation may be a very early step in the progression to cancer ( Jin and Robertson , 2013 ) . Such silencing is proposed to act similarly to a germline mutation in a DNA repair gene and predisposes these cells to cancer . This may occur through deficiency in DNA repair . This would allow for accumulation of DNA damage causing increased errors during DNA synthesis , leading to mutations that can give rise to cancer . This may further contribute to the tumor suppressor functions of cavin3 . Finally , the example of cavin3 leads us to propose a general model for cell stress sensing mediated by cavins when they are released from caveolae to interact with intracellular targets . Rigorous control of such a pathway would require that cytosolic levels of cavins be kept low under steady-state conditions . Recent work shows that this can be achieved by ubiquitination of a conserved phosphoinositide-binding patch on cavins that is only exposed when cavins are released from caveolae ( Tillu et al . , 2015 ) . In the absence of stabilizing interactions , the released cavin protein will undergo proteasomal degradation , but , as shown here , interaction with BRCA1 stabilizes cavin3 , preventing degradation . We propose that the interaction of cavin3 with BRCA1 in response to short-term stress can facilitate DNA repair . With a prolonged stress , this can trigger apoptosis as a protective mechanism . This forms a novel signaling pathway to protect cells against many cellular stresses and represents a new paradigm in cellular signaling that can explain the evolutionary conservation of caveolae and their involvement in multiple signal transduction pathways . In view of the loss of cavin3 in numerous cancers ( Carén et al . , 2011; Kim et al . , 2014; Lee et al . , 2008; Lee et al . , 2011; Martinez et al . , 2009; Tong et al . , 2010; Xu et al . , 2001; Zöchbauer-Müller et al . , 2005 ) and the crucial role of BRCA1 as a tumor suppressor ( King and Marks , 2003; Miki et al . , 1994; Venkitaraman , 2002 ) , these studies describing a new functional partner for BRCA1 suggest that cavin3 should be considered in future cancer diagnostic and therapeutic strategies . Dulbecco’s modified Eagle’s medium ( DMEM , Cat# 10313-021 ) , Z150 L-glutamine 100× ( Cat# 25030-081 ) , and Trypsin-EDTA ( 0 . 05% ) phenol red ( Cat# 25300062 ) were from Gibco by Life Technologies , Australia . SERANA fetal bovine serum ( FBS ) ( Cat# FBS-AU-015 , batch no . 18030416 ) was from Fisher Biotechnology , Australia . cOmplete , mini EDTA-free protease inhibitor cocktail ( Cat# 11836170001 ) , PhosSTOP Phosphatase Inhibitors ( Cat# 4906837001 ) , hydrogen peroxide 30% ( w/w ) solution ( Cat# H1009 ) , AZD2461 ( Cat# SML 1858 ) , and MG132 ( Z-Leu-Leu-Leu-al , Cat# C2211 ) were from Sigma-Aldrich . The following antibodies were used: rabbit anti-53BP1 ( Cat# GTX 112864 , GeneTex , WB 1:1000 ) , rabbit anti-ACCA antibody ( Cell Signaling , Cat# 3662 , RRID:AB_2219400 , WB 1:5000 ) , mouse anti-Actin antibody ( Millipore , Cat# MAB1501 , RRID:AB_2223041 , WB 1:5000 ) , rabbit anti-ACLY antibody ( Sigma-Aldrich , Cat# HPA028758 , RRID:AB_10603575 , WB 1:2000 ) , mouse anti-Aurora kinase antibody ( BD Biosciences , Cat# 611082 , RRID:AB_2227708 , PLA 1:100 ) , mouse-anti-BARD1 E-11 antibody ( Santa Cruz , Cat# sc-74559 , RRID:AB_2061237 , WB 1:500 ) , rabbit anti-BRCA1 20 antibody ( Santa Cruz , Cat# sc-642 , RRID:AB_630944 , WB 1:500 , IF 1:100 , PLA 1:100 ) , mouse anti-BRCA1 MS110 antibody ( Abcam , Cat# ab16780 , RRID:AB_2259338 , WB 1:1000 , IF 1:100 , PLA 1:100 ) , mouse-anti-BRCA1 D-9 antibody ( Santa Cruz , Cat# sc-6964 , RRID:AB_626761 , IF 1:50 ) , rabbit-anti-BRCA1 antibody ( Millipore , Cat# 07-434 , RRID:AB_2275035 , WB 1:2000 ) , rabbit-anti-BRCA1 antibody ( Proteintech , Cat# 22363-1-AP , RRID:AB_2879090 , WB 1:1000 ) , rabbit anti-BRCA2 antibody ( BioVision , Cat# 3675-30T , RRID:AB_2067764 , WB 1:2000 ) , rabbit anti-BRCC36 antibody ( ProScience , Cat# 4311 , WB 1:1000 ) , rabbit anti-BRCC45 antibody ( GeneTex , Cat# GTX105364 , RRID:AB_1949757 , WB 1:2000 ) , mouse anti-Caldesmon antibody ( BD Biosciences , Cat# 610660 , WB 1:3000 ) , mouse anti-alpha catenin antibody ( Cell Signaling , Cat# 2131 , WB 1:3000 ) , mouse anti-gamma catenin antibody ( Cell Signaling , Cat# 2309 , WB 1:3000 ) , rabbit anti-CAV1 antibody ( BD Biosciences , Cat# 610060 , WB 1:5000 ) , mouse anti-cavin1 antibody ( Abmart , China , 1:100 PLA ) , and rabbit anti-cavin1 antibody were raised as described previously and used for immunofluorescence ( Bastiani et al . , 2009 ) , rabbit anti-cavin1 antibody ( Sigma-Aldrich , Cat# AV36965 , RRID:AB 1855947 , WB 1:2000 ) , mouse anti-cavin3 antibody ( Novus , Cat# H00112464-MO4 , PLA 1:200 ) , rabbit anti-cavin3 antibody ( Proteintech , Millennium Sciences , Pty , Ltd , Cat# 16250-1-AP , RRID:AB_2171897 , WB 1:2000 , IF 1:300 , PLA 1:200 ) , rabbit anti-CHD3 antibody ( GeneTex , Sapphire Bioscience , Cat# GTX131779 , RRID:AB_2886520 , WB 1:500 ) , rabbit anti-DDX21 antibody ( Novus , Cat# NBP1-88310 , RRID:AB_11027665 , WB 1:2000 ) , rabbit anti-EGFR Clone LA22 antibody ( Millipore , Cat# 05-104 , RRID:AB_11210086 , WB 1:4000 ) , mouse-anti-FANCD2 antibody ( GeneTex , Cat# GTX116037 , RRID:AB2036898 , WB 1:500 ) , mouse anti-Flotillin Clone 18 antibody ( BD Biosciences , Cat# 610821 , RRID:AB_398140 , PLA 1:100 ) , mouse anti-GFP antibody ( Roche , Cat# 11814460001 , RRID:AB_390913 , WB 1:4000 , PLA 1:300 ) , rabbit anti-Histone H2A . X-Chip Grade ( Abcam , Cat# ab20669 , RRID:AB_445689 , WB 1:1000 ) , rabbit phospho-Histone H2A . X ( Ser 139 ) ( 20E3 ) antibody ( Cell Signaling Technology , Cat# 9718 , RRID:AB_2118009 , IF 1:500 ) , rabbit phospho-Histone H2A . X CHIP Grade antibody ( Abcam , Cat# ab2893 , RRID:AB_303388 , WB: 1:3000 ) , rabbit anti-HLTF antibody ( Proteintech , Cat# 14286-1-AP , WB 1:2000 ) , rabbit anti-MDC1 antibody ( Novus , Cat# 10056657SS , RRID:AB_838567 , WB 1:100 ) , sheep anti-Merit40 antibody ( R&D Systems , Cat# AF6604SP , RRID:AB_10717577 , WB 1:500 ) , rabbit anti-PARP1 antibody ( GeneTex , Cat# GTX112864 , RRID:AB_11173565 , WB 1:1000 ) , mouse anti-PCNA antibody ( Millipore , Cat# NA03T , RRID:AB_2160357 , PLA 1:100 ) , rabbit anti-PGK1 antibody ( GeneTex , Cat# GTX107614 , RRID:AB_2037666 , WB 1:3000 ) , rabbit anti-PKM antibody ( GeneTex , Cat# GTX107977 , RRID:AB_1951264 , WB 1:3000 ) , mouse anti-Rad51 antibody ( Novus , Cat# 100-184 , RRID:AB_350083 , WB 1:1000 ) , rabbit anti-RAP80 D1T6Q antibody ( Cell Signaling Technology , Cat# 14466 , RRID:AB_2798487 , WB 1:1000 , IF 1:100 ) , rabbit anti-RNF168 antibody ( GeneTex , Cat# GTX118147 , RRID:AB_11169617 , WB 1:1000 ) , and mouse anti-Tubulin DM1A antibody ( Abcam , Cat# ab7291 , RRID:AB_2241126 , WB 1:4000 ) . Secondary antibodies for immunofluorescence were Alexa Fluor 488 Goat anti-Rabbit IgG ( H + L ) ( Thermo Fisher Scientific , Cat# A-11034 , RRID:AB_141637 , IF 1:500 ) , Alexa Fluor 546 Goat anti-Mouse IgG ( H + L ) ( Thermo Fisher Scientific , Cat# A-11030 , RRID:AB_2534089 , IF 1:500 ) , Alexa Fluor 594 Donkey anti-Rabbit IgG ( H + L ) ( Thermo Fisher Scientific , Cat# A-21207 , RRID:AB_141637 , IF 1:500 ) , and Alexa Fluor 594 Goat anti-Mouse IgG ( H + L ) ( Thermo Fisher Scientific , Cat# A-21203 , RRID:AB_141633 , IF 1:500 ) . Secondary antibodies for western blotting were Goat anti-Mouse IgG ( H + L ) cross adsorbed secondary antibody , HRP ( Thermo Fisher Scientific , Cat# G-21040 , RRID:AB_2536527 , WB 1:5000 ) , Goat anti-Mouse IgG ( H + L ) cross adsorbed secondary antibody , HRP ( Thermo Fisher Scientific , Cat# G-21234 , RRID:AB_2536527 , WB 1:5000 ) , and Rabbit anti-Sheep IgG ( H + L ) ( Abcam , Cat# ab97130 , RRID:AB_2536530 , WB 1:2000 ) . MCF7 cells , a human adenocarcinoma cell line with a low invasive phenotype ( ATTC HBT-22 , RRID:CVCL_0031 ) , were subjected to STR profiling ( QIMR Berghofer Cancer Research Institute ) . MDA-MB231 cells ( ATCC HTB-26 , RRID:CVCL_0062 ) , a human adenocarcinoma cell line , and A431 cells ( ATCC CRL-1555 , RRID:CVCL_0037 ) , HeLa cells ( ATCC CRM-CCL2 , RRID:CVCL_0030 ) , and HeLa KO for cavin3 were cultured in DMEM supplemented with 10% ( vol/vol ) FBS , 100 units/ml penicillin , and 100 μg/ml streptomycin . All cell lines were routinely tested for mycoplasma . MCF7 cells were seeded at 1 × 106 cells and were transfected with 5 μg pEGFP DNA , pEGFP-cavin1 , pEGFP-cavin2 , pEGFP-cavin3 , or pEGFP-CAV1 DNA using Lipofectamine 2000 ( Invitrogen ) according to the manufacturer's instructions . G418 ( Sigma-Aldrich , Cat# 472788001 ) was used as a selection drug at 500 μg/ml . The HeLa cavin3 KO cell line was generated as follows according to the protocol published previously ( Stroud et al . , 2016 ) . Targeting was to the first exon at the second in-frame ATG about one-third through the exon as this was easy for targeting . In brief , MCF7 , MDA-MB231 , and A431 cells seeded onto glass coverslips at 70% confluence were washed once in PBS and were then fixed in 4% ( vol/vol ) PFA in PBS for 20 min at room temperature ( RT ) . Coverslips were washed three times in excess PBS and were permeabilized in 0 . 1% ( vol/vol ) Triton X-100 in PBS for 7 min and blocked in 1% ( vol/vol ) bovine serum albumin ( BSA ) ( Sigma-Aldrich ) in PBS for 30 min at RT . The primary antibodies were diluted in 1% ( vol/vol ) BSA in PBS and incubated for 1 hr at RT . Secondary antibodies ( Molecular Probes ) were diluted in 1% ( vol/vol ) BSA in PBS and incubated for 1 hr at RT . Washes were performed in PBS . Coverslips were rinsed in distilled water and mounted in Mowiol ( Mowiol 488 , Hoechst AG ) in 0 . 2 M Tris-HCl , pH 8 . 5 . The images were taken on a laser-scanning microscope ( LSM 510 META , Carl Zeiss , Inc ) using a 63× oil lens , NA 1 . 4 . Adjustments of brightness and contrast were applied using ImageJ software ( NIH ) . The LUT of images for PLA were inverted for better visualization of PLA dots in cells . HeLa WT and cavin3 KO cells were pre-permeabilized with CSK buffer ( 10 mM HEPES , 100 mM NaCl , 300 mM sucrose , 3 mM MgCl2 , 0 . 7% Triton X-100 ) for 5 min and were then fixed with 4% PFA/PBS for 15 min , permeabilized with 0 . 5% Triton X-100 solution for 15 min followed by blocking for 1 hr RT . Cells were then immunostained with primary antibodies against mouse BRCA1 alone ( Santa Cruz , Cat# sc-6954 , RRID:AB_626761 , IF 1:50 ) , Rap80 alone ( Cell Signaling Technology , Cat# 14466 , RRID:AB_2798487 , IF 1: 50 ) , γH2AX alone ( Abcam , Cat# 20669 , RRID:AB_445689 , IF 1:100 ) , and the appropriate Alexa Fluor 488 Goat anti-Rabbit IgG ( H + L ) ( Thermo Fisher Scientific , Cat# A-11034 , IF 1:500 ) conjugated secondary antibodies . Images were taken with a Zeiss microscope . Quantification of the percent of cells was based on foci formation ( more than 5 foci/nucleus ) was determined from more than 500 cells/experimental condition from 2 to 3 independent experiments using an automated plugin for ImageJ . Detection of an interaction between BRCA1 and the cavin or CAV1 proteins was assessed using the Duolink II Detection Kit ( Sigma-Aldrich ) according to the manufacturer's specifications . The Duolink In situ PLA Probe Anti-Rabbit MINUS ( Sigma-Aldrich , DUO92005 , RRID:AB_2810942 ) and Duolink In situ PLA Probe anti-Mouse PLUS ( Sigma-Aldrich , DUO92001 , RRID:AB_281039 ) and Duolink In situ detection reagents Orange ( DUO 92007 ) were used in all PLA experiments . The primary antibodies used were mouse monoclonal GFP ( 1:500 ) and rabbit polyclonal BRCA1 ( 1:200 ) , rabbit cavin3 ( 1:200 ) and mouse PCNA ( 1:100 ) , rabbit cavin3 ( 1:200 ) and mouse Aurora Kinase ( 1:100 ) , rabbit cavin3 ( 1:200 ) and Flotillin ( 1:100 ) , and cavin3 ( 1:200 ) and mouse cavin 1 ( 1:100 ) . The signal was visualized as a distinct fluorescent spot and was captured on an Olympus BX-51 upright Fluorescence Microscope . The number of PLA signals in a cell was quantified in ImageJ using a Maximum Entropy Threshold and Particle Analysis where 50 cells in each treatment group were analyzed from at least three independent experiments . For SDS-PAGE , cells were harvested , rinsed in PBS , and were lysed in lysis buffer containing 50 mM Tris pH 7 . 5 , 150 mM NaCl , 5 mM EDTA pH 8 . 0 , 1% Triton X-100 with protease and phosphatase inhibitors . Lysates were collected by scraping and cleared by centrifugation at 4°C . The protein content of all extracts was determined using the Pierce BCA Protein Assay Kit ( Cat# 23225 , Thermo Fisher Scientific ) using BSA as the standard . 30 μg of cellular protein were resolved by 10% SDS-PAGE and were transferred to Immobilin P 0 . 45 mm PVDF membrane ( Merck ) . Bound IgG was visualized with horseradish peroxidase-conjugated secondary antibodies and the Clarity Western ECL Substrate ( Cat# 1705061 , Bio-Rad , Gladesville , New South Wales , Australia ) . A431 or MDA-MB231 cells were plated on coverslips at 70% confluency . Cells were either left untreated or were treated with 200 μM H2O2 for 30 min , 90% hypo-osmotic media for 10 min , or UV treatment for 2 min without media with a UV germicidal light source ( UV-C 254 nm ) and allowed to recover for 30 min in complete cell culture medium as previously described in McMahon et al . , 2019 . All cells were fixed and processed for cavin3 and BRCA1 or cavin3 and cavin1 using the PLA as described . HeLa WT and cavin3 KO cells were counted using a hemocytometer and seeded into 96-well plate at 1000 cells/well ( eight wells for each treatment ) in 90 ml medium per well . Cells were either left untreated or were treated with 90% hypo-osmotic media ( 90% water in DMEM ) , UV treatment for 2 min without media with a UV germicidal light source ( UV-C 254 nm ) or 200 μM H2O2 . After stress addition , 10 ml of PrestoBlue Viability Reagent ( 10× ) ( Absorbance wavelength: 600 nm ) ( Thermo Fisher Scientific ) was added to cells . The PrestoBlue reagent was incubated constantly in wells over a time course from 2 hr to 24 hr . Control wells containing only cell culture media ( no cells ) were included in triplicate on each plate for background fluorescence calculations . Plates were returned to a 37°C incubator . Both absorbance values at 570 nm and 600 nm were measured for each plate in a TECAN Infinite 200 Pro reader ( Millennium Science ) , where 570 nm was used as the experimental wavelength and 600 nm as normalization wavelength . For PARP inhibitor experiments , cells were either left untreated or were treated with PARP inhibitor ( AZD2461 5 nM ) for 6 days after which 10 μl of PrestoBlue Viability Reagent ( 10× ) ( absorbance wavelength: 600 nm ) ( Thermo Fisher Scientific ) was added to cells . Control wells containing only cell culture media ( no cells ) were included in triplicate on each plate for background fluorescence calculations . Plates were returned to a 37°C incubator . Both absorbance values at 570 nm and 600 nm were measured for each plate in a TECAN Infinite 200 Pro reader , where 570 nm was used as the experimental wavelength and 600 nm as normalization wavelength . Raw data was processed to evaluate the percent reduction of PrestoBlue reagent for each well by using the following equation referring to the manufacturer’s protocol:%ReductioninPrestoblue= ( 117216×A1 ) − ( 80586×A2 ) ( 155677×A1 ) − ( 14652×A2 ) ×100where A1 is the absorbance of test wells at 570 nm , A2 is the absorbance of test wells at 600 nm , N1 is the absorbance of media-only wells at 570 nm , and N2 is the absorbance of media-only wells at 600 nm . Human cavin3 Stealth siRNAs ( set of three – HSS174185 , 150811 , 150809 ) and Human BRCA1 Stealth siRNAs ( set of three – HSS101089 , 186096 , 186097 ) were purchased from Life Technologies Australia Pty Ltd . Two siRNA oligonucleotides to cavin3 or BRCA1 were found to reduce protein levels ( oligo 1 and oligo 2 ) and were transfected into cells at 24 hr and 48 hr after plating using Lipofectamine 2000 reagent ( Invitrogen ) with a ratio of 6 μl Lipofectamine to 150 pmol siRNA . Cells were split and harvested after 72–96 hr for further analysis . WT and cavin3 KO cells lacking CHD3 , FANCD2 , PARP1 , and TP53BP1 were generated using the Alt-R CRISPR-Cas9 system ( Integrated DNA Technologies ) . The following predesigned Alt-R CRISPR-Cas9 gRNAs were used: Each RNA oligo ( Alt-R CRISPR Cas9 cRNA , tracrRNA ) was resuspended in Nuclease-Free IDTE Buffer . The crRNA and tracrRNA were mixed in equimolar concentrations , heated at 95°C for 5 min , followed by cooling to RT . To produce the RNP complex for each well of a 96-well plate , the following was combined: 1 . 5 μl of 1 μM Guide RNA oligos , 1 . 5 μl of 1 μM diluted Cas9 enzyme with 0 . 6 μl of Cas9 PLUS Reagent from Lipofectamine CRISPRMAX kit and 21 . 4 μl of Opti-MEM Media followed by incubation at RT for 5 min to assemble the RNP complexes . The RNP was further mixed with 1 . 2 μl of CRISPRMAX transfection reagent in Opti-MEM for a further 20 min to form the transfection complexes . This was then added to 40 , 000 HeLa WT or cavin3 KO cells/ml that were seeded in a well of a 96-well tissue culture plate . The plates containing the transfection complexes and cells were returned to a tissue culture incubator for 72 hr . These cells were then subjected to single-cell plating for clonal selection . Loss of each of the proteins was verified by western blot analysis of cell lysates using the following antibodies: CHD3 ( GeneTex , Cat# GTX131779 , RRID:AB_2886520 , WB: 1:500 ) , FANCD2 ( GeneTex , Cat# GTX116037 , RRID:AB_2036898 , WB: 1:500 ) , PARP1 ( GeneTex , Cat# GTX112864 , RRID:AB_11173565 , WB: 1:1000 ) , and 53BP1 ( GeneTex , Cat# GTX70310 , WB 1:1000 ) . WT HeLa and cavin3 KO cells were seeded at low density ( 500 cells/well ) in six-well plates , left untreated or treated with 5 nM concentrations of PARP ( AZD2461 ) , and were allowed to form colonies for 6 days . Colonies were fixed and stained with 0 . 5% crystal violet/20% ethanol and counted . Results were normalized to plating efficiencies where the Comet microscopes slides were prepared the day before the assay by melting low melting point 0 . 5% agarose in a microwave until the agarose was completely molten . Thoroughly cleaned glass microscope slides were layered with the agarose . Slides were left on a flat surface to air-dry overnight where a transparent agarose film formed after drying . Coated slides were placed at 37°C before use . HeLa WT and cavin3 KO cells either left untreated or treated with UV ( 2 min ) and a 4 hr recovery time were trypsinized , and cells were suspended at 2 × 105 cell/mL in 1× PBS . The cell samples were prepared immediately before starting the assay , and all samples were handled in a dimmed environment to prevent DNA damage from light . The cell suspension was mixed with 0 . 5% molten low melting point agarose ( at 37°C ) at a ratio 1:10 ( v/v ) . Cells were mixed gently by pipetting up and down and immediately added on top of the agarose layer on the glass slides . The side of the pipette tip was used to spread the agarose/cell mixture to ensure the formation of a thin layer . Slides were then placed at 4°C in the dark for 30 min . Slides were then carefully immersed in lysis buffer ( 2 . 5M NaCl , 0 . 1 M EDTA pH 8 . 0 , 10 mM Tris , where the pH was adjusted to 10 . 0 with NaOH pellets and chilled before use ) at 4°C in the dark for 1 hr . Slides were then immersed in alkaline solution at 4°C in the dark for 30 min . Slides were gently removed from the alkaline solution and then gently immersed in chilled 1× TBE solution for 10 min in the dark . Prechilled TBE buffer was added in the electrophoresis slide tray , and the slides were placed inside for electrophoresis . The power supply was set to voltage of 1 V/cm ( the length between electrodes ) and run for 15 min at 4°C . Excess buffer was removed from the slides , which were then immersed in three changes of chilled dH2O for 2 min . Slides were then gently immersed in chilled 70% ethanol for 5 min at RT in the dark . Slides were then allowed to dry . 50 µl green fluorescent nucleic acid staining solution ( Vista green ) was then added onto each slide and was stained for 15 min at RT in the dark . The visualization and quantification of DNA breaks was based on epifluorescence microscopy . Randomly captured images from the stained comet slides were from a fluorescence microscope with a 10× objective lens . The DNA damage was quantified by measuring the displacement between the genetic material of the nucleus ( 'comet head' ) and the resulting 'tail' using ImageJ software . At least 50–100 cells were analyzed per sample from three independent experiments . The following equations were used in the analysis: Equal numbers of subconfluent MCF7 cells expressing GFP alone , cavin1-GFP , cavin2-GFP , cavin3-GFP , and CAV1-GFP were seeded on coverslips . Twenty-four hours later , cells were subject to UV-C exposure for 2 min without media . Complete medium lacking phenol red was added to the cells that were left at 37°C to recover . LDH release assay was measured in triplicate samples from 50 μl of conditioned media expressing cells using the Cytotoxicity Detection KitPLUS ( LDH ) from Sigma-Aldrich according to the manufacturer's instructions . Post-nuclear supernatant from UV exposure cells was also prepared and subjected to western blot analysis with antibodies to BRCA1 ( WB 1:500 ) , GFP ( WB 1:3000 ) , and Tubulin ( WB 1:5000 ) . For knockdown experiments of cavin3 and BRCA1 , after 72 hr of knockdown , cells were left untreated or were further transfected with BRCA1-GFP or cavin3-GFP overnight , respectively , and were then subjected to UV exposure 2 min and a recovery time of 6 hr . LDH release was then measured from the cell supernatant in triplicate as indicated in the respective figure legends . Single-molecule spectroscopy was performed . Leishmania cell-free lysates were prepared according to Kovtun et al . , 2011; McMahon et al . , 2019; Mureev et al . , 2009 . Where indicated , MDA-MB231 or MCF7 cells were transiently cotransfected with BRCA1-GFP and mCherry alone as the control , cavin1-Cherry , cavin2-Cherry , cavin3-Cherry , or CAV1-Cherry constructs . A PNS fraction from the MDA-MB231 and MCF7 cells was prepared in 1× PBS with protease and phosphatase inhibitors for analysis . Single-molecule coincidence measurements were performed using pairs of tagged proteins to ascertain their interaction . One protein of the pair was tagged with GFP , and the other with mCherry , and both were diluted to single-molecule concentrations ( ~1 nM ) . Two lasers , with wavelengths of 488 nm and 561 nm ( to excite GFP and mCherry , respectively ) , were focused to a confocal volume using a 40×/1 . 2 NA water immersion objective . The fluorescence signal from the fluorophores was collected and separated into two channels with a 565 nm dichroic . The resulting GFP and mCherry signals were measured after passing through a 525/20 nm bandpass and 580 nm long-pass filter , respectively . The signal from both channels was recorded simultaneously with a time resolution of 1 ms , and the threshold for positive events was set at 50 photons/ms . The coincidence ratio ( C ) for each event was calculated as C = mCherry/ ( GFP + mCherry ) , after subtracting a 6% leakage of the GFP signal into the mCherry channel . Coincident events corresponded to ~0 . 25 < C < 0 . 75 . After normalizing for the total number of events ( >1000 in all cases ) , a histogram of the C values for the protein pair was fitted with 3 Gaussians , corresponding to signals from solely GFP ( green ) , coincidence ( yellow ) , and solely mCherry ( red ) . Samples were prepared for mass spectrometry and analyzed as previously described ( Stroud et al . , 2016 ) . Briefly , cells were lysed in 1% ( w/v ) sodium deoxycholate , 100 mM Tris-HCl ( pH 8 . 1 ) , Tris ( 2-carboxyethy ) phosphine ( TCEP ) , 20 mM chloroacetamide , and incubated at 99 °C for 10 min . Reduced and alkylated proteins were digested into peptides using trypsin by incubation at 37 °C overnight , according to the manufacturer’s instructions ( Promega ) . Detergent was removed from the peptides using SDB-RPS stage tips as described ( Stroud et al . , 2016 ) . Peptides were reconstituted in 0 . 1%% trifluoroacetic acid ( TFA ) , 2% ACN , and analyzed by online nano-HPLC/electrospray ionization-MS/MS on a Q Exactive Plus connected to an Ultimate 3000 HPLC ( Thermo Fisher Scientific ) . Peptides were loaded onto a trap column ( Acclaim C18 PepMap nano Trap × 2 cm , 100 μm I . D , 5 μm particle size , and 300 Å pore size; Thermo Fisher Scientific ) at 15 μl/min for 3 min before switching the pre-column in line with the analytical column ( Acclaim RSLC C18 PepMap Acclaim RSLC nanocolumn 75 μm × 50 cm , PepMap100 C18 , 3 μm particle size 100 Å pore size; Thermo Fisher Scientific ) . The separation of peptides was performed at 250 nl/min using a nonlinear ACN gradient of buffer A ( 0 . 1% FA , 2% ACN ) and buffer B ( 0 . 1% FA , 80% ACN ) , starting at 2 . 5% buffer B to 35 . 4% followed by ramp to 99% over 278 min . Data were collected in positive mode using Data Dependent Acquisition using m/z 375–1575 as MS scan range , HCD for MS/MS of the 12 most intense ions with z ≥ 2 . Other instrument parameters were MS1 scan at 70 , 000 resolution ( at 200 m/z ) , MS maximum injection time 54 ms , AGC target 3E6 , normalized collision energy was at 27% energy , isolation window of 1 . 8 Da , MS/MS resolution 17 , 500 , MS/MS AGC target of 2E5 , MS/MS maximum injection time 54 ms , minimum intensity was set at 2E3 , and dynamic exclusion was set to 15 s . Thermo raw files were processed using the MaxQuant platform ( Tyanova et al . , 2016 ) version 1 . 6 . 5 . 0 using default settings for a label-free experiment with the following changes . The search database was the UniProt human database containing reviewed canonical sequences ( June 2019 ) and a database containing common contaminants . ‘Match between runs’ was enabled with default settings . Maxquant output ( proteinGroups . txt ) was processed using Perseus ( Tyanova et al . , 2016 ) version 1 . 6 . 7 . 0 . Briefly , identifications marked ‘Only identified by site , ’ ‘Reverse , ’ and ‘Potential Contaminant’ were removed along with identifications made using <2 unique peptides . Log2 transformed LFQ Intensity values were grouped into control and knockout groups , each consisting of three replicates . Proteins not quantified in at least two replicates from each group were removed from the analysis . Annotations ( Gene Ontology Biological Process [GOBP] ) were loaded through matching with the majority protein ID . A two-sample , two-sided t-test was performed on the values with significance determined using permutation-based FDR statistics ( FDR 5% , S0 = 1 ) . Enrichment analysis of GOBP terms was performed on significantly altered proteins using a significance threshold of 4% FDR . Statistical analyses were conducted using Microsoft Excel and Prism ( GraphPad ) . Statistical significance was determined either by two-tailed Student's t-test , one-way ANOVA using the Bonferroni comparisons test with a 95% confidence interval or nested ANOVA , as indicated in the figure legends . Significance was calculated , where * indicates p<0 . 05 , indicates p<0 . 01 , * indicates p<0 . 001 , and **** indicates p<0 . 0001 .	When cells become cancerous they often stop making certain proteins . This includes a protein known as cavin3 which resides in bulb-shaped pits of the membrane that surrounds the cell called caveolae . These structures work like stress detectors , picking up changes in the membrane and releasing proteins , such as cavin3 , into the cell’s interior . Past studies suggest that cavin3 might interact with a protein called BRCA1 that suppresses the formation of tumors . Cells with mutations in the gene for BRCA1 struggle to fix damage in their DNA , and have to rely on other repair proteins , such as PARPs ( short for poly ( ADP-ribose ) polymerases ) . Blocking PARP proteins with drugs can kill cancer cells with problems in their BRCA1 proteins . However , it was unclear what role cavin3 plays in this mechanism . To investigate this , McMahon et al . exposed cells grown in the laboratory to DNA-damaging UV light to stimulate the release of cavin3 from caveolae . This revealed that cavin3 interacts with BRCA1 when cells are under stress , and helps stabilize the protein so it can perform DNA repairs . Cells without cavin3 showed decreased levels of the BRCA1 protein , but compensated for the loss of BRCA1 by increasing the levels of their PARP proteins . These cells also had increased DNA damage following treatment with drugs that block PARPs , similar to cancer cells carrying mutations in the gene for BRCA1 . These findings suggest that cavin3 helps BRCA1 to suppress the formation of tumors , and therefore should be considered when developing new anti-cancer treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2021	Cavin3 released from caveolae interacts with BRCA1 to regulate the cellular stress response
Dengue and chikungunya are increasing global public health concerns due to their rapid geographical spread and increasing disease burden . Knowledge of the contemporary distribution of their shared vectors , Aedes aegypti and Aedes albopictus remains incomplete and is complicated by an ongoing range expansion fuelled by increased global trade and travel . Mapping the global distribution of these vectors and the geographical determinants of their ranges is essential for public health planning . Here we compile the largest contemporary database for both species and pair it with relevant environmental variables predicting their global distribution . We show Aedes distributions to be the widest ever recorded; now extensive in all continents , including North America and Europe . These maps will help define the spatial limits of current autochthonous transmission of dengue and chikungunya viruses . It is only with this kind of rigorous entomological baseline that we can hope to project future health impacts of these viruses . The mosquitoes Aedes aegypti [= Stegomyia aegypti] and Aedes albopictus [= Stegomyia albopicta] ( Reinert et al . , 2009 ) are vectors of several globally important arboviruses , including dengue virus ( DENV ) ( Simmons et al . , 2012 ) , yellow fever virus ( Jentes et al . , 2011 ) , and chikungunya virus ( CHIKV ) ( Leparc-Goffart et al . , 2014 ) . The public health impact of DENV and CHIKV has increased dramatically over the last 50 years , with both diseases spreading to new geographic locations and increasing in incidence within their range ( Weaver , 2014 ) . The remaining burden of vaccine-preventable yellow fever is similarly likely to be dramatically underestimated ( Garske et al . , 2014 ) . DENV , with a nearly ubiquitous distribution in the tropics and more recently introduced to Europe ( ECDC , 2014; Schaffner and Mathis , 2014 ) , is the most prevalent human arboviral infection causing 100 million apparent annual infections world-wide with almost half of the world's population at risk of infection ( Brady et al . , 2012; Bhatt et al . , 2013 ) . CHIKV recently received considerable public health attention due to the outbreaks in Réunion in 2005–2006 ( 225 , 000 infections ) ( Borgherini et al . , 2007 ) , Italy in 2007 ( 205 infections ) ( Rezza et al . , 2007 ) , and France in 2010 and 2014 ( 2 and 11 locally transmitted cases , respectively ) ( La Ruche et al . , 2010; Grandadam et al . , 2011; Paty et al . , 2014 ) as well as its recent invasion into the Americas with over 1 million cases recorded to date ( Cauchemez et al . , 2014; Johansson et al . , 2014; Morens and Fauci , 2014 ) . Increases in distribution and intensity of transmission are compounded by the lack of commercially available antivirals or vaccines for either disease ( Simmons et al . , 2012; Roy et al . , 2014 ) , although new therapeutics and vaccines are in development ( McArthur et al . , 2013; Powers , 2014; Villar et al . , 2015 ) . Similarly , while yellow fever infections have been on the decline due to extensive vector control and an effective vaccine developed more than 70 years ago , it still causes a significant disease burden in Africa and South America ( Poland et al . , 1981; World Health Organization , 1990; Garske et al . , 2014 ) . Given the public health impact of these diseases and their rapid global spread , understanding the current and future distribution , and determining the geographic limits of transmission and transmission intensity , will enable more efficient planning for disease control ( Carrington and Simmons , 2014; Semenza et al . , 2014; Messina et al . , 2015 ) . Because these diseases can only persist where their mosquito vectors , Ae . aegypti and Ae . albopictus are present , understanding the distributions of these two species underpins this strategy . The global expansion of these arboviruses was preceded by the global spread of their vectors ( Charrel et al . , 2014 ) . Ae . aegypti originated in Africa where its ancestral form was a zoophilic treehole mosquito named Ae . aegypti formosus ( Brown et al . , 2014 ) . The domestic form Ae . aegypti is genetically distinct with discrete geographic niches ( Brown et al . , 2011 ) . It was hypothesised that due to harsh conditions coupled with the onset of the slave trade , Ae . aegypti were introduced into the New World from Africa , from where it subsequently spread globally to tropical and sub-tropical regions of the world ( Brown et al . , 2014 ) . Ae . albopictus , originally a zoophilic forest species from Asia , spread to islands in the Indian and Pacific Oceans ( Delatte et al . , 2009 ) . During the 1980s it rapidly expanded its range to Europe , the United States and Brazil ( Medlock et al . , 2012; Carvalho et al . , 2014 ) . Today both Ae . aegypti and Ae . albopictus are present in most Asian cities and large parts of the Americas ( Lambrechts et al . , 2011 ) . Ae . aegypti feed almost exclusively on humans in daylight hours and typically rest indoors ( Scott and Takken , 2012 ) . In contrast Ae . albopictus is usually exophagic and bites humans and animals opportunistically ( Paupy et al . , 2009 ) but has also been shown to exhibit strongly anthropophilic behavior similar to Ae . aegypti in specific contexts ( Ponlawat and Harrington , 2005; Delatte et al . , 2010 ) . A number of previous studies have mapped the global or regional distributions of Ae . aegypti and Ae . albopictus by focusing on different aspects of their ecology . The majority examined the impacts of climatic conditions , often with an exclusive focus on temperature . Kobayashi et al . ( 2002 ) and Nawrocki and Hawley ( 1987 ) used results from laboratory studies to identify potential limits of establishment in Japan and Asia suggesting a minimum mean temperature in the coldest months of −2°C and −5°C respectively limits their distribution . Brady et al . ( 2013 ) extended that work by modeling the adult survival of both species under laboratory and field conditions , indicating that Ae . albopictus has higher survival rates than Ae . aegypti , though adults of the latter can tolerate a wider range of temperatures . Applying these results to global temperature data , Brady et al . ( 2014 ) produced maps indicating areas where the temperature is suitable for these vectors to persist . Whilst temperature is clearly a crucial factor constraining the distribution of the two species , these results alone are not sufficient to discriminate between areas where the species can and cannot persist . Other studies went further using statistical models , predicting the distributions of both species ( though particularly Ae . albopictus ) using a broader range of climatic variables including precipitation ( Benedict et al . , 2007; Medley , 2010; Fischer et al . , 2011; Caminade et al . , 2012; Khormi and Kumar 2014; Campbell et al . , 2015 ) . Whilst these studies incorporated several generic climatic factors to predict the current and future distribution of the species , we were able to integrate a bespoke species-specific temperature suitability covariate and account for anthropogenic factors that are known to influence Ae . aegypti and Ae . albopictus distributions ( Reiter et al . , 2003 ) . Both species are container-inhabiting but differ in their behaviour and biology so that they occupy different niches ( Eisen and Moore , 2013 ) . A few local studies showed , however , that local spread of Ae . albopictus and declining Ae . aegypti populations might be linked to inter-species competition ( O'Meara et al . , 1995; Daugherty et al . , 2000; Juliano et al . , 2007 ) and/or non-reciprocal cross-species inseminations ( Bargielowski et al . , 2013 ) . Socio-economic factors affecting the distribution of the Aedes mosquitoes other than the use of containers to store water , include the use of air-conditioning , housing quality , and the rate of urbanisation ( Ramos et al . , 2008; Aström et al . , 2012 ) . In addition to exclusively focusing on meteorological factors in determining the spatial extent of the Aedes mosquitoes , many models used small sets of input occurrence data , which were biased towards particular countries with well-developed surveillance systems , such as , Brazil and Taiwan ( Benedict et al . , 2007; Medley , 2010; Fischer et al . , 2011; Campbell et al . , 2015 ) . In this context , we set out to model the global distribution of these two important vector species , compiling the most comprehensive occurrence dataset to date from published literature and national entomological surveys . To overcome previous modelling limitations , a probabilistic species distribution model using Boosted Regression Trees ( BRT ) was produced for each vector . Our models combine environmental and , for the first time , land-cover variables to predict the global distribution of both species at high spatial resolution . Importantly , the models quantify prediction uncertainty and aim at identifying key contributing factors and inter-species differences in their environmental niches . In total , data collection yielded 19 , 930 and 22 , 137 spatially unique occurrence records for Ae . aegypti and Ae . albopictus respectively , which were used to train the distribution models . This includes up-to date records from national entomological surveys from Brazil and Taiwan for both species ( Carvalho et al . , 2014; Yang et al . , 2014 ) . For Ae . aegypti , >60% of all occurrence records are from Asia and Oceania , 35% are from the Americas and only 575 unique occurrences are available for Africa and Europe ( Table 1a ) . Similarly for Ae . albopictus , most of the occurrences are from Asia ( 75% ) , 23% are from the Americas and only 542 records are available from Europe and Africa ( Table 1b ) . For each continent the top 10 countries in terms of occurrences recorded are shown for both species ( Table 1 ) . The geographic distribution of the occurrence records is the widest ever recorded with particularly high spatial and temporal resolution in Taiwan and Brazil for both species and in the United States for Ae . albopictus . All occurrence data have been made openly available through an online data repository to ensure consistency and reproducibility ( Pigott and Kraemer , 2014; Kraemer et al . , 2015a ) . 10 . 7554/eLife . 08347 . 003Table 1 . The geographic distribution of spatially unique occurrence records for the Americas , Europe/Africa , and Asia/OceaniaDOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 003CountryOccurrencesCountryOccurrencesCountryOccurrencesAe . aegyptiAmericasBrazil5 , 044Europe/AfricaSenegal112Asia/OceaniaTaiwan9 , 490USA436Cameroon55Indonesia603Mexico411Kenya52Thailand495Cuba177United Republic of Tanzania44India423Argentina170Côte d'Ivoire40Australia282Trinidad and Tobago152Nigeria35Viet Nam223Venezuela130Madagascar28Malaysia112Colombia128Gabon27Singapore44Puerto Rico120Mayotte20Philippines36Peru89Sierra Leone20Cambodia29Ae . albopictusAmericasBrazil3 , 441Europe/AfricaItaly203Asia/OceaniaTaiwan15 , 339USA1 , 594Madagascar58Malaysia186Mexico50Cameroon42Indonesia161Cayman Islands15France37India150Haiti13Gabon27Japan97Guatemala12Albania22Thailand82Venezuela7Mayotte21Singapore44Colombia3Greece18Lao People's Democratic Republic26Cuba3Israel17Philippines22Puerto Rico3Lebanon15Viet Nam18Top 10 countries in terms of occurrence records for each continent are shown for Ae . aegypti ( a ) and Ae . albopictus ( b ) . Maps showing the predicted global distribution for Ae . aegypti and Ae . albopictus are presented in Figures 1 , 2 , respectively . The distributions of the two species differ markedly in a number of places . Ae . aegypti is predicted to occur primarily in the tropics and sub-tropics , with concentrations in northern Brazil and southeast Asia including all of India , but with relatively few areas of suitability in Europe ( only Spain and Greece ) and temperate North America . In Australia , however , Ae . aegypti shows a wider geographic distribution than Ae . albopictus , which is confined to the east coast , largely reflecting the known historic distribution of Ae . aegypti . By contrast , the distribution of Ae . albopictus extends into southern Europe ( Figure 3A ) , northern China , southern Brazil , northern United States ( 3b ) , and Japan . Again , this reflects the current and historic distribution of Ae . albopictus and the ability of the species to tolerate lower temperatures ( Tsuda and Takagi , 2001; Lounibos et al . , 2002; Thomas et al . , 2012; Brady et al . , 2014 ) . 10 . 7554/eLife . 08347 . 004Figure 1 . Global map of the predicted distribution of Ae . aegypti . The map depicts the probability of occurrence ( from 0 blue to 1 red ) at a spatial resolution of 5 km × 5 km . DOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 00410 . 7554/eLife . 08347 . 005Figure 1—figure supplement 1 . Effect plots of covariates used in this study showing the marginal effect of each covariate on probability of presence for Ae . aegypti ( 1 ) and Ae . albopictus ( 2 ) : enhanced vegetation index ( EVI ) annual mean ( A ) ; Enhanced vegetation index—range ( B ) ; annual monthly maximum precipitation ( C ) ; annual monthly minimum precipitation ( D ) ; temperature suitability ( E ) ; urban areas ( F ) ; peri-urban areas ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 00510 . 7554/eLife . 08347 . 006Figure 1—figure supplement 2 . Set of covariate layers used to predict the ecological niche of Ae . aegypti and Ae . albopictus described in detail in the ‘Materials and methods’ section; ( A ) enhanced vegetation index ( EVI ) annual mean , ( B ) EVI annual range , ( C ) annual monthly maximum precipitation , ( D ) annual monthly minimum precipitation , ( E ) temperature suitability for Ae . albopictus , ( F ) temperature suitability for Ae . aegypti , ( G ) rural , peri-urban and urban classification layer . DOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 00610 . 7554/eLife . 08347 . 007Figure 1—figure supplement 3 . Visualization of pixel level uncertainty calculated using the upper and lower bounds of the 95% confidence intervals associated with the prediction maps for Ae . aegypti ( A ) and Ae . albopictus ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 00710 . 7554/eLife . 08347 . 008Figure 1—figure supplement 4 . The distribution of the occurrence database for Ae . aegypti ( A ) and Ae . albopictus ( B ) plotted on the underlying prediction surface . DOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 00810 . 7554/eLife . 08347 . 009Figure 2 . Global map of the predicted distribution of Ae . albopictus . The map depicts the probability of occurrence ( from 0 blue to 1 red ) at a spatial resolution of 5 km × 5 km . DOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 00910 . 7554/eLife . 08347 . 010Figure 3 . Predicted probability of occurrence of Ae . albopictus in Europe ( A ) and the United States ( B ) , regions in which Ae . albopictus is rapidly expanding its range . Points represent known occurrences ( transient [triangles] or established [circles] ) until the end of 2013 . DOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 010 In Europe , the predicted potential distribution of Ae . albopictus contains most of the known occurrence points , but suitability is also predicted in Portugal and the west of Spain , and in much of south-eastern Europe and the Balkans , where the species has yet to be reported . Similarly , in China Ae . albopictus has yet to be reported from much of the area predicted to be environmentally suitable . By contrast , in the United States the species has been reported from almost all of the predicted suitable areas , with the exception of a small band of predicted suitability on the western slope of the Sierra Nevada . Due to the relatively sparse reporting from Africa it remains uncertain whether areas predicted to be highly suitable are already infested or have yet to be colonized by the species . Ae . albopictus for example has only been reported from some West African countries ( Nigeria , Cameroon , Gabon , the Central African Republic , Congo , Côte d'Ivoire ) and Madagascar , and South Africa ( as well as some islands in the Indian Ocean ) . The distribution of Ae . aegypti in Africa seems to be much wider , with reports of species occurrence in over 30 countries . For both species , the most important predictor was temperature . Temperature suitability indices had high relative influence statistics for both species; this variable was selected in approximately half of regression tree decisions for Ae . aegypti ( 54 . 9% , CI = 53 . 7–56% ) and Ae . albopictus ( 44 . 3% , CI = 42 . 7–45 . 6% ) . The full definition of a relative influence statistic is given in the ‘Materials and methods’ section under the heading Predictive performance and relative influence of covariates . Precipitation and vegetation indices made up the remainder of predictors . Urban land cover made very little contribution to either model ( Table 2 ) . Model evaluation statistics under cross-validation were high ( AUC: 0 . 87 and 0 . 9 respectively ) for both model ensembles , indicating high predictive performance of the model . Effect plots for each covariate are shown in Figure 1—figure supplement 2 . Maps of uncertainty associated with these predictions are presented in Figure 1—figure supplement 3 . 10 . 7554/eLife . 08347 . 011Table 2 . Relative contribution of environmental covariates predicting the global distribution of Ae . aegypti and Ae . albopictusDOI: http://dx . doi . org/10 . 7554/eLife . 08347 . 011Mean contribution Ae . aegypti ( % ) 95% confidence interval Ae . aegypti ( % ) Mean contribution Ae . albopictus ( % ) 95% confidence interval Ae . albopictus ( % ) Temperature suitability54 . 953 . 7–5644 . 342 . 7–45 . 6Maximum precipitation13 . 612 . 6–14 . 613 . 912 . 7–14 . 9Enhanced vegetation index ( mean ) 12 . 111 . 3–12 . 915 . 314 . 5–16 . 3Minimum precipitation9 . 18 . 5–1016 . 115 . 2–16 . 9Enhanced vegetation index ( range ) 8 . 37 . 7–99 . 18 . 3–10 . 1Urbanicity21 . 3–2 . 41 . 10 . 7–1 . 7 By combining the most comprehensive dataset of occurrence records with an advanced modelling approach and a bespoke set of environmental and land-cover correlates , we have produced contemporary high-resolution probability of occurrence maps for Ae . aegypti and Ae . albopictus , two of the most important disease vectors globally . Dengue and chikungunya , pathogens transmitted by these vectors and rapidly expanding in their distributions , are increasingly prominent in public health agendas and pose significant health threats to humans ( Staples et al . , 2009; Gardner et al . , 2012; Bhatt et al . , 2013; Weaver and Lecuit , 2015 ) . In common with previous work to map the global distributions of the dominant vectors of malaria ( Sinka et al . , 2010a , 2010b , 2011 ) , the maps will improve efforts to understand the spatial epidemiology of associated arboviruses , and to predict how these could change in the future . Specifically , these maps may be used to prioritize surveillance for these vector species and the diseases caused by the viruses they transmit in areas where disease and entomological reporting remains poor . For example , in parts of Asia and Africa where there is a mismatch between predicted environmental suitability and reported occurrences , these maps could be used to determine whether the vector has yet to fill its niche or if it is present but has not been reported due to limited entomological surveillance . They may also be used to identify areas where the species could persist but has yet to be reported , in order to proactively prevent vector establishment . The relative contributions of each of the environmental covariates to the global models concur with our theoretical and experimental understanding of each species' biology . Both species' distributions are highly dependent on the limiting factor temperature places on survival of the adult mosquitoes and on the gonotrophic cycle ( Brady et al . , 2013 ) ( Table 2 ) . The inclusion of a bespoke temperature suitability index ( Brady et al . , 2014 ) , both in defining the pseudo-absences and as a covariate , allowed us to capture both geographic and temporal variations in the species-specific effects of temperature in a single variable , leading to improved predictive skill of the models . As both Ae . aegypti and Ae . albopictus lay their eggs in small water-filled containers ( Morrison et al . , 2004 ) , it is encouraging that precipitation also has a strong influence on the model's predictions . The stronger influence of minimum precipitation for Ae . albopictus than for Ae . aegypti ( 16 . 1% vs 9 . 1% , Table 2 ) may reflect the former species' preference for non-domestic juvenile habitats , which are solely reliant on filling via precipitation . By contrast , Ae . aegypti primarily inhabits domestic water-holding containers ( Scott et al . , 2000 ) that are maintained in low-precipitation environments by water storage activities . The greater importance of enhanced vegetation index ( EVI ) for Ae . albopictus than for Ae . aegypti ( 15 . 3% vs 12 . 1% , Table 2 ) also supports the hypothesis that Ae . albopictus tends to prefer non-domestic juvenile sites ( Morrison et al . , 2004 ) . This does not , however , rule out the possibility that the two species can overlap . Additional finer scale studies need to be conducted to investigate if competitive exclusion for hosts and/or habitat occurs between Ae . aegypti and Ae . albopictus . The effect of urbanicity was surprisingly low for both species ( 2% and 1 . 1% for Ae . albopictus and Ae . aegypti , respectively ) . As both species have been shown to inhabit a wide variety of urban and peri-urban settings with various degrees of intensity ( Powell and Tabachnick , 2013; Li et al . , 2014 ) , it is likely that the simple urban/rural distinction of our urbanicity covariate did not sufficiently capture this variation and instead continuous covariates such as EVI allow to better distinguish the respective habitat types and were thus chosen more frequently by the model . Incorporating a larger set of covariates allowed us to investigate not only the effect of temperature on survival but for additional variance as shown in the relative influence plots ( Figure 1—figure supplement 1 ) . Future Aedes species distribution models could be improved by including a comprehensive global covariate that distinguishes human settlements using complex satellite imagery processing tools ( Schneider , 2012 ) . Our maps are based on covariates where each 5 km × 5 km pixel represents yearly mean average values . We therefore produce maps that represent the long-term average distribution of both species . However , this does not allow us to directly infer seasonal patterns of distributions which might be of importance on the periphery of the species distributions . With a more temporally resolved dataset it may be possible to capture the effects of intra-annual seasonality on the species' distributions . Adding mechanistic determinants , such as survival , have previously been used to combine seasonal patterns with global distribution maps ( Johansson et al . , 2014 ) . To make best use of the comprehensive set of data collected , we construct models and maps at a global scale , allowing the model to share information across the whole spectrum of environmental regions . However , given the scale at which this study was performed , there is always the possibility that variation in microclimate or local adaptive strategies of both species may have a significant impact in some locations . Previous studies have discussed the risk of pathogen importation and autochthonous transmission of DENV and CHIKV in Europe and the Americas without comprehensively accounting for the distribution of the vectors ( Bogoch et al . , 2014; Schaffner and Mathis , 2014 ) . These freely available vector distributions maps ( http://goo . gl/Zl2P7J ) can now be used as covariates to refine these studies and to generate high-resolution maps of the risk of possible local DENV and CHIKV transmission in currently non-endemic settings . Such maps would be useful for prioritizing surveillance in areas where there is a risk of disease importation . This will be especially important in areas where sporadic cases of related viruses have been reported , such as Europe , the United States , Argentina , and China ( Rezza et al . , 2007; Otero and Solari , 2010; Wu et al . , 2010; Johansson , 2015 ) . Both Ae . aegypti and Ae . albopictus have a history of global expansion associated with trade and travel ( Tatem et al . , 2006; Brown et al . , 2014; Gloria-Soria et al . , 2014 ) . Introductions of the species over long distances and between continents has been associated with international trade routes via shipping and overland spread driven by human movement and transport routes , both facilitated by the endophilic behavior of the two species ( Nawrocki and Hawley , 1987; Tatem et al . , 2006; Hofhuis et al . , 2009 ) . The global spread of the associated pathogens has undoubtedly been a consequence of increasing global connectedness . As these processes continue and the world becomes increasingly connected and urbanized , risk of importation and subsequent autochthonous transmission of DENV and CHIKV will continue to increase ( Allwinn et al . , 2008; Tomasello and Schlagenhauf , 2013; Khan et al . , 2014; Messina et al . , 2015 ) . The true distribution of both species is influenced by a variety of factors , not just the ones presented here . Nevertheless , this study represents an important baseline for further refinements . For instance , our maps can be used to indicate areas where the species are likely to become established if introduced . Accurately predicting the future distributions of these species will also require model-based estimates of the rate at which these species colonize new areas . Such predictions can be informed by human and trade mobility patterns between endemic and non-endemic regions as well as data on the past spread of the vectors . Improving our ability to predict rates of vector importation will therefore be crucial to inferring future risk ( Seebens et al . , 2013 ) . Previous studies have provided crucial information on genetic variation both within and between populations of these two vector species ( Brown et al . , 2011 ) . As the volume of georeferenced information on the population genetics of Ae . aegypti and Ae . albopictus increases , the potential to incorporate this information into mapping analyses to understand the current and future distribution of disease risk also increases . Phylogeographic analyses offer a unique way to infer the recent patterns of vector spread and to identify the major routes of importation ( Allicock et al . , 2012 ) . This information is crucial to inform models that predict the risk of vector introductions . Phylogenetic information could also be used to inform future iterations of the species distribution models used here by enabling the model to characterize and map environmental suitability for different vector subspecies . This could be particularly useful in the case of Ae . albopictus where genetic variation is known to underlie the ability to undergo diapause and therefore to overwinter in colder locations ( Takumi et al . , 2009 ) . Mapping the distributions of distinct genetic subgroups could also improve our understanding of the complex interactions between mosquito vector populations and virus strains and how this relates to spatial variation in transmission intensity ( Tsetsarkin et al . , 2007; Vazeille et al . , 2007; Tsetsarkin and Weaver , 2011; Zouache et al . , 2014 ) . The maps presented comprise a contemporary estimate of the current and potential future distribution of Ae . aegypti and Ae . albopictus . As more occurrence data become available , these maps can be refined to incorporate recent importation and establishment events and corresponding improvements in predictions . By disseminating both the occurrence data and the predictive maps on an open-access basis we hope to facilitate both the future development of these maps and their uptake by the global public health community . While the niche of a species is determined by a host of environmental , ecological and socio-economic factors of unknown influence and interaction strength , it is possible to exclude parts of the niche if the direct effects of one factor on a step rate-limiting to population persistence are well known . One such example for mosquito population persistence is whether temperature permits adult females to survive long enough to complete their first gonotrophic cycle and thus oviposit . Both adult female longevity and length of first gonotrophic cycle are temperature dependent . Combining these two relationships with a dynamic population-level simulation , Brady et al . ( 2013 , 2014 ) evaluated the thermal limits to persistence of Ae . aegypti and Ae . albopictus populations on a global scale . The binary outputs of this model are used as a mask to sample pseudo-absence points in locations known to be unsuitable–thereby informing the statistical model using mechanistic model outputs . The temperature suitability index developed by Brady et al . is also used in a continuous variable form ( i . e . , the relative number of ovipositions of parous females permitted by temperature ) as a covariate in the BRT model . The database used for this study contains information on the known global occurrences of the adults , pupae , larvae or eggs of Ae . aegypti and Ae . albopictus globally from 1960–2014 . We included data from a variety of sources , including ( 1 ) published literature and ( 2 ) primary and unpublished occurrence data from national and international entomological surveys . To our knowledge this is the largest , most comprehensive global dataset for both Ae . aegypti and Ae . albopictus . Confirmed Aedes occurrences were entered in the database after a comprehensive literature search using methods described elsewhere ( Kraemer et al . , 2015a; Kraemer et al . , 2015b; http://dx . doi . org/10 . 5061/dryad . 47v3c ) . In short , this included extracting all available location ( latitude and longitude ) information from the relevant articles , primarily using Google Maps ( http://www . google . com/maps ) so that it matched the spatial resolution of our covariate datasets of approximately 5 km × 5 km . Primary and unpublished data sources were obtained from Brazil , Europe , Indonesia , Taiwan , and the United States . After consolidating all data into two large databases for each species , independently they underwent spatial and temporal standardization . An occurrence record was defined as a single occurrence at a given unique location within one calendar year . This was important to avoid over-representation in regions where multiple surveys per year were performed , such as Taiwan or Brazil . To ensure the accuracy of the data we overlaid the geolocated occurrence points with a raster that distinguished land from water . Any records that were positioned outside the land area were subsequently removed . In total we assembled 19 , 930 and 22 , 137 occurrence records for Ae . aegypti and Ae . albopictus respectively . The distribution of occurrence points are plotted in Figure 1—figure supplement 4 . The distribution of both species considered in this study are known to be influenced by environmental factors such as temperature and demographic factors such as urbanisation ( Lounibos , 2002; Brown et al . , 2014 ) . Global gridded maps of such variables are becoming ever more available and have been commonly applied in SDM and disease mapping ( Hijmans et al . , 2005; Hay et al . , 2006; Gething et al . , 2011; Bhatt et al . , 2013; Pigott et al . , 2014a , 2014b ) . The rationale for the inclusion of each variable we used is described below . BRT models consistently outperform other species distribution models such as maximum entropy ( Maxent ) , GARP , and BIOCLIM in their predictive performance ( Elith et al . , 2006; Leathwick et al . , 2006 ) . BRT combines the strengths of regression trees ( i . e . , the omission of irrelevant variables and the ability to model complex interactions ) with machine learning techniques ( i . e . , the building of an ensemble of models that approximate the true response surface [Elith and Leathwick , 2009] ) . To prevent overfitting , the model used a penalized forward stepwise search and cross-validation method to identify the optimal number of decision trees ( Elith et al . , 2008 ) . Modelling was performed using the gbm , dismo , raster and seegSDM R packages using the R v 3 . 1 . 1 environment ( Ridgeway , 2013; Golding , 2014; Hijmans , 2014; R Core Team , 2014 ) .	Mosquitoes spread many disease-causing viruses and parasites between people and other animals , including viral infections such as dengue and chikungunya . Both infections cause high fevers often accompanied with excruciating joint pain or other flu-like symptoms . Dengue and chikungunya have become growing public health problems over the last fifty years . Today about half of the world's population is at risk of dengue infection , while chikungunya outbreaks , which were previously limited to Africa and Asia , have recently been reported in the Caribbean , South America and Europe . The dengue and chikungunya viruses are transmitted between people by two species of mosquitoes called Aedes aegypti and Ae . albopictus . Therefore it is important to work out where these mosquito species are found around the globe to identify the areas at risk . It is also important to predict where these species could become established if they were introduced , in order to identify areas that could become at risk in the future . Kraemer et al . now provide updated predictions about the distribution of these two mosquito species around the globe . These predictions are based upon the most up-to-date data on the known locations of the species combined with information on environmental conditions across the globe . The updated maps show that these Aedes mosquitoes are now found across all continents , including North America and Europe . Aedes albopictus mosquitoes in particular are rapidly expanding their territory around the globe . Kraemer et al . used their new maps to show that , unlike in the United States , many of the areas in Europe and China that could support this mosquito species do not yet appear to have been colonized . These findings provide a map of the distribution of both species as it stands at the moment . Further work is now needed to better understand which factors are contributing to the rapid expansion of these mosquitoes' range and what might be done to control this spread .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "epidemiology", "and", "global", "health" ]	2015	The global distribution of the arbovirus vectors Aedes aegypti and Ae. albopictus
HIV has been reported to be cytotoxic in vitro and in lymph node infection models . Using a computational approach , we found that partial inhibition of transmissions of multiple virions per cell could lead to increased numbers of live infected cells . If the number of viral DNA copies remains above one after inhibition , then eliminating the surplus viral copies reduces cell death . Using a cell line , we observed increased numbers of live infected cells when infection was partially inhibited with the antiretroviral efavirenz or neutralizing antibody . We then used efavirenz at concentrations reported in lymph nodes to inhibit lymph node infection by partially resistant HIV mutants . We observed more live infected lymph node cells , but with fewer HIV DNA copies per cell , relative to no drug . Hence , counterintuitively , limited attenuation of HIV transmission per cell may increase live infected cell numbers in environments where the force of infection is high . HIV infection is known to result in extensive T cell depletion in lymph node environments ( Sanchez et al . , 2015 ) , where infection is most robust ( Brenchley et al . , 2004; Doitsh et al . , 2010; Doitsh et al . , 2014; Finkel et al . , 1995; Galloway et al . , 2015; Mattapallil et al . , 2005 ) . Depletion of HIV infectable target cells , in addition to onset of immune control , is thought to account for the decreased replication ratio of HIV from an initial peak in early infection ( Bonhoeffer et al . , 1997; Nowak and May , 2000; Perelson , 2002; Phillips , 1996; Quiñones-Mateu and Arts , 2006; Ribeiro et al . , 2010; Wodarz and Levy , 2007 ) . This is consistent with observations that individuals are most infectious in the initial , acute stage of infection , where the target cell population is relatively intact and produces high viral loads ( Hollingsworth et al . , 2008; Wawer et al . , 2005 ) . T-cell death occurs by several mechanisms , which are either directly or indirectly mediated by HIV infection . Accumulation of incompletely reverse transcribed HIV transcripts is sensed by interferon-γ–inducible protein 16 ( Monroe et al . , 2014 ) and leads to pyroptotic death of incompletely infected cells by initiating a cellular defence program involving the activation of caspase 1 ( Doitsh et al . , 2010; Doitsh et al . , 2014; Galloway et al . , 2015 ) . HIV proteins Tat and Env have also been shown to lead to cell death of infected cells through CD95-mediated apoptosis following T-cell activation ( Banda et al . , 1992; Westendorp et al . , 1995a; Westendorp et al . , 1995b1995 ) . Using SIV infection , it has been shown that damage to lymph nodes due to chronic immune activation leads to an environment less conducive to T-cell survival ( Zeng et al . , 2012 ) . Finally , double strand breaks in the host DNA caused by integration of the reverse transcribed virus results in cell death by the DNA-PK-mediated activation of the p53 response ( Cooper et al . , 2013 ) . The lymph node environment is conducive to HIV infection due to: ( 1 ) presence of infectable cells ( Deleage et al . , 2016; Embretson et al . , 1993; Tenner-Racz et al . , 1998 ) ; ( 2 ) proximity of cells to each other and lack of flow which should enable cell-to-cell HIV spread ( Baxter et al . , 2014; Dale et al . , 2011; Groot et al . , 2008; Groppelli et al . , 2015; Gummuluru et al . , 2002; Hübner et al . , 2009; Jolly et al . , 2004; Jolly et al . , 2011; Münch et al . , 2007; Sherer et al . , 2007; Sourisseau et al . , 2007; Sowinski et al . , 2008 ) ; ( 3 ) decreased penetration of antiretroviral therapy ( ART ) ( Fletcher et al . , 2014a ) . Multiple infections per cell have been reported in cell-to-cell spread of HIV ( Baxter et al . , 2014; Boullé et al . , 2016; Dang et al . , 2004; Del Portillo et al . , 2011; Dixit and Perelson , 2004; Duncan et al . , 2013; Law et al . , 2016; Reh et al . , 2015; Russell et al . , 2013; Sigal et al . , 2011; Zhong et al . , 2013 ) . In this mode of HIV transmission , an interaction between the infected donor cell and the uninfected target results in directed transmission of large numbers of virions ( Baxter et al . , 2014; Groppelli et al . , 2015; Hübner et al . , 2009; Sowinski et al . , 2008 ) . This is in contrast to cell-free infection , where free-floating virus finds target cells through diffusion . Both modes occur simultaneously when infected donor cells are cocultured with targets . However , the cell-to-cell route is thought to be the main cause of multiple infections per cell ( Hübner et al . , 2009 ) . In the lymph nodes , several studies showed multiple infections ( Gratton et al . , 2000; Jung et al . , 2002; Law et al . , 2016 ) while another study did not ( Josefsson et al . , 2013 ) . One explanation for the divergent results is that different cell subsets are infected to different degrees . For example , T cells were shown not to be multiply infected in the peripheral blood compartment ( Josefsson et al . , 2011 ) . However , more recent work investigating markers associated with HIV latency in the face of ART found that the average number of HIV DNA copies per cell is greater than one in 3 out of 12 individuals . This occurred in the face of ART in the CD3-positive , CD32a high CD4 T-cell subset ( Descours et al . , 2017 ) . In the absence of suppressive ART , it would be expected that the number of HIV DNA copies per cell would be higher . Multiple viral integration attempts per cell may increase the probability of death . One consequence of HIV-mediated death may be that attenuation of infection may increase viral replication by increasing the number of live targets . Indeed , it has been suggested that more attenuated HIV strains result in more successful infections in terms of the ability of the virus to replicate in the infected individual ( Ariën et al . , 2005; Nowak and May , 2000; Payne et al . , 2014; Quiñones-Mateu and Arts , 2006; Wodarz and Levy , 2007 ) . Here , we experimentally examined the effect of attenuating cell-to-cell spread by using HIV inhibitors . We observed that partially inhibiting infection with drug or antibody resulted in an increase in the number of live infected cells in both a cell line and in lymph node cells . This is , to our knowledge , the first experimental demonstration at the cellular level that attenuation of HIV infection can result in an increase in live infected cells under specific infection conditions . We introduce a model of infection where each donor to target transmission leads to an infection probability r and death probability q per infection attempt . In our experimental system , one infection attempt is measured as one HIV DNA copy , whether integrated or unintegrated . The probability of successful infection of a target cell given n infection attempts is 1- ( 1 r ) n ( Sigal et al . , 2011 ) . We define Ln as the probability of a cell to survive infection in the face of n infection attempts . Assuming infection attempts act independently , Ln= ( 1-q ) n . The probability of a cell to be infected and not die after it has been exposed to n infection attempts is therefore:Pn= ( 1− ( 1−r ) n ) ( 1−q ) n This model makes several simplifying assumptions: ( 1 ) all infection attempts have equal probabilities to infect targets . ( 2 ) The probability for a cell to die from each transmission is equal between transmissions . ( 3 ) Infection attempts act independently , and productive infection and death are independent events . In this model , r and q capture the probabilities for a cell to be infected or die post-reverse transcription . For example , mutations which reduce viral fitness by decreasing the probability of HIV to integrate would reduce r , while mutations which reduce the probability of successful reverse transcription would reduce n . If the number of infection attempts n is Poisson distributed with mean λ , the probability for a cell to be infected is 1-e-rλ and the probability of a cell to live is Ln = e-qλ ( see Supplementary file 1 for parameters and definitions ) . As derived in Appendix 1 , the probability that a cell is productively infected will be:Pλ=e-λq ( 1-e-λr ( 1-q ) ) Since antiretroviral drugs lead to a reduction in the number of infection attempts by , for example , decreasing the probability of reverse transcription in the case of reverse transcriptase inhibitors , we introduced a drug strength value d , where d = 1 in the absence of drug and d > 1 in the presence of drug . In the presence of drug , λ is decreased to λ/d . The drug therefore tunes λ , and if the antiretroviral regimen is fully suppressive , λ/d is expected to be below what is required for ongoing replication . The probability of a cell to be infected and live given drug strength d is therefore:Pλ/d=e-λq/d ( 1-e-λr ( 1-q ) /d ) Analysis of the probability of a cell to survive and be infected as a function of r and q shows that at each drug strength d/λ , Pλ increases as the probability of infection r increases ( Figure 1 ) . Hence , the value of r strongly influences the amplitude of Pλ . How Pλ behaves when drug strength d/λ increases depends on the parameter values of r and q . A subset of parameter values results in a peak in the number of infected cells at intermediate d/λ , decreasing as drug strength increases further ( Figure 1 ) . We refer to such a peak in infected numbers as an infection optimum . As q increases , the cost of multiple infections per cell increases , and the infection optimum shifts to higher d/λ values . A fall from the infection optimum at decreasing d/λ is driven by increasing cell death as a result of increasing infection attempts per cell . This slope is therefore shallower , and peaks broader , at low q values ( Figure 1 ) . Our model assumes that cellular infection and death due to an HIV infection attempt are independent processes . This is based on observations that support a role for cell death as a cellular defence mechanism which may occur before productive infection , such as programmed cell death triggered by HIV integration induced DNA damage ( Cooper et al . , 2013 ) . An alternative model is that HIV-mediated cell death depends on productive infection . This would be consistent with cell death due to , for example , expression of viral proteins ( Westendorp et al . , 1995b1995 ) . Since the concentration of viral proteins may also scale with the number of infections per cell , we derived the mathematical model for such a process in the supplementary mathematical analysis . The models are equivalent , showing that independence of cell death and infection is not a necessary condition for an infection optimum to occur in the presence of inhibitor . Given that an infection optimum is dependent on parameter values , we next examined whether these parameter values occur experimentally in HIV infection . We therefore first tested for an infection optimum in the RevCEM cell line engineered to express GFP upon HIV Rev protein expression ( Wu et al . , 2007 ) . We subcloned the cell line to maximize the frequency of GFP-positive cells upon infection ( Boullé et al . , 2016 ) . We needed to detect the number of infection attempts per cell λ . To estimate this , we used PCR to detect the number of reverse transcribed copies of viral DNA in the cell by splitting each individual infected cell over multiple wells . We then detected the number of wells with HIV DNA by PCR amplification of the reverse transcriptase gene . Hence , the number of positive wells indicated the minimum number of viral DNA copies per cell , since more than one copy can be contained within the same well ( Josefsson et al . , 2011; Josefsson et al . , 2013 ) . We first measured the number of viral DNA copies in ACH-2 cells , reported to contain a single inactive HIV integration per genome ( Chun et al . , 1997; O'Doherty et al . , 2002 ) . We sorted a total of 166 ACH-2 cells at one cell per well into lysis buffer and subdivided single-cell lysates into four wells ( Figure 2—figure supplement 1A ) . About one quarter of cells showed a PCR product of the expected size . Cells with more than one HIV copy per cell were very rare and may reflect either errors in cell sorting or dividing cells ( Figure 2—figure supplement 1B ) . Similar frequencies were obtained when the ACH-2 cell line was subcloned or split over 10 wells ( Figure 2—figure supplement 1C ) . Given that each ACH-2 cell contains one HIV DNA copy , the frequency of detection indicated our detection efficiency per HIV DNA copy . To investigate the effect of multiple infection attempts per cell , we used coculture infection , where infected ( donor ) cells are co-incubated with uninfected ( target ) cells and lead to cell-to-cell spread . We used approximately 2% infected donor cells as our input , and detected the number of HIV DNA copies per cell by flow cytometric sorting of individual GFP-positive cells followed by splitting each cell lysate over 10 wells . Wells were then amplified by PCR and visualized on an agarose gel ( Figure 2A ) . We assayed 60 cells and obtained a wide distribution of viral DNA copies per cell , which ranged from 0 to 9 copies ( Figure 2B ) . The range of HIV DNA copies per cell fit a Poisson distribution with two means better than either a single mean Gaussian or Poisson distribution . However , the fit of the two mean Poisson distribution did not show two obvious peaks , and instead seemed to fit the data better due to the addition of fit parameters ( Figure 2—figure supplement 2 ) . Hence we cannot conclude that the distribution is bimodal . We also detected the HIV copy number in 30 GFP-positive cells infected by cell-free HIV . HIV in cell-free form was obtained by filtering supernatant from HIV producing cells to exclude cells or cell fragments , then infecting target cells with the filtered virus . Infection with this virus is defined here as cell-free infection . In this case , we detected either zero or one HIV copy per cell ( Figure 2B inset ) . The frequency of single HIV DNA copies was 0 . 23 , identical to the measured result in the ACH-2 cell line . We computationally corrected the detected number of DNA copies in coculture infection for the sensitivity of our PCR reaction as determined by the ACH-2 results ( Materials and methods ) . On average we obtained 15 ± 3 copies per cell after correction . To tune λ , we added the HIV reverse transcriptase inhibitor efavirenz ( EFV ) to infections . To calculate d , we used cell-free infection ( Figure 2C , see Figure 2—figure supplement 3 for logarithmic y-axis plot ) , which as verified above , results in single HIV copies per cell . For cell-free infection , we approximate d = 1/Tx , where Tx is defined as the number of infected cells with drug divided by the number of infected cells without drug with single infection attempts ( see Materials and methods and [ ( Sigal et al . , 2011] ) . This is equivalent to 1-ε in a commonly used model describing the effect of inhibitors on infection . In this model , ε is drug effectiveness , with the 50% inhibitory drug concentration ( IC50 ) and the Hill coefficient for drug action as parameter values ( Canini and Perelson , 2014; Shen et al . , 2008 ) . We fit the observed response of infection to EFV using this approach to estimate d across a range of EFV concentrations . Fit of the model to the cell-free data using wild type , EFV-sensitive HIV showed a monotonic decrease with IC50 = 2 . 9 nM and Hill coefficient of 2 . 1 ( Figure 2C , black line ) . We next dialed in EFV to tune λ/d in coculture infection . To obtain the number of infected target cells , and specifically exclude donor cells or donor-target cell fusions , target cells were marked by the expression of mCherry . Donor cells were stained with the vital stain Cell Trace Far Red ( CTFR ) . The concentration of live infected cells was determined after 2 days in coculture with infected donors . Live infected cells were identified based on the absence of cell death indicator dye DAPI fluorescence , and presence of GFP . The input of infected donor cells was excluded from the count of infected cells based on the absence of mCherry fluorescence . Donor-target cell fusions were excluded by excluding CTFR-positive cells ( see Figure 2—figure supplement 4 for gating strategy ) . While the percent of infected cells was reduced with drug , the concentration of live infected cells increased ( Figure 2—figure supplement 4 ) . We observed a peak in the number of live infected target cells at 4 nM EFV ( Figure 2D ) . We then fit the number of live infected cells using Equation ( 3 ) , where Pλ was multiplied by the input number of target cells per ml ( 106 cells/ml ) to obtain the predicted number of live infected cells per ml of culture . This was done to constrain r in the model , which strongly determines the amplitude of Pλ/d as described above . Equation ( 3 ) best fit the behaviour of infection when r = 0 . 22 and q = 0 . 17 , resulting in a peak at 4 . 8 nM EFV ( Figure 2D , black line ) . Hence an infection optimum is present in the cell line infection system . In order to determine whether the fitted r and q values were within a reasonable range , we measured these values experimentally . To measure r , we infected with cell-free HIV to avoid the broad distribution of HIV copy numbers observed in cell-to-cell spread , and determined the fraction of live infected cells Pλ ( Figure 2—figure supplement 5A ) . We then determined the mean number of HIV copies per cell λ for the same set of experiments corrected by the efficiency of detection ( Figure 2—figure supplement 5B ) . The parameter r was calculated as -ln ( 1-Pλ ) /λ ( Supplementary file 2 ) . To measure q , we blocked cell division using serum starvation to measure differences in cell concentration due to cell death only , and not due to proliferation of uninfected cells ( Figure 2—figure supplement 5C ) . We then infected with cell-free HIV and measured Lλ , defined as the fraction of live cells remaining upon infection with λ HIV DNA copies relative to infection blocked with EFV ( see below ) . To specifically detect the decrease in live cells as a result of events downstream of reverse transcription , we compared infected cells to cells exposed to the same virus concentration but treated with 40 nM EFV , a drug concentration where infection by cell-free virus is negligible ( Figure 2—figure supplement 3 ) . q was then calculated as -ln ( Lλ ) /λ , where Lλ was the probability of a cell to live given transmission with λ copies ( Supplementary file 2 ) . Measured r and q values were 0 . 28 ± 0 . 08 and 0 . 15 ± 0 . 07 ( mean ±standard deviation ) , respectively . The solution to Equation ( 3 ) using these values showed similar behavior to the solution with the fitted values for wild-type HIV infection , indicating that the fitted values gave a reasonable approximation of the behavior of the system ( Figure 2D , dashed green line ) . In order to investigate the dynamics of cell depletion due to cell-to-cell HIV spread and its modulation by the addition of an inhibitor , we performed time-lapse microscopy over a two day infection window . While infection parameters were different due to the constraints of visualizing cells ( Materials and methods ) , the general trend from the data was deterioration in the number of live cells in the time-lapse culture starting at 1 day post-infection when no drug was added . The deterioration in live cell numbers was averted by the addition of EFV ( Figure 2—figure supplement 6 ) . We next investigated whether an infection optimum occurs with EFV-resistant HIV . To derive the resistant mutant , we cultured wild-type HIV in our reporter cell line in the presence of EFV . We obtained the L100I partially resistant mutant . We then replaced the reverse transcriptase of the wild-type molecular clone with the mutant reverse transcriptase gene ( Materials and methods ) . We derived dmut for the L100I mutant using cell-free mutant infection ( Figure 3A , see Figure 3—figure supplement 1 for logarithmic y-axis plot ) . The L100I mutant was found to have an IC50 = 29 nM EFV and a Hill coefficient of 2 . 0 ( Figure 3A , black line ) . We next performed coculture infection ( see Figure 3—figure supplement 2 for gating strategy ) . Similarly to wild-type HIV coculture infection , there was a peak in the number of live infected target cells for the L100I mutant infection . However , the peak in live infected cells was shifted to 40 nM EFV ( Figure 3B ) . Fits were obtained to Equation ( 3 ) using dmut values and λ measured for wild-type infection . The fits recapitulated the experimental results when r = 0 . 29 and q = 0 . 13 , with a fitted peak at 45 nM EFV ( Figure 3B , black line ) . The solution to Equation ( 3 ) using the measured values for r and q showed a similar pattern to that obtained with the fitted values ( Figure 3B , dashed green line ) . We note that both wild type and mutant coculture infection has data points above the fit line at the highest drug concentrations . This may be a limitation of our model at drug values much higher than observed at the infection optimum . In this range of drug values , our model predicts a more pronounced decline in the number of infected cells than is observed experimentally . In order to examine whether a peak in live infected targets can be obtained with an unrelated inhibitor , we used the HIV neutralizing antibody b12 . This antibody is effective against cell-to-cell spread of HIV ( Baxter et al . , 2014; Reh et al . , 2015 ) . We obtained a peak in live infected cells at 5 ug/ml b12 ( Figure 4 ) . The b12 concentration that resulted in a peak number of live infected cells was the same for wild-type virus and the L100I mutant , showing that L100I mutant fitness gain was EFV specific . In contrast , cell-free infection in the face of b12 showed a sharp and monotonic drop in live infected cells for both wild type and mutant virus ( Figure 4—figure supplement 1 ) . While the RevCEM cell line is a useful tool to illustrate the principles governing the formation of an infection optimum , the sensitivity of such an optimum to parameter values would make its presence in primary HIV target cells speculative . We therefore investigated whether a fitness optimum occurs in primary human lymph node cells , the anatomical site which would be most likely to have a high force of infection . We derived human lymph nodes from HIV-negative individuals from indicated lung resections ( Supplementary file 3 ) , cellularized the lymph node tissue using mechanical separation , and infected the resulting lymph node cells with HIV . A fraction of the cells was infected by cell-free virus and used as infected donor cells . We added these to uninfected target cells from the same lymph node to test coculture infection , and detected the number of live infected cells 4 days post-infection with the L100I EFV-resistant mutant in the face of EFV . We detected the number of live infected cells by the exclusion of dead cells with the fixable death detection dye eFluor660 followed by single cell staining for HIV Gag using anti-p24 antibody ( Figure 5A ) . In each of the lymph nodes tested , we observed a peak in live infected cells at intermediate EFV concentrations . Lymph node cells from participant 205 showed a peak of live infected cells at 100 nM EFV ( Figure 5A , first row ) . The infection optimum in the lymph node cells of study participant 257 was visible as a plateau between 50 and 200 nM EFV . In the presence of EFV , there was a decrease in the fraction of dead cells that was offset by a similar increase in the fraction of live infected cells for lymph nodes from all participants . There were more overall detectable cells with EFV , resulting in differences in the absolute concentrations of live infected cells being larger than the differences in the fractions of live infected cells between EFV and non-drug-treated cells ( right two columns in Figure 5A , with absolute number of live infected cells shown in parentheses in the flow cytometry plots ) . This is most likely due to cells which died early becoming fragments and so being excluded from the total population in the absence of EFV . Peaks in the number of live infected cells in the face of drug may be specific to lymph node derived cells . Cell-to-cell infection of peripheral blood mononuclear cells ( PBMC ) with wild-type HIV showed a slight peak at a very low EFV concentration in cells from one blood donor , which was not repeated in cells from two other donors ( Figure 5—figure supplement 1 ) . We used a lymph node from study participant 251 , where we obtained more cells , to examine the number of HIV DNA copies per cell . Cells from this lymph node showed an infection optimum at 50 nM EFV ( Figure 5A , third row ) . To detect the effect of EFV on integrations per cell , we sorted single cells based on p24-positive signal , de-crosslinked to remove the fixative ( Materials and methods ) , then divided each cell lysate into four wells . Using fewer wells saved reagents without changing sensitivity , as demonstrated in the ACH-2 cell line ( Figure 2—figure supplement 1C ) . We detected HIV DNA copies by PCR 2 days post-infection . We observed multiple DNA copies in EFV-untreated lymph node cells . The number of copies decreased with EFV ( Figure 5B ) . We corrected for sensitivity of detection as quantified in ACH-2 cells ( Materials and methods ) . The corrected numbers were 21 HIV DNA copies with no drug , and five copies in the presence of EFV at the infection optimum ( Figure 5C , see Figure 5—figure supplement 2 for histograms of raw HIV DNA copy numbers per cell ) . Hence , the decrease in the number of copies still results in sufficient copies to infect the cell . Since L100I does not often occur in the absence of other drug resistance mutations according to the Stanford HIV Drug Resistance Database ( Rhee et al . , 2003 ) , we repeated the experiment with the K103N mutant , a frequently observed mutation in virologic failure with a higher level of resistance to EFV relative to the L100I mutant . We used cell-free infection to obtain drug inhibition per virion at each level of EFV , which we denote d103 ( Figure 6A , see Figure 6—figure supplement 1 for logarithmic y-axis plot ) . The fits showed a monotonic decrease with IC50 = 26 . 0 nM and Hill coefficient of 1 . 5 ( Figure 6A , black line ) . We then proceeded to use the K103N mutant in coculture infection , using cells from two different lymph nodes in different experiments ( see Figure 6—figure supplement 2 for results of individual experiments ) . We observed an infection optimum with EFV in lymph node cells . The peak in the number of live infected cells in the presence of drug was between 80 and 160 nM EFV ( Figure 6B ) . We fit the experimental data with Equation ( 3 ) using d103 values and the number of DNA copies in the absence of drug measured for L100I infection . We did not calculate the predicted number of infected cells for Pλ/d values since the lymph node is a complex environment containing different cell subsets ( Sallusto et al . , 1999 ) and the number of infectable target cells at the start of infection is difficult to determine . Hence , we normalized both the experimental number of live infected cells and the Pλ/d values from Equation ( 3 ) to the maximum value in each case . The fits recapitulated the experimental results when r = 0 . 91 and q = 0 . 15 , with a fitted peak at 90 nM EFV ( Figure 6B , black line ) . The q value matched the measured result in the cell line , while the r value was much higher . However , the fitted r value in this case is not expected to be accurate since we were unable to constrain it with the number of infected cells relative to the starting number of target cells . To examine if the observed peak in live cells may be due to EFV alone , we measured cell viability in lymph node cells from one of the study participants used in the above experiment as a function of EFV without infection . No clear dependence on EFV in the absence of infection was detected ( Figure 6—figure supplement 3 ) . The optimal virulence concept in ecology proposes that virulence needs to be balanced against host survival for optimal pathogen spread ( Bonhoeffer et al . , 1996; Bonhoeffer and Nowak , 1994; Gandon et al . , 2001; Jensen et al . , 2006 ) . At the cellular level , this implies that the number of successfully infected cells may increase when infection virulence is reduced . The current study is , to our knowledge , the first to address this question experimentally at the level of individual cells infected with HIV . Using a model where cells are infected and die in a probabilistic way , we found that there were two possible outcomes of partially inhibiting infection . In the case where cells were infected by single infection attempts , inhibition always led to a decline in the number of live infected cells , since inhibition reduced the number of infections per cell from one to zero . In contrast , in the case of multiple infection attempts per cell , the possibility existed that inhibition reduced the number of integrating HIV DNA copies , without extinguishing infection of the cell completely . If each HIV DNA copy increases the probability of cell death , reducing the number of HIV DNA copies without eliminating infection should lead to an increased probability of infected cell survival . This would consequently lead to an increase in the number of live infected cells . We investigated the outcome of partial inhibition of infection in both a cell line and primary lymph node cells . In both systems , we observed that there was a peak in live infected cell number at intermediate inhibitor concentrations . This correlated to a decreased number of viral DNA copies per cell . Further increasing inhibitor concentration led to a decline in live infected cell numbers , and infecting with EFV resistant mutants shifted the peak in live infected number to higher EFV concentrations . Our model as described by Equation ( 3 ) reproduced the essential behaviour of the experimental results . Construction of the model assumed independence of productive infection and cell death . However , as shown in the Appendix 1 , an equivalent model can be constructed assuming a dependence of cell death on infection . Neither model accurately captures infection dynamics at high-drug concentrations , away from the infection optimum . In this range , where the number of infection attempts per cell is much lower than 1 , infection declined more slowly with drug than predicted . The model can be further refined using a distribution for the number of DNA copies per cell . Moreover , the probability of death per HIV DNA copy we denote q may be dependent on how many infection attempts preceded the current infection attempt , and the model can be improved by measuring this dependence . Physiologically , an infection optimum in the face of an antiretroviral drug may be important in HIV infection of lymph node cells and may be less pronounced in cells from peripheral blood . We used EFV in our study since it is a common component of first line antiretroviral therapy , with frequent drug resistance mutations . However , the infection optimum we describe should occur with other classes of antiretroviral drugs , since all drugs should decrease the multiplicity of infection between cells . In terms of modeling , a future therapy component such as the integrase inhibitor dolutegravir would exert its effect on r and not λ in our model . However , the effect is symmetrical since e- ( λ/d ) r = e-λ ( r/d ) . The more complex outcome of partial inhibition of infection should also be considered in other infections where multiple pathogens infect one cell and host cell death is a possible outcome ( Mahamed et al . , 2017 ) . These observations reinforce previous results showing that successful completion of reverse transcription leads to cellular cytotoxicity . In addition to HIV cytotoxicity caused by viral integrations through the mechanism of double strand breaks ( Cooper et al . , 2013 ) , other mechanisms of HIV-induced death are also present , including IFI16-dependent innate immune system sensing of abortive reverse transcripts following non-productive infection of resting T cells ( Doitsh et al . , 2014; Monroe et al . , 2014 ) . The experiments presented here reflect the effect of partial inhibition on productive infection of HIV target cells , which mostly consist of activated T-cell subsets , not resting T cells . More complex models would be needed to decipher the effect of partial inhibition of HIV infection on resting T-cell numbers and the outcome of this in terms of available T-cell targets in future infection cycles . This study may have implications for the establishment of viral reservoirs in the context of poorly controlled infections , infections with some degree of drug resistance , or infections where some replication may take place in the face of ART , since infected cell survival is a pre-requisite for long-term persistence . The clinical implications of an infection optimum in the presence of EFV with EFV-sensitive HIV strains are likely to be negligible , since the drug concentrations at which the infection optimum occurs are extremely low . However , for EFV-resistant HIV , the infection optimum shifts to the range of EFV concentrations observed in lymph nodes ( ~100 nM ) ( Fletcher et al . , 2014b ) , and can be expected to shift to even higher EFV concentrations with more resistant mutants . As EFV has a longer half-life than the other antiretroviral drugs co-formulated with it , it may be the only agent present in partially adherent individuals for substantial periods of time ( Taylor et al . , 2007 ) . Therefore , partial inhibition of HIV infection with EFV may provide a surprising advantage to EFV resistant mutants , and may allow individuals failing therapy to better transmit drug resistant strains . Lymph nodes were obtained from the field of surgery of participants undergoing surgery for diagnostic purposes and/or complications of inflammatory lung disease . Informed consent was obtained from each participant , and the study protocol approved by the University of KwaZulu-Natal Institutional Review Board ( approval BE024/09 ) . Blood for PBMC was obtained from healthy blood donors under the same study protocol . The following reagents were obtained through the AIDS Research and Reference Reagent Program , National Institute of Allergy and Infectious Diseases , National Institutes of Health: the antiretroviral EFV; RevCEM cells from Y . Wu and J . Marsh; HIV molecular clone pNL4-3 from M . Martin; ACH-2 cells from T . Folks . Cell-free viruses were produced by transfection of HEK293 cells with pNL4-3 using TransIT-LT1 ( Mirus , Madison , WI ) or Fugene HD ( Roche , Risch-Rotkreuz , Switzerland ) transfection reagents . Virus containing supernatant was harvested after 2 days of incubation and filtered through a 0 . 45 µm filter ( Corning , New York , NY ) . b12 antibody was produced from transfecting HEK293 cells with a b12 expression plasmid ( expressed under a CMV promoter on a pHAGE6 lentiviral plasmid backbone , gift from A . Balazs ) , followed by harvesting of cell supernatant and purification at the California Institute of Technology protein expression core . The number of virus genomes in viral stocks was determined using the RealTime HIV-1 viral load test ( Abbott Diagnostics , Santa Clara , CA ) . For r and q measurement , 0 . 45 µm filtered cell-free supernatants from infected RevCEM cells were used , to include any secreted factors which may modulate cell-death . The L100I and K103N mutants were evolved by serial passages of wild-type NL4-3 in RevCEM cells in the presence of 20 nM EFV . After 16 days of selection , the reverse transcriptase gene was cloned from the proviral DNA and the mutant reverse transcriptase gene was inserted into the NL4-3 molecular clone . RevCEM clones E7 and G2 used in this study were generated as previously described ( Boullé et al . , 2016 ) . Briefly , the E7 clone was generated by subcloning RevCEM cells at single-cell density . Surviving clones were subdivided into replicate plates . One of the plates was screened for the fraction of GFP expressing cells upon HIV infection using microscopy , and the clone with the highest fraction of GFP-positive cells was selected . To generate the G2 clone , E7 cells were stably infected with the mCherry gene under the EF-1α promoter on a pHAGE2-based lentiviral vector ( gift from A . Balazs ) , subcloned , and screened for >99% mCherry-positive cells . All cell lines not authenticated , and mycoplasma negative . Cell culture and experiments were performed in complete RPMI 1640 medium supplemented with L-Glutamine , sodium pyruvate , HEPES , non-essential amino acids ( Lonza , Basel , Switzerland ) , and 10% heat-inactivated FBS ( GE Healthcare Bio-Sciences , Pittsburgh , PA ) . Lymph node cells were obtained by mechanical separation of lymph nodes and frozen at 5 × 106 cells/ml in a solution of 90% FBS and 10% DMSO with 2 . 5 μg/ml Amphotericin B ( Lonza ) . Cells were stored in liquid nitrogen until use , then thawed and resuspended at 106 cells/ml in complete RPMI 1640 medium supplemented with L-Glutamine , sodium pyruvate , HEPES , non-essential amino acids ( Lonza ) , 10% heat-inactivated FBS ( Hyclone ) , and IL-2 at 5 ng/ml ( PeproTech ) . Phytohemagglutinin at 10 µg/ml ( Sigma-Aldrich , St Louis , MO ) was added to activate cells . PBMCs were isolated by density gradient centrifugation using Histopaque 1077 ( Sigma-Aldrich ) and cultured at 106 cells/ml in complete RPMI 1640 medium supplemented with L-Glutamine , sodium pyruvate , HEPES , non-essential amino acids ( Lonza ) , 10% heat-inactivated FBS ( GE Healthcare Bio-Sciences , Pittsburgh , PA ) , and IL-2 at 5 ng/ml ( PeproTech , Rocky Hill , NJ ) . Phytohemagglutinin at 10 µg/ml ( Sigma-Aldrich ) was added to activate cells . For both primary cell types , donor cells for coculture infection were cultured for one day then infected by cell-free virus , while target cells were cultured for three days and infected with either cell-free HIV or infected donor cells . Cells from the parental ACH-2 cell line were diluted to 10 cells/ml in conditioned medium , with conditioned medium generated by culturing ACH-2 cells to 106 cells/ml , then filtering through a 0 . 22 µm filter ( Corning ) . 25 μl of the diluted cell suspension was then distributed to each well of a Greiner μClear 384-well plate ( mean of 0 . 5 cells per well ) . Clones were cultured for 3 weeks , where each week an additional 25 μl of conditioned medium was added to each well . Clones were detected in 5% of wells and two clones , designated D6 and C3 , were randomly chosen and further expanded . For a cell-free infection of RevCEM clones , PBMC and lymph node cells , 106 cells/ml were infected with 2 × 108 NL4-3 viral copies/ml ( ~20ng p24 equivalent ) for 2 days . For coculture infection , infected cells from the cell-free infection were used as the donors and cocultured with 106 cells/ml target cells . For RevCEM clones , 2% infected donor cells were added to uninfected targets and cocultured for 2 days in tissue culture experiments , and 20% infected donor cells were added to uninfected targets and cocultured for 2 days for time-lapse experiments . For lymph node cells and PBMCs , a ratio of 1:4 donor to targets cells was used . Infection was over 2 days in PBMC infection and for 4 days for infection of lymph node cells . To determine the number of live infected cells in reporter cell line experiments , E7 RevCEM reporter cells were infected as above used as donor cells . Prior to co-incubation with target cells , donor cells were stained with CellTrace Far Red ( CTFR , Thermo Fisher Scientific , Waltham , MA ) at 1 µM and washed according to manufacturer’s instructions . The G2 mCherry-positive reporter cells were used as infection targets , and cocultured with 2% infected donor cells for 2 days . The coculture infection was pulsed with 100 ng/ml DAPI ( Sigma-Aldrich ) immediately before flow cytometry and the number of live infected targets cells was determined by the number of DAPI negative , CTFR negative and mCherry and GFP double positive cells on a FACSAria Fusion machine ( BD Biosciences , Sparks , MD ) using the 355 , 488 and 633 nm laser lines . For cell-free infections where fewer fluorescence channels were used , a pulse of 300 nM of the far-red live cell impermeable dye DRAQ7 ( Biolegend , San Diego , CA ) immediately before flow cytometry was substituted for DAPI , and live infected cells detected as the number of DRAQ7-negative , GFP-positive cells on a FACSCaliber machine using 488 and 633 nm laser lines . Lymph node cells were resuspended in 1 ml of phosphate buffered saline ( PBS ) and stained at a 1:1000 dilution of the eFlour660 dye ( Thermo Fisher Scientific ) according to the manufacturer’s instructions . Cells were then fixed and permeabilized using the BD Cytofix/Cytoperm Fixation/Permeabilization kit ( BD Biosciences ) according to the manufacturer’s instructions . Cells were then stained with anti-p24 FITC conjugated antibody ( KC57 , Beckman Coulter , Brea , CA ) . Live infected lymph node cells were detected as the number of eFluor660-negative , p24-positive cells . Cells were acquired with a FACSAriaIII or FACSCaliber machine ( BD Biosciences ) using 488 and 633 nm laser lines . Results were analysed with FlowJo 10 . 0 . 8 software . For single-cell sorting to detect the number of HIV DNA copies per cell , cells were single-cell sorted using 85 micron nozzle in a FACSAriaIII machine . GFP-positive , DRAQ7-negative RevCEM clones were sorted 1 day post-infection into 96 well plates ( Biorad , Hercules , CA ) containing 30 µl lysis buffer ( 2 . 5 µl 0 . 1M Dithiothreitol , 5 µl 5% NP40 and 22 . 5 µl molecular biology grade water [Kurimoto et al . , 2007] ) . For experiments to determine the number of HIV DNA copies to measure r and q , the DRAQ7-negative subset was sorted . Fixed , p24-positive , eFluor660-negative lymph node cells were single-cell sorted two days post-infection into 96-well plates containing 5 µl of PKD buffer ( Qiagen , Hilden , Germany ) with 1:16 proteinase K solution ( Qiagen ) ( Thomsen et al . , 2016 ) . Sorted plates were snap frozen and kept at −80°C until ready for PCR . For analysis by flow cytometry , a minimum of 50 , 000 cells were collected per data point . For imaging infection by time-lapse microscopy , cell density was reduced to 5 × 104 cells/ml and cells were attached to ploy-l-lysine ( Sigma-Aldrich ) coated optical six-well plates ( MatTek , Ashland , MA ) . Infections with and without EFV were imaged in tandem using a Metamorph-controlled Nikon TiE motorized microscope with a Yokogawa spinning disk with a 20x , 0 . 75 NA phase objective in a biosafety level three facility . Excitation sources were 488 ( GFP ) and 561 ( mCherry ) laser lines and emission was detected through a Semrock Brightline quad band 440–40/521–21/607-34/700-45 nm filter . Images were captured using an 888 EMCCD camera ( Andor , Belfast , UK ) . Temperature ( 37°C ) , humidity and CO2 ( 5% ) were controlled using an environmental chamber ( OKO Labs , Naples , Italy ) . Fields of view were captured every 20 min . To facilitate automated image analysis of time-lapse experiment data , mCherry expressing G2 clone cells were used as targets and E7 clone cells used as infected donors . The number of live cells was measured as the number of cells expressing mCherry since intracellular mCherry protein is soluble and hence lost upon cell death when cellular membrane integrity is compromised . The number of live infected cells was measured as the number of cells expressing both mCherry and GFP . Three independent experiments were performed . Movies were analyzed using custom code developed with the Matlab R2014a Image Analysis Toolbox . Images in the mCherry channel were thresholded and the imfindcircle function used to detect round objects within the cell radius range . Cell centers were found . GFP signal underwent the same binary thresholding . The number of mCherry-positive 16 pixel2 squares around the cell centers was used as the as the number of total target cells at each time-point , and the number of squares double positive for fluorescence in the GFP channel was used as the number of infected target cells . 96-well plates of cells previously sorted at 1 cell per well were thawed at room temperature and spun down . Fixed cells were de-crosslinked by incubating in a thermocycler at 56°C for 1 hr . The lysate from each well was split equally over 10 wells ( 2 . 5 µl each well after correction for evaporation ) for E7 RevCEM or four wells ( 6 . 8 µl each well after correction for evaporation ) for lymph nodes , containing 50 µl of Phusion hot start II DNA polymerase ( New England Biolabs , Ipswich , MA ) PCR reaction mix ( 10 µl 5X Phusion HF buffer , 1 µl dNTPs , 2 . 5 µl of the forward primer , 2 . 5 µl of the reverse primer , 0 . 5 µl Phusion hot start II DNA polymerase , 2 . 5 µl of DMSO and molecular biology grade water to 50 µl reaction volume ) . Two rounds of PCR were performed . The first round reaction amplified a 700 bp region of the reverse transcriptase gene using the forward primer 5’ CCTACACCTGTCAACATAATTGGAAG 3’ and reverse primer 5’ GAATGGAGGTTCTTTCTGATG 3’ . Cycling program was 98°C for 30 s , then 34 cycles of 98°C for 10 s , 63°C for 30 s and 72°C for 15 s with a final extension of 72°C for 5 min . 1 µl of the first round product was then transferred into a PCR mix as above , with nested second round primers ( forward 5’ TAAAAGCATTAGTAGAAATTTGTACAGA 3’ , reverse 5’ GGTAAATCCCCACCTCAACAGATG 3’ ) . The second round PCR amplified a 550 bp product which was then visualized on a 1% agarose gel . PCR reactions were found to work best if sorted plates were thawed no more than once , and plates which underwent repeated freeze-thaw cycles showed poor amplification . A stochastic simulation in Matlab was used to generate a distribution for the number of positive wells per cell for each mean number of DNA copies per cell λ . The probability for a DNA copy to be present within a given well and be detected was set as σ/w , where σ was the detection sensitivity calculated as the number of ACH-2 with detectable integrations divided by the total number of ACH-2 cells assayed ( 38/166 , σ = 0 . 23 ) , and w was the number of wells . A random number m representing DNA copies per cell from a Poisson distribution with a mean λ was drawn , and a vector R of m random numbers from a uniform distribution was generated . If there existed an element Ri of the vector with a value between 0 and σ/w , the first well was occupied . If an element existed with a value between σ/w+γ and 2 ( σ/w ) , where γ <<1 , the second well was occupied , and if between ( σ/w+γ ) ( n-1 ) and n ( σ/w ) , the nth well was occupied . The sum of wells occupied at least once was determined , and the process repeated j times for each λ , where j was the number of cells in the experimental data . A least squares fit was performed to select λ which best fit the experimental results across well frequencies , and mean and standard deviation for λ was derived by repeating the simulation 10 times . To obtain d , we normalized Equation ( 2 ) by the fraction of infected cells in the absence of drug ( Sigal et al . , 2011 ) to obtain Tx = ( infected targets with EFV ) / ( infected targets no EFV ) = ( ( 1- ( 1 r ) λ/d ) ( 1-q ) λ/d ) / ( ( 1- ( 1 r ) λ ) ( 1-q ) λ ) . We approximate the result at small r , q to Tx = ( 1- e-rλ/d ) e-qλ/d/ ( 1- e-rλ ) e-qλ=eqqλ ( 1-1/d ) ( ( 1- e-rλ/d ) / ( 1- e-rλ ) ) . Expanding the exponentials we obtain Tx = ( 1 + qλ ( 1–1/d ) ) ( ( -rλ/d ) /-rλ ) = ( 1 + qλ ( 1–1/d ) ) ( 1/d ) . We note that at λ <1 , qλ ( 1–1/d ) <<1 , and hence Tx ≅ 1/d . Tx was measured from the experiments to obtain d values at the EFV concentrations used for cell-free infection , where λ < 1 . To obtain a fit of d as a function of the concentration of drug that gives half-maximal inhibition ( IC50 ) and Hill coefficient ( h ) for EFV , we used the relation for the fraction cells remaining infected in the face of drug ( Canini and Perelson , 2014 ) , whose definition is equivalent to Tx at λ < 1: ( 4 ) 1d=1− =1−[EFV]h[EFV]h+IC50h . Cell-free supernatant used in infection was derived as follows: 106 cells/ml were infected with 2 × 108 NL4-3 viral copies/ml ( ~20ng p24 equivalent ) for 2 days . Thereafter 0 . 2% of the infected cells from the cell-free infection were added to 106 cells/ml target cells . The infected supernatant from the coculture 2 days post-infection was filtered using a 0 . 45 µm filter ( Corning ) and added to cells at a 1:8 dilution , where the dilution was calibrated to result in non-saturating infection in terms of GFP expression . A fraction of the cells were sorted into lysis buffer at one cell per well 1 day post infection , split over four wells , and PCR performed as described above to determine HIV copy number per cell . The remaining cells from the same infection were used to determine frequency of DRAQ-7-negative , GFP-positive cells 2 days post-infection using flow cytometry . Cell-free supernatant used in infection was derived as for r , except that 1 day before harvesting of the viral supernatant from infected cells , infected cells were washed twice with PBS and serum-free growth medium added . At the same time , the target cells for the infection were washed twice with PBS and serum-free growth medium added . Cells were split into two wells , and EFV to a final concentration of 40 nM was added to one of the wells . Supernatant was harvested and filtered as described for r , and added to cells at a 1:2 dilution . A fraction of the cells were sorted into lysis buffer at one cell per well 1 day post-infection , split over four wells , and PCR performed as described above to determine HIV copy number per cell . The remaining cells from the same infection were used to determine the frequency of live and dead cells two days post-infection . The concentration of live cells was measured using the TC20TM automated cell counter ( Bio-Rad ) with trypan blue staining ( Lonza ) .	The HIVvirus infects cells of the immune system . Once inside , it hijacks the cellular molecular machineries to make more copies of itself , which are then transmitted to new host cells . HIV eventually kills most cells it infects , either in the steps leading to the infection of the cell , or after the cell is already producing virus . HIV can spread between cells in two ways , known as cell-free or cell-to-cell . In the first , individual viruses are released from infected cells and move randomly through the body in the hope of finding new cells to infect . In the second , infected cells interact directly with uninfected cells . The second method is often much more successful at infecting new cells since they are exposed to multiple virus particles . HIV infections can be controlled by using combinations of antiretroviral drugs , such as efavirenz , to prevent the virus from making more of itself . With a high enough dose , the drugs can in theory completely stop HIV infections , unless the virus becomes resistant to treatment . However , some patients continue to use these drugs even after the virus they are infected with develops resistance . It is not clear what effect taking ineffective , or partially effective , drugs has on how HIV progresses . Using efavirenz , Jackson , Hunter et al . partially limited the spread of HIV between human cells grown in the laboratory . The experiments mirrored the situation where a partially resistant HIV strain spreads through the body . The results show that the success of cell-free infection is reduced as drug dose increases . Yet paradoxically , in cell-to-cell infection , the presence of drug caused more cells to become infected . This can be explained by the fact that , in cell-to-cell spread , each cell is exposed to multiple copies of the virus . The drug dose reduced the number of viral copies per cell without stopping the virus from infecting completely . The reduced number of viral copies per cell made it more likely that infected cells would survive the infection long enough to produce virus particles themselves . Viruses that can kill cells , such as HIV , must balance the need to make more of themselves against the speed that they kill their host cell to maximize the number of infected cells . If transmission between cells is too effective and too many virus particles are delivered to the new cell , the virus may not manage to infect new hosts before killing the old ones . These findings highlight this delicate balance . They also indicate a potential issue in using drugs to treat partially resistant virus strains . Without care , these treatments could increase the number of infected cells in the body , potentially worsening the effects of living with HIV .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "microbiology", "and", "infectious", "disease" ]	2018	Incomplete inhibition of HIV infection results in more HIV infected lymph node cells by reducing cell death
Paying attention to one speaker in a noisy place can be extremely difficult , because to-be-attended and task-irrelevant speech compete for processing resources . We tested whether this competition is restricted to acoustic-phonetic interference or if it extends to competition for linguistic processing as well . Neural activity was recorded using Magnetoencephalography as human participants were instructed to attend to natural speech presented to one ear , and task-irrelevant stimuli were presented to the other . Task-irrelevant stimuli consisted either of random sequences of syllables , or syllables structured to form coherent sentences , using hierarchical frequency-tagging . We find that the phrasal structure of structured task-irrelevant stimuli was represented in the neural response in left inferior frontal and posterior parietal regions , indicating that selective attention does not fully eliminate linguistic processing of task-irrelevant speech . Additionally , neural tracking of to-be-attended speech in left inferior frontal regions was enhanced when competing with structured task-irrelevant stimuli , suggesting inherent competition between them for linguistic processing . The seminal speech-shadowing experiments conducted in the 50 s and 60 s set the stage for studying one of the primary cognitive challenges encountered in daily life: how do our perceptual and linguistic systems deal effectively with competing speech inputs ? ( Cherry , 1953; Broadbent , 1958; Treisman , 1960 ) . Over the past decades , a wealth of behavioral and neural evidence has accumulated showing that when only one speech-stream is behaviorally relevant , auditory and linguistic resources are devoted to its preferential encoding at the expense of other task-irrelevant input . Consequentially , this so-called ‘attended’ message can be repeated , comprehended , and remembered , whereas very little of competing task-irrelevant speech is explicitly recalled ( Glucksberg , 1970; Ambler et al . , 1976; Neely and LeCompte , 1999; Oswald et al . , 2000; Brungart et al . , 2001 ) . This attentional selection is accompanied by attenuation of speech-tracking for unattended speech in auditory regions ( Mesgarani and Chang , 2012; Horton et al . , 2013; Zion Golumbic et al . , 2013a; O'Sullivan et al . , 2015; Fiedler et al . , 2019; Teoh and Lalor , 2019 ) as well as language-related regions ( Zion Golumbic et al . , 2013b ) , particularly for linguistic-features of the speech ( Brodbeck et al . , 2018a; Broderick et al . , 2018; Ding et al . , 2018; Brodbeck et al . , 2020a ) . However , as demonstrated even in the earliest studies , the content of task-irrelevant speech is probably not fully suppressed and can affect listener behavior in a variety of ways ( Moray , 1959; Bryden , 1964; Yates , 1965 ) . Indeed , despite decades of research , the extent to which concurrent speech are processed and the nature of the competition for resources between ‘attended’ and ‘task-irrelevant’ input in multi-speaker contexts is still highly debated ( Kahneman , 1973; Driver , 2001; Lachter et al . , 2004; Bronkhorst , 2015 ) . Fueling this debate are often-conflicting empirical findings regarding whether or not task-irrelevant speech is processed for semantic and linguistic content . Many studies fail to find behavioral or neural evidence for processing task-irrelevant speech beyond its acoustic features ( Carlyon et al . , 2001; Lachter et al . , 2004; Ding et al . , 2018 ) . However , others are able to demonstrate that at least some linguistic information is gleaned from task-irrelevant speech . For example , task-irrelevant speech is more distracting than non-speech or incomprehensible distractors ( Rhebergen et al . , 2005; Iyer et al . , 2010; Best et al . , 2012; Gallun and Diedesch , 2013; Carey et al . , 2014; Kilman et al . , 2014; Swaminathan et al . , 2015; Kidd et al . , 2016 ) , and there are also indications for implicit processing of the semantic content of task-irrelevant speech , manifest through priming effects or memory intrusions ( Tun et al . , 2002; Dupoux et al . , 2003; Rivenez et al . , 2006; Beaman et al . , 2007; Carey et al . , 2014; Aydelott et al . , 2015; Schepman et al . , 2016 ) . Known as the ‘Irrelevant Sound Effect’ ( ISE ) , these are not always accompanied by explicit recall or recognition ( Lewis , 1970; Bentin et al . , 1995; Röer et al . , 2017a ) , although in some cases task-irrelevant words , such as one’s own name , may also ‘break in consciousness’ ( Cherry , 1953; Treisman , 1960; Wood and Cowan , 1995; Conway et al . , 2001 ) . Behavioral findings indicating linguistic processing of task-irrelevant speech have been interpreted in two opposing ways . Proponents of Late-Selection attention theories understand them as reflecting the system’s capability to apply linguistic processing to more than one speech stream in parallel , albeit mostly pre-consciously ( Deutsch and Deutsch , 1963; Parmentier , 2008; Parmentier et al . , 2018; Vachon et al . , 2020 ) . However , others maintain an Early-Selection perspective , namely , that only one speech stream can be processed linguistically due to inherent processing bottlenecks , but that listeners may shift their attention between concurrent streams giving rise to occasional ( conscious or pre-conscious ) intrusions from task-irrelevant speech ( Cooke , 2006; Vestergaard et al . , 2011; Fogerty et al . , 2018 ) . Adjudicating between these two explanations experimentally is difficult , due to the largely indirect-nature of the operationalizations used to assess linguistic processing of task-irrelevant speech . Moreover , much of the empirical evidence fueling this debate focuses on detection of individual ‘task-irrelevant’ words , effects that can be easily explained either by parallel processing or by attention-shifts , due to their short duration . In attempt to broaden this conversation , here we use objective neural measures to evaluate the level of processing applied to task-irrelevant speech . Using a previously established technique of hierarchical frequency-tagging ( Ding et al . , 2016; Makov et al . , 2017 ) , we are able to go beyond the question of detecting individual words and probe whether linguistic processes that require integration over longer periods of time – such as syntactic structure building – are applied to task-irrelevant speech . To study this , we recorded brain activity using Magnetoencephalography ( MEG ) during a dichotic listening selective-attention experiment . Participants were instructed to attend to narratives of natural speech presented to one ear , and to ignore speech input from the other ear ( Figure 1 ) . Task-irrelevant stimuli consisted of sequences of syllables , presented at a constant rate ( 4 Hz ) , with their order manipulated to either create linguistically Structured or Non-Structured sequences . Specifically , for the Non-Structured syllables were presented in a completely random order , whereas in the Structured stimuli syllables were ordered to form coherent words , phrases , and sentences . In keeping with the frequency-tagging approach , each of these linguistic levels is associated with a different frequency ( words - 2 Hz , phrases - 1 Hz , sentences - 0 . 5 Hz ) . By structuring task-irrelevant speech in this way , the two conditions were perfectly controlled for low-level acoustic attributes that contribute to energetic masking ( e . g . loudness , pitch , and fine-structure ) , as well as for the presence of recognizable acoustic-phonetic units , which proposedly contributes to phonetic interference during speech-on-speech masking ( Rhebergen et al . , 2005; Shinn-Cunningham , 2008; Kidd et al . , 2016 ) . Rather , the only difference between the conditions was in the order of the syllables which either did or did not form linguistic structures . Consequentially , if the neural signal shows peaks at frequencies associated with linguistic-features of Structured task-irrelevant speech , as has been reported previously when these type of stimuli are attended or presented without competition ( Ding et al . , 2016; Ding et al . , 2018; Makov et al . , 2017 ) , this would provide evidence that integration-based processes operating on longer time-scales are applied to task-irrelevant speech , for identifying longer linguistic units comprised of several syllables . In addition , we also tested whether the neural encoding of the to-be-attended speech itself was affected by the linguistic structure of task-irrelevant speech , which could highlight the source of potential tradeoffs or competition for resources when presented with competing speech ( Zion Golumbic et al . , 2013b; O'Sullivan et al . , 2015; Fiedler et al . , 2019; Teoh and Lalor , 2019 ) . We measured MEG recordings from 30 ( 18 females , 12 males ) native Hebrew speakers . Participants were adult volunteers , ages ranging between 18 and 34 ( M = 24 . 8 , SD = ± 4 . 2 ) , and all were right-handed . Sample size was determined a priori , based on a previous study from our group using a similar paradigm and electrophysiological measures ( Makov et al . , 2017 ) , where significant effects were found in a sample of n = 21 participants . Exclusion criteria for participation included: non-native Hebrew speakers , a history of neurological disorders or ADHD ( based on self-report ) or the existence of metal implants ( which would disrupt MEG recordings ) . The study was approved by the IRB committee at Bar-Ilan University and all participants provided their written consent for participation prior to the experiment . Natural speech stimuli were narratives from publicly available Hebrew podcasts and short audio stories ( duration: 44 . 53 ± 3 . 23 s ) . These speech materials were chosen from an existing database in the lab , that were used in previous studies and for which the behavioral task had already been validated ( see Experimental Procedure ) . The stimuli originally consisted of narratives in both female and male voices . However , since it is known that selective attention to speech is highly influenced by whether the competing voices are of the same/different sex ( Brungart et al . , 2001; Rivenez et al . , 2006; Ding and Simon , 2012 ) , and since the task-irrelevant stimuli were recorded only in a male voice ( see below ) , we transformed narratives that were originally recorded in a female voice to a male voice ( change-gender function in Praat; Boersma , 2011 , http://www . praat . org ) . To ensure that the gender change did not affect the naturalness of the speech and to check for abnormalities in the materials , we conducted a short survey among 10 native Hebrew speakers . They all agreed that the speech sounded natural and normal . Sound intensity was equated across all narratives . Stimuli examples are available at: https://osf . io/e93qa . These natural speech narratives served as the to-be-attended stimuli in the experiment . For each participant , they were randomly paired with task-irrelevant speech ( regardless of condition ) , to avoid material-specific effects . A bank of individually recorded Hebrew syllables were used to create two sets of isochronous speech sequences . Single syllables were recorded in random order by a male actor , and remaining prosodic cues were removed using pitch normalization in Praat . Additional sound editing was performed to adjust the length of each syllable to be precisely 250 ms either by truncation or silence padding at the end ( original mean duration 243 . 6 ± 64 . 3 ms , range 168–397 ms ) . In case of truncation , a fading out effect was applied to the last 25 ms to avoid clicks . Sound intensity was then manually equated for all syllables . These syllables were concatenated to create long sequences using custom-written scripts in MATLAB ( The MathWorks; code available at https://osf . io/e93qa ) , equated in length to those of the natural speech segments ( 44 . 53 ± 3 . 23 seconds ) . Sequences could either be Non-Structured , with syllables presented in a fully random order without creating meaningful linguistic units , or they could be linguistically Structured . Structured sequences were identical to those used in a previous study from our group ( Makov et al . , 2017 ) , and were formed as follows: Every two syllables formed a word , every two words formed a phrase , and every two phrases formed a sentence . Because syllables were grouped hierarchically into linguistic constituents with no additional acoustic gaps inserted between them , different linguistic hierarchies are associated with fixed periodicities throughout the stimuli ( syllables at 4 Hz , words at 2 Hz , phrases at 1 Hz , and sentences at 0 . 5 Hz; Figure 1b ) . Structured stimuli also contained no prosodic cues or other low-level acoustic indications for boundaries between linguistics structures , nor did Structured sentences include rhymes , passive form of verbs , or arousing semantic content . See Supplementary Material for more information on the construction of Structured and Non-Structured stimuli . The modulation spectrum of both types of task-irrelevant stimuli is shown in Figure 1d . It was calculated using a procedure analogous to the spectral analysis performed on the MEG data , in order to ensure maximal comparability between the spectrum of the stimuli and the spectrum of the neural response . Specifically , ( 1 ) the broadband envelope of each sequence was extracted by taking the root-mean-square of the audio ( 10 ms smoothing window ) ; ( 2 ) the envelope was segmented into 8 s long segments , which was identical to the segmentation of the MEG data; ( 3 ) a fast Fourier transform ( FFT ) was applied to each segment; and ( 4 ) averaged across segments . As expected , both stimuli contained a prominent peak at 4 Hz , corresponding to the syllable-rate . The Structured stimuli also contain a smaller peak at 2 Hz , which corresponds to the word-rate . This is an undesirable side-effect of the frequency-tagging approach , since ideally these stimuli should not contain any energy at frequencies other than the syllable-rate . As shown in the Supplementary Material , the 2 Hz peak in the modulation spectrum reflects the fact that a consistently different subset of syllables occurs at each position within the sequence ( e . g . at the beginning\end of words; for similar acoustic effects when using frequency-tagging of multi-syllable words see Luo and Ding , 2020 . A similar 2 Hz peak is not observed in the Non-Structured condition , where the syllables are randomly positioned throughout all stimuli . Given this difference in the modulation spectrum , if we were to observe a 2 Hz peak in the neural response to Structured vs . Non-Structured stimuli , this would not necessarily provide conclusive evidence for linguistic ‘word-level’ encoding ( although see Makov et al . , 2017 ) and Supplementary Material for a way to control for this ) . As it happened , in the current dataset we did not see a 2 Hz peak in the neural response in either condition ( see Results ) , therefore this caveat did not affect the interpretability of the data in this specific instance . Importantly , neither the Structured nor the Non-Structured stimuli contained peaks at frequencies corresponding to other linguistic levels ( 1 Hz and 0 . 5 Hz ) , hence comparison of neural responses at these frequencies remained experimentally valid . Stimuli examples are available at: https://osf . io/e93qa . The experiment used a dichotic listening paradigm , in which participants were instructed to attend to a narrative of natural Hebrew speech presented to one ear , and to ignore input from the other ear where either Structured or Non-Structured task-irrelevant speech was presented . The experiment included a total of 30 trials ( 44 . 53 ± 3 . 23 seconds ) , and participants were informed at the beginning of each trial which ear to attend to , which was counterbalanced across trials . The sound intensity of the task-irrelevant stimuli increased gradually during the first three seconds of each trial , to avoid inadvertent hints regrading word-boundaries , and the start of each trial was excluded from data analysis . After each trial , participants answered four multiple choice questions about the content of the narrative they were supposed to attend to ( 3-answers per question; chance level = 0 . 33 ) . Some of the questions required recollection of specific details ( e . g . ‘what color was her hat ? ” ) , and some addressed the ‘gist’ of the narrative , ( e . g . ‘why was she sad ? ” ) . The average accuracy rate of each participant ( % questions answered correctly ) was calculated across all questions and narratives , separately for trials in the Structured and Non-Structured condition . This task was chosen as a way to motivate and guide participants to direct attention toward the to-be-attended narrative and provide verification that indeed they listened to it . At the same time , we recognize that this task is not highly sensitive for gauging the full extent of processing the narrative for two reasons: ( 1 ) its sparse sampling of behavior ( four questions for a 45 s narrative ) ; and ( 2 ) since accuracy is affected by additional cognitive factors besides attention such as short-term memory , engagement , and deductive reasoning . Indeed , behavioral screening of this task showed that performance was far from perfect even when participants listened to these narratives in a single-speaker context ( i . e . , without additional competing speech; average accuracy rate 0 . 83 ± 0 . 08; n = 10 ) . Hence , we did not expect performance on this task to necessarily reflect participants’ internal attentional state . At the same time , this task is instrumental in guiding participants' selective attention toward the designated speaker , allowing us to analyze their neural activity during uninterrupted listening to continuous speech , which was the primarily goal of the current study . Based on previous experience , the Structured speech materials are not immediately recognizable as Hebrew speech and require some familiarization . Hence , to ensure that all participants could identify these as speech , and in order to avoid any perceptual learning effects during the main experiment , they underwent a familiarization stage prior to the start of the main experiment , inside the MEG . In this stage , participants heard sequences of 8 isochronous syllables , which were either Structured – forming a single sentence – or Non-Structured . After each trial participants were asked to repeat the sequence out loud . The familiarity stage continued until participants correctly repeated five stimuli of each type . Structured and Non-Structured stimuli were presented in random order . At the end of the experiment , a short auditory localizer task was performed . The localizer included hearing tones in five different frequencies: 400 , 550 , 700 , 850 , and 1000 Hz , all 200 ms long . The tones were presented with random ISIs: 500 , 700 , 1000 , 1200 , 1400 , 1600 , 1800 , and 2000 ms . Participants listened to the tones passively and were instructed only to focus on the fixation mark in the center of the screen . MEG recordings were conducted with a whole-head , 248-channel magnetometer array ( 4D Neuroimaging , Magnes 3600 WH ) in a magnetically shielded room at the Electromagnetic Brain Imaging Unit , Bar-Ilan University . A series of magnetometer and gradiometer reference coils located above the signal coils were used to record and subtract environmental noise . The location of the head with respect to the sensors was determined by measuring the magnetic field produced by small currents delivered to five head coils attached to the scalp . Before the experimental session , the position of head coils was digitized in relation to three anatomical landmarks ( left and right preauricular points and nasion ) . The data was acquired at a sample rate of 1017 . 3 Hz and an online 0 . 1 to 200 Hz band-pass filter was applied . The 50 Hz power line noise fluctuations were recorded directly from the power line as well as vibrations using a set of accelerometers attached to the sensor in order to remove the artifacts on the MEG recordings . Preprocessing was performed in MATLAB ( The MathWorks ) using the FieldTrip toolbox ( http://www . fieldtriptoolbox . org ) . Outlier trials were identified manually by visual inspection and were excluded from analysis . Using independent component analysis ( ICA ) we removed eye movements ( EOG ) , heartbeat and vibrations of the building . The clean data was then segmented into 8 s long segments , which corresponds to four sentences in the Structured condition . Critically , these segments were perfectly aligned such that they all start with the onset of a syllable , which in the Structured condition will also be the onset of a sentence . Source estimation was performed in Python ( http://www . python . org ) using the MNE-python platform ( Gramfort et al . , 2013; Gramfort et al . , 2014 ) . Source modeling was performed on the pre-processed MEG data , by computing Minimum-Norm Estimates ( MNEs ) . In order to calculate the forward solution , and constrain source locations to the cortical surface , we constructed a Boundary Element Model ( BEM ) for each participant . BEM was calculated using the participants’ head shape and location relative to the MEG sensors , which was co-registered to an MRI template ( FreeSurfer; surfer . nmr . mgh . harvard . edu ) . Then , the cortical surface of each participant was decimated to 8194 source locations per hemisphere with at least 5 mm spacing between adjacent locations . A noise covariance matrix was estimated using the inter-trial intervals in the localizer task ( see Additional Tasks ) , that is , periods when no auditory stimuli were presented . Then , an inverse operator was computed based on the forward solution and the noise covariance matrix , and was used to estimate the activity at each source location . For visualizing the current estimates on the cortical surface , we used dynamic Statistical Parametric Map ( dSPM ) , which is an F-statistic calculated at each voxel and indicating the relationship between MNE amplitude estimations and the noise covariance ( Dale et al . , 2000 ) . Finally , individual cortical surfaces were morphed onto a common brain , with 10 , 242 dipoles per hemisphere ( Fischl et al . , 1999 ) , in order to compensate for inter-subject differences . The behavioral score was calculated as the average correct response across trials ( four multiple-choice question per narrative ) for each participant . In order to verify that participants understood and completed the task , we performed a t-test between accuracy rates compared to chance-level ( i . e . 0 . 33 ) . Then , to test whether performance was affected by the type of task-irrelevant speech presented , we performed a paired t-test between the accuracy rates in both conditions . We additionally performed a median-split analysis of the behavioral scores across participant , based on their neural response to task-irrelevant speech ( specifically the phrase-level response; see MEG data analysis ) , to test for possible interactions between performance on the to-be-attended speech and linguistic neural representation of task-irrelevant speech . Behavioral results reflecting participants response accuracy on comprehension questions about narratives were significantly above chance ( M = 0 . 715 , SD = ± 0 . 15; t ( 28 ) =26 . 67 , p<0 . 001 ) . There were no significant differences in behavior as a function of whether task-irrelevant speech was Structured or Non-Structured ( t ( 28 ) =−0 . 31 , p=0 . 75; Figure 2 ) . Additionally , to test for possible interactions between answering questions about the to-be-attended speech and linguistic neural representation of task-irrelevant speech , we performed a median-split analysis of the behavioral scores across participants . Specifically , we used the magnitude of the ITPC value at 1 Hz in the Structured condition ( averaged across all sensors ) , in order to split the sample into two groups – with high and low 1 Hz responses . We performed a between-group t-test on the behavioral results in the Structured condition , and also on the difference between conditions ( Structured – Non-Structured ) . Neither test showed significant differences in performance between participants whose 1 Hz ITPC was above vs . below the median ( Structured condition: t ( 27 ) = −1 . 07 , p=0 . 29; Structured – Non-Structured: t ( 27 ) = −1 . 04 , p=0 . 15 ) . Similar null-results were obtained when the median-split was based on the source-level data . Scalp-level spectra of the Inter-trial phase coherence ( ITPC ) showed a significant peak at the syllabic-rate ( 4 Hz ) in response to both Structured and Non-Structured hierarchical frequency-tagged speech , with a four-pole scalp-distribution common to MEG recorded auditory responses ( Figure 3a ) ( p<10^−9; large effect size , Cohen's d > 1 . 5 in both ) . As expected , there was no significant difference between Structured and Non-Structured condition in the 4 Hz response ( p=0 . 899 ) . Importantly , we also observed a significant peak at 1 Hz in the Structured condition ( p<0 . 003; moderate effect size , Cohen's d = 0 . 6 ) , but not in the Non-Structured condition ( p=0 . 88 ) . Comparison of the 1 Hz ITPC between these conditions also confirmed a significant difference between them ( p=0 . 045; moderate effect size , Cohen's d = 0 . 57 ) . The scalp-distribution of the 1 Hz peak did not conform to the typical auditory response topography , suggesting different neural generators . No other significant peaks were observed at any other frequencies , including the 2 Hz or 0 . 5 Hz word and sentence-level rates , nor did the responses at these rates differ significantly between conditions . In order to determine the neural source of the 1 Hz peak in the Structured condition , we repeated the spectral analysis in source-space . An inverse solution was applied to individual trials and the ITPC was calculated in each voxel . As shown in Figure 3b , the source-level ITPC spectra , averaged over each hemisphere separately , is qualitatively similar to that observed at the scalp-level . The only significant peaks were at 4 Hz in both conditions ( p<10−8; large effect size , Cohen's d > 1 . 7 in both conditions and both hemispheres ) and at 1 Hz in the Structured condition ( left hemisphere p=0 . 052 , Cohen's d = 0 . 43; right hemisphere p<0 . 007 , Cohen's d = 0 . 6 ) , but not in the Non-Structured condition . Statistical comparison of the 1 Hz peak between conditions revealed a significant difference between the Structured and Non-Structured condition over the left hemisphere ( p=0 . 026 , Cohen's d = 0 . 57 ) , but not over the right hemisphere ( p=0 . 278 ) . Figure 3c shows the source-level distribution within the left hemisphere of the difference in 1 Hz ITPC between the Structured and Non-Structured condition . The effect was observed primarily in frontal and parietal regions . Statistical testing evaluating the difference between conditions was performed in 22 pre-determined ROIs per hemisphere , using a permutation test . This indicated significant effects in several ROIs in the left hemisphere including the inferior-frontal cortex and superior parietal cortex , as well as the mid-cingulate and portions of the middle and superior occipital gyrus ( cluster-corrected p=0 . 002 ) . No ROIs survived multiple-comparison correction in the right hemisphere ( cluster-corrected p=0 . 132 ) , although some ROIs in the right cingulate were significant at an uncorrected level ( p<0 . 05 ) . With regard to the 4 Hz peak , it was localized as expected to bilateral auditory cortex and did not differ significantly across conditions in either hemisphere ( left hemisphere: p=0 . 155 , right hemisphere: p=0 . 346 ) . We therefore did not conduct a more fine-grained analysis of different ROIs . As in the scalp-level data , no peaks were observed at 2 Hz and no significant difference between conditions ( left hemisphere: p=0 . 963 , right hemisphere: p=0 . 755 ) . Speech tracking analysis of responses to the to-be-attended narrative yielded robust TRFs ( scalp-level predictive power r = 0 . 1 , p<0 . 01 vs . permutations ) . The TRF time-course featured two main peaks , one ~ 80 ms and the other ~140 ms , in line with previous TRF estimations ( Akram et al . , 2017; Fiedler et al . , 2019; Brodbeck et al . , 2020b ) . Both the scalp-level and source-level TRF analysis indicated that TRFs were predominantly auditory – showing the common four-pole distribution at the scalp-level ( Figure 4a ) and in the source-level analysis was localized primarily to auditory cortex ( superior temporal gyrus and sulcus; STG/STS ) as well as left insula/IFG ( Figure 4b; 140 ms ) . When comparing the TRFs to the to-be-attended speech as a function of whether the competing task-irrelevant stimulus was Structured vs . Non-Structured , some interesting differences emerged . Spatial-temporal clustering permutation test on the scalp-level TRFs revealed significant differences between the conditions between 70–180 ms ( p<0 . 05; cluster corrected ) , including both the early and late TRF peaks , at a large number of sensors primarily on the left ( Figure 4a and c ) . Specifically , TRF responses to the to-be-attended speech were enhanced when the task-irrelevant stimulus was Structured vs . Non-Structured . The effect was observed at the scalp-level with opposite polarity in frontal vs . medial sensors , and was localized at the source-level to a single cluster in the left inferior-frontal cortex , that included portions of the insula and orbito-frontal cortex ( Figure 4c; spatial-temporal clustering p=0 . 0058 ) . Top-down attention is an extremely effective process , by which the perceptual and neural representation of task-relevant speech are enhanced at the expense of task-irrelevant stimuli , and speech in particular ( Horton et al . , 2013; Zion Golumbic et al . , 2013b; O'Sullivan et al . , 2015; Fiedler et al . , 2019; Teoh and Lalor , 2019 ) . However , the question still stands: what degree of linguistic processing is applied to task-irrelevant speech ? One prominent position is that attention is required for linguistic processing and therefore speech that is outside the focus of attention is not processed beyond its sensory attributes ( Lachter et al . , 2004; Brodbeck et al . , 2018a; Ding et al . , 2018 ) . However , several lines of evidence suggest that linguistic features of task-irrelevant speech can be processed as well , at least under certain circumstances . For example , task-irrelevant speech is more disruptive to task performance if it is intelligible , as compared to unintelligible noise-vocoded or rotated speech ( Marrone et al . , 2008; Iyer et al . , 2010; Best et al . , 2012; Gallun and Diedesch , 2013; Swaminathan et al . , 2015; Kidd et al . , 2016 ) or a foreign language ( Freyman et al . , 2001; Rhebergen et al . , 2005; Cooke et al . , 2008; Calandruccio et al . , 2010; Francart et al . , 2011 ) . This effect , referred to as informational masking , is often attributed to the detection of familiar acoustic-phonetic features in task-irrelevant speech , that can lead to competition for phonological processing ( 'phonological interference' ) ( Durlach et al . , 2003; Drullman and Bronkhorst , 2004; Kidd et al . , 2008; Shinn-Cunningham , 2008; Rosen et al . , 2013 ) . However , the phenomenon of informational masking alone is insufficient for determining the extent to which task-irrelevant speech is processed beyond identification of phonological units . Other lines of investigation have focused more directly on whether task-irrelevant speech is represented at the semantic level . Findings that individual words from a task-irrelevant source are occasionally detected and recalled , such as one’s own name , ( Cherry , 1953; Wood and Cowan , 1995; Conway et al . , 2001; Röer et al . , 2017b; Röer et al . , 2017a ) , have been taken as evidence that task-irrelevant inputs can be semantically processed , albeit the information may not be consciously available . Along similar lines , a wealth of studies demonstrate the ‘Irrelevant Sound Effect’ ( ISE ) , showing that the semantic content of task-irrelevant input affects performance on a main task , mainly through priming effects and interference with short-term memory ( Lewis , 1970; Bentin et al . , 1995; Surprenant et al . , 1999; Dupoux et al . , 2003; Beaman , 2004; Rivenez et al . , 2006; Beaman et al . , 2007; Rämä et al . , 2012; Aydelott et al . , 2015; Schepman et al . , 2016; Vachon et al . , 2020 ) . However , an important caveat precludes interpreting these findings as clear-cut evidence for semantic processing of task-irrelevant speech: Since these studies primarily involve presentation of arbitrary lists of words ( mostly nouns ) , usually at a relatively slow rate , an alternative explanation is that the ISE is simply a result of occasional shifts of attention toward task-irrelevant stimuli ( Carlyon , 2004; Lachter et al . , 2004 ) . Similarly , the effects of informational masking discussed above can also be attributed to a similar notion of perceptual glimpsing , that is gleaning bits of the task-irrelevant speech in the short ‘gaps’ in the speech that is to-be-attended ( Cooke , 2006; Kidd et al . , 2016; Fogerty et al . , 2018 ) . These claims – that effects of task-irrelevant speech are not due to parallel processing but reflect shifts of attention – are extremely difficult to reject empirically , as they would require insight into the listeners’ internal state of attention , which at present is not easy to operationalize . In attempt to broach the larger question of processing task-irrelevant speech , the current study takes a different approach by focusing not on detection of single words , but on linguistic processes that operate over longer timescales . To this end the stimuli used here , in both the to-be-attended and the task-irrelevant ear , was continuous speech rather than word-lists whose processing requires accumulating and integrating information over time , which is strikingly different than the point-process nature of listening to word-lists ( Fedorenko et al . , 2016 ) . Using continuous speech is also more representative of the type of stimuli encountered naturally in the real world ( Hill and Miller , 2010; Risko et al . , 2016; Matusz et al . , 2019; Shavit-Cohen and Zion Golumbic , 2019 ) . In addition , by employing hierarchical frequency-tagging , we were able to obtain objective and direct indications of which levels of information were detected within task-irrelevant speech . Indeed , using this approach we were able to identify a phrase-level response for Structured task-irrelevant speech , which serves as a positive indication that these stimuli are indeed processed in a manner sufficient for identifying the boundaries of syntactic structures . An important question to ask is whether the phrase-level response observed for task-irrelevant speech can be explained by attention shifts ? Admittedly , in the current design participants could shift their attention between streams in an uncontrolled fashion , allowing them to ‘glimpse’ portions of the task-irrelevant speech , integrate and comprehend ( portions of ) it . Indeed , this is one of the reasons we refrain from referring to the task-irrelevant stream as ‘unattended’: since we have no principled way to empirically observe the internal loci or spread of attention , we chose to focus on its behavioral relevance rather than make assumptions regarding the participants’ attentional-state . Despite the inherent ambiguity regarding the underlying dynamics of attention , the fact that here we observe a phrase-level response for task-irrelevant speech is direct indication that phonetic-acoustic information from this stream was decoded and integrated over time , allowing the formation of long-scale representations for phrasal boundaries . If this is a result of internal ‘glimpsing’ , this would imply either that ( a ) the underlying hierarchical speech structure was detected and used to guide ‘glimpses’ in a rhythmic-manner to points in time that are most informative; or ( b ) that ‘glimpses’ occur irregularly , but that sufficient information is gleaned through them and stored in working-memory to allow the consistent detection phrasal boundaries in task-irrelevant speech . Both of these options imply a sophisticated multiplexed encoding-scheme for successful processing of concurrent speech , that relies on precise temporal control and working-memory storage . Another possibility , of course , is that there is no need for attention-shifts and that the system has sufficient capacity to process task-irrelevant speech in parallel to focusing primarily on the to-be-attended stream . As mentioned above , the current data cannot provide insight into which of these listening-strategies underlies the generation of the observed phrase-level response to task-irrelevant speech . However , we hope that future studies will gain empirical access into the dynamic of listeners’ internal attentional state and help shed light on this pivotal issue . The current study is similar in design to another recent study by Ding et al . , 2018 where Structured frequency-tagged speech was presented as a task-irrelevant stimulus . In contrast to the results reported here , they did not find significant peaks at any linguistic-related frequencies in the neural response to task-irrelevant speech . In attempt to resolve this discrepancy , it is important to note that these two studies differ in an important way – in the listening effort that was required of participants in order to understand the to-be-attended speech . While in the current experiment to-be-attended speech was presented in its natural form , mimicking the listening effort of real-life speech-processing , in the study by Ding et al . , 2018 to-be-attended speech was time-compressed by a factor of 2 . 5 and naturally occurring gaps were removed , making the comprehension task substantially more effortful ( Nourski et al . , 2009; Müller et al . , 2019 ) . Load Theory of Attention proposes that the allocation of processing resources among competing inputs can vary as a function of the perceptual traits and cognitive load imposed by the task ( Lavie et al . , 2004; Murphy et al . , 2017 ) . Accordingly , it is plausible that these divergent results are due to the extreme difference in the perceptual load and listening effort in the two studies . Specifically , if understanding the to-be-attended speech imposes relatively low perceptual and cognitive load , then sufficient resources may be available to additionally process aspects of task-irrelevant speech , but that this might not be the case as the task becomes more difficult and perceptually demanding ( Wild et al . , 2012; Gagné et al . , 2017; Peelle , 2018 ) . More broadly , the comparison between these two studies invites re-framing of the question regarding the type/level of linguistic processing applied to task-irrelevant speech , and propels us to think about this issue not as a yes-or-no dichotomy , but perhaps as a more flexible process that depends on the specific context ( Brodbeck et al . , 2020b ) . The current results provide a non-trivial positive example for processing task-irrelevant speech that is indeed processed beyond its acoustic attributes , in an experimental context that closely emulates the perceptual and cognitive load encountered in real-life ( despite the admitted unnatural nature of the task-irrelevant speech ) . At the same time , they do not imply that this is always the case , as is evident from the diverse results reported in the literature regarding processing task-irrelevant speech , as discussed at length above . Rather , they invite adopting a more flexible perspective of processing bottlenecks within the speech processing system , that takes into consideration the perceptual and cognitive load imposed in a given context , in line with load theory of attention ( Mattys et al . , 2012; Lavie et al . , 2014; Fairnie et al . , 2016; Gagné et al . , 2017; Peelle , 2018 ) . Supporting this perspective , others have also observed that the level of processing applied to task-irrelevant stimuli can be affected by task demands ( Hohlfeld and Sommer , 2005; Pulvermüller et al . , 2008 ) . Moreover , individual differences in attentional abilities , and particularly the ability to process concurrent speech , have been attributed partially to working-memory capacity , a trait associated with the availability of more cognitive resources ( Beaman et al . , 2007; Forster and Lavie , 2008; Naveh-Benjamin et al . , 2014; Lambez et al . , 2020 ) but cf . ( Elliott and Briganti , 2012 ) . As cognitive neuroscience research increasingly moves toward studying speech processing and attention in real-life circumstances , a critical challenge will be to systematically map out the perceptual and cognitive factors that contribute to , or hinder , the ability to glean meaningful information from stimuli that are outside the primary focus of attention . The phrase-level neural response to task-irrelevant Structured speech was localized primarily to two left-lateralized clusters: one in the left anterior fronto-temporal cortex and the other in left posterior-parietal cortex . The fronto-temporal cluster , which included the IFG and insula , is known to play an important role in speech processing ( Dronkers et al . , 2004; Humphries et al . , 2006; Brodbeck et al . , 2018b; Blank and Fedorenko , 2020 ) . The left IFG and insula are particularly associated with linguistic processes that require integration over longer periods of time , such as syntactic structure building and semantic integration of meaning ( Fedorenko et al . , 2016; Matchin et al . , 2017; Schell et al . , 2017 ) , and are also recruited when speech comprehension requires effort , such as for degraded or noise-vocoded speech ( Davis and Johnsrude , 2003; Obleser and Kotz , 2010; Davis et al . , 2011; Hervais-Adelman et al . , 2012 ) . Accordingly , observing a phrase-level response to task-irrelevant speech in these regions is in line with their functional involvement in processing speech under adverse conditions . With regard to the left posterior-parietal cluster , the interpretation for why a phrase-level response is observed there is less straightforward . Although some portions of the parietal cortex are involved in speech processing , these are typically more inferior than the cluster found here ( Hickok and Poeppel , 2007; Smirnov et al . , 2014 ) . However , both the posterior-parietal cortex and inferior frontal gyrus play an important role in verbal working-memory ( Todd and Marois , 2004; Postle et al . , 2006; Linden , 2007; McNab and Klingberg , 2008; Edin et al . , 2009; Østby et al . , 2011; Rottschy et al . , 2012; Gazzaley and Nobre , 2012; Ma et al . , 2012; Meyer et al . , 2014; Meyer et al . , 2015; Yue et al . , 2019; Fedorenko and Blank , 2020 ) . Detecting the phrasal structure of task-irrelevant speech , while focusing primarily on processing the to-be-attended narratives , likely requires substantial working-memory for integrating chunks of information over time . Indeed , attention and working-memory are tightly linked constructs ( McNab and Klingberg , 2008; Gazzaley and Nobre , 2012; Vandierendonck , 2014 ) , and as mentioned above , the ability to control and maintain attention is often associated with individual working-memory capacity ( Cowan et al . , 2005; Beaman et al . , 2007; Forster and Lavie , 2008; Naveh-Benjamin et al . , 2014; Lambez et al . , 2020 ) . Therefore , one possible interpretation for the presence of a phrase-level response to task-irrelevant speech in the left posterior-parietal cortex and inferior frontal regions , is their role in forming and maintaining a representation of task-irrelevant stimuli in working-memory , perhaps as a means for monitoring the environment for potentially important events . Although in the current study we found significant neural response to task-irrelevant speech at the phrase-rate , we did not see peaks at the word- or at the sentence-rate . Regarding the sentence-level response , it is difficult to determine whether the lack of an observable peak indicates that the stimuli were not parsed into sentences , or if this null-result is due to the technical difficulty of obtaining reliable peaks at low-frequencies ( 0 . 5 Hz ) given the 1/f noise-structure of neurophysiological recordings ( Pritchard , 1992; Miller et al . , 2009 ) . Hence , this remains an open question for future studies . Regarding the lack of a word-level response at 2 Hz for Structured task-irrelevant stimuli , this was indeed surprising , since in previous studies using the same stimuli in a single-speaker context we observe a prominent peak at both the word- and the phrase-rate ( Makov et al . , 2017 ) . Although we do not know for sure why the 2 Hz peak is not observed when this speech was presented as task-irrelevant concurrently with another narrative , we can offer some speculations for this null-result: One possibility is that the task-irrelevant speech was indeed parsed into words as well , but that the neural signature of 2 Hz parsing was not observable due to interference from the acoustic contributions at 2 Hz ( see Supplementary Materials and Luo and Ding , 2020 ) . However , another possibility is that the lack of a word-level response for task-irrelevant speech indicates that it does not undergo full lexical analysis . Counter to the linear intuition that syntactic structuring depends on identifying individual lexemes , there is substantial evidence that lexical and syntactic processes are separable and dissociable cognitive processes , that rely on partially different neural substrates ( Friederici and Kotz , 2003; Hagoort , 2003; Humphries et al . , 2006; Nelson et al . , 2017; Schell et al . , 2017; Pylkkänen , 2019; Morgan et al . , 2020 ) . Indeed , a recent frequency-tagging study showed that syntactic phrasal structure can be identified ( generating a phrase-level peak in the neural spectrum ) even in the complete absence of lexical information ( Getz et al . , 2018 ) . Hence , it is possible that when speech is task-irrelevant and does not receive full attention , it is processed only partially , and that although phrasal boundaries are consistently detected , task-irrelevant speech does not undergo full lexical analysis . This matter regarding the depth of lexical processing of task-irrelevant speech , and its interaction with syntactic analysis , remains to be further explored in future research . Besides analyzing the frequency-tagged neural signatures associated with encoding the task-irrelevant stimuli , we also looked at how the neural encoding of to-be-attended speech was affected by the type of task-irrelevant speech it was paired with . In line with previous MEG studies , the speech-tracking response ( estimated using TRFs ) was localized to auditory temporal regions bilaterally and left inferior frontal regions ( Ding and Simon , 2012; Zion Golumbic et al . , 2013a; Puvvada and Simon , 2017 ) . The speech tracking response in auditory regions was similar in both conditions; however , the response in left inferior-frontal cortex was modulated by the type of task-irrelevant speech presented and was enhanced when task-irrelevant speech was Structured vs . when it was Non-Structured . This pattern highlights the nature of the competition for resources triggered by concurrent stimuli . When the task-irrelevant stimulus was Non-Structured , even though it was comprised of individual phonetic-acoustic units , it did not contain meaningful linguistic information and therefore did not require syntactic and semantic resources . However , the Structured task-irrelevant speech poses more of a competition , since it constitutes fully intelligible speech . Indeed , it is well established that intelligible task-irrelevant speech causes more competition and therefore are more distracting than non-intelligible speech ( Rhebergen et al . , 2005; Iyer et al . , 2010; Best et al . , 2012; Gallun and Diedesch , 2013; Carey et al . , 2014; Kilman et al . , 2014; Swaminathan et al . , 2015; Kidd et al . , 2016 ) . A recent EEG study found that responses to both target and distractor speech are enhanced when the distractor was intelligible vs . unintelligible ( Olguin et al . , 2018 ) , although this may depend on the specific type of stimulus used ( Rimmele et al . , 2015 ) . However , in most studies it is difficult to ascertain the level ( s ) of processing where competing between the inputs occurs , and many effects can be explained by variation in the acoustic nature of maskers ( Ding and Simon , 2014 ) . The current study is unique in that all low-level features of Structured and Non-Structured speech stimuli were perfectly controlled , allowing us to demonstrate that interference goes beyond the phonetic-acoustic level and also occurs at higher linguistic levels . The findings that the speech tracking response of the to-be-attended narratives is enhanced when competing with a Structured task-irrelevant speech , specifically left inferior-frontal brain regions , where we also observed tracking of the phrase-structure of task-irrelevant speech , pinpoints the locus of this competition to these dedicated speech-processing regions , above and beyond any sensory-level competition ( Davis et al . , 2011; Brouwer et al . , 2012; Hervais-Adelman et al . , 2012 ) . Specifically , they suggest that the enhanced speech tracking response in IFG reflects the investment of additional listening effort for comprehending the task-relevant speech ( Vandenberghe et al . , 2002; Gagné et al . , 2017; Peelle , 2018 ) . Since the neural response to the to-be-attended speech was modulated by the type of competition it faced , then why was this not mirrored in the current behavioral results as well ? In the current study participants achieved similar accuracy rates on the comprehension questions regardless of whether the natural-narratives were paired with Structured or Non-Structured stimuli in the task-irrelevant ear , and there was no significant correlation between the neural effects and performance . We attribute the lack of a behavioral effect primarily to the insensitivity of the behavioral measures used here , that consisted of asking four multiple-choice questions after each 45 s long narrative . Although numerous previous studies have been able to demonstrate behavioral ‘intrusions’ of the task-irrelevant stimuli on performance of an attended-task , these have been shown using more constrained experimental paradigms , that have the advantage of probing behavior at a finer scale , but are substantially less ecological ( e . g . memory-recall for short lists of words or priming effects; Tun et al . , 2002; Dupoux et al . , 2003; Rivenez et al . , 2006; Rivenez et al . , 2008; Carey et al . , 2014; Aydelott et al . , 2015 ) . In moving toward studying speech processing and attention under more ecological circumstances , using natural continuous speech , we face an experimental challenge of obtaining sufficiently sensitive behavior measures without disrupting listening with an ongoing task ( e . g . target detection ) or encroaching too much on working-memory . This is a challenge shared by many previous studies similar to ours , and is one of the main motivations for turning directly to the brain and studying neural activity during uninterrupted listening to continuous speech , rather than relying on sparse behavioral indications ( Ding et al . , 2016; Makov et al . , 2017; Brodbeck et al . , 2018a; Broderick et al . , 2018; Broderick et al . , 2019; Donhauser and Baillet , 2020 ) . The current study contributes to ongoing efforts to understand how the brain deals with the abundance of auditory inputs in our environment . Our results indicate that even though top-down attention effectively enables listeners to focus on a particular task-relevant source of input ( speech in this case ) , this prioritization can be affected by the nature of task-irrelevant sounds . Specifically , we find that when the latter constitutes meaningful speech , left fronto-temporal speech-processing regions are engaged in processing both stimuli , potentially leading to competition for resources and more effortful listening . Additional brain regions , such as the PPC , are also engaged in representing some aspects of the linguistic structure of task-irrelevant speech , which we interpret as maintaining a representation of what goes on in the ‘rest of the environment’ , in case something important arises . Importantly , similar interactions between the structure of task-irrelevant sounds and responses to the to-be-attended sounds have been previously demonstrated for non-verbal stimuli as well ( Makov and Zion Golumbic , 2020 ) . Together , this highlights the fact that attentional selection is not an all-or-none processes , but rather is a dynamic process of balancing the resources allocated to competing input , which is highly affected by the specific perceptual , cognitive and environmental aspects of a given task .	We are all familiar with the difficulty of trying to pay attention to a person speaking in a noisy environment , something often known as the ‘cocktail party problem’ . This can be especially challenging when the background noise we are trying to filter out is another conversation that we can understand . In order to avoid being distracted in these kinds of situation , we need selective attention , the cognitive process that allows us to attend to one stimulus and to ignore other irrelevant sensory information . How the brain processes the sounds in our environment and prioritizes them is still not clear . One of the central questions is whether we can take in information from several speakers at the same time or whether we can only understand speech from one speaker at a time . Neuroimaging techniques can shed light on this matter by measuring brain activity while participants listen to competing speech stimuli , helping researchers understand how this information is processed by the brain . Now , Har-Shai Yahav and Zion Golumbic measured the brain activity of 30 participants as they listened to two speech streams in their native language , Hebrew . They heard each speech in a different ear and tried to focus their attention on only one of the speakers . Participants always had to attend to natural speech , while the sound they had to ignore could be either natural speech or unintelligible syllable sequences . The activity of the brain was registered using magnetoencephalography , a non-invasive technique that measures the magnetic fields generated by the electrical activity of neurons in the brain . The results showed that unattended speech activated brain areas related to both hearing and language . Thus , unattended speech was processed not only at the acoustic level ( as any other type of sound would be ) , but also at the linguistic level . In addition , the brain response to the attended speech in brain regions related to language was stronger when the competing sound was natural speech compared to random syllables . This suggests that the two speech inputs compete for the same processing resources , which may explain why we find it difficult to stay focused in a conversation when there are other people talking in the background . This study contributes to our understanding on how the brain processes multiple auditory inputs at once . In addition , it highlights the fact that selective attention is a dynamic process of balancing the cognitive resources allocated to competing information rather than an all-or-none process . A potential application of these findings could be the design of smart devices to help individuals focus their attention in noisy environments .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "neuroscience" ]	2021	Linguistic processing of task-irrelevant speech at a cocktail party
Cancer cells usually exhibit aberrant cell signaling and metabolic reprogramming . However , mechanisms of crosstalk between these processes remain elusive . Here , we show that in an in vivo tumor model expressing oncogenic Drosophila Homeodomain-interacting protein kinase ( Hipk ) , tumor cells display elevated aerobic glycolysis . Mechanistically , elevated Hipk drives transcriptional upregulation of Drosophila Myc ( dMyc; MYC in vertebrates ) likely through convergence of multiple perturbed signaling cascades . dMyc induces robust expression of pfk2 ( encoding 6-Phosphofructo-2-kinase/fructose-2 , 6-bisphosphatase; PFKFB in vertebrates ) among other glycolytic genes . Pfk2 catalyzes the synthesis of fructose-2 , 6-bisphosphate , which acts as a potent allosteric activator of Phosphofructokinase ( Pfk ) and thus stimulates glycolysis . Pfk2 and Pfk in turn are required to sustain dMyc protein accumulation post-transcriptionally , establishing a positive feedback loop . Disruption of the loop abrogates tumorous growth . Together , our study demonstrates a reciprocal stimulation of Myc and aerobic glycolysis and identifies the Pfk2-Pfk governed committed step of glycolysis as a metabolic vulnerability during tumorigenesis . In the 1920s , Otto Warburg first discovered that cancer cells vigorously take up glucose and preferentially produce lactate even in the presence of oxygen , a phenomenon now widely termed the Warburg effect or aerobic glycolysis ( Warburg et al . , 1927 ) . Despite his pioneering work , the Warburg effect was largely disregarded for the subsequent decades ( Liberti and Locasale , 2016; Koppenol et al . , 2011 ) . After the discoveries of oncogenes and tumor suppressor genes , cancers are generally considered as genetic diseases rather than metabolic ones ( Wishart , 2015; Seyfried et al . , 2014 ) . Not until the 1980s did the revisiting of the Warburg effect in connection with oncogenes spark substantial research in cancer metabolism ( Koppenol et al . , 2011 ) , laying the foundation for 2-deoxy-2- ( 18F ) fluoro-D-glucose ( 18F-FDG ) positron emission tomography ( PET ) in clinical cancer diagnosis ( Vander Heiden et al . , 2009; Ben-Haim and Ell , 2008 ) , recognition of metabolic reprogramming as a hallmark of cancer ( Hanahan and Weinberg , 2011; Ward and Thompson , 2012 ) and development of anti-cancer agents targeting aerobic glycolysis ( Ganapathy-Kanniappan and Geschwind , 2013; Pelicano et al . , 2006; Granchi and Minutolo , 2012 ) . How the Warburg effect arises and contributes to tumor progression have always been the center in the field of cancer metabolism ( Lu et al . , 2015 ) . Warburg hypothesized that mitochondrial impairment is the cause of aerobic glycolysis and cancer ( Warburg , 1956 ) , which has sparked much controversy among scientists ( Senyilmaz and Teleman , 2015 ) . Contrary to his idea , the current , widely-accepted view is that oncogenic drivers such as RAS , MYC , hypoxia-inducible factors ( HIFs ) and steroid receptor coactivators ( SRCs ) promote cancer cell proliferation and directly stimulate aerobic glycolysis through regulating transcriptional expression or catalytic activities of metabolic enzymes ( Koppenol et al . , 2011 ) . Several explanations for the Warburg effect have been put forward , including rapid adenosine triphosphate ( ATP ) production and de novo biosynthesis of macromolecules ( Liberti and Locasale , 2016 ) . Recently , Warburg effect functions are being re-evaluated as a growing body of evidence shows that metabolic rewiring in cancers impacts cell signaling and epigenetics ( Liberti and Locasale , 2016; Lu and Thompson , 2012 ) . Drosophila has proven to be a powerful genetic model organism for studying tumorigenesis in vivo largely due to high conservation of genes and signaling cascades between human and flies and reduced genetic redundancy ( Gonzalez , 2013; Herranz and Cohen , 2017 ) ( Supplementary file 1 ) . In a recent study , we showed that elevation of Drosophila Hipk causes an in vivo tumor model characterized by tissue overgrowth , loss of epithelial integrity and invasion-like behaviors ( Blaquiere et al . , 2018 ) . The tumorigenic roles of Drosophila Hipk seem to be conserved in mammals as the four members of the HIPK family ( HIPK1-4 ) are also implicated in certain cancers ( Reviewed in Blaquiere and Verheyen , 2017 ) . For example , HIPK1 is highly expressed in breast cancer cell lines , colorectal cancer samples and oncogenically-transformed mouse embryo fibroblasts ( Kondo et al . , 2003; Rey et al . , 2013 ) . Also , HIPK2 is elevated in certain cancers including cervical cancers , pilocytic astrocytomas , colorectal cancer cells and in other proliferative diseases , such as thyroid follicular hyperplasia ( Al-Beiti and Lu , 2008; Cheng et al . , 2012; Yu et al . , 2009; D'Orazi et al . , 2006; Saul and Schmitz , 2013; Deshmukh et al . , 2008; Lavra et al . , 2011 ) . To gain a better understanding of cancer metabolism , we set out to use the fly Hipk tumor model to investigate whether and how cellular metabolism is altered in tumor cells . We find that Hipk-induced tumorous growth is accompanied by elevated aerobic glycolysis . Furthermore , we identify novel feedback mechanisms leading to prolonged dMyc expression and hence tumorigenesis . Our study reveals potential metabolic vulnerabilities that could be exploited to suppress tumor growth . Larval imaginal epithelia such as wing and eye-antennal imaginal discs , which give rise to wing and eye structures in adult flies , respectively , are extensively used as tumor models to study human carcinomas ( Herranz et al . , 2016 ) . To generate the Hipk tumor model , we used the Gal4-UAS system ( Brand and Perrimon , 1993 ) to induce overexpression of hipk in larval wing discs ( full genotype: dpp-Gal4>UAS-RFP + UAS-hipk , abbreviated as dpp>RFP + hipk ) ( Figure 1c–d ) . Fluorescent proteins , for example red fluorescent proteins ( RFP ) or green fluorescent proteins ( GFP ) , were co-expressed to label the transgene-expressing cells . As previously reported ( Blaquiere et al . , 2018 ) , in contrast to control discs ( dpp>RFP ) ( Figure 1a–b ) , tumorous growth and severe tissue distortions were evident in hipk-expressing discs ( dpp>RFP + hipk ) ( Figure 1c–d ) . To ask if elevated Hipk alters cellular metabolism , we first used a fluorescently labeled glucose analog , 2-deoxy-2-[ ( 7-nitro-2 , 1 , 3-benzoxadiazol-4-yl ) amino]-D-glucose ( 2-NBDG ) ( Figure 1j ) . Like D-glucose , 2-NBDG is transported into cells through glucose transporters and phosphorylated by hexokinases ( Yoshioka et al . , 1996; O'Neil et al . , 2005 ) . However , due to the presence of the fluorescent amino group at the C-2 position , 2-NBDG-6-phosphate cannot proceed further through glycolysis and is consequently trapped within the cells , rendering it a probe for monitoring glucose uptake ( O'Neil et al . , 2005 ) . 2-NBDG is often used in Drosophila to measure glucose uptake . For example , overexpression of glut1 ( encoding Glucose transporter 1 ) is sufficient to promote the accumulation in 2-NBDG ( Niccoli et al . , 2016 ) . Increased import of 2-NBDG is also found in dMyc-expressing cells ( de la Cova et al . , 2014 ) and RasV12scrib−/− tumor cells ( Katheder et al . , 2017 ) , indicating that such cells acquire enhanced glucose metabolism . Similarly , we found that cells with elevated Hipk ( RFP positive ) in hipk-expressing wing discs ( dpp>RFP + hipk ) exhibited statistically significant 2-NBDG accumulation when compared with the neighboring wild-type cells ( RFP negative ) ( Figure 1c–d and i ) or cells ( either RFP positive or negative ) in control discs ( dpp>RFP ) ( Figure 1a–b ) . A comparable phenomenon was seen in larval eye-antennal discs ( Figure 1—figure supplement 1a–b ) . Furthermore , we detected a mild upregulation of glut1 ( 1 . 6-fold ) in hipk-expressing discs by quantitative real-time PCR ( qRT-PCR ) ( Figure 2a ) . Knockdown of glut1 using RNA interference ( RNAi ) moderately reduced 2-NBDG incorporation in Hipk tumor cells ( Figure 1—figure supplement 1c–d ) . The knockdown efficiency of glut1-RNAi was tested using a ubiquitous Gal4 driver , act5c-Gal4 ( Figure 1—figure supplement 2a ) . Together , these results suggest that elevated Hipk facilitates the import of glucose into cells at least in part through augmenting Glut1 expression . Next , we evaluated the intracellular glucose levels using a FRET ( Förster resonance energy transfer ) -based glucose sensor , which is composed of a glucose-binding domain ( GBD ) fused with cyan fluorescent protein ( CFP ) and yellow fluorescent protein ( YFP ) ( Volkenhoff et al . , 2018; Fehr et al . , 2003 ) ( Figure 1l ) . The binding of glucose to GBD induces conformational changes , leading to increased FRET efficiency ( the ratio of FRET to CFP ) . Intriguingly , a lower FRET/CFP ratio indicative of reduced glucose concentration was observed in hipk-expressing cells ( Figure 1g–h and k ) when compared with control discs ( Figure 1e–f and k ) . This implies that despite enhanced glucose uptake , glucose is substantially utilized in the rapidly proliferating Hipk tumor cells . Glycolysis is a catabolic process that breaks down glucose into pyruvate via multiple enzymatic steps . To ask if glycolysis is upregulated , we examined the expression profile of glycolytic genes by qRT-PCR and changes greater than 1 . 5-fold were considered significant . We isolated RNA from whole wing discs expressing either GFP ( control ) or hipk under the dpp-Gal4 driver control . qRT-PCR analyses revealed that genes encoding Hexokinases ( Hex-A and Hex-C , 2 . 2-fold and 4 . 3-fold respectively ) , Phosphoglucose isomerase ( Pgi , 2 . 4-fold ) , Phosphofructokinase 2 ( Pfk2 , also known as Pfrx , 2 . 7-fold ) and Lactate dehydrogenase ( Ldh , also known as Impl3 , 3 . 4-fold ) are significantly upregulated in hipk-expressing discs ( Figure 2a–b ) . Expression of other glycolytic genes , on the other hand , remained relatively unchanged . As the Dpp domain comprises 10–20% of the wing disc , the glycolytic gene upregulation within hipk-expressing cells is likely understated . In light of such a limitation , we used a GFP-based enhancer trap to monitor Ldh transcription in vivo ( Ldh-GFP ) ( Quiñones-Coello et al . , 2007 ) . In control wing discs , little Ldh-GFP was detected ( Figure 2c ) , indicating that Ldh is minimally expressed under physiological conditions as previously described ( Wang et al . , 2016 ) . On the contrary , we observed robust Ldh-GFP expression in hipk-expressing wing discs ( Figure 2d ) , confirming that elevated Hipk induces Ldh expression . Hexokinases phosphorylate glucose to form glucose-6-phosphate ( G6P ) , which is the first essentially irreversible step in glycolysis ( Figure 2b ) . Pgi catalyzes the reversible conversion of G6P and fructose-6-phosphate . Pfk2 is a bifunctional enzyme that synthesizes fructose-2 , 6-bisphosphate ( F2 , 6-BP ) . F2 , 6-BP is a potent allosteric activator of Pfk , which governs the committed , second irreversible step in glycolysis . Ldh , similar to LDHA in vertebrates ( Rechsteiner , 1970 ) , favors the reduction of pyruvate to lactate , regenerating nicotinamide adenine dinucleotide ( NAD+ ) such that glycolysis continues unabated ( Valvona et al . , 2016 ) . The robust upregulation of these glycolytic enzymes in hipk-expressing discs is in agreement with previous data implying that glucose consumption is stimulated . Taken together , the results support the hypothesis that Hipk tumor cells exhibited elevated glycolytic activities , resembling the Warburg effect seen in cancers . To elucidate the underlying mechanisms for the metabolic alterations in the Hipk tumors , we examined gene expression of dMyc and Sima ( HIF1-α in vertebrates ) since these two transcription factors are well-known inducers of glycolysis in cancer cells ( Miller et al . , 2012a; Marín-Hernández et al . , 2009 ) . qRT-PCR analyses showed that dMyc was upregulated in hipk-expressing discs whereas sima expression remained unchanged ( Figure 3a ) . Larval wing discs are mainly composed of three regions known as the notum , hinge and pouch , corresponding to the adult structures they give rise to ( Figure 3b ) . Using a dMyc-lacZ enhancer trap ( encoding β-galactosidase ( β-gal ) ) to monitor the transcriptional control of dMyc in vivo , we observed that dMyc was expressed at high levels in the pouch region of control discs ( Figure 3c ) . In the hinge and notum areas of control discs , dMyc was minimally and weakly expressed , respectively ( Figure 3c ) . In hipk-expressing discs , we found intense β-galactosidase staining , confirming that Hipk cell-autonomously drives transcriptional upregulation of dMyc . Interestingly , the increase in the β-gal signal intensity appeared to be the largest in the wing hinge ( Figure 3d , arrowhead ) where the tumor cells proliferated most rapidly , while the intensity faded when moving away from the hinge region ( Figure 3d ) . Immunofluorescent staining using antibodies against dMyc confirmed the accumulation of dMyc proteins in Hipk tumor cells especially in the wing hinge ( Figure 3e–f ) , which positively correlates with the marked increase in dMyc mRNA levels in the same region ( Figure 3d ) . dMyc proteins appeared to not accumulate in hipk-expressing cells in the wing pouch ( Figure 3f ) , possibly due to weak induction in dMyc transcription ( Figure 3d ) . The variability in Hipk-induced dMyc accumulation suggests a potential region-specificity in dMyc upregulation . To assess this , we generated random flip-out clones of hipk-expressing cells . The clones were marked by RFP . We observed that hipk-expressing clones located in the dorsal hinge ( Figure 4c , arrowhead , Figure 4—figure supplement 1a ) , the lateral hinge ( Figure 4—figure supplement 1b ) as well as the epithelium on the peripodial side exhibited marked dMyc accumulation ( Figure 4—figure supplement 1c ) . On the contrary , when hipk-expressing clones were located in the wing pouch , dMyc levels appeared unchanged ( Figure 4—figure supplement 1d ) . Altogether , our data imply that elevated Hipk drives dMyc upregulation in a region-specific manner . We further investigated the molecular mechanisms underlying the upregulation of dMyc in Hipk tumor cells . Earlier studies have demonstrated that Hipk is a versatile regulator of numerous signaling pathways . For example , Hipk stimulates the transcriptional activity of Yorkie ( Yki , a Drosophila orthologue of YAP/TAZ ) , the effector of the Hippo tumor suppressor pathway ( Chen and Verheyen , 2012; Poon et al . , 2012 ) . Hipk also functions as a positive regulator of both Wnt/Wingless ( Wg ) and Hedgehog signaling by stabilizing Armadillo ( β-catenin in vertebrates ) and Cubitus interruptus ( Ci; Gli in vertebrates ) , respectively ( Lee et al . , 2009a; Swarup and Verheyen , 2011 ) . In addition , Hipk has been found to antagonize Groucho , a global co-repressor , to promote Notch signal transduction ( Lee et al . , 2009b ) . If Hipk-induced dMyc upregulation is through deregulation of multiple signaling pathways , we would expect to see dMyc accumulation when the signaling pathways are perturbed . We first focused on the dorsal hinge area since ectopic dMyc expression and tumor growth are most striking in such region in hipk-expressing discs . Flip-out clones expressing a constitutively active form of Yki ( Yki-S168A ) to inactivate the Hippo pathway were usually round and most of them in the dorsal hinge displayed dMyc upregulation ( Figure 4d , arrowheads ) . Wing discs with flip-out clones expressing an active form of Arm ( Arm-S10 ) featured an expansion in the hinge area and some dMyc accumulated at the interface between the clone and the neighboring wild-type cells ( Figure 4e , arrowheads ) but not at the center of the clone . Flip-out clones expressing Ci to activate Hedgehog signaling in the hinge region were round or irregular in shape and exhibited elevated levels of dMyc ( Figure 4f , arrowheads ) . Consistent with previous findings ( Djiane et al . , 2013 ) , expression of activated Notch appeared to induce dMyc accumulation in both cell and non-cell autonomous fashions in the hinge area likely involving a signaling relay ( Figure 4g ) . Therefore , individual perturbation of signaling pathways that have been previously shown to be regulated by Hipk was sufficient to trigger ectopic dMyc expression in the hinge region . In the wing pouch region , activation of Yki failed to cause any dMyc upregulation ( Figure 4—figure supplement 1e ) . This is in contrast to previous findings that overexpression of Yki increased dMyc protein levels in both hinge and pouch regions ( Ziosi et al . , 2010 ) . Such discrepancy could be due to the different expression levels of the UAS-Yki constructs . Flip-out clones expressing Arm-S10 or activated Notch displayed reduced dMyc expression ( Figure 4—figure supplement 1f and h ) , which is consistent with earlier studies showing that Wg and Notch signaling negatively regulate dMyc expression in the pouch region ( Duman-Scheel et al . , 2004; Herranz et al . , 2008 ) . Clones with Ci overexpression were few and tiny in the wing pouch and the dMyc levels in the clones seemed comparable to those in the neighboring wild-type cells ( Figure 4—figure supplement 1g ) . Our data suggest that the region specificity in dMyc upregulation applies to not only Hipk tumor cells , but also cells with other tumorigenic stimuli . In summary , our data show that in the wing hinge region where Hipk tumor growth is most prominent , dMyc protein markedly accumulates . Individual perturbations in signaling pathways that have been shown to be regulated by Hipk also result in ectopic dMyc expression in the same region . Hence , it is likely that in the rapidly proliferating Hipk tumor cells , aberrant signals propagate and converge at the transcriptional control of dMyc . In other words , the modulation of dMyc levels by Hipk seems to be a cumulative effect of alterations in multiple signaling outputs . In addition to growth control , dMyc/MYC plays a conserved role in upregulating glycolysis in vertebrates and Drosophila during normal development and in cancers ( de la Cova et al . , 2014; Gallant , 2013 ) . As expected and previously reported ( de la Cova et al . , 2014 ) , overexpression of dMyc caused marked 2-NBDG uptake ( Figure 5a and d ) and promoted glycolytic gene expression ( Figure 5e ) . More importantly , knockdown of dMyc by RNAi in hipk-expressing discs reversed the 2-NBDG accumulation ( Figure 5b–d ) and moderately suppressed glycolytic gene upregulation ( Figure 5e ) , suggesting that dMyc is essential for the enhanced glycolysis in Hipk tumor cells . Additionally , knockdown of dMyc significantly suppressed the tumorous growth and tissue distortions in hipk-expressing discs , as seen by the decrease in GFP positive cells ( Figure 5f–g ) . Using an independent RNAi construct targeting dMyc ( dMyc-RNAi #2 ) , we observed a similar effect upon silencing of dMyc ( Figure 5h ) . dMyc immunofluorescent staining showed the efficient knockdown when dMyc-RNAi constructs were expressed ( Figure 5g–i ) . Wing discs with dMyc knockdown alone appeared wild-type ( Figure 5i ) . Altogether , our results imply that dMyc-induced glycolysis may be a key driver for Hipk tumor growth . Intriguingly , overexpression of dMyc alone ( Figure 5a ) , despite the enhanced glycolysis , failed to reproduce the tumorous phenotypes observed in hipk-expressing discs ( Figure 5f ) . Thus , while upregulation of dMyc is indispensable for Hipk-induced glycolysis and tumor growth , additional mechanisms are involved in the Hipk tumors to trigger the tumorigenic process . Earlier studies have shown that flies incompetent at glycolysis displayed reduced viability as deletion or ubiquitous knockdown of the glycolytic genes pfk , ald , pgk , pglym78 , eno or pyk resulted in lethality ( Volkenhoff et al . , 2015; Gerber et al . , 2006; Miller et al . , 2012b; Tennessen et al . , 2011 ) . Likewise , we observed that flies ubiquitously depleted of pgk ( act5c>pgk-RNAi ) failed to reach the larval stage . Flies lacking pfk ( act5c>pfk-RNAi ) or pyk ( act5c>pyk-RNAi ) reached the larval stage but could not pupariate . Flies with pfk2 knockdown ( act5c>pfk2-RNAi ) died at the early pupal stage but did not differentiate further to become late pupae . Most flies lacking glut1 ( act5c>glut1-RNAi ) or pgi ( act5c>pgi-RNAi ) succeeded in emerging as adults . The timing of lethality following knockdown targeting glycolytic genes could presumably be influenced by maternal contribution , knockdown efficiency or the importance of the glycolytic step which the particular enzyme controls . To confirm the regulatory roles of Pfk and Pfk2 in glycolytic flux , we measured the relative pyruvate content in the knockdown larvae . Consistent with previous work ( Li et al . , 2018 ) , knockdown of pfk significantly decreased pyruvate levels ( Figure 6—figure supplement 1 ) . Similarly , knockdown of pfk2 induced an approximately 60% drop in the pyruvate content ( Figure 6—figure supplement 1 ) . The knockdown efficiencies of all RNAi constructs used in this study to silence glycolytic genes were evaluated by qRT-PCR analyses ( Figure 1—figure supplement 2 ) . To examine the roles of glycolytic activation in Hipk tumor cells , we used RNAi to deplete key glycolytic genes . Knockdown of hex-A or hex-C had negligible effects on Hipk tumor growth; as did hex-A and hex-C double knockdown ( Figure 6—figure supplement 2 ) . Similarly , knockdown of glut1 or pgi failed to block Hipk-induced tumorigenesis ( Figure 6—figure supplement 3a–b ) . Knockdown of pfk2 , on the contrary , markedly prevented tumor growth and restored tissue architecture in hipk-expressing discs ( Figure 6b ) , suggesting that the committed step of glycolysis determines Hipk tumorigenesis . Indeed , the effect of pfk knockdown was comparable to that of pfk2 knockdown ( Figure 6c ) , even though we did not detect significant pfk upregulation ( <1 . 5 fold ) in hipk-expressing discs ( Figure 2a ) . Knockdown of pgk , pyk or Ldh appeared to slightly restrict Hipk tumor growth but the tissue morphology remained distorted ( Figure 6d , Figure 6—figure supplement 3c–d ) . We also examined dMyc expression in Hipk tumor cells when glycolysis was inhibited . Intriguingly , dMyc accumulation was largely abolished in hipk-expressing discs when pfk2 or pfk was knocked down ( Figure 6a–c ) . In the absence of hipk overexpression , neither knockdown of pfk nor pfk2 affected the normal dMyc expression pattern nor led to defects in tissue growth ( Figure 6—figure supplement 4 ) . Knockdown of glut1 , pgi , pgk , pyk or Ldh in Hipk tumor cells did not rescue dMyc upregulation ( Figure 6d , Figure 6—figure supplement 3 ) . Together , our results suggest that dMyc accumulation and the associated Hipk tumor growth are most sensitive to the committed step of glycolysis governed by Pfk2-Pfk . Using the dMyc-lacZ reporter , we found that dMyc transcriptional activation remained robust when pfk2 or pfk was depleted in Hipk tumor cells even though the tumor growth was reduced and dMyc protein levels were reduced ( Figure 6e–g ) . Thus , our data indicate that Pfk2 and Pfk are required to sustain ectopic dMyc expression specifically in Hipk tumor cells through post-transcriptional regulation . The control of Hipk tumor growth by Pfk2 or Pfk seemed not to be a consequence of altered Hipk protein stability or functions because of the following reasons . First , we did not detect a noticeable change in Hipk protein levels in discs depleted of pfk2 or pfk when compared with discs expressing hipk alone ( Figure 6e–g ) , indicating that blocking glycolysis does not trigger Hipk protein degradation . Second , the robust stimulation of the dMyc-lacZ reporter in hipk-expressing discs with pfk2 or pfk knockdown suggests that Hipk activities remain intact . Indeed , when we examined other signaling targets downstream of Hipk as readouts for Hipk activity , we found that Hipk-mediated stabilization of Armadillo ( Wnt/Wg signaling ) , and induction of Cyclin E or expanded-lacZ ( Hippo signaling ) remained unaffected by knockdown of pfk2 or pfk in a Hipk overexpression background , indicating Hipk is still functional ( Figure 6—figure supplement 5 ) . Altogether , our data suggest that while dMyc upregulation promotes glycolytic activities , the upregulation of Pfk2 , along with subsequent F2 , 6-BP biosynthesis and Pfk activation , is required to foster the accumulation of dMyc proteins . In other words , a positive feedback loop between dMyc and aerobic glycolysis is formed , coordinating growth signals with metabolic states to reinforce tumor progression . Multiple approaches and tools have been developed to measure glycolysis including measurement of the extracellular acidification rate ( ECAR ) , fluorescent probes , biosensors , fluorescent/colorimetric assays and mass spectrometry ( Bittner et al . , 2010; TeSlaa and Teitell , 2014; Tennessen et al . , 2014 ) . In the Hipk tumor model , the tumor cells are surrounded by wild-type cells . Our study primarily used fluorescence imaging such that we could examine changes in metabolite levels , gene and protein expression between tumor cells and the adjacent wild-type cells . Our work delineates the causes and significance of metabolic changes in tumor cells . Drosophila tumor models frequently acquire metabolic changes , especially the Warburg effect . For instance , epidermal growth factor receptor ( EGFR ) -driven tumors , tumors with activated platelet-derived growth factor ( PDGF ) /vascular endothelial growth factor ( VEGF ) receptor ( Pvr ) and tumors with polarity loss ( such as scrib or dlg mutant tumors ) feature robust upregulation of Ldh among other glycolytic genes ( Wang et al . , 2016; Eichenlaub et al . , 2018; Bunker et al . , 2015 ) . Depletion of Ldh was shown to reduce growth of EGFR-driven tumors but not tumors with polarity loss ( Eichenlaub et al . , 2018; Bunker et al . , 2015 ) . In RasV12scrib−/− tumors , elevated glucose uptake was evident , but its significance was not evaluated ( Katheder et al . , 2017 ) . Similar to the previously described models , Hipk tumor cells exhibited elevated glucose metabolism ( Figures 1–2 ) . Metabolic reprogramming was driven by dMyc upregulation ( Figures 3–5 ) . Although the transcript levels of another glycolytic inducer sima remained unchanged ( Figure 3a ) , we could not eliminate the possibility that Sima protein levels are altered in the tumor cells . Thus , whether Sima is involved in the tumor growth warrants further studies . Genetical inhibition of Pfk or Pfk2 was sufficient to block Hipk-induced tumorigenesis whereas depletion of Pgk , Pyk or Ldh at most slightly reduced tumor growth ( Figure 6; Figure 6—figure supplement 3 ) . Therefore , our data suggest that targeted but not generic inhibition of glycolysis is required to abrogate tumor growth . Although MYC functions in cancer metabolism are widely recognized , MYC is generally considered ‘undruggable’ largely due to its nuclear localization , lack of an enzymatic ‘active site’ , and indispensable roles during normal development ( Soucek and Evan , 2010 ) . Hence , limited therapeutic strategies have been developed , and identification of the ‘druggable’ regulatory proteins of MYC becomes critical . Our study reveals two modes of dysregulation of endogenous dMyc specifically during Hipk-induced tumorigenesis . The first mode is transcriptional stimulation of dMyc likely as a consequence of convergence of multiple signaling outputs ( Figures 3–4 ) . In our previous study , we found that genetically targeting individual signaling cascades failed to restrain Hipk tumor growth ( Blaquiere et al . , 2018 ) . This is likely due to the redundancy of signaling cascades in inducing dMyc upregulation . It is also interesting to note that dMyc upregulation and the associated tumor growth are most striking in the wing hinge ( Figures 3–4 ) . The pouch region , however , seems more refractory to such alterations . We also observed the lowest intracellular glucose levels ( Figure 1g–h ) and robust Ldh upregulation ( Figure 2d ) in the hinge area . dMyc-dependent glucose uptake ( Figure 5b–c ) , on the other hand , was enhanced in both hinge and pouch regions even though dMyc protein buildup is undetectable in the wing pouch , suggesting that glucose uptake is particularly sensitive to changes in dMyc levels . Our data therefore imply that the Warburg effect and the tumor growth phenotype may be linked to the dMyc levels being induced . The region-specific susceptibility to tumorigenic stimuli has been previously described and the hinge region was coined a ‘tumor hotspot’ because of its unique epithelial cell architectures and high endogenous JAK/STAT signaling ( Tamori et al . , 2016 ) . It is possible that such features in the hinge region also contribute to the sensitivity of dMyc upregulation by elevated Hipk and other signaling pathways . Further studies are required to verify this proposition . The second mode of dMyc regulation is the metabolic control of dMyc protein levels by aerobic glycolysis . Specifically , we found that the rate-limiting enzymes Pfk2 and Pfk are required to perpetuate dMyc buildup in tumor cells in a post-transcriptional manner ( Figure 6 ) . Similar to the effects on tumor growth , while pfk2/pfk knockdown prevented dMyc accumulation , knockdown of other glycolytic enzymes had little effect on sustaining dMyc accumulation . Such a disparity could possibly be explained by the potency of the glycolytic enzymes in controlling glycolytic flux . While Pfk2 and Pfk govern the rate-limiting , committed step of glycolysis , RNAi targeting other glycolytic genes may fail to render the corresponding steps rate-limiting , especially in tumors with elevated aerobic glycolysis . In other words , knockdown of other glycolytic enzymes may not reach the threshold that would restrict glycolytic flux as potently as pfk2/pfk knockdown . Given that Pfk2 is mis-expressed in tumor cells but not in normal cells ( Figure 2a ) and that dMyc accumulation is most sensitive to the Pfk2-Pfk mediated committed step of glycolysis specifically in tumor cells ( Figure 6 ) , targeting Pfk2 may be a favorable , selective metabolic strategy in the treatment of cancers , especially those displaying ectopic MYC expression . The human homologs of fly Pfk and Pfk2 are PFK and PFK-2/FBPase ( PFKFB ) , respectively . The functions of mammalian PFK and PFKFB in controlling glycolytic flux are well-defined . The allosteric regulation of PFK by F2 , 6-BP seems to be conserved among mammals and insects as F2 , 6-BP is able to activate insect Pfk , likely due to the conserved amino acid residues for F2 , 6-BP binding ( Nunes et al . , 2016 ) . This suggests that Pfk2 can induce Pfk activation and hence glycolytic flux through biosynthesizing F2 , 6-BP . A previous report shows that flies lacking pfk2 exhibited levels of circulating glucose and reduced sugar tolerance , providing a functional link between Pfk2 and glucose metabolism ( Havula et al . , 2013 ) . We provide evidence that both pfk2 and pfk knockdown larvae exhibited significant decreases in pyruvate contents , confirming a conserved role of Pfk2 in regulating glycolytic flux . Recently , the roles of PFK and PFKFB in the regulation of transcription factors or cofactors have drawn considerable attention . For example , vertebrate PFK was shown to physically interact with TEAD factors to stimulate YAP/TAZ activity , thus promoting proliferation and malignancy in cancer cells ( Enzo et al . , 2015 ) . PFKFB4 ( an isoform of PFK2 ) phosphorylates and activates oncogenic SRC-3 to promote breast cancers through stimulating purine synthesis ( Dasgupta et al . , 2018 ) . These studies point to the fact that PFK and PFKFB functions are not restricted to metabolic regulation . Thus , it is tempting to speculate that Pfk2/Pfk may sustain ectopic dMyc through a direct interaction with or even phosphorylation of dMyc . That being said , as knockdown of Pfk2 or Pfk had no effects on dMyc protein levels under normal conditions ( Figure 6—figure supplement 4 ) , we tend to believe that a tumorigenic environment with active aerobic glycolysis is necessary for the metabolic control of dMyc in particular through the committed step governed by Pfk2/Pfk . In summary , we and others show that glycolytic enzymes Pfk and Pfk2 , likely through their core biochemical role in stimulating glycolytic flux , glycolysis-independent actions or both , act as key players in coupling metabolic demands with growth signals to achieve cancer progression . Drosophila melanogaster flies were raised on standard cornmeal-molasses food . Stocks were kept at 25°C and crosses at 29°C unless otherwise indicated . Three Gal4 fly lines , ( 1 ) dpp-Gal4/TM6B , ( 2 ) hh-Gal4/TM6B and ( 3 ) act5c-Gal4/TM6B ( BL #3954 ) were used to induce transgene expression . To generate heat-shock inducible actin flip-out clones , ( 4 ) yw , hsFlp[122];;act>CD2>Gal4 , UAS-RFP/TM6B was used . To mark the transgene-expressing cells , ( 5 ) UAS-GFP ( BL #5431 ) and ( 6 ) UAS-RFP ( BL #7118 ) were used . To generate the Hipk tumor model , ( 7 ) UAS-HA-hipk-3M ( abbreviated as UAS-hipk ) ( Swarup and Verheyen , 2011 ) and ( 8 ) dpp-Gal4 , UAS-hipk/TM6B were used . RNAi fly strains used include ( 9 ) UAS-dMyc-RNAi ( BL #25783 ) , ( 10 ) UAS-dMyc-RNAi #2 ( BL #25784 ) , ( 11 ) UAS-glut1-RNAi ( BL #40904 ) , ( 12 ) UAS-hex-A-RNAi ( BL #35155 ) , ( 13 ) UAS-hex-C-RNAi ( BL #57404 ) , ( 14 ) UAS-pfk2-RNAi ( BL #35380 ) , ( 15 ) UAS-pfk-RNAi ( BL #34336 ) , ( 16 ) UAS-Ldh-RNAi ( BL #33640 ) , ( 17 ) UAS-pgi-RNAi ( BL #51804 ) , ( 18 ) UAS-pgk-RNAi ( BL #33632 ) and ( 19 ) UAS-pyk-RNAi ( BL #35218 ) . To perturb signaling pathway activities , the following strains were used: ( 20 ) UAS-yki-S168A-GFP ( BL #28836 ) , ( 21 ) UAS-Arm-S10 ( Mirkovic et al . , 2011 ) , ( 22 ) UAS-Ci-5M , ( 23 ) UAS-N-act . Other strains used include ( 24 ) myc-lacZ/FM7c ( BL #12247 ) and ( 25 ) UAS-dMyc ( BL #9674 ) . ( 26 ) UAS-FLII12Pglu-700µδ6 ( Glucose sensor ) is a generous gift from Dr . Stefanie Schirmeier ( Volkenhoff et al . , 2018 ) . ( 27 ) Ldh-GFP is a generous gift from Dr . Ingrid Lohmann . Strains with BL stock number were obtained from Bloomington Drosophila Stock Center ( Bloomington , IN , USA ) . A 10–15 min heat shock ( 37°C ) was applied to larvae grown for two to three days after egg laying ( 25°C ) . After heat shock , the larvae were grown at 29°C and dissected when they reached third instar larval stage . Larval imaginal discs were dissected in PBS and fixed in 4% paraformaldehyde ( PFA ) for 15 min at room temperature . Samples were washed with PBS with 0 . 1% Triton X-100 ( PBST ) . After blocking with 5% BSA in PBST for 1 hr at room temperature , samples were incubated with primary antibodies overnight at 4°C . The following primary antibodies were used: ( 1 ) mouse anti-β-Galactosidase ( 1:50 , DSHB 40-1a ) , ( 2 ) rabbit anti-dMyc ( d1-717 ) ( 1:500 , Santa Cruz Biotechnology , Inc sc-28207 ) , ( 3 ) rabbit anti-Hipk ( 1:200; Blaquiere et al . , 2018 ) , ( 4 ) mouse anti-Armadillo ( 1:50 , diluted form , DSHB N2 7A1 ) , ( 5 ) rabbit anti-Cyclin E ( d-300 ) ( 1:100 , Santa Cruz Biotechnology , Inc sc-33748 ) . After washing with PBST , samples were incubated with Cy3- and/or Alexa Fluor 647-conjugated secondary antibodies ( 1:500 , Jackson ImmunoResearch Laboratories , Inc ) , DAPI ( 4' , 6-Diamidino-2-Phenylindole , Dihydrochloride ) ( final concentration: 0 . 2 μg per mL , Invitrogen D1306 ) and/or Rhodamine phalloidin ( 1:500 , Molecular Probes R415 ) for 2 hr at room temperature . Samples were mounted in 70% Glycerol/PBS after wash . Images were taken on a Nikon Air laser-scanning confocal microscope and processed by Image J . Imaginal discs from third instar larva were dissected in PBS and temporarily stored in RNAlater Stabilization solution ( Invitrogen AM7020 ) . Total RNA was extracted using RNeasy Mini Kits ( Qiagen 74101 ) . First strand cDNA was synthesized using OneScript Plus cDNA Synthesis Kit ( Abm G236 ) . qRT-PCR were performed using SensiFast SYBR Hi-ROS Kit ( Bioline 92005 ) on StepOne Real-time PCR System ( Applied Biosystems ) . Imaginal discs from third instar larva were dissected in PBS , followed by incubation with 2-NBDG ( Invitrogen N13195 ) at 0 . 3 mM for 15 min in the dark at room temperature . After rinsing with PBS , samples were fixed and processed according to standard immunofluorescence staining protocol . The glucose sensor UAS-FLII12Pglu-700µδ6 was generated by Dr . Schirmeier’s group ( Volkenhoff et al . , 2018 ) . Imaginal discs expressing the sensor with or without overexpression of hipk were mounted and imaged on the same day . The same confocal settings as previously described ( Volkenhoff et al . , 2018 ) were used to record the CFP and FRET images . FRET efficiency ( FRET/CFP ) was calculated as quotient of FRET and CFP using Image J . Since we were not determining the actual values of intracellular glucose concentrations , we did not take into account a correction for spectral bleed through and thus uncorrected FRET/CFP ratios were used to compare glucose levels in discs of different genotypes . The values of FRET efficiency were displayed using the Fire Lookup Table ( LUT ) in figures and shown in boxplots . Whole-body larvae were homogenized in PBS , followed by centrifugation at 14 , 000 g for 10 min . The supernatants were used for the pyruvate assay . The pyruvate assay was performed using EnzyChrom Pyruvate Assay Kit ( BioAssay Systems EPYR-100 ) according to the manufacturer’s instructions . Absorbance at 570 nm was measured in a SpectraMax M5 fluorescent microplate reader ( Molecular Devices ) . Relative pyruvate levels were determined after normalizing with the protein levels . Protein quantification was performed using the Bio-Rad protein assay ( Bio-Rad 500–0006 ) and absorbance at 595 nm was measured . Experiments were conducted with two biological replicates . Each assay was performed with two technical replicates . For data analyses , unpaired two-tailed Student t-tests were used to determine p-values using Microsoft Excel . Bar graphs were generated by Microsoft Excel . Boxplots were generated using BoxPlotR with data points included ( Spitzer et al . , 2014 ) . We used the Spear definition to show the maximum and minimum values by the upper and lower whiskers , respectively . The third quartile , median and the first quartile are shown in the box accordingly to a standard boxplot .	Cancer arises when cells in the body divide and grow excessively . These cells will often also take up more glucose than normal cells and break it down into another chemical known as lactate faster . This change to the chemical reactions happening within the cell , also called a metabolic change , is required to help produce the extra DNA , proteins and fatty molecules that it needs to grow . Elevated levels of certain proteins can trigger the changes that lead to the growth of tumors . MYC ( called dMyc in fruit flies ) is one of these proteins . It controls cell division and increases the production of enzymes that break down glucose . Hipk is another protein that can induce tumor growth in fruit flies , but how it does so was unknown . Here , Wong et al . show that high levels of Hipk boost glucose metabolism and accumulation of dMyc protein in fruit fly cells . They also describe the link between increased glucose metabolism and uncontrolled cell division . First , fruit fly cells were fed a fluorescent molecule similar to glucose that cannot be broken down by the cells . This experiment established that glucose uptake increases in cells with too much Hipk . These cells also break down glucose faster , confirming that they have increased glucose metabolism . Cells with high levels of Hipk also activate the genes that generate the enzymes involved in glucose breakdown , and increase the activity of the gene coding for dMyc . Levels of the dMyc protein are higher in these cells , which was shown by staining the cells with fluorescent molecules that specifically bind the dMyc protein . It is this buildup of dMyc protein that activates the genes coding for the enzymes responsible for glucose breakdown . PFK2 is one of these enzymes . Finally , Wong et al . inhibited the production of the enzymes that are elevated in cells with high Hipk . Stopping the production of PFK2 prevents both tumor growth and the accumulation of dMyc protein . This shows that high levels of dMyc increase PFK2 levels , leading to further dMyc buildup , and creating a loop that links the uncontrolled cell division caused by too much dMyc and the shift to higher glucose metabolism . These results highlight new potential targets for cancer therapy , showing that targeting glucose metabolism may reduce , or even stop , tumor growth .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2019	A positive feedback loop between Myc and aerobic glycolysis sustains tumor growth in a Drosophila tumor model
In placebo hypoalgesia research , the strength of treatment expectations and experiences are key components . However , the reliability or precision of expectations had been mostly ignored although being a likely source for interindividual differences . In the present study , we adopted a Bayesian framework , naturally combining expectation magnitudes and precisions . This postulates that expectations ( prior ) are integrated with incoming nociceptive information ( likelihood ) and both are weighted by their relative precision to form the pain percept and placebo effect . Sixty-two healthy subjects received heat pain during fMRI . Placebo effects were more pronounced in subjects with more precise treatment expectations and correlated positively with the relative precision of the prior expectation . Neural correlates of this precision were observed in the periaqueductal gray and the rostral ventromedial medulla , indicating that already at the level of the brainstem the precision of an expectation can influence pain perception presenting strong evidence for Bayesian integration in placebo hypoalgesia . Placebo effects and concomitant hypoalgesia in the pain context exemplify the substantial influence that expectation and experience can have on treatment outcomes and have therefore been intensely investigated over the last decades ( Atlas et al . , 2010; Colloca and Benedetti , 2006; de la Fuente-Fernández et al . , 2001; Enck et al . , 2013; Kirsch , 1999; Reicherts et al . , 2016; Rief et al . , 2011; Schenk et al . , 2014; Stone et al . , 2005; Wager et al . , 2004 ) . Various studies have identified factors that modulate these effects such as value ( Geuter et al . , 2013; Waber et al . , 2008 ) , treatment history ( Kessner et al . , 2014; Kessner et al . , 2013 ) , doctor-patient relationship ( Benedetti , 2013 ) , and context effects ( Blasi et al . , 2001 ) . Importantly , these modulators are likely to affect the precision or inverse variance of those expectations , which might explain – at least in part – the large interindividual differences observed in placebo hypoalgesia studies ( Vase et al . , 2009; Wager et al . , 2011 ) . For that reason , it is presumed that different precision levels in expectations and prior treatment experience seem to potentially change treatment outcomes ( Büchel et al . , 2014 ) . Pollo and colleagues presented different verbal instructions to manipulate patients’ treatment response expectancies and observed the largest analgesic effect in the ‘sure treatment’ group compared to two other groups being either non-informed or having only a 50% chance of receiving a painkiller ( Pollo et al . , 2001 ) . This manipulation of the probability of treatment indicates that this is likely to have an effect on the ensuing placebo response . The more precise expectations are the more probable a specific treatment outcome might be presumed . More recent approaches or frameworks in pain research already focused on accounting for several modulating factors influencing pain perception to provide a broader view on this very subjective topic ( Anchisi and Zanon , 2015; Büchel et al . , 2014; Petzschner et al . , 2017; Wager et al . , 2013; Wiech et al . , 2014 ) . For that reason , this was a starting point to formally develop a model which adequately combines the precision and magnitude of prior expectations on placebo effects ( Büchel et al . , 2014 ) and test this in a group of individuals . At the heart of this model is optimal Bayesian integration ( Knill and Pouget , 2004; Körding and Wolpert , 2004; O'Reilly et al . , 2012 ) , a framework that explicitly accounts for variability of prior information such as expectations and might thus be helpful to explain the integration of expectations and sensory information in placebo hypoalgesia ( Büchel et al . , 2014 ) . Implicit in this idea is the assumption , that the brain constantly integrates incoming sensory input with expectations and , as a consequence , generates new expectations about the environment to minimize future surprise ( Feldman and Friston , 2010; Friston , 2010; Friston and Kiebel , 2009 ) . In essence , Bayesian integration optimally integrates previous expectations ( prior ) with incoming sensory information ( likelihood ) and makes a prediction about the outcome of a certain event ( posterior ) : the posterior is proportional to the product of prior and likelihood . Importantly , in this framework , both terms are weighted by their relative precision to estimate the posterior . Putting this into the placebo context , previous treatment experience and expectations serve as the Bayesian prior being illustrated by a probability distribution reflecting pain relief as well as treatment efficacy precision . A new incoming untreated pain experience reflects the likelihood distribution . By integrating the two Bayesian key components , prior and likelihood , this framework ( Büchel et al . , 2014 ) offers the opportunity to explain the outcome of a new treatment experience , including the placebo effect , by predicting ones perceived pain as the model posterior ( Figure 1A , B ) : ( 1 ) p ( pain\|sensoryinput ) ∝p ( pain ) ∗p ( sensoryinput\|pain ) A highly precise prior treatment expectation should lead to a stronger placebo effect as compared to a highly variable expectation . By collecting subjective pain perception in the form of explicit ratings ( visual analogue scale , VAS ) , both , the mean and precision level ( reflected by the variance ) of these subjective reports translate into predictable changes in the pain percepts . These pain ratings are the basis of the aforementioned probability distributions . Furthermore , numerous imaging studies identified brain regions being involved in processing placebo hypoalgesia ( for a review see Wager and Atlas , 2015 ) highlighting the need for a multidimensional approach to investigate the underlying mechanisms . Models that can parsimoniously account for several modulating factors ( Anchisi and Zanon , 2015; Büchel et al . , 2014; Wager et al . , 2013; Wiech et al . , 2014 ) are essential to better understand and relate behavioral as well as neuronal aspects of sensory processing such as pain . This can be achieved through model-based functional magnetic resonance imaging , fMRI ( Gläscher and O'Doherty , 2010 ) by using various features of these models to identify related neural correlates . In the present study , we directly tested the hypothesis that Bayesian optimal integration is a possible mechanism by which expectation is integrated with sensory information in the context of placebo hypoalgesia . To test this framework , we investigated the naturally occurring variability in expectations and also explicitly varied levels of expectation precision in two experimental groups ( Figure 1A , B ) . This manipulation especially addressed the fact , that a pain treatment experience is usually not ( only ) verbally induced or influenced by a physician but mostly experienced by the perceived pain relief of the individual compared to the untreated nociceptive sensation . Therefore , treatment instructions had been identical for both groups but the treated painful sensation was manipulated in terms of the precision levels of prior treatment expectation . Signaled by two visual cues , participants of both groups received heat pain in a placebo treatment and untreated control condition ( Figure 1C , D ) . Via different temperature variations across trials during a conditioning phase ( Figure 1D ) , one group experienced the placebo treatment as variable ( low treatment precision ) , whereas the other group experienced it as constant ( high treatment precision ) . Assuming these two different priors represent different precision levels concerning expected treatment outcomes , we are able to compare them in terms of behavioral influences on observed placebo treatment outcomes . Importantly , we used fMRI to also investigate neural correlates of this mechanism in the brain adopting a model-based fMRI approach ( Gläscher and O'Doherty , 2010 ) . With this combined approach , we not only sought to better explain placebo treatment outcomes on a behavioral level , but also identify neural correlates which contribute to Bayesian integration of treatment expectations . We focused our fMRI analysis primarily on the periaqueductal gray ( PAG ) and the underlying processing concerning the influence of precision levels of prior expectations on placebo treatment outcomes . This a priori region of interest ( ROI ) hypothesis was especially based on a previous study that implicated the PAG in processing of the precision of vicarious information ( Yoshida et al . , 2013 ) . Supporting this , another study investigated pain avoidance prediction error coding ( Roy et al . , 2014 ) and showed that the PAG , among others , was involved in the modulation of expected probability of pain in this context . Combining this finding with the assumption that prediction error and precision level coding are distinct but share related aspects of modulatory functions , this is also hinting to the PAG being involved in the processing of prior treatment precision levels in a hypoalgesia context . Although the literature discusses several other regions such as lateral orbitofrontal ( OFC ) and rostral anterior cingulate cortex ( rACC ) to be involved in aversive prediction error coding ( Seymour et al . , 2004 , 2005; Shackman et al . , 2011; Zeidan et al . , 2015 ) , yet , no other regions were specifically related to precision level or variability processing . For that reason , this study focused the imaging analysis on the PAG . In a first step , we compared VAS ratings between the two groups during the conditioning and test phase respectively using a mixed-effects analysis . In this analysis , it was focused on the experimentally induced precision level of prior treatment expectation resulting in the use of a subset of participants ( Nsub = 49 ) ensuring that the intended conditioning manipulation of high vs . low prior treatment precision was induced according to the respective group assignment of subjects ( high treatment precision - HTP: nsub-HTP = 23 , low treatment precision - LTP: nsub-LTP = 26 , see Material and methods for further description and generating procedure of sub-samples ) . As expected , for the conditioning phase ( Figure 2A ) , a main effect of condition ( placebo vs . control ) was observed ( F ( 2 , 1172 ) = 24 . 55 , p<0 . 001 ) reflecting the two distinct temperatures used for creating the treatment experience . Neither a main effect of group ( p=0 . 963 ) nor an interaction effect of group and condition ( p=0 . 885 ) was revealed indicating that the conditioning procedure did not differ between the two groups . Subject-specific standard deviations for placebo conditioning ratings ( Figure 2B ) and therefore the variability ( i . e . inverse precision ) of prior treatment expectations was larger for the LTP compared to the HTP group ( LTP 22 . 46 ± 5 . 33 vs . HTP 13 . 63 ± 4 . 76 , t ( 47 ) = 6 . 12 , p<0 . 001 ) reflecting that in their levels of placebo treatment precision the two groups differed . It is important to note that this effect of group differences concerning precision levels was driven by the use of a sub-sample in this analysis . This served as a proof of concept for the conditioning manipulation and was done to then investigate the precision level effect on the test phase placebo response ( in other words , precision level being the independent compared to placebo effect being the dependent variable ) . Concerning the test phase ( Figure 2C , D ) , in which all subjects received identical heat stimuli for both conditions , again , a main effect of condition was observed ( F ( 2 , 1172 ) = 4 . 49 , p<0 . 001 ) . This reflects the significant placebo effect by indicating less painfulness for treatment compared to control stimuli even though participants received identical heat stimulation in both conditions . Importantly , the interaction between group and condition in the test phase reflects group differences concerning the placebo effect by accounting for possible confounding inter-individual differences ( e . g . pain sensitivity ) . This interaction of group and condition became significant ( F ( 2 , 1172 ) = 2 . 72 , p=0 . 007 ) revealing , as hypothesized , that the HTP group ( mean VAS difference of the two conditions 6 . 95 ± SD 11 . 31 ) showed a larger placebo effect than the LTP group ( 1 . 16 ± 25 . 43 ) . These results suggest that levels of prior treatment precision modulate the placebo effect . No main effect of group was observed ( p=0 . 240 ) . To evaluate the Bayesian framework , we modeled distributions of subjective ratings for every subject separately . Maximum likelihood estimates for the Bayesian model parameters of prior and likelihood were obtained by fitting Gaussian density functions to the data resulting in predictions of the test phase placebo ratings as displayed in Equation ( 2 ) . By integrating the model parameters of prior and likelihood a prediction of the test phase placebo ratings is presented not only involving the mean but , more importantly , including the respective precision level of prior treatment expectations and sensory inputs reflected in the variance of both normal distributions . As our Bayesian framework incorporates prior and likelihood precisions in a continuous manner , the whole sample ( ignoring experimental groups ) was analyzed . As expected , the predicted placebo effect ( likelihoodμ−posteriorμ ) correlated significantly with the observed placebo effect ( meantestcontrol−meantestplacebo ) with r = 0 . 441 , p<0 . 001 . To explain placebo treatment outcomes , the Null model ( see Materials and methods ) assumed no influence of the treatment experience during conditioning . This was contrasted to the Bayesian integration model hypothesizing the integration of prior expectations and experiences with the new incoming sensory observations ( likelihood ) . By using a random effects ( RFX ) Bayesian model selection approach ( Rigoux et al . , 2014; Stephan et al . , 2009 ) to estimate the overall posterior model probability across subjects , the better explanation of the given data was provided by the Bayesian integration model , reflected in a greater posterior model probability ( Figure 3A and B , see also Figure 3—figure supplements 1 and 2 for single subject fits ) . The RFX conditional expectations of model probabilities of 0 . 913 ( exceedance probability φ1 ≈ 100% ) for the Bayes model compared to 0 . 087 ( exceedance probability φ0 ≈ 0% ) for the Null model reflect this result ( Figure 3B ) . This finding indicates that the model incorporating the variance of treatment expectation performed better than the Null model . In more detail , 31 subjects ( HTP 17 , LTP 14 ) showed a Bayes factor ( BF10 ) larger than three in favor of the Bayesian integration model ( Figure 3A ) , indicating at least a moderate evidence ( Kass and Raftery , 1995; Lee and Wagenmakers , 2013 ) . Concerning the Null model , seven subjects showed a Bayes factor ( BF01 ) larger than three ( HTP 4 , LTP 3 ) . This led to a positive evidence ratio of PER10 = 31/7 = 4 . 43 . Importantly , no difference between the groups was observed meaning that in both groups , a comparable number of model fits was significantly better described by the Bayesian integration model in contrast to the Null model . In 24 subjects , none of the two models was favored significantly over the other model . Investigating the mean of treatment expectations ( μprior ) and the relationship with placebo effect magnitudes , we did not observe a correlation ( p=0 . 997 ) which makes it more likely that differences in treatment precision levels are a possible modulator of placebo treatment outcomes . To investigate this further , we used attraction weight wprior ( see Materials and methods , Equation ( 3 ) ) , a parameter that considers the precision level ( i . e . inverse variance ) of both , prior and likelihood , irrespective of the influence of the mean parameters of the two . Importantly , the attraction weight reflects a relative , integrated precision measure of prior and likelihood and includes the assumption that a certain level of treatment variability is necessary to induce a placebo effect as full predictability of a treatment outcome would not induce expectation processes ( de la Fuente-Fernández et al . , 2004 ) . In other words , it reflects the relative influence of prior and likelihood on the posterior . Attraction weight was positively correlated with the placebo effect ( r = 0 . 306 , p=0 . 016 , Figure 3C ) . This positive relationship indicates that in subjects with higher precision ( i . e . less variability ) in prior treatment expectation a larger placebo effect magnitude was observed , whereas subjects with higher precision for the perceived sensory inputs ( likelihood ) showed smaller magnitudes . Using a multiple linear regression approach , our analysis showed that the placebo effect was better predicted by the precision of treatment expectations ( σprior ) compared to the precision of perceived sensory inputs ( σlike ) ( F ( 2 , 59 ) = 6 . 83 , p=0 . 002 , R²=0 . 188 ) . The prediction of subjects’ placebo effect magnitude was equal to 15 . 005–0 . 883* ( σprior ) +0 . 562* ( σlike ) . A negative regression weight for σprior indicates that the placebo effect magnitude is expected to decrease for subjects with less precise ( more variable ) prior treatment expectations , after controlling for the precision level of sensory inputs ( σlike ) . Precision of prior treatment expectation was a significant predictor of placebo effect magnitudes ( β = −0 . 883 , t ( 59 ) = 3 . 642 , p=0 . 001 ) and precision of sensory inputs showed a trend effect ( β = 0 . 562 , t ( 59 ) = 1 . 842 , p=0 . 070 ) . Only including precision of prior treatment expectation explained less variance ( R²=0 . 141 , adj . R²=0 . 117 ) than entering both attraction weight components ( R²=0 . 188 , adj . R²=0 . 161 ) . This suggests that higher precision or less variability within individual treatment expectations ( prior ) compared to perceived sensory inputs ( likelihood ) leads to greater placebo effects . In a next step , we investigated how precision-related treatment outcomes are reflected at the neural level . As we had a strong hypothesis concerning the periaqueductal gray ( PAG ) ( Roy et al . , 2014; Yoshida et al . , 2013 ) , we specifically looked for brain activity associated with Bayesian integration parameters within this area . Most importantly , we were interested in the neural correlates of the attraction weight ( wprior ) as this value reflects the relative precision of both , expectations and incoming sensory inputs , without any influence of simple intensity coding of pain ( no involvement of the mean ) . By using wprior as a covariate for the test phase placebo condition , we observed an activation in the PAG ( coordinates [2 -26 -8] , kE = 8 , t ( 60 ) = 4 . 16 , pFWE = 0 . 001; Figure 4 and Figure 4—figure supplement 1 ) . Higher BOLD signals in the PAG were related to smaller attraction weight values ( r = −0 . 457 ) . This more detailed description of the PAG finding is visualized in Figure 4B . In other words , the less precise prior treatment expectations are relative to sensory inputs , the stronger the PAG BOLD signal . Interestingly , the same contrast also revealed a cluster in the rostral ventromedial medulla ( RVM ) but this was not significant when correcting the p-value for the entire volume ( coordinates [2 -36 -46] , kE = 16 , t ( 60 ) = 4 . 06 , puc <0 . 001 ) . Here , also a negative correlation with the relative precision was observed ( r = −0 . 451 ) . Additional , more explorative , results reflecting the neural correlates of μprior , log⁡σprior , and the posterior model probability for the Bayesian over the Null model ( Figure 3A , blue bars ) can be found in the Supplement ( Figure 4—figure supplements 2 , 3 and 4 ) . Our results provide evidence that a Bayesian integration mechanism in the context of placebo hypoalgesia can account for placebo effects on a behavioral level . More importantly , our fMRI data revealed that key parameters of this mechanism are represented in the periaqueductal gray ( PAG ) . These results add to other approaches also proposing alternative ways of analyzing perceptual experiences such as pain ( Anchisi and Zanon , 2015; Büchel et al . , 2014; Wager et al . , 2013; Wiech et al . , 2014 ) indicating the need for models that can parsimoniously account for several modulating factors . Through model-based fMRI , these models allow to identify neural correlates of various aspects of pain perception and placebo hypoalgesia . On the behavioral level , we observed that placebo effects were smaller in subjects with less precision in their prior treatment experience ( higher variability ) and more pronounced for those who perceived the treatment as more constant ( higher precision ) relative to the incoming sensory stimuli . The observed placebo effect did not correlate with the simple mean of prior treatment expectations which reflects the pain intensity during treatment conditioning . This intensity mean of the prior was not able to describe the strength of the placebo effect via a correlative relationship , whereas the precision level of the prior was . By showing that , we were able to present a framework that indicates strong evidence for placebo effects being explained by optimal Bayesian integration ( Anchisi and Zanon , 2015; Büchel et al . , 2014 ) . Also , our results showed that Bayesian integration predicted placebo treatment outcome based on various individual prior and likelihood distributions . The framework did not favor high over low prior precision or vice versa and predicted the different cases equally well . As placebo effect magnitudes are often highly variable ( Vase et al . , 2009 ) our results reflect the usefulness of a Bayesian approach that can account for variable as well as precise prior treatment expectations . In some subjects , the Bayesian integration approach did not describe the treatment outcome significantly better than the Null model . Reasons for that are highly speculative and include , for example , that not all individuals may combine information of prior experiences and new sensory inputs in an optimal way . Finding markers to predict which individual uses optimal Bayesian integration and which does not would be a subject for future research . Additionally , the experimental manipulation used in this study provides new insights in the modulation of different levels of precision of prior treatment expectations by holding the test phase stimuli constant in both groups . The present study used test phase stimuli matching prior expectations concerning treatment precision levels in the HTP but not in the LTP group as induced by the different conditioning procedures . This was done to investigate the specific effect of precision levels of prior treatment expectations without interfering variability of the test phase stimulation . However , for that reason , our study cannot answer the question whether a modulation of precision levels of the test phase stimuli may influence the placebo effect and pain perception in a different way . This might be answered in future studies with both matching or mismatching conditioning and test phase precision levels . Our fMRI data indicated that the PAG signal change represents relative expectation precision of the Bayesian integration process in a placebo hypoalgesia study . This brainstem area is part of the ascending and descending pain system and is crucial for pain modulation ( Fairhurst et al . , 2007 ) , pain avoidance prediction error coding ( Roy et al . , 2014 ) , and the processing of precision of vicarious information ( Yoshida et al . , 2013 ) . The PAG is not only known to mediate pain inhibition ( Jones and Gebhart , 1988 ) , but also involved in pain facilitatory processes ( Vanegas and Schaible , 2004 ) which makes it a key structure of anti- as well as pro-nociceptive effects . Our behavioral data shows that pain modulation underlying placebo hypoalgesia depends on the precision of the prior expectation . Therefore , our data suggests that the PAG is an area crucially involved in precision biased integration processes due to its opposing modulatory properties . Previous research already hinted at a representation of uncertainty in the PAG during a painful vicarious observation task ( Yoshida et al . , 2013 ) . They observed a potent hyperalgesia effect during this vicarious observation task in subjects who showed high susceptibility to induced variability and related this to an increased BOLD signal in the periaqueductal gray ( PAG ) . In this study , the results of the winning uncertainty-hyperalgesia model suggested a strong effect of social assimilation and hyperalgesic uncertainty , but were not able to clearly determine whether this was driven by an underlying linear mean effect of pain or the uncertainty that was induced during the observation task . By using an optimally integrated precision weight to explicitly investigate the influence of variability irrespective of the mean intensity of pain in the placebo treatment context , our data indicates a significant influence of the PAG in variability coding . However , our data does not negate the influence of the mean intensity of pain as the significant behavioral main effect of condition during our placebo test phase illustrates a clear relationship . We additionally observed that the rostral ventromedial medulla ( RVM ) is involved in this process . Again , facilitatory as well as inhibitory projections from the PAG relay in the RVM ( Vanegas and Schaible , 2004 ) which underpins and adds to the relationship of variability coding in the PAG as both brainstem areas are involved in opposing modulatory processes during pain . In both , PAG and RVM , a stronger activation was observed when the treatment expectation was less precise ( i . e . high variability ) relative to the precision of the incoming sensory stimuli . Dependent on the context , including anticipatory and/or direct responses to nociceptive stimuli , the placebo effect literature reports PAG activation patterns being both , increased or decreased in placebo hypoalgesia reflecting the mediating effect of this area ( Eippert et al . , 2009; Geuter et al . , 2013; Peciña et al . , 2013; Scott et al . , 2008; Wager et al . , 2004; Wager et al . , 2007; Wager and Atlas , 2015; Zubieta et al . , 2005 ) , whereas the RVM shows mainly increased activation in placebo hypoalgesia ( Eippert et al . , 2009 ) . These activation patterns are likely being related to variability coding as seen in our results . Variability during pain most probably introduces uncertainty about future painfulness ( Seidel et al . , 2015 ) . In this context , our data suggests that less precise prior treatment expectations based on missing information lead to higher activation in the PAG probably due to processing less predictable ( Fairhurst et al . , 2007 ) outcomes . It seems that the PAG’s modulatory processing ( Jones and Gebhart , 1988; Linnman et al . , 2012; Vanegas and Schaible , 2004 ) is reflected by this signal increase in the presence of non-precise information about a treatment . Also , some studies were able to show that PAG placebo-induced signal increases were related to the strength of the analgesic effect connecting this to the opioidergic descending pain control system ( Eippert et al . , 2009; Peciña et al . , 2013; Wager et al . , 2004 ) . Our results additionally indicate that PAG signal increase is related to less precise prior treatment expectations ( prior ) compared to more precise incoming sensory stimuli ( likelihood ) during a placebo treatment , importantly not being related to simple pain intensity but rather optimally integrated variability coding . Therefore , an influence of prior treatment precision levels on opioidergic descending pain modulations is speculated as we also observed the same trend-wise activation pattern in the RVM . This is supported by a previous finding of the anticipated analgesic effect being positively correlated with a signal increase in the PAG ( Scott et al . , 2008 ) , reflecting the responsiveness of this brainstem area to placebo-induced expectations . In both areas , PAG and RVM , placebo-induced BOLD-activations can be significantly reduced by naloxone ( Eippert et al . , 2009 ) , an opioid-antagonist known to impair placebo-dependent pain reduction ( Amanzio and Benedetti , 1999; Grevert et al . , 1983; Levine and Gordon , 1984; Levine , 1978 ) , which reflects their involvement in the opioidergic descending pain control system . As these are the two brain regions we also observed during the coding of expectation precision under a placebo treatment condition , an involvement of the opioidergic descending pain system is likely . Additionally , the PAG-RVM circuit is known to be involved in the aforementioned anti- but also in pro-nociceptive responses such as cholecystokinin ( CCK ) antagonizing opiate analgesia ( Watkins et al . , 1984 ) , nocebo hyperalgesia ( Benedetti et al . , 2006 ) , opiate hyperalgesia ( Xie et al . , 2005 ) , and safety signal-mediated hyperalgesia ( Wiertelak et al . , 1992 ) . As it was previously speculated that the PAG-RVM modulation in pro-nociceptive responses may likely by involved in the generation and maintenance of discomforting and painful functional disorders such as chronic pain , irritable bowel syndrome , and fibromyalgia ( Tracey and Dunckley , 2004 ) , a substantial relevance of our findings also connecting the PAG-RVM circuit to treatment precision level coding may shed more light on presumably important aspects of these disorders . A neuronal modulation via the precision level of prior treatment expectation concerning these disorders could generate new hypotheses about interindividual differences between patients . At this point , these interpretations are highly speculative though and need further investigation in the future . Due to the study’s specific experimental manipulation , the current findings suggest a strong influence of perceptual modulations concerning different levels of prior treatment precision . However , this does not limit the interpretation of the results to only a perceptual but also possible changes in homeostatic regulatory actions as a recent probabilistic model suggested ( Petzschner et al . , 2017; Stephan et al . , 2016 ) . The authors propose that belief updates are constantly influencing the performance of the interoceptive-allostatic circuit ( allostatic self-efficacy ) which might be able to better explain differential diagnosis in disorders such as fatigue and depression . This model supports the findings of our study in a way that the precision of prior beliefs predicted placebo treatment outcomes better than the precision of sensory inputs relating a stronger influence of expectations compared to new sensory information to these disorders . To summarize , our results add to the existing literature and frameworks , as we observed that not in the cortex but already at the brainstem level , Bayesian integration was able to explain signal changes in placebo hypoalgesia . This nicely compliments previous research suggesting that Bayesian integration during the processing of placebo treatments already takes place at a very basic level of pain perception ( Anchisi and Zanon , 2015 ) , but , as investigated in a behavioral study , was not able to reveal the underlying neural mechanisms . In a clinical context , our data clearly indicates that the level of precision of prior treatment experiences and associated expectations is a crucial determinant of placebo effects in treatment outcomes . This mechanism can clinically be exploited by providing precise a priori information concerning a treatment , which will help to create precise prior expectations . Additionally , as there is growing evidence that the chronification of pain may be related to a dysregulation of the descending pain modulatory system ( for a review see Ossipov et al . , 2014 ) , our Bayesian framework and the related neural findings not only add to the placebo literature but also inform other clinical areas investigating disruptions of the modulatory circuits during pain processing . Seventy healthy male right-handed subjects with no history of psychiatric or neurological illness were assigned to two groups using a randomized double-blind allocation . Eight subjects had to be excluded due to incomplete data collection or technical difficulties ( four in each group ) . Data analysis was performed on the remaining 62 participants ( mean age ± SD: 24 . 60 ± 3 . 77 years , range: 19–34 years ) . Both groups only differed in variability levels within the placebo treatment conditioning ( high treatment precision - HTP , no induced variability; low treatment precision - LTP , high induced variability , n = 31 each ) . The groups did not differ significantly in age ( years: HTP 24 . 71 ± 3 . 88; LTP 24 . 48 ± 3 . 71; t ( 60 ) = 0 . 234 , p=0 . 816 ) or basic pain thresholds ( just painful °C: HTP 44 . 0 ± 2 . 4; LTP 44 . 0 ± 2 . 3; t ( 60 ) = 0 . 005 , p=0 . 996 ) . They were tested on 2 days with an average break of approximately 5 days between testing ( days: HTP 4 . 55 ± 3 . 33; LTP 5 . 03 ± 3 . 74; t ( 60 ) = 0 . 593 , p=0 . 592 ) . The study was approved by the Ethics Committee of the Medical Board Hamburg , Germany . All subjects were remunerated for participation and gave written informed consent in accordance with the Declaration of Helsinki prior to the experiment . This included information about exclusion criteria ( neurological and/or pain related diseases , psychological disorders , skin afflictions , substance abuse , current medication , physical and/or emotional stress ) and all experimental procedures such as MRI measurements , thermal and electrical stimulation as well as possible adverse reactions . They were not informed about the purpose of the study investigating placebo hypoalgesia until post-experimental debriefing . Each subject signed a second informed consent after debriefing that the acquired data was not withdrawn from the study . A between-subjects design was used to create an overall sample that shows sufficient variability of prior precision concerning placebo treatment expectations . All subjects were told that TENS is well-established and known to reduce pain and that the aim of this study was to investigate the underlying neural mechanisms of TENS . To create treatment expectation and let the subjects experience the analgesic effect , both verbal suggestion and conditioning components were used similar to other placebo studies ( Colloca and Benedetti , 2006; Montgomery and Kirsch , 1997; Price et al . , 1999; Wager et al . , 2004 ) . Experimental instructions and suggestions concerning the putative treatment were identical for both groups and delivered in written and oral format by an experimenter unaware of the individual’s group assignment . The experimenter always presented herself wearing a white coat . Day 1 was only used to identify potential abnormal pain perception and familiarize the subject with the painful heat stimulation and the VAS rating procedure as well as to check the subject’s MR applicability by a physician . First , basic pain thresholds were assessed performing a limits procedure by slowly increasing temperature until the heat was reported as just painful by the subject . This was done three times and the mean was used as an anchor point for the actual calibration procedure trials . Heat calibration consisted of 16 different intensity trials presented in a pseudorandomized order on the right volar forearm . Trials and rating procedure were presented as explained above . To calculate individual temperatures for corresponding VAS ratings of 30 , 50 , and 70 , a sigmoidal function was fitted to the ratings . This ensured that individual pain ratings , despite different intensities , were comparable across participants . The whole procedure during day 1 lasted about 1 hr . On day 2 , the fMRI experiment was performed including a conditioning and test phase . The four skin patches and two TENS electrode positions ( Figure 1C ) were marked on the left volar forearm using a stencil . Each position was used for one of the four experimental sessions ( Figure 1D ) in a pseudorandom order during scanning . Two electrodes for the sham TENS were placed beside the four skin patches . Before going into the scanner , all subjects answered a mood ( Steyer et al . , 1997 ) and TENS questionnaire to assess their possible foreknowledge regarding TENS as a medical treatment . To account for possible context effects of the MR environment ( Ellerbrock and May , 2015 ) , the 16-trial-calibration procedure from day one was repeated on the right volar forearm with the subject laying in the scanner without acquiring BOLD data . The new calculated temperatures corresponding to 30 ( in °C mean ± SD: HTP 44 . 4 ± 0 . 7; LTP 44 . 3 ± 0 . 6 ) , 50 ( HTP 44 . 9 ± 0 . 6; LTP 45 . 0 ± 0 . 6 ) , and 70 ( HTP 45 . 5 ± 0 . 6; LTP 45 . 7 ± 0 . 6 ) VAS were used for scanning which immediately started after pain calibration . For every trial of the whole experiment , all subjects expected a painful heat intensity of 70% of their pain tolerance . During the conditioning phase , all participants underwent two sessions ( placebo ‘TENS on’ skin patch and control ‘TENS off’ skin patch ) with 12 trials each . The order of stimulation ( placebo or control session first ) and patch position were counterbalanced across participants and matched for the two groups to minimize possible order confounds . Unbeknownst to the participants , for both groups , the average placebo treatment temperature was set to 30% pain tolerance . Despite that , only within this placebo condition , the two groups differed from each other regarding the level of treatment variability ( Figure 1D ) . The HTP group experienced the placebo treatment as consistently effective meaning that they were always presented with the same pain intensity of VAS 30 . In contrast , the LTP group received a pain relieving placebo treatment with varying temperatures ( SD = 0 . 55°C; mean VAS 30; range around 30 VAS temperature: ± 0° to ± 0 . 8°C ) across placebo treatment trials . Importantly , during the conditioning , the untreated control stimuli were identical for both groups and included heat stimulation of 70% pain tolerance intensity without any induced variability across trials . This manipulation procedure served to enhance expectations regarding the placebo treatment and its effectiveness concerning heat pain relief . Each trial ( Figure 1C ) consisted of an inter-trial-interval ( ITI , 12–20 s ) , followed by an anticipation phase ( 5 . 5–8 s ) , the painful heat stimulation ( 8 s ) , a delay ( 2 s ) , and the VAS rating procedure ( 7 s ) . The ITI was represented by a white fixation cross . For the anticipation cue , either a red fixation cross ( control condition ) or a red fixation cross surrounded by a yellow circle ( placebo condition ) was presented . The respective cue remained during heat stimulation and disappeared after cooling down leaving a blank screen for the post-stimulus delay . Subsequently , the VAS appeared and subjects rated and confirmed their perceived painfulness . At the end of each session , participants additionally rated the subjective average painfulness of all trials received during this session . The conditioning phase was directly followed by the test phase in which the created treatment expectation was compared to the non-manipulated control condition . Importantly , during the test phase , 12 identical heat stimuli of 50% pain tolerance intensity were applied for both placebo and control session , respectively ( Figure 1D ) . As stimulation was physically identical in both sessions , placebo effects were assessed by directly comparing VAS ratings of the two conditions . Ratings were expected to be decreased in the placebo compared to the control condition and even more reduced in the HTP compared to the LTP group . Outside the scanner , participants completed several questionnaires to assess personality dimensions and mood components . This also included a post-experimental TENS questionnaire assessing the subject’s experience during the treatment prior to debriefing ( see Figure 1—source data 1 and Figure 1—figure supplement 1 ) . The whole experimental procedure during day 2 lasted about 3 hr . For stimulus presentation , triggering and , recording of pain ratings , Matlab ( Mathworks , Natick , MA ) and the open-source Matlab based Psychophysics Toolbox 3 ( Brainard , 1997; Pelli , 1997 ) was used . Skin conductance was acquired on the distal and proximal hypothenar of the left hand , placing both electrodes on dermatome C8 . Additionally during scanning , respiration and heart rate was recorded by using the Expression patient monitoring system ( Invivo Corporation , Orlando , FL ) . A CED 2502 ( Cambridge Electronic Design Limited , Cambridge , UK ) was used to amplify and a CED micro 1401 to digitalize skin conductance signal at 1000 Hz . The data was recorded by the CED software Spike 2 . Magnetic resonance imaging ( MRI ) data were acquired using a 3T Magnetom Trio scanner ( Siemens , Erlangen , Germany ) equipped with a 32-channel head coil . BOLD responses were measured using a T2* sensitive echo planar imaging ( EPI ) sequence . Each volume consisted of 38 transversal slices with a voxel size of 2 × 2×2 mm³ and a 1 mm gap ( repetition time: 2 . 35 s , echo time: 26 ms , flip angle: 80° , field of view 224 × 224 mm , GRAPPA PAT-factor: 2 , reference lines: 48 ) . Volumes were tilted approximately 30° relative to AC-PC line to allow coverage of most of the brainstem area . Considering T1 saturation , the first 4 volumes of every session were discarded . To account for B0 inhomogeneity , prior to each session , B0 field maps were also acquired ( 40 slices , voxel size: 3 × 3×3 mm³ , repetition time: 398 ms , short echo time: 4 . 31 ms , long echo time: 6 . 77 ms , flip angle: 40° , field of view 216 × 216 mm ) . Additionally , a high-resolution anatomical T1-weighted image was acquired for each subject ( MPRAGE sequence , voxel size: 1 × 1×1 mm³ ) . Concerning the statistical analysis , a frequentist approach was applied to test for main and interaction effects of the experiment , and computational modeling was used to predict placebo treatment outcomes and compare them with the observed data . FMRI data and statistical analyses were performed using statistical parametric mapping ( SPM12 , Wellcome Trust Centre for Neuroimaging , London , UK ) . The first four images of each run were discarded prior to further analyses . Preprocessing consisted of motion correction ( realignment and field map correction ) , coregistration of the anatomical T1 image to the functional scans , segmentation of the anatomical T1 image producing DARTEL-imported native tissue class images and in the next step a flow field of the T1 image in Montreal Neurological Institute ( MNI ) standard space ( IXI555_MNI152 template of VBM12 toolbox ) using the DARTEL toolbox as implemented in SPM12 . First-level analysis was performed in subject-specific native space . Data were high-pass filtered with a 128 s cut-off period and corrected for temporal autocorrelations using a first-order autoregressive model . Functional MRI data analysis was based on a general linear model ( GLM ) approach as it is implemented in SPM12 . The first-level design matrix of each subject consisted of 10 regressors for each session , resulting in a total of forty regressors: anticipation cue onset ( 5 . 5–8 s ) , pain onset ( 8 s ) , VAS rating ( 7 s ) , six motion regressors obtained during realignment , and one session constant . Each regressor was modeled using a boxcar function and subsequently convolved with the hemodynamic response function . After model estimation , t statistics for each voxel were calculated . All ensuing output images were then normalized to MNI space using the previously obtained subject-specific DARTEL flow field , smoothed with an 6 mm ( FWHM ) isotropic Gaussian kernel , and then used for second-level analyses . For that , it was investigated whether behavioral Bayesian model parameters reflecting the processing of variability would predict placebo-induced changes in brain signals in the PAG during the test phase placebo condition . Therefore , the attraction weight ( wprior ) was used as a covariate in a one-sample t test , testing whether variability variations in prior and likelihood would explain changes in BOLD responses . To complement information about the relationship between Bayesian integration and possible related BOLD responses concerning prior treatment expectation , we additionally performed one-sample t tests either using μprior , log⁡σprior , or the posterior model probability for the Bayesian over the Null model ( as in Figure 3A , blue bars ) as respective covariates ( see Figure 4—figure supplements 2 , 3 and 4 ) . Figure 4—figure supplement 5 shows the main effect of pain and placebo , respectively . For imaging data analyses , results were considered significant after multiple comparison correction using a family-wise error rate ( FWE ) approach and a threshold of p<0 . 05 . Correction was based on a small-volume approach using a 6 mm sphere around coordinates ( MNI , x y z: 1 –29 −12 ) obtained from previous studies on the PAG ( Linnman et al . , 2012 ) . Other areas found during the analyses were considered significant on a whole brain corrected level of p<0 . 05 . Activations that did not survive whole brain correction but met the criteria of p<0 . 001 uncorrected and were located in either pain or placebo relevant areas are considered informative and are also reported . Single subject parameter extraction of the BOLD signal and the respective covariate was done by first correcting for multiple comparisons , extracting beta values for each surviving voxel of the corrected ROI cluster per subject , and then averaging subject-wise over all surviving voxels . Statistical maps are presented with a threshold of p<0 . 001 uncorrected ( uc ) , masked with the field of view of data acquisition , and overlaid on the mean structural image of all subjects . Activations are reported in mm ( x y z ) using MNI standard space .	On a battlefield in World War II , surgeon Henry Beecher ran out of morphine . To his surprise , he found that replacing the missing morphine with saltwater allowed him to continue operating on wounded soldiers . Although saltwater contains no active pain-relieving ingredients , it reduced the soldiers’ pain . This is an example of the placebo effect . Placebos have been shown to reduce autonomic responses to pain , such as sweating . They also modulate activity in brain regions that process pain . But why do some of us experience larger placebo effects than others ? Grahl et al . propose that the size of the placebo effect depends on our expectations about a treatment . More specifically , it depends on how precise those expectations are . Imagine two people who have taken the same treatment many times , and who have experienced the same average reduction in pain . But for one person , the treatment reduced their pain by roughly the same amount each time . For the other , the treatment sometimes reduced their pain by a large amount and other times hardly at all . The first person will have more precise expectations than the second about how effective the treatment will be in future . Grahl et al . propose that the first person will thus experience a greater placebo effect in response to a ‘fake’ version of the treatment . To test this idea , Grahl et al . applied painful heat to the forearms of healthy volunteers lying inside a brain scanner . On half the trials , the volunteers were told that they would also receive an electrical pain-relieving therapy . In reality , this treatment was never applied . After each trial , the volunteers rated the intensity of the pain they had experienced . As expected , the volunteers reported less pain when they thought they were receiving a pain-relieving treatment . Moreover , those volunteers with more precise expectations about the treatment reported greater pain relief than volunteers with less precise expectations . The former group also showed less activity in one of the brain’s major pain-processing centers , the periaqueductal gray . These findings help shed light on why some people experience larger placebo effects than others . They suggest that helping patients form precise expectations about their treatment , by giving them precise information about its likely effectiveness , may boost the placebo effect . Further studies are needed to determine whether this phenomenon also occurs in patients with pain disorders . If it does , it could help such patients manage their pain using fewer active painkillers .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2018	The periaqueductal gray and Bayesian integration in placebo analgesia
Mammalian sirtuin 6 ( Sirt6 ) is a conserved NAD+-dependent deacylase and mono-ADP ribosylase that is known to be involved in DNA damage repair , metabolic homeostasis , inflammation , tumorigenesis , and aging . Loss of Sirt6 in mice results in accelerated aging and premature death within a month . Here , we show that haploinsufficiency ( i . e . , heterozygous deletion ) of Trp53 dramatically extends the lifespan of both female and male Sirt6-deficient mice . Haploinsufficiency of Trp53 in Sirt6-deficient mice rescues several age-related phenotypes of Sirt6-deficient mice , including reduced body size and weight , lordokyphosis , colitis , premature senescence , apoptosis , and bone marrow stem cell decline . Mechanistically , SIRT6 deacetylates p53 at lysine 381 to negatively regulate the stability and activity of p53 . These findings establish that elevated p53 activity contributes significantly to accelerated aging in Sirt6-deficient mice . Our study demonstrates that p53 is a substrate of SIRT6 , and highlights the importance of SIRT6-p53 axis in the regulation of aging . Mammalian sirtuin 6 ( Sirt6 ) , a conserved NAD+-dependent deacylase and mono-ADP ribosylase , has been implicated in a range of regulatory pathways involved in premature aging and other age-associated pathologies ( Kugel and Mostoslavsky , 2014 ) . Mice deficient in Sirt6 exhibit severe premature aging phenotypes , including metabolic defects , lymphopenia , osteoporosis , genomic instability , and early death by 4 weeks of age ( Mostoslavsky et al . , 2006 ) . Additionally , the ectopic expression of Sirt6 has been shown to extend the lifespan of male mice by 15% ( Kanfi et al . , 2012 ) . Sirt6 was first identified to deacetylate histone H3 at lysine 9 and 56 ( in a NAD+-dependent manner ) at the telomeres , promoter regions of target genes and globally as well , to mediate DNA damage repair , maintain telomeric metabolism , suppress NF-κB pathway in promoting longevity and regulate cell cycle ( Kawahara et al . , 2009; Michishita et al . , 2008; Michishita et al . , 2009; Yang et al . , 2009 ) . The intricate roles of Sirt6 in DNA damage repair and metabolic regulation have been reported to stem from its inherent NAD+-dependent deacetylating activity on histones , CtIP , GCN5 and others , apart from its NAD+-dependent mono-ADP ribosylase activity ( Dominy et al . , 2012; Kaidi et al . , 2010; Mao et al . , 2011 ) . These and other foundational studies have established the importance of Sirt6 in the regulation of processes contributing to aging and longevity . Although Sirt6-mediated regulation has been reported for several signaling pathways ( for example , NF-κB , AKT , and IGF1 ) ( Kanfi et al . , 2012; Kawahara et al . , 2009; Pan et al . , 2016; Xiao et al . , 2010 ) , the key mechanism underlying the severe acceleration of aging and premature death in Sirt6-deficient mice remains elusive . Trp53 , also denoted as p53 , is a tumor suppressor gene that triggers cell cycle arrest , apoptosis , and/or senescence ( Rufini et al . , 2013; Vousden and Prives , 2009 ) . p53 has been widely implicated in premature senescence and aging ( Hinkal et al . , 2009; Poyurovsky and Prives , 2010; Varela et al . , 2005 ) . p53 was the first non-histone protein identified to undergo acetylation as a post-translational modification ( Gu and Roeder , 1997 ) . Ever since , a range of lysine residues have been identified across the domains of p53 , with majority of acetylated lysine residues being concentrated in the C-terminus region of p53 ( Brooks and Gu , 2011; Gu and Zhu , 2012; Gu and Roeder , 1997; Kruse and Gu , 2009 ) . Lysine 382 , localized in the C-terminus of p53 , is identified to undergo marked acetylation followed by lysine 381 and 373 ( Gu and Roeder , 1997 ) . Given the stunted half-life of around 30 min for p53 , these acetylations have been reported to impart stability to p53 , which then mediates a range of downstream functions in response to DNA damage , oncogenic stress , age-associated abnormalities , tumor suppression and others ( Donehower , 2009; Lozano , 2010; Vousden and Prives , 2009 ) . In laminopathy-based progeria , mutant lamin A results in elevated p53 signaling and reducing p53 level significantly ameliorates premature aging phenotypes and extends lifespan ( Varela et al . , 2005 ) , thus suggesting that activation of p53-dependent downstream signaling accelerates aging . Interestingly , lamin A is an endogenous activator of SIRT6 and mutant lamin A results in compromised SIRT6 deacetylase activity ( Ghosh et al . , 2015 ) . Given that functional defects in Sirt6 and activation of p53 are both associated with laminopathy-based premature aging , it is conceivable to speculate that loss of Sirt6 may also impact p53 signaling . To determine whether aberrant p53 signaling contributes to the severe progeroid phenotypes in Sirt6-deficient mice , we generated Sirt6-deficient mice with haploinsufficiency ( i . e . , heterozygous deletion ) of p53 and monitored the growth and survival of the compound mutant mice along with their wild-type and Sirt6 single knockout ( KO ) littermates . We assessed and compared a range of premature aging-associated abnormalities in our generated compound mutant mice ( Sirt6-/-Trp53+/- mice ) , which have been previously reported in Sirt6 single KO ( Sirt6-/-Trp53+/+ ) mice ( Mostoslavsky et al . , 2006 ) . Our findings , revealing the direct link between Sirt6 and p53 , underscore the critical role of the Sirt6-p53 regulatory axis in aging and age-associated diseases , which has potential therapeutic implications in healthy aging management . To determine whether p53 signaling is involved in the causation of premature senescence upon loss of Sirt6 , we first analyzed the expression of some classical downstream targets of p53 in Sirt6-proficient ( Sirt6+/+Trp53+/+ ) and deficient ( Sirt6-/-Trp53+/+ ) primary mouse embryonic fibroblasts ( MEFs ) obtained from littermates , and explored the possibility that functional cross-talk may occur between these two proteins . To this end , 3 batches of primary MEF cells collected from independent batches of embryos were analyzed . To avoid mutations in p53 pathway occurring spontaneously over serial passaging of MEFs , only primary MEFs of P1 or P2 passage were used in experiments , unless stated otherwise . Intriguingly , a range of downstream targets of p53 , including p21 , Puma , Noxa , Bax and Ddit4 ( Poyurovsky and Prives , 2010 ) , were upregulated in Sirt6-/-Trp53+/+mouse embryonic fibroblasts ( MEFs ) as compared to Sirt6+/+Trp53+/+MEFs ( Figure 1A ) . Consistent with previous reports stating that loss of Sirt6 results in premature senescence ( Mao et al . , 2012; Mostoslavsky et al . , 2006 ) , Sirt6-/-Trp53+/+ MEFs exhibited enhanced senescence-associated β-galactosidase activity and senescence-like flattened cellular morphology at passage 6 ( P6 ) ( Figure 1B and C and Figure 1—figure supplement 1A ) . Sirt6-/-Trp53+/+ MEFs also exhibited enhanced p16 levels at P6 ( Figure 1—figure supplement 1B ) . However , haploinsufficiency of Trp53 significantly rescued the senescence-associated phenotypes in Sirt6-/- MEFs at P6 ( Figure 1B and C and Figure 1—figure supplement 1A , 3 independent batches of MEFs have been used in the study which were obtained from independent batches of littermate embryos ) . Also , Trp53 haploinsufficiency significantly attenuated p16 levels and the enhanced expression of the downstream targets of p53 in Sirt6-/-Trp53+/- MEFs ( Figure 1—figure supplement 1B , C ) . Enhancement of sensitivity to DNA damage upon loss of Sirt6 has been previously reported ( Mostoslavsky et al . , 2006 ) . Again , the increased sensitivity to DNA damage in Sirt6-/-Trp53+/+ MEFs treated with gamma-irradiation was substantially attenuated in Sirt6-/-Trp53+/- MEFs ( Figure 1D and Figure 1—figure supplement 1D ) . The decreased cell viability of Sirt6-/-Trp53+/+ MEFs was significantly improved in Sirt6-/-Trp53+/- MEFs ( Figure 1E ) . To further investigate the effects of partial ablation of p53 in Sirt6 knockout ( KO ) background at the organismal level , we used compound heterozygous mating strategy to generate Sirt6 KO mice with haploinsufficiency of p53 ( Sirt6-/-Trp53+/- mice ) as well as Sirt6-/-Trp53+/+ ( Sirt6 KO ) and Sirt6+/+Trp53+/+ ( wild-type ) littermates . The internal organs , such as kidneys , liver and spleen from these mice were collected for further analyses . Consistent with our findings in MEFs ( Figure 1A ) , there was a significant upregulation of the expression of several downstream targets of p53 in the liver , kidneys , and spleen of Sirt6-/-Trp53+/+ mice ( Figure 1F and G and Figure 1E ) . However , Trp53 haploinsufficiency significantly suppressed the expression of those downstream targets of p53 in the liver , kidneys , and spleen of Sirt6-/-Trp53+/- mice ( Figure 1F and G and Figure 1—figure supplement 1E ) . This further suggests that the upregulation of these targets upon loss of Sirt6 is indeed a consequence of p53 activation . Next , we examined whether haploinsufficiency of Trp53 could rescue the accelerated aging phenotypes of Sirt6-deficient mice . To this end , we employed compound heterozygous mating strategy to mate Sirt6+/-Trp53+/+ mice ( in pure FVB background ) with Sirt6+/+Trp53+/- mice ( in pure C57BL/6 background ) to generate Sirt6+/-Trp53+/- mice and interbred them to generate Sirt6-/-Trp53+/- mice along with Sirt6-/-Trp53+/+ and Sirt6+/+Trp53+/+ littermate mice . Given that mixed strains could produce false positive results , we used the litters containing all Sirt6+/+Trp53+/+ , Sirt6-/-Trp53+/+ and Sirt6-/-Trp53+/- mice for further analysis . Consistent with a previous report ( Mostoslavsky et al . , 2006 ) , majority of the Sirt6-/-Trp53+/+ mice exhibited severe premature aging and were much smaller than their wild-type littermates at 3 weeks of age ( Figure 2A ) . Interestingly , the Sirt6-/-Trp53+/- mice appeared much healthier than the Sirt6-/-Trp53+/+ littermates post 3 weeks of birth ( Figure 2A ) . Though smaller than their littermate wild-type ( Sirt6+/+Trp53+/+ ) controls , both female and male Sirt6-/-Trp53+/- mice were significantly larger in body size than their Sirt6-/-Trp53+/+ littermates ( Figure 2A n=25-35 for all three genotypes ) . The mean body weights of both female and male Sirt6-/-Trp53+/- mice were also significantly more than that of Sirt6-/-Trp53+/+ littermates ( Figure 2B and C ) . The Sirt6-deficient mice mostly died within 4 weeks of birth , whereas the Sirt6-/-Trp53+/- littermate mice exhibited striking extensions in lifespan: female mice exhibited a 16-fold increase in maximum lifespan and 11-fold increase in median lifespan ( Figure 2D ) ; males showed a 14-fold extension in maximum lifespan and a 7 . 5-fold extension in median lifespan ( Figure 2E ) . Given only 4 weeks of lifespan in Sirt6-/-Trp53+/+mice , the average lifespan of 30 and 44 weeks in male and female Sirt6-/-Trp53+/- mice , respectively , represents a significant lifespan extension by Trp53 haploinsufficiency . Also , the lifespan extension observed in the compound mutant mice in our study ( Figure 2A–E ) was much higher than the lifespan extension observed because of mixed background strains ( Peshti et al . , 2017 ) , thus suggesting that the observed rescue in longevity of sirt6-deficient mice is indeed a result of Trp53 heterozygosity . Consistent with previous reports ( Jacks et al . , 1994; Tyner et al . , 2002 ) , we found that Sirt6+/+Trp53+/- mice had median lifespan of 56–64 weeks ( Figure 2D and E ) , and mostly succumbed to tumors by 72–80 weeks . However , at four weeks of age , the Sirt6+/+Trp53+/- mice exhibited no significant differences from Sirt6+/+Trp53+/+ ( wild-type ) mice in either body size or body weight . The body weights of Sirt6-/-Trp53+/- mice ( both females and males ) increased steadily with age as did their wild-type ( WT ) littermates , although they always weighed less than their WT littermate mice ( Figure 2F and G ) . At 3 weeks of age , the Sirt6-/-Trp53+/- mice were more active than the Sirt6-/-Trp53+/+mice but were less active than their WT littermates . While comparing p53 expression levels in the cells and tissues of 3-week old mice , p53 protein level in Sirt6-/-Trp53+/- mice was observed to be decreased as compared to Sirt6-/-Trp53+/+ mice , but was more than that of Sirt6+/+Trp53+/+ mice ( Figure 1—figure supplement 1F ) . Next , we analyzed the GH/IGF1 axis in the mice by checking their serum IGF1 levels . As previously reported ( Mostoslavsky et al . , 2006 ) , the serum IGF1 levels were drastically reduced in Sirt6-/-Trp53+/+ mice as compared to Sirt6+/+Trp53+/+ mice at the age of around 24 days . Interestingly , the serum IGF1 levels in Sirt6-/-Trp53+/- mice were significantly rescued with respect to Sirt6-/-Trp53+/+ mice , but were less than that of their wild-type littermates ( Figure 2—figure supplement 1A ) . Also , the serum IGF1 levels of Sirt6-/-Trp53+/- mice were less than that of their wild-type littermates during their terminal stage ( 10-16 months ) ( Figure 2—figure supplement 1B ) . Apart from serum IGF1 levels , we also analyzed glucose levels in the mice . The Sirt6-/-Trp53+/+ mice displayed a minor drop in their serum glucose levels , while the compound mutant ( Sirt6-/-Trp53+/- ) mice exhibited an almost comparable glucose level to their wild-type littermates at the age of 24 days ( Figure 2—figure supplement 1C ) . Although Sirt6-/-Trp53+/- mice , in their terminal lifespan , had slightly reduced serum glucose levels than their wild-type littermates , the values did not reach significance ( Figure 2—figure supplement 1D ) . The dramatic extension of lifespan in Sirt6-deficient mice upon haploinsufficiency of Trp53 encouraged us to further examine whether the premature aging-associated abnormalities previously reported in Sirt6-deficient mice ( Mostoslavsky et al . , 2006 ) are rescued or not . As expected , at around 24 days of age , the sizes of the internal organs of Sirt6-/-Trp53+/- mice , such as spleen , thymus , kidney , liver , lungs and heart , were markedly larger than those of Sirt6-/-Trp53+/+ mice , although they were smaller than those of wild-type ( Sirt6+/+Trp53+/+ ) littermates ( Figure 3A ) . Lordokyphosis ( increased curvature of the spine ) , a prominent phenotype of Sirt6-/-Trp53+/+ mice ( Mostoslavsky et al . , 2006 ) , was significantly improved in Sirt6-/-Trp53+/-mice , irrespective of gender ( Figure 3B and Figure 3—figure supplement 1A ) . Colitis , a characteristic feature previously reported in Sirt6-deficient mice ( Mostoslavsky et al . , 2006 ) , was largely rescued in Sirt6-/-Trp53+/- mice when compared to the Sirt6-/-Trp53+/+ littermates ( Figure 3C ) . When senescence was evaluated , there was significantly less senescence-associated β-galactosidase activity in the organs from Sirt6-/-Trp53+/- mice , such as spleen , liver and kidneys , as compared to Sirt6-/-Trp53+/+ mice ( Figure 3D ) , thus suggesting a marked rescue of premature senescence in the compound mutant mice upon haploinsufficiency of Trp53 . Taken together , these results are in line with our observation of lifespan extension in Sirt6-deficient mice by haploinsufficiency of Trp53 , and further confirm that elevated p53 activity plays a significant role in accelerated aging and premature death in Sirt6-deficient mice . Regarding cancer incidence , some of the Sirt6+/+Trp53+/+ ( WT ) mice ( approximately 6–8% ) developed tumors at around 2–2 . 5 years , as expected . Approximately 26–28% of Sirt6+/+Trp53+/- mice developed tumors by 17–20 months , regardless of gender as already reported ( Jacks et al . , 1994; Tyner et al . , 2002 ) . The Sirt6-/-Trp53+/+ mice mostly died within 4 weeks of birth , and no incidence of cancer was observed . Interestingly , around 32% of the Sirt6-/-Trp53+/- ( compound mutant ) mice developed tumors during their terminal lifespan . However , the incidence of cancer in Sirt6-/-Trp53+/- mice was earlier ( 9-16 months ) than the Sirt6+/+Trp53+/- mice ( 17-20 months ) . The Sirt6+/+Trp53+/- mice mostly developed sarcoma by 17-20 months ( Figure 3—figure supplement 1B ) . However , the Sirt6-/-Trp53+/- mice developed other types of tumors , such as kidney tumor ( 19% ) , prostate tumor ( 17% ) , bladder tumor ( 16% ) , and pancreatic tumor ( 8% ) ( Figure 3—figure supplement 1C–F respectively ) , while only 3% of them exhibited incidence of sarcoma . Taken together , sirt6 depletion enhanced the onset and diversified the type of tumors in the Trp53+/- background , given that Sirt6 is a tumor suppressor itself ( Lerrer and Cohen , 2013; Sebastián et al . , 2012 ) . Apart from severe body wasting and loss of vigour in Sirt6-/-Trp53+/- mice in their terminal stage , some female Sirt6-/-Trp53+/- mice succumbed to the disorder of incontinence at around 14-16 months , while around 95% of the Sirt6-/-Trp53+/- male mice exhibited penile protrusion at around 9-11 months of age ( Figure 3—figure supplement 1G ) . Also , the Sirt6-/-Trp53+/- mice remained sterile throughout their lifespan . Stem cell depletion is one of the hallmarks of premature aging and heterozygosity of Trp53 has been reported to improve stem cell maintenance ( Amir et al . , 2017; Belle et al . , 2015; López-Otín et al . , 2013; Lu et al . , 2015 ) . We therefore sought to examine the status of mesenchymal and hematopoietic stem cells in the bone marrow of Sirt6+/+Trp53+/+ , Sirt6-/-Trp53+/+and Sirt6-/-Trp53+/- mice ( isolation and purification of bone marrow-derived stem cells have been described in the methods ) . Consistent with the age-associated decline of bone marrow stem cells , we observed a notable drop in the number of mesenchymal and hematopoietic stem cells derived from the bone marrow of both male and female Sirt6-/-Trp53+/+ mice ( Figure 4A–D ) . However , upon Trp53 haploinsufficiency , the percentage of mesenchymal and hematopoietic stem cells in the bone marrow of Sirt6-/-Trp53+/- mice ( both females and males ) was partially , yet significantly restored as compared to their littermate Sirt6-/-Trp53+/+ mice ( Figure 4A–D ) . It has been previously reported that CD4+CD8+ double-positive cells undergo drastic reduction in the thymus of Sirt6-/-Trp53+/+mice post 3 weeks of birth ( Mostoslavsky et al . , 2006 ) . Consistently , we also observed a marked decline in the thymic CD4+CD8+ double-positive cells of Sirt6-/-Trp53+/+ mice at around 24 days of age ( Figure 4E and F ) . However , there was a striking increase in the number of CD4+CD8+ double-positive cells in the thymus of Sirt6-/-Trp53+/- mice as compared to that in the Sirt6-/-Trp53+/+ littermates ( Figure 4E and F ) . We confirmed the previous observation that apoptotic responses are upregulated in the thymus of Sirt6-deficient mice ( Mostoslavsky et al . , 2006 ) ( Figure 4G ) . Interestingly , this was partially rescued in both thymus and bone marrow stromal cells ( BMSCs ) of Sirt6-/-Trp53+/- mice when compared to Sirt6-/-Trp53+/+ mice ( Figure 4G and H ) . In addition , there was a significant increase in the number of B220+ cells ( characterizing B cells across all developmental stages ) and CD11b+ cells ( representing monocytes ) in the spleen of Sirt6-/-Trp53+/- mice than that in Sirt6-/-Trp53+/+ littermates ( Figure 4I and J ) , thus indicating that haploinsufficiency of Trp53 rescues the total B cell count and the total monocyte count , respectively , in Sirt6-deficient scenario . To further investigate the mechanistic link between Sirt6 deficiency and p53 activation and understand how p53 is involved in the accelerated aging of Sirt6-deficient mice , we first tested if these two proteins physically interact . To this end , we performed reciprocal co-immunoprecipitation experiments to evaluate interactions between SIRT6 and p53 in HEK293 cells without any external DNA damage , since DNA damage-mediated p53 upregulation could taint the results on interaction . Via co-immunoprecipitation with specific antibodies against SIRT6 and p53 , we observed interaction between the two proteins at the endogenous level ( Figure 5A ) . To further confirm this interaction , we ectopically expressed FLAG-tagged SIRT6 and p53 in HEK293 cells individually , and could observe endogenous p53 and SIRT6 being pulled down in FLAG immunoprecipitates respectively ( Figure 5—figure supplement 1A and B ) . Next , to rule out the possibility of cell-specific effect on the interaction between SIRT6 and p53 , we used recombinant human SIRT6 ( rh SIRT6 ) and FLAG-p53 eluted from HEK293 cells for in vitro binding experiment and analyzed direct physical association between SIRT6 and p53 . Indeed , p53 could pull down SIRT6 in vitro , suggesting a direct interaction between SIRT6 and p53 ( Figure 5B ) . Acetylation is one of the major post-translational modifications regulating the stability and activity of p53 ( Marouco et al . , 2013; Reed and Quelle , 2014 ) . A previous study has indicated that SIRT6 , as a NAD+-dependent deacetylase , is not responsible for p53 deacetylation at lysine ( K ) 382 ( Michishita et al . , 2005 ) . However , given that multiple potential acetylation sites exist in p53 protein ( Marouco et al . , 2013; Reed and Quelle , 2014 ) , it is conceivable to speculate that SIRT6 may deacetylate p53 at sites other than K382 . To test this , we performed in vitro deacetylation assays using eluted FLAG-p53 ( purified from HEK293 cells ) and recombinant SIRT6 ( rh SIRT6 ) . As predicted , we observed that SIRT6 could deacetylate p53 in vitro ( Figure 5C ) . In line with the nature of SIRT6 as a NAD+-dependent deacetylase , SIRT6-mediated p53 deacetylation in vitro was abrogated in the absence of NAD+ ( Figure 5C and D ) . In addition , this deacetylation was largely diminished in the presence of the sirtuin inhibitor nicotinamide ( Figure 5—figure supplement 1C and D ) . In accordance with existing literature ( Michishita et al . , 2005 ) , we observed no alteration in p53 acetylation at lysine ( K ) 382 even with increasing concentration of rhSIRT6 in vitro ( Figure 5E ) . However , when checking the other two major sites of acetylation in the C-terminus of p53 , that is lysine 381 and 373 ( Gu and Roeder , 1997 ) , the pan-acetylation ( i . e . , total acetylation ) and acetylation at K381 of p53 was found to significantly reduce with increasing concentrations of SIRT6 ( Figure 5E and F ) . No significant change in the acetylation of p53 at K373 was observed in the presence of rhSIRT6 ( Figure 5E and F ) . Consistently , p53 acetylation at K381 was upregulated in SIRT6-deficient cells ( SIRT6 KO cells ) in comparison with mock CRISPR control cells , when assessed with western blotting ( Figure 5G ) . Again , no significant change in p53 acetylation at K382 was observed in SIRT6-deficient cells when compared with mock CRISPR control cells ( Figure 5G ) . In agreement with Western blotting results , immunofluorescence staining also showed that p53 acetylation was upregulated at K381 , but not K382 in SIRT6-deficient cells ( Figure 5H and Figure 5—figure supplement 1E ) . When p53 lysine 381 was mutated to arginine ( K381R , non-acetylatable mutant ) , p53 acetylation at this site was no longer detectable , thus confirming the epitope specificity of the antibodies against p53 acetylation at K381 ( Figure 5—figure supplement 1F ) . Additionally , expression of the FLAG-tagged wild-type ( WT ) form of SIRT6 attenuated the upregulated p53 K381 acetylation in SIRT6-deficient HEK293 cells ( Figure 5I ) . However , such attenuation in p53 acetylation at K381 was not observed with the ectopic expression of the catalytically inactive form of SIRT6 ( H133Y ) ( Figure 5J ) . Although 90–95% of the SIRT6 KO cells were stained with antibodies against p53 acetylated at K382 , ectopically expressed SIRT6 ( both active and catalytically inactive forms ) had no effect on p53 acetylation at K382 ( Figure 5—figure supplement 1G ) . Next , we analyzed immunoprecipitated FLAG-tagged WT p53 and K381R mutant p53 for acetylation in mock CRISPR control cells and SIRT6-deficient cells . Consistent with data shown in Figure 5G , pan-acetylation of WT p53 was increased in the SIRT6-deficient cells over control cells ( Figure 5K and L ) and the K381R mutation caused a significant reduction in pan-acetylation of p53 in both control and SIRT6-deficient cells ( Figure 5K ) . Surprisingly , the increase in pan-acetylation of p53 was largely diminished for K381R mutant in SIRT6-deficient cells as compared to control cells ( Figure 5K and L ) . These findings support the idea that the increased p53 pan-acetylation in the absence of SIRT6 is largely attributable to the acetylation status at lysine 381 of p53 ( Figure 5K and L ) . Also , endogenous p53 immunoprecipitated from SIRT6 KO HEK293 cells exhibited more acetylation at K381 as compared to mock CRISPR cells , and this increased acetylation of p53 at K381 went down with ectopic expression of wild-type ( WT ) SIRT6 ( Figure 5—figure supplement 1H ) . To further substantiate the observation of upregulation of p53 acetylation at K381 upon Sirt6 deficiency , we tested the same in cells and tissues of mice , and similarly observed upregulation of p53 acetylation at K381 in Sirt6-/-Trp53+/+ mice as compared to Sirt6+/+Trp53+/+ littermate mice ( Figure 5—figure supplement 1I and J ) . It is known that acetylation enhances the stability of p53 ( Gu and Zhu , 2012; Kruse and Gu , 2009 ) . Indeed , HEK293 cells lacking SIRT6 ( SIRT6 KO cells ) had approximately two-fold increases in the total p53 protein level when compared with mock CRISPR control cells ( Figure 6A and B ) . Immunofluorescence staining further confirmed the increased endogenous levels of p53 in SIRT6 KO cells as compared to control cells ( Figure 6C ) . Similarly , liver and kidney tissues from Sirt6-deficient mice exhibited a significant increase in p53 protein levels ( Figure 6D and E ) . This increase in p53 expression was also observed in Sirt6 null ( Sirt6-/- ) primary mouse embryonic fibroblasts ( passage P2 MEFs were used to avoid alterations because of serial passaging ) ( Figures Figure 6—figure supplement 1A and B ) . Treatment with the protein synthesis inhibitor cycloheximide further confirmed the enhanced stability of p53 in SIRT6-deficient cells ( Figures Figure 6—figure supplement 1C and D ) . However , we did not observe increased phosphorylation of p53 at serine 15 in SIRT6-deficient HEK293 cells compared with wild-type cells without irradiation ( Figure 6F ) . This phosphorylation of p53 robustly increased in both wild type and Sirt6-deficient cells upon irradiation ( Figure 6F ) . This suggests that the upregulation of p53 in SIRT6-deficient cells is unlikely a direct consequence of increased DNA damage in the absence of SIRT6 ( Figure 6F ) . Furthermore , we observed no significant increase in p53 mRNA levels in the absence of Sirt6 in HEK293 cells , MEF cells or mice tissues ( Figure 6G–J ) . Hence , it is likely that SIRT6 regulates p53 at the post-translational level . Since we observed that SIRT6 deacetylates p53 at lysine 381 and loss of SIRT6 conferred stability to p53 , we further sought to examine the importance of p53 acetylation at K381 on the stability of p53 . To address this , we generated another mutant construct of p53 , with lysine 381 mutated to glutamine ( K381Q ) by site-directed mutagenesis , which mimics the acetyl form of p53 at K381 . FLAG-tagged WT p53 , K381R mutant ( lysine to arginine , non-acetylatable mutant ) , and K381Q mutant ( lysine to glutamine , acetyl mimic mutant ) were ectopically expressed in HEK293 cells individually , followed by treatment with cycloheximide to examine the stability of different mutants of p53 . As predicted , we observed a significant drop in the stability of p53 K381R mutant , which was further intensified upon cycloheximide treatment ( Figure 6—figure supplement 2A and B ) . On the other hand , p53 K381Q mutant displayed improved stability , as evidenced by the comparable protein expression levels before and after cycloheximide treatment ( Figure 6—figure supplement 2A and B ) . These findings reinstate that loss of SIRT6 leads to upregulation of p53 acetylation at K381 , which further confers stability to p53 . In agreement with increase in the expression of senescent biomarker p16 in SIRT6 KO cells with respect to control cells , we also observed an increase in the expression of p16 in cells with ectopic expression of p53 K381Q mutant ( Figure 6—figure supplement 2C and D ) . This data suggests that hyperacetylation of p53 at K381 imparts senescence-like properties to cells , thus contributing to the premature senescence observed in Sirt6-deficient cells . Although recent studies have reported that p53 upregulates SIRT6 expression ( Jung et al . , 2016; Zhang et al . , 2014 ) , the mechanistic explanation of this regulation remains largely unclear . On the other hand , there is no existing literature about the regulation of p53 by SIRT6 . In this study , we show that loss of SIRT6 results in increased acetylation of p53 , leading to elevated apoptosis and senescence at the cellular level and to accelerated aging in mice . Haploinsufficiency of Trp53 significantly reduced premature senescence , substantially restored immune cell and stem cell populations , and dramatically extended the lifespan of Sirt6-deficient mice . These data identified a novel SIRT6-p53 axis in the regulation of senescence and aging . Our study reveals a direct link between SIRT6 and p53 , wherein the stability of p53 is regulated through SIRT6-mediated deacetylation of p53 at K381 . Hence , these findings not only establish p53 as a substrate for SIRT6 , but also identify SIRT6-mediated p53 deacetylation as a critical mechanism to suppress cellular senescence and organismal aging . Although we have identified lysine 381 of p53 as a major site of deacetylation by SIRT6 , we cannot exclude the possibility that SIRT6 might target other lysine residues of p53 for deacetylation or other post-translational modifications . This would require further investigation in the future . Sirt6-/-Trp53+/+ mice have been reported to die because of acute aging-associated degenerative processes , such as severe organ degeneration , body wasting , abruptly lowered serum IGF-1 levels , gut inflammation and chronic metabolic disorders ( Mostoslavsky et al . , 2006 ) . The Sirt6-/-Trp53+/- mice showed signs of severe body wasting during their terminal lifespan ( around 12-16 months ) . Around 32% of the Sirt6-/-Trp53+/- ( compound mutant ) mice also developed cancer during their terminal lifespan . Some Sirt6-/-Trp53+/- female mice succumbed to the disorder of incontinence , that is uncontrolled urination at around 14-16 months , while around 95% of Sirt6-/-Trp53+/- male mice exhibited penile protrusion at around 9-11 months of age ( Figure 1G ) . Taken together , the Sirt6-/-Trp53+/- mice died of disorders which are both overlapping with and distinct from Sirt6-/-Trp53+/+ mice . Although mixed genetic background has been reported to extend the lifespan of a percentage of Sirt6-deficient mice ( Peshti et al . , 2017 ) , all the analyses in our study have been done on littermate Sirt6-/-Trp53+/+ and Sirt6-/-Trp53+/- mice ( both on mixed backgrounds ) , where the Sirt6-/-Trp53+/+ mice had reduced lifespan of around 4 weeks and the Sirt6-/-Trp53+/- mice exhibited extended maximal lifespan over the littermate Sirt6-/-Trp53+/+ mice ( Figure 2A–E ) . This limits the possibility of mixed genetic background effects in our study . Moreover , the median lifespan for both male and female Sirt6-/-Trp53+/- mice on mixed background observed in our study is significantly longer than that of the reported Sirt6-/-Trp53+/+ mice on mixed background ( Peshti et al . , 2017 ) . This suggests that the observed lifespan extension of compound mutant mice is indeed attributable largely , if not all , to Trp53 heterozygosity . Moreover , Peshti et al . ( 2017 ) reported gender-specific differences in lifespan extension upon Sirt6 deficiency in mixed genetic background . Although we did observe that the maximal lifespan of the female Sirt6-/-Trp53+/- mice was slightly more than the male Sirt6-/-Trp53+/- mice ( Figure 2D and E ) , the differences were not as striking as reported , suggesting further lifespan extension in Sirt6-deficient mice by Trp53 haploinsufficiency reduces the percentage of gender-specific difference in lifespan . However , retinal disorders were prominently observed in our Sirt6-/-Trp53+/- mice similar with that reported in Peshti et al . ( 2017 ) , consistent with a previous report stating that Sirt6 plays critical roles in the maintenance of retinal functions ( Silberman et al . , 2014 ) . Another possibility cannot be eliminated that the very short lifespan of Sirt6 KO mice could result from oversensitivity to p53 activation associated with the genetic backgrounds used in the respective studies , thus resulting in enhanced cell death even at low DNA damage levels , thereby accelerating aging . Reducing Trp53 levels by haploinsufficiency likely reduces this effect and hence delays aging , thereby allowing longer lifespan . Regarding cancer incidence , the types of tumors exhibited by Sirt6-/-Trp53+/- and Sirt6+/+Trp53+/- mice were very distinct . Given that Sirt6 has already been established as a potent tumor suppressor by several independent studies ( Lerrer and Cohen , 2013; Sebastián et al . , 2012 ) , this explains as to why Sirt6-/-Trp53+/- mice exhibit an earlier onset of tumorigenesis and mostly develop tumors other than sarcoma . Hence , sirt6 depletion enhanced the onset and diversified the type of tumors in the Trp53+/- background . The dramatic lifespan extension observed in the mouse model reported in this study is significantly longer than any other rescue reported so far in mammalian aging models . The lifespan extension reported here is six times longer than the highest rescue of premature aging achieved so far in Sirt6-deficient mice ( Kawahara et al . , 2009 ) . This suggests that the activated p53 pathway may be one of the most critical mechanisms underlying the accelerated aging and premature death in the absence of Sirt6 . This further reinstates the importance of the SIRT6-p53 axis in cellular senescence and organismal aging . To also note here is that , though significant , the lifespan extension and phenotype amelioration by Trp53 haploinsufficiency in Sirt6-deficient mice represent a partial rescue when compared with the wild-type mice . Given that Sirt6 regulates a range of pathways such as NF-κB , IGF1 , AKT and others ( Dominy et al . , 2012; Kanfi et al . , 2012; Kawahara et al . , 2009; Pan et al . , 2016; Xiao et al . , 2010 ) , it is tempting to speculate that these pathways may work in orchestra to contribute to the severe premature aging in mice in the absence of Sirt6 . It is conceivable that Sirt6 null mice with Trp53 haploinsufficiency can serve as a better model in evaluating the contribution of other signaling pathways to the premature aging in the absence of Sirt6 . In addition , targeting multiple signaling pathways at the same time may further extend the lifespan in Sirt6-deficient mice . Given the recent report stating that inhibition of the interaction between FOXO4 and p53 rescues p53-mediated senescence and aging ( Baar et al . , 2017 ) , it would be interesting to analyze if a similar administration of FOXO4 inhibitor could ameliorate aging-associated abnormalities in Sirt6-deficient mice . Although the mammalian SIRT1 protein is also known to deacetylate p53 ( Langley et al . , 2002 ) , the lack of any significant rescue of the abnormalities of Sirt1-deficient mice by loss of p53 ( Kamel et al . , 2006 ) clearly suggests , when viewed alongside our results , distinct roles for SIRT1 and for SIRT6 in the modulation of p53 activity and aging/longevity . In addition , the upregulation of p53 stability and activity in Sirt6-null scenario and the severe premature aging observed in Sirt6-deficient mice , clearly suggest the insufficiency of endogenous SIRT1 in compensating the hyperactivity of p53 under Sirt6-deficient conditions . Apart from a better understanding of the contribution of different pathways regulated by SIRT6 to the aging process at the organismal level , the extended lifespan in Sirt6-deficient mice with Trp53 haploinsufficiency also provides an opportunity to understand the in vivo role for Sirt6 in development , particularly during the maturation as the compound mutant mice survived remarkably long enough with growth retardation ( Figure 2 ) . On the other hand , given both SIRT6 and p53 are potent tumor suppressors ( Donehower and Lozano , 2009; Sebastián et al . , 2012 ) and their regulatory roles in aging , this study opens a new window for future study in the control of choice between aging and cancer development by the balance and cross-talk of p53 and SIRT6 . p53 has been widely implicated in premature senescence and aging and is the focus of a large number of studies to develop therapies . For example , independent studies have reported the efficacy of p53 inhibitors in promoting neuroprotection against age-associated neurodegenerative diseases ( Culmsee et al . , 2001; Duan et al . , 2002; Zhu et al . , 2002 ) . Hence , identification of SIRT6 as a negative regulator of p53 unveils new avenues of research not only for basic scientists but also for those working to develop intervention and therapies against cancer and aging . HEK293 cells have been purchased from ATCC . HEK293 cells and mouse embryonic fibroblasts ( MEFs ) were cultured in DMEM supplemented with 10% fetal bovine serum ( FBS ) in 37°C incubators with 5% CO2 and atmospheric oxygen conditions . HEK293 SIRT6 KO cells were a kind gift from Dr . Baohua Liu ( Shenzhen University , China ) . Cell lines have been authenticated by short tandem repeat ( STR ) profile analysis and genotyping . Mycoplasm contamination were routinely examined by PCR . For analyzing effects of knocking down Trp53 ( alternatively denoted as p53 ) , we generated Sirt6-/-Trp53+/- compound mutant MEFs along with their WT ( Sirt6+/+Trp53+/+ ) and Sirt6 KO ( Sirt6-/-Trp53+/+ ) littermates at E12 . 5 . Mycoplasma contamination was analyzed by DNA staining with DAPI . Transfection was performed with X-tremeGENE HP DNA Transfection Reagent ( Roche , USA ) and Lipofectamine 3000 ( Invitrogen , USA ) . FLAG-tagged full length SIRT6 construct was obtained from addgene . Flag-tagged SIRT6 catalytically inactive mutant was kindly provided by Dr . Katrin Chua ( Stanford school of Medicine , USA ) . FLAG-p53 and HA-P300 constructs were provided by Dr . Zhenkun Lou ( Mayo clinic , USA ) . The point mutant constructs ( p53 K381R , p53 K381Q ) were generated using a site-directed mutagenesis kit ( Agilent Technologies , QuikChange II XL ) . Antibodies against rabbit anti SIRT6 , rabbit anti Acetyl p53 K373 , rabbit anti Acetyl p53 K382 were obtained from Cell Signaling . Antibodies against rabbit anti Acetyl p53 K381 and rabbit anti H3 were purchased from Abcam . Rabbit anti H3K9ac , mouse anti γ-H2AX , and rabbit anti-PAN acetyl lysine antibodies were purchased from Millipore ( Bedford , MA , USA ) . Rabbit anti H3K56ac antibody was purchased from Upstate . Anti p53 ( DO-1 ) , p53 ( FL-393 ) , and IgG antibodies were purchased from Santa Cruz ( Santa Cruz , CA , USA ) . Mouse anti FLAG M2 antibody was purchased from Sigma . PE anti-mouse CD105 and FITC anti-mouse CD34 antibodies were purchased from eBiosciences . PE anti-mouse CD11b , APC anti-mouse CD44 , PE-Cy5 anti-mouse CD8 , and FITC anti-mouse CD4 antibodies were purchased from Biolegend . PE anti-mouse Sca-1 , APC anti-mouse c-Kit , FITC anti-mouse CD31 , and PerCP anti-mouse B220 antibodies were purchased from BD pharmingen . G agarose beads were purchased from Invitrogen , USA and anti-FLAG M2 affinity beads were purchased from Sigma . Cycloheximide ( purchased from Sigma Aldrich ) was used at a working concentration of 150 µg/ml . Recombinant Human SIRT6 protein was expressed in BL21 ( DE3 ) strain using pET28a-sumo vector and was purified using Superdex200 gel-filtration column . SIRT6-mediated deacetylation was assayed on eluted p53 ( overexpressed as a FLAG-tagged protein along with HA-tagged P300 and immunoprecipitated from HEK293 cells using FLAG antibodies , followed by elution with 3X FLAG peptide from Sigma ) using the assay buffer as described in SIRT6 direct fluorescent screening assay kit by Cayman chemical ( USA ) . Eluted acetyl p53 was incubated with recombinant human SIRT6 ( rhSIRT6 ) for 45 min at 37°C , in the presence or absence of NAD+ and nicotinamide . Acetylation of p53 was detected using antibodies against PAN-acetyl lysine ( K ) , acetyl K381 , acetyl K382 and acetyl K373 . All animal works were performed with permission from local animal ethic committee ( CULATR ) and according to the guidelines and regulations . The Sirt6co floxed mutant mice have been purchased from Jackson laboratories and mated with β-actin-cre mice to generate Sirt6+/-Trp53+/+ mice . Sirt6-/-Trp53+/- compound mutant mice were generated by compound heterozygous mating strategy since Sirt6 KO mice die within a month and p53 KO mice develop tumors very early and die prematurely . Briefly , Sirt6+/-Trp53+/+ and Sirt6+/+Trp53+/- mice were bred to generate Sirt6+/-Trp53+/- mice . Then Sirt6+/-Trp53+/- mice were interbred to generate Sirt6-/-Trp53+/- mice . Litters containing WT ( Sirt6+/+Trp53+/+ ) , Sirt6 KO ( Sirt6-/-Trp53+/+ ) and compound mutant mice ( Sirt6-/-Trp53+/- ) were used for further analysis . Cells were harvested , washed with 1x PBS and resuspended in suspension buffer ( 0 . 1M NaCl , 10 mM Tris-HCl , pH 7 . 5 , 1 mM EDTA , 1 mM DTT , pH 8 . 0 , protease inhibitors , phosphatase inhibitors ) followed by addition of an equal volume of laemmli buffer ( 0 . 1 M Tris-HCl , pH 7 . 0 , 4% SDS , 20% glycerol , 1 mM DTT , protease inhibitors , phosphatase inhibitors ) and boiled for 10 min . Tissue samples were minced and dounced thoroughly with suspension buffer , followed by addition of an equal volume of laemmli buffer and boiled immediately for 15 min . Western blotting was done as previously illustrated ( Liu et al . , 2005 ) . Relative band intensity was measured by Image J and normalized to corresponding controls . Statistical analysis was performed using at least three independent immunoblots , which were quantified and two-tailed student's T test was used for calculating P values . Co-immunoprecipitation analyses were performed as described below . Briefly , cells were lysed with pre-chilled RIPA buffer containing 250 mM or 500 mM NaCl , protease inhibitors and phosphatase inhibitors . Primary antibodies or appropriate control IgGs were added to the lysates and incubated for 2 hr at 4°C on a rocking platform followed by addition of agarose beads and incubation was done O/N at 4°C . The beads were washed thrice with RIPA buffer ( 500 mM NaCl ) , resuspended with laemmli buffer and boiled for 10 min . The protein suspension was collected by centrifugation and stored at −80°C for western analysis . Total RNA from cells or tissues ( after douncing ) was isolated using Trizol ( Invitrogen ) and 2 µl of the total extracted RNA was used for reverse-transcription reaction to generate cDNA using PrimeScript RT mastermix from Takara . Relative expression of the target genes was measured by qPCR and were normalized against respective Gapdh expression levels . Cells were grown on chamber slides , fixed with 4% paraformaldehyde ( in PBS ) , washed once in PBS , washed again with PBTr ( PBS containing 0 . 1% Triton X-100 ) and blocked with 5% serum ( FBS ) in PBTr for 1 hr at RT ( room temperature ) . Primary antibody was then diluted in PBTr and incubated O/N at 4°C . The slides were washed three times in PBTr , incubated with FITC- , TRITC-coupled secondary antibodies , or with Alexa-fluor 488/562 ( donkey anti-rabbit ) , diluted in PBTr for 60–75 min at RT , washed three times with PBTr , then two times with PBS followed by mounting with SlowFadeÂ Gold antifade reagent with DAPI ( Invitrogen , USA ) , sealed with nail polish and subjected for confocal microscopic analysis at room temperature . Images were obtained using 63X , 1 . 4 NA oil objective ( Carl Zeiss LSM 700 inverted confocal microscope equipped with ZEN 2010 software version 6 . 0 . 0 . 309 ) with 405 nm , 488 nm and 555 nm laser illumination ( standard excitation and emission filter sets ) . Images were processed using ZEISS ZEN lite software . Senescence-associated β-galactosidase staining kit from Cell Signaling was used . Tissues were fixed with gelatin and cryopreserved before cutting sections . These cryosections were then immersed in 1x PBS and incubated at 37˚C for 20–30 min , followed by a wash with 1x PBS . Then the slides were incubated with fixative solution for 10–15 min , washed one with 1x PBS , and incubated with β-galactosidase staining solution for O/N at 37˚C in dark . The slides were then washed once with 1x PBS and mounted with SlowFadeÂ Gold antifade reagent ( Invitrogen ) . For cells , 105 cells were plated onto six well plates , grown O/N in 37˚C incubator , followed by aspiration of growth medium and washing twice with 1x PBS . The rest of the procedures were similar to that of staining of slides as mentioned above . Serum glucose and IGF-1 levels in mice were assayed using kits from Abcam and R and D Systems respectively , following the manufacturer's protocols . 107 bone marrow cells were resuspended in 1x RBC lysis solution and kept on ice for 5 min , followed by centrifugation at 500 g for 5 min and washing with ice cold 1x PBS . The bone marrow stromal cells were then stained with 0 . 5 µl each of PE-Sca1 , APC-cKit and FITC-CD34 for hematopoietic stem cell profile analysis and 0 . 5 µl each of PE-CD105 , APC-CD44 and FITC-CD31 for mesenchymal stem cell analysis . Only Sca-1+c-Kit+CD34- cells were counted as hematopoietic stem cells and only CD105+CD44+CD31- cells were counted as mesenchymal stem cells in flow cytometric analysis . Mono-stained cells and cells stained with isotype controls were also analyzed simultaneously . Staining was done for 20–30 min on ice in dark , followed by washing with pre-chilled 1x PBS , and resuspension in 1x PBS and then taken for FACS analysis using BD FACSCantoII Analyzer . Similarly , thymic cells were stained with 0 . 5 µl of FITC-CD4 and PE/Cy5-CD8 , keeping mono-stained and isotype control stained cells following the protocol mentioned above . Also , splenic cells were stained with 0 . 5 µl of PerCP-B220 and PE-CD11b separately keeping isotype controls , and similar procedures were followed to prepare for FACS analysis . Analysis of apoptosis was done using TACS Annexin V-Biotin Apoptosis Detection kit from R and D systems . Cells from thymus and bone marrow stromal cells from tibia and femur of mice were flushed out using ice-cold 1x PBS . 106 cells were washed once with ice cold 1x PBS , stained with propidium iodide and Annexin-V Biotin for 15 min in dark at RT , followed by centrifugation at 500 g for 5 min . Then the cells were incubated with FITC labelled Streptavidin antibody ( BD Pharmingen , BD Biosciences , San Jose , USA ) for 15 min in dark at RT , followed by centrifugation at 500 g for 5 min , resuspension in pre-chilled 1x PBS and flow cytometric analysis . After sacrificing the mice by cervical dislocation , both femurs and tibias were dissected , their ends were cut with a sharp and sterile scissor , and the bone marrow stromal cells were flushed out with pre-chilled 1x PBS using a 231/2 needle and were passed through 70-µm cell strainer for filtering out the clustered cells . The viable cell number was then calculated using hemocytometer after dilution with Trypan blue . Thymic and splenic cells were also isolated by flushing the thymus and spleen of mice using pre-chilled 1x PBS and cell number was calculated in a similar way . Measurement of cell viability was performed using TACS Cell Proliferation Assay Kit ( R and D systems ) . Briefly , around 5000 MEF cells were seeded onto 96 well plates in triplicates and grown for 48 hr . 10 µl of MTT reagent was added to each well , keeping triplicate controls of non-treated blank sets for each cell type . The cells were placed back in 37˚C incubator for 4 hr , and development of purple coloration was monitored . Then 100 µl of detergent reagent was added to each well , and the plate was left covered in dark for 2–4 hr , followed by measurement of absorbance at 570 nm in a microplate reader . Average values of blanks were subtracted from average values from triplicate readings and graph was plotted to analyze the same . X-ray imaging of the whole body of mice was done using UltraFocus1000 ( fully-shielded X-ray cabinet ) from Faxitron . Quantifications have been represented as mean ± SEM , n >= 3 for all experiments performed . Statistical significance was analyzed by two-tailed unpaired Student's t-test and one-way ANOVA using GraphPad Prism software . Kaplan-Meier Survival curves have been plotted using GraphPad Prism software , and Log-rank ( Mantel-Cox ) test has been employed to calculate statistical significance . Since our sample sizes varied from 3 to 6 , it was difficult to conclude whether normal distribution was achieved or not . Given that parametric tests are more sensitive and powerful in determining statistically significant differences which non-parametric tests might overlook , we either used Student's t-test or ANOVA to calculate statistical significance in our results .	Almost without exception , mammals age as they grow older . Older mammals are at greater risk of diseases like cancer , and have fewer stem cells that would otherwise help to keep their organs healthy . Some of the proteins that regulate and impact upon aging have been identified . One of these is an enzyme called SIRT6 , which is thought to promote longevity . Mice without any SIRT6 suffer from severe premature aging , and rather than living up to two years like normal mice , SIRT6-deficient mice die within one month of their birth . The mutant mice also lose stem cells and exhibit signs of organ degeneration and body wasting . Another protein called p53 is well known for having the opposite effect to SIRT6: it accelerates aging and helps to prevent tumor growth . However , it was unclear if p53 is also involved in the processes that lead to the premature death of mice without SIRT6 . Now , Ghosh et al . report that mouse cells and tissues without SIRT6 have more p53 compared to control samples . Biochemical experiments showed that the SIRT6 and p53 proteins physically interact , and that SIRT6 could use its enzymatic activity to remove a chemical modification , called acetylation , from p53 . Without this specific acetylation , p53 became less stable and its levels dropped . Consequently , p53 was stabilized in the SIRT6-deficient cells . When Ghosh et al . deleted one copy of the gene that codes for p53 in SIRT6-deficient mice , mutant mice that had before only lived for a month now lived for up to sixteen months . Additionally , the mice were healthier , showing fewer signs of aging: for example , they had more immune cells and stem cells , straighter spines , and showed less gut inflammation and less body wasting . These findings suggest that SIRT6 does indeed inhibit p53 to counteract the normal aging process . Future experiments may explore if this regulation also holds true in human cells . Detailed knowledge of these molecular interactions could also open up more research into therapies against cancer and aging .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "biochemistry", "and", "chemical", "biology" ]	2018	Haploinsufficiency of Trp53 dramatically extends the lifespan of Sirt6-deficient mice
Centrosomes comprise a pair of centrioles surrounded by pericentriolar material ( PCM ) . The PCM expands dramatically as cells enter mitosis , but it is unclear how this occurs . In this study , we show that the centriole protein Asl initiates the recruitment of DSpd-2 and Cnn to mother centrioles; both proteins then assemble into co-dependent scaffold-like structures that spread outwards from the mother centriole and recruit most , if not all , other PCM components . In the absence of either DSpd-2 or Cnn , mitotic PCM assembly is diminished; in the absence of both proteins , it appears to be abolished . We show that DSpd-2 helps incorporate Cnn into the PCM and that Cnn then helps maintain DSpd-2 within the PCM , creating a positive feedback loop that promotes robust PCM expansion around the mother centriole during mitosis . These observations suggest a surprisingly simple mechanism of mitotic PCM assembly in flies . Centrosomes help regulate many cell processes , including cell shape , cell polarity , and cell division ( Doxsey et al . , 2005; Bettencourt-Dias and Glover , 2007 ) , and centrosome defects have been implicated in several human pathologies ( Nigg and Raff , 2009; Zyss and Gergely , 2009 ) . Centrosomes are the major microtubule ( MT ) -organising centres ( MTOCs ) in many animal cells . They form when centrioles assemble a matrix of pericentriolar material ( PCM ) around themselves . Several hundred proteins are concentrated in the PCM , including many MT-organising proteins , cell-cycle regulators , and checkpoint and signalling proteins ( Müller et al . , 2010 ) ; thus , the centrosome appears to function as an important co-ordination centre in the cell ( Doxsey et al . , 2005 ) . Although centrioles usually organize only small amounts of PCM in interphase cells , the PCM expands dramatically as cells prepare to enter mitosis—a process termed centrosome maturation ( Khodjakov and Rieder , 1999 ) . Many proteins have been implicated in mitotic PCM assembly . These include ( 1 ) centriole-associated proteins , such as ‘Asl/Cep152’ ( Bonaccorsi et al . , 1998; Varmark et al . , 2007; Dzhindzhev et al . , 2010 ) and ‘Sas-4/CPAP’ ( Cho et al . , 2006; Gopalakrishnan et al . , 2011 ) , ( 2 ) proteins that have a centriole-associated fraction and a fraction that spreads out into the PCM , such as ‘Pericentrin/D-PLP’ ( Martinez-Campos et al . , 2004; Zimmerman et al . , 2004; Fu and Glover , 2012; Lawo et al . , 2012; Mennella et al . , 2012 ) and ‘DSpd-2/Cep192’ ( Pelletier et al . , 2004; Dix and Raff , 2007; Gomez-Ferreria et al . , 2007; Giansanti et al . , 2008; Zhu et al . , 2008; Joukov et al . , 2010; Decker et al . , 2011; Joukov et al . , 2014 ) , ( 3 ) proteins that reside in the PCM , such as ‘Cnn/Cdk5Rap2’ ( Megraw et al . , 1999; Lucas and Raff , 2007; Fong et al . , 2008; Barr et al . , 2010; Conduit et al . , 2010 ) , ‘DGp71WD/NEDD1’ ( Haren et al . , 2006 , 2009; Lüders et al . , 2006; Manning et al . , 2010 ) , and ‘γ-tubulin’ ( Sunkel et al . , 1995; Hannak et al . , 2002 ) , and ( 4 ) mitotic protein kinases , such as ‘Polo/Plk1’ and ‘Aurora A’ ( Barr and Gergely , 2007; Petronczki et al . , 2008 ) . In recent super-resolution microscopy studies , several of these proteins appeared to be highly organized around interphase centrioles , but the organisation of proteins within the extended mitotic PCM was much less apparent ( Fu and Glover , 2012; Lawo et al . , 2012; Mennella et al . , 2012; Sonnen et al . , 2012 ) . It has long been thought that the mitotic PCM is assembled on an underlying scaffold structure ( Dictenberg et al . , 1998; Schnackenberg et al . , 1998 ) . We recently showed that Drosophila Centrosomin ( Cnn ) can form such a scaffold around centrioles and that this scaffold is assembled from the inside out ( Conduit et al . , 2014 ) : Cnn molecules continuously incorporate into the scaffold around the centrioles and the scaffold then fluxes slowly outward , away from the centrioles . This inside out assembly mechanism could be important , as it potentially allows the assembly of the mitotic PCM to be regulated by the centrioles . Many PCM proteins , however , can be recruited to mitotic centrosomes in the absence of Cnn , albeit at reduced levels ( Lucas and Raff , 2007 ) , suggesting that at least one other protein must be able to form a scaffold around centrioles that can recruit other PCM components . We reasoned that such a scaffold might also be assembled from the inside out . To identify such a protein ( s ) , we analyzed the dynamic behaviour of the eight Drosophila centrosomal proteins that , in addition to Cnn , have been most strongly implicated in mitotic PCM recruitment: Asl , Sas-4 , D-PLP , DSpd-2 , γ-tubulin , DGp71WD , Polo , and Aurora A . We found that only DSpd-2 behaves like Cnn , as it incorporates into the PCM close to the centrioles and then spreads slowly outward to form a scaffold-like structure that recruits other PCM components . Importantly , in the absence of either DSpd-2 or Cnn , PCM recruitment is diminished , but in the absence of both proteins , it is abolished . We show that Asl appears to initiate the recruitment of DSpd-2 and Cnn exclusively to the mother centrioles; DSpd-2 then helps to recruit more Cnn , while Cnn helps to maintain DSpd-2 within the PCM , thus creating a positive feedback loop that promotes the scaffold assembly . Thus , mitotic PCM assembly appears to be a surprisingly simple process in flies: Asl initiates the recruitment of Spd-2 and Cnn to mother centrioles , and these proteins then assemble into scaffolds that spread out from the mother centriole and form a platform upon which most , if not all , other PCM proteins ultimately assemble . We used spinning disk confocal microscopy to perform fluorescence recovery after photobleaching ( FRAP ) experiments in combination with radial-profiling to quantify the spatio-temporal dynamics of various GFP-fusion proteins in Drosophila syncytial embryos . We selected the nine proteins , including Cnn , which have been most strongly implicated in the PCM recruitment in flies: Asl-GFP , AurA-GFP , GFP-Cnn , DGp71WD-GFP , D-PLP-GFP , DSas-4-GFP , DSpd-2-GFP , γ-tubulin-GFP , and Polo-GFP ( Mennella et al . , 2013 ) . We assessed the expression level of these fusion proteins relative to their endogenous proteins by Western blotting ( Figure 1—figure supplement 1 ) . Prior to photobleaching , the fusion proteins displayed different centrosomal distributions ( Figure 1A ) . Asl-GFP and DSas-4-GFP are known to be closely associated with centrioles ( Fu and Glover , 2012; Mennella et al . , 2012 ) , and their fusions were tightly localized in the centre of the PCM; they exhibited a fluorescence intensity profile similar to that of sub-resolution ( 170 nm ) beads ( Figure 1A ) , indicating that their true distribution was below the resolution of our microscope system . The other proteins were all distributed more broadly throughout the PCM to varying extents . 10 . 7554/eLife . 03399 . 003Figure 1 . Centrosomal DSpd-2 displays an unusual dynamic behaviour . ( A ) Graph shows the centrosomal fluorescence intensity profiles of various centrosomal proteins , along with the profile of 170 nm sub-resolution beads . ( B and C ) Images show the FRAP behaviour of AurA-GFP ( B ) and DSpd-2-GFP ( C ) ; time before and after photobleaching ( t = 0 ) is indicated . Note how AurA-GFP fluorescence appears to recover evenly throughout the region it originally occupied , whereas DSpd-2-GFP fluorescence appears to initially recover only in the centre of the PCM and then spread outward . ( D–G ) Quantification of the recovery dynamics of AurA-GFP ( D and E ) and DSpd-2-GFP ( F and G ) . Graphs show the average fluorescence intensity profile of at least 10 centrosomes at the selected time-points after photobleaching: ( D ) and ( F ) show the pre-bleached profiles ( blue lines ) and successive ‘raw’ recovery profiles ( various shades of red ) , whereas ( E ) and ( G ) show the pre-bleached profiles and successive normalized recovery profiles ( various shades of pink/purple—normalized so that their peak intensity is equal to the peak intensity of the pre-bleached profile ) . The normalized recovery curves of AurA-GFP are essentially identical to the pre-bleached profile at all time-points ( E ) , while for DSpd-2-GFP they are initially narrower and spread outward over time ( G ) . ( H and I ) Schematics illustrate the dynamic behaviour of AurA-GFP ( and most other PCM components ) ( H ) and DSpd-2-GFP ( and GFP-Cnn ) ( I ) . Cytoplasmic AurA-GFP molecules exchange with binding sites spread throughout the PCM ( H ) , whereas DSpd-2-GFP molecules are recruited by binding sites located close to the centrioles; once released from these binding sites the molecules spread slowly outward into the more peripheral regions of the PCM ( I ) . See also Figure 1—figure supplements 1 and 2 , and Video 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 00310 . 7554/eLife . 03399 . 004Figure 1—figure supplement 1 . An analysis of the expression levels of several GFP-tagged centrosomal proteins . Western blots compare the levels of various centrosomal proteins ( as indicated ) from wild-type embryo extracts ( lanes 1 and 3 ) or from extracts of embryos expressing a GFP-fused version of the protein ( lanes 2 , 4 , and 5 ) . Lanes 1 to 4 were probed with antibodies raised against the centrosomal protein in question; lane 5 was probed with anti-GFP antibodies . Lanes 1 and 2 were loaded with 5 μl of extract; lanes 3 to 5 were loaded with 10 μl of extract . The bottom panel shows a Western blot for GAGA factor , which acts as a loading control ( all blots were controlled in this way , but only one example is shown here ) . GFP-Cnn , DSpd-2-GFP , and Asl-GFP were expressed in a null mutant background; the other GFP-fusions were expressed in a wild-type background . Note how GFP-Cnn and D-PLP-GFP appear to be expressed at slightly higher levels than the endogenous protein , but that the other fusion proteins are expressed either at similar or at slightly lower levels than the endogenous protein . We previously showed that increasing the cytoplasmic concentration of GFP-Cnn increases the rate at which it is incorporated into the PCM , but does not change the inside out incorporation behaviour ( Conduit et al . , 2010 ) . Note also that GFP-Cnn and DSpd-2-GFP are overexpressed and underexpressed , respectively , but both display a similar dynamic behaviour ( Figure 1F , G , Figure 1—figure supplement 2M , N ) , again suggesting that the expression level of a GFP-fusion protein does not significantly affect the mode in which it is incorporated into the PCM . We have not been able to analyze the relative expression level of Polo-GFP , as we were unable to get anti-Polo antibodies to work on Western blots . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 00410 . 7554/eLife . 03399 . 005Figure 1—figure supplement 2 . An analysis of the dynamic behaviour of several GFP-tagged centrosomal proteins . Quantification of the FRAP recovery dynamics of DGp71WD-GFP ( A and B ) , Polo-GFP ( C and D ) , γ-tubulin-GFP ( E and F ) , Asl-GFP ( G and H ) , DSas-4-GFP ( I and J ) , D-PLP-GFP ( K and L ) , and GFP-Cnn ( M and N ) . Graphs show the average fluorescence intensity profile of ≥10 centrosomes ( ‘Materials and methods’ ) at selected time-points after photobleaching ( t = 0 ) . Graphs on the left of each panel display pre-bleached profiles ( blue lines ) and ‘raw’ recovery profiles ( various shades of red lines ) . Graphs on the right of each panel display pre-bleached profiles and recovery profiles ( various shades of purple lines ) that have all been normalized so that their peak intensity is equal to the peak intensity of the pre-bleach profile . Note how the normalized recovery profiles of DGp71WD-GFP ( B ) , Polo-GFP ( D ) , and γ-tubulin-GFP ( F ) are very similar to their pre-bleached profiles , indicating that they are all incorporated evenly throughout the PCM domain they originally occupied . This is also true for Asl-GFP ( H ) and DSas-4-GFP ( J ) , but these proteins have the same distribution as sub-resolution beads ( Figure 1A ) , so their true distribution cannot be resolved on this microscope system . The recovery dynamics of D-PLP-GFP ( K and L ) are complicated by the presence of a slow-exchanging centriole fraction and a fast-exchanging PCM fraction ( Martinez-Campos et al . , 2004 ) , which cannot be properly distinguished at this resolution . It appears that the slow recovery rate of the centriole fraction compared to the PCM fraction causes the normalized recovery profiles to be wider than the pre-bleached profile . It can be seen , however , that the PCM fraction of D-PLP shows no sign of outward movement through the PCM ( L ) . The normalized recovery profiles of GFP-Cnn ( N ) are narrower than the pre-bleached profile and spread outwards over time , reflecting the fact that Cnn molecules are initially incorporated only in the centre of the PCM and then move slowly outward ( Conduit et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 005 After photobleaching , the Asl-GFP , Sas-4-GFP , AurA-GFP , DGp71WD-GFP , γ-tubulin-GFP , and Polo-GFP fluorescence all recovered at different rates , but each protein appeared to recover evenly throughout the domain that it originally occupied ( Figure 1B , D; Figure 1—figure supplement 2; Video 1A , B ) . This even recovery was confirmed when the recovery profiles were normalized so that their peak fluorescence intensity at each time-point equalled one; this showed that , at all time-points , the shape of each normalized recovery curve closely matched the shape of its respective pre-bleached profile ( Figure 1E , Figure 1—figure supplement 2; Video 1C ) . We conclude that most PCM proteins are recruited to centrosomes by binding sites that are already distributed throughout the PCM volume that each protein occupies ( Figure 1H ) . These observations strongly support the idea that most PCM proteins are recruited by an underlying scaffold structure . 10 . 7554/eLife . 03399 . 006Video 1 . DSpd-2-GFP molecules are initially incorporated into the centre of the PCM . ( related to Figure 1 ) . All Videos shown here are maximum intensity projections of image stacks . These videos illustrate the dynamic behaviour of AurA-GFP ( A–C ) or DSpd-2-GFP ( D–F ) at centrosomes in Drosophila embryos . Time before and after photobleaching ( t = 0 s ) is shown at the top right of ( C ) and ( F ) . The graphs in ( B ) and ( E ) are the line profiles representing the distribution of the AurA-GFP and the DSpd-2-GFP centrosomal fluorescence , respectively , through time: the blue lines represent the pre-bleached profiles , and the red lines represent the recovering profiles . Note how the AurA-GFP fluorescence signal appears to recover evenly throughout the PCM domain it originally occupied , whereas the DSpd-2-GFP fluorescence signal appears to initially recover in the centre of the PCM and then spreads outwards . This is most clearly seen in ( C ) and ( F ) where the recovery profiles at all time-points have been normalized so that their peak intensities are equal to 1 ( purple lines ) . Note how the normalized AurA-GFP recovery profiles all closely overlap with the pre-bleached profile , whereas the normalized DSpd-2-GFP recovery profiles are all narrower than the pre-bleached profile and spread out over time . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 006 We note that the dynamics of D-PLP-GFP are complicated by the presence of two distinct centrosomal fractions of D-PLP: one fraction tightly localized at the centriole that turns over slowly , and another that localizes to the PCM and turns over rapidly ( Martinez-Campos et al . , 2004 ) ( Figure 1—figure supplement 2K , L ) . Nevertheless , it was clear from recovery profiles that the PCM fraction of D-PLP-GFP did not initially recover centrally and then move outwards , but rather recovered throughout the PCM volume . Unlike the other PCM components we tested , DSpd-2-GFP fluorescence initially recovered only in the central region of the PCM and then gradually began to recover in the more peripheral regions ( Figure 1C , F , G; Video 1D–F ) . This suggested that , like Cnn ( Conduit et al . , 2014; Figure 1—figure supplement 2M , N ) , DSpd-2 is recruited to binding sites that are only located close to the centrioles; once released from these sites , it then spreads slowly outward into the more peripheral regions of the PCM ( Figure 1I ) . A comparison of DSpd-2-GFP recovery kinetics in the central and peripheral regions strongly supported this interpretation ( Figure 2A–C ) . The central region of DSpd-2-GFP fluorescence exhibited a typical , logarithmic-shaped , FRAP recovery curve , with a fast initial recovery rate that gradually slowed over time ( green line , Figure 2B ) . In contrast , the peripheral regions exhibited an unusual FRAP recovery curve , with a slow initial recovery rate that gradually increased over time ( solid red line , Figure 2B ) ; this was best seen when the peripheral recovery curve was normalized so that the pre-bleached signal equalled one ( dotted red line , Figure 2B ) . These unusual recovery dynamics at the periphery of the PCM can most easily be explained if DSpd-2-GFP molecules are constantly moving from the centre of the centrosome to the periphery: as the number of fluorescent molecules in the centre increases with time after photobleaching , so the number of fluorescent molecules moving from the centre into the periphery gradually increases , explaining why the peripheral recovery rate speeds up over time . 10 . 7554/eLife . 03399 . 007Figure 2 . DSpd-2-GFP molecules spread away from the centrioles . ( A ) Graph displays the pre-bleached profile of DSpd-2-GFP at centrosomes in Drosophila embryos ( blue line , average of 10 centrosomes ) . Boxes highlight the central region of the PCM ( green box ) and peripheral regions of the PCM ( red boxes ) that were analyzed in a FRAP experiment . ( B ) Graph displays the average fluorescence intensity through time in the centre ( green line ) and periphery ( red and dotted red lines ) of the PCM after photobleaching . Arrows indicate times of photobleaching . The dotted red line represents the peripheral recovery after it has been normalized so that its initial pre-bleached value is equal to the initial pre-bleached value of the central recovery curve . Note how the central DSpd-2-GFP fluorescence recovery exhibits a typical logarithmic-shaped FRAP curve , with a fast initial rate that slows over time . In contrast , peripheral DSpd-2-GFP fluorescence recovery exhibits a very unusual behaviour as it is initially slow and then speeds up over time . This strongly suggests that DSpd-2-GFP molecules move from the centre to the periphery of the PCM ( see main text ) . ( C ) Graph compares the initial recovery kinetics in the central ( green lines ) or peripheral ( red lines ) regions of the PCM after a first ( light lines ) and then second ( dark lines ) photobleaching event . The second photobleaching event took place when the rate of recovery in the centre was slow and the rate of recovery in the periphery was fast ( see t = 200 s in B ) . Note that after the second bleaching event the central recovery returned to its original fast rate and the peripheral recovery returned to its original slow rate , showing that the increasing rate of recovery in the periphery was not due to peripheral binding sites gradually exchanging faster over time . ( D–G ) Graphs display the pre-bleached profiles ( D and F ) and recovery kinetics ( E and G ) of AurA-GFP ( D and E ) and Polo-GFP ( F and G ) at centrosomes in embryos ( average of 11 centrosomes ) in the same format as shown for DSpd-2-GFP ( A and B ) . Note how the peripheral recovery curves of AurA-GFP ( red and dotted red lines in E ) and Polo-GFP ( red and dotted red lines in G ) both have a similar shape to their central recovery curves , indicating that the unusual peripheral kinetics of DSpd-2-GFP ( red and dotted red lines in B ) are not observed with other proteins that have a similar distribution around the centrioles . Error bars = SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 007 We previously used super-resolution structured illumination microscopy ( 3D-SIM ) in living fly embryos to show that the GFP-Cnn scaffold radiates away from the centrioles and forms large projections that extend outwards along the centrosomal MTs ( Conduit et al . , 2014; Figure 3A ) . DSpd-2-GFP had a similar distribution with several spoke-like projections extending from a central ring ( Figure 3C ) , although , in agreement with our radial profiling data , much less DSpd-2 extended into the more peripheral regions of the PCM ( Figure 3C ) . Two-colour 3D-SIM in living embryos confirmed that DSpd-2-GFP and mCherry-Cnn extensively overlapped in the more central regions of the PCM ( Figure 3E , Figure 3—figure supplement 1 ) and revealed that the more peripheral Cnn projections usually contained small amounts of DSpd-2 ( arrowheads , Figure 3E and Figure 3—figure supplement 1B–D ) . 10 . 7554/eLife . 03399 . 008Figure 3 . DSpd-2-GFP appears to form scaffold-like structure around the centrioles that partially co-localizes with the Cnn scaffold . ( A–D ) 3D-SIM images of centrosomes from embryos expressing either GFP-Cnn ( A and B ) or DSpd-2-GFP ( C and D ) where the MTs are either present ( A and C ) or have been depolymerized by colchicine injection ( B and D ) . ( A ) In untreated embryos , large projections of GFP-Cnn extend outwards ( red arrowheads ) from a central hollow ( red arrow ) , which presumably contains the mother centriole . ( B ) After MT depolymerisation , the GFP-Cnn scaffold collapses into a largely amorphous structure; presumably , the molecular detail of the scaffold cannot be resolved even at this high resolution . The slightly larger central ‘hollow’ in the GFP-Cnn signal ( red arrow ) likely reflects the ability of the Cnn molecules to move a short distance away from the centre of the centrosome in the absence of microtubules ( Conduit et al . , 2014 ) ; these molecules then get ‘trapped’ in the more peripheral regions of the PCM , as they cannot efficiently leave the centrosome in the absence of MTs ( Conduit et al . , 2014 ) . ( C ) In untreated embryos , DSpd-2-GFP appears as a series of spoke-like projections ( red arrowheads ) that extend away from a central ring ( red arrow ) , which presumably surrounds the mother centriole; some of these projections weakly extend into the peripheral PCM . ( D ) After MT de-polymerisation , DSpd-2-GFP retains a large degree of its structure: there is a clear central ring ( red arrow ) with several spoke-like projections extending outwards ( red arrowheads ) ; these projections , however , no longer appear to extend into the more peripheral regions of the PCM . ( E ) Two-colour 3D-SIM images of centrosomes in untreated embryos co-expressing DSpd-2-GFP ( green ) and mCherry-Cnn ( red ) . The panels on the right are enlargements of the boxed centrosome in the panel on the left; note the clear hollow in the centre of the centrosome ( arrows ) , and how the mCherry-Cnn signal extends further away from the centrioles than the DSpd-2-GFP signal , although weak DSpd-2-GFP fluorescence can often be observed in the same region as the more peripheral mCherry-Cnn ( arrowheads ) . See also Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 00810 . 7554/eLife . 03399 . 009Figure 3—figure supplement 1 . An approximate quantification of mCherry-Cnn and DSpd-2-GFP co-localisation . The mCherry-Cnn ( red ) and DSpd-2-GFP ( green ) fluorescence intensity along the three lines that run through a 3D-SIM image of a centrosome in ( A ) are plotted on the graphs in ( B ) , ( C ) , and ( D ) . Green lines represent the fluorescent intensity of DSpd-2-GFP , and red lines represent the fluorescent intensity of mCherry-Cnn . The graphs have been normalized so that the peak value in the central region of the PCM is equal to 1 . Arrows within the graphs indicate the central regions of the centrosome where mCherry-Cnn and DSpd-2-GFP extensively co-localize . Arrowheads indicate the more peripheral regions of the centrosome that contain high levels of mCherry-Cnn fluorescence but only low levels of DSpd-2-GFP fluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 009 Unsurprisingly , the GFP–Cnn projections ( Figure 3A ) were no longer visible when the MTs were depolymerized with colchicine , and the GFP–Cnn scaffold became more compact and amorphous in appearance ( Figure 3B ) . The small amounts of DSpd-2 extending into the peripheral PCM ( Figure 3C ) were also lost when the MTs were depolymerized ( Figure 3D ) , but the more central DSpd-2-GFP region retained a clear organisation , with several spoke-like projections of DSpd-2-GFP emanating outward into the PCM from a central ring ( Figure 3D ) . We conclude that DSpd-2-GFP is part of a scaffold-like structure that forms around the centrioles and that retains some macromolecular structural integrity even in the absence of MTs . In our 3D-SIM images , we noticed that DSpd-2 and , to a lesser extent , Cnn both seemed to emanate from a single toroidal structure that we presume contains the mother centriole ( arrows , Figure 3A , C ) . To test if this structure was really the source of the centrosomal DSpd-2 and Cnn , we modified our 3D-SIM system to enable us ( to our knowledge for the first time ) to combine 3D-SIM with FRAP ( ‘Materials and methods’ ) . This analysis revealed that after photobleaching DSpd-2-GFP fluorescence initially recovered in a toroidal pattern ( t = 30 s , Figure 4A ) and then moved slowly outwards on dynamic projections ( t = 120 s to t = 450 s , Figure 4A; Video 2A ) . The initial toroidal distribution was very similar to that previously reported for Asl , one of the several proteins that form a toroid specifically around the mother centriole ( Mennella et al . , 2013 ) ( compare the recovering DSpd-2-GFP signal at t = 30 s in Figure 4A to the unbleached Asl-GFP signal shown in the inset; also compare the average profile of Asl-GFP to the average initial recovery profile of DSpd-2-GFP , Figure 4C ) . Thus , DSpd-2-GFP is initially incorporated into the PCM by binding sites that are tightly concentrated around only one of the two centrioles—almost certainly the mother—and the spatial distribution of these sites extensively overlaps with the distribution of Asl . 10 . 7554/eLife . 03399 . 010Figure 4 . 3D-SIM FRAP analysis of DSpd-2-GFP and GFP-Cnn behaviour at centrosomes . ( A and B ) 3D-SIM images show the dynamic behaviour of DSpd-2-GFP ( A ) and GFP-Cnn ( B ) at centrosomes in living Drosophila embryos after FRAP; time before and after photobleaching ( t = 0 ) is indicated . ( A ) DSpd-2-GFP fluorescence initially recovers in a toroid shape around the centriole ( A , t = 30 s ) , which has similar dimensions to unbleached Asl-GFP ( yellow inset ) . The protein then moves slowly outwards , forming dynamic projections that spread away from the centriole . ( B ) GFP-Cnn fluorescence initially recovers in a broader region around the centrioles ( t = 60 s in B ) , which has similar dimensions to the pre-bleached DSpd-2-GFP signal ( t = −30 s in A ) . ( C ) Graph compares the average prebleached ( dotted blue line ) and initial recovery ( dotted pink line ) profiles of DSpd-2-GFP to the average unbleached profile of Asl-GFP ( grey line ) ; all profiles were normalized so that their peak value is equal to 1 . Note how the initial DSpd-2-GFP recovery profile is essentially identical to the un-bleached Asl-GFP profile . ( D ) Graph compares the average pre-bleached ( solid blue line ) and initial recovery ( solid pink line ) profiles of GFP-Cnn , to the average pre-bleached profile of DSpd-2-GFP ( dotted blue line ) , and the average unbleached profile of Asl-GFP ( grey line ) ; all profiles were normalized so that their peak value is equal to 1 . Note how the initial GFP-Cnn recovery profile is very similar ( particularly in the more peripheral regions ) , to the pre-bleached DSpd-2-GFP profile , but is quite distinct from the pre-bleached Asl-GFP profile . See also Video 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 01010 . 7554/eLife . 03399 . 011Video 2 . Super-resolution 3D-SIM FRAP experiments reveal the differences between how DSpd-2 and Cnn are incorporated into the PCM . ( Related to Figure 4 ) . All Videos shown here are maximum intensity projections of image stacks . Live cell SD-SIM illustrating the dynamic behaviour of DSpd-2-GFP ( A ) and GFP-Cnn ( B ) at centrosomes in Drosophila embryos . The employed OMX Blaze 3D-SIM system enables sub-diffraction live cell imaging at high frame rates with ∼two-fold better xy- and z-resolution compared to conventional microscopy . Time before and after photobleaching ( t = 0 s ) is shown at the top right of each panel . Note how , prior to photobleaching , GFP-Cnn has a broader distribution that DSpd-2-GFP within the PCM . Immediately after photobleaching , DSpd-2-GFP fluorescence recovers in the shape of a toroid around the centriole , supporting our conclusion that Asl , which is distributed as a toroid around the centriole , is the major recruiter of DSpd-2 to centrosomes . In contrast , GFP-Cnn fluorescence recovers in a broader region around the centrioles , supporting our conclusion that DSpd-2 is the major recruiter of Cnn to centrosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 011 Interestingly , when we performed 3D-SIM FRAP with GFP-Cnn , the fluorescence initially recovered in the central region of the PCM ( t = 60 s , Figure 4B ) , but this region was significantly broader than the toroidal region in which DSpd-2-GFP initially recovered ( see t = 30 s , Figure 4A ) . Moreover , centrosomal GFP-Cnn fluorescence recovered more slowly than that of DSpd-2-GFP ( compare Figure 1F to Figure 1—figure supplement 2M ) . Thus , although the centrosomal binding sites for DSpd-2 and Cnn appear to be concentrated around the mother centriole , they do not precisely overlap and they recruit DSpd-2 and Cnn at different rates , strongly suggesting that DSpd-2 and Cnn are not incorporated into the PCM together as part of the same complex . We wondered how DSpd-2 might be recruited to the toroidal structure surrounding the mother centriole . The centriole proteins DSas-4 , Asl , and D-PLP all localize in a toroid pattern around mother centrioles ( Fu and Glover , 2012; Mennella et al . , 2012 ) and have all been implicated in PCM recruitment , suggesting that they could help recruit DSpd-2 . Because DSas-4 and Asl mutants lack centrioles ( Basto et al . , 2006; Blachon et al . , 2008 ) , we tested this possibility by inhibiting the function of these proteins in embryos using antibody-injection ( Conduit et al . , 2010 ) . Anti-Asl antibodies bound to centrosomes and reduced the rate of DSpd-2-GFP incorporation into the PCM by ∼75% , whereas anti-DSas-4 and anti-D-PLP antibodies had little effect ( Figure 5 ) —although these antibodies bound to centrioles/centrosomes and at least partially disrupted the function of their cognate protein ( Basto et al . , 2008; Novak et al . , 2014; unpublished data ) . These observations are consistent with the previous finding that Asl is required for the centrosomal localization of DSpd-2 ( Giansanti et al . , 2008 ) . Moreover , our yeast two-hybrid ( Y2H ) analysis revealed several strong direct interactions between Asl and DSpd-2 ( Figure 6 , Figure 6—source data 1 ) . Although we cannot be certain that Asl and DSpd-2 interact directly in vivo , collectively our data indicate that Asl has an important , and potentially direct , role in recruiting DSpd-2 to mother centrioles in fly embryos . 10 . 7554/eLife . 03399 . 012Figure 5 . Inhibiting Asl function strongly perturbs DSpd-2 incorporation into the PCM . Images ( A–D ) show results from FRAP experiments monitoring how DSpd-2-GFP ( green ) incorporation into the PCM is affected by inhibiting the function of various centriole-associated components with injected Texas-red-labelled antibodies ( red—as indicated ) . The antibodies bind to their cognate protein at centrosomes close to the injection site ( B–D ) , but not to those far from the injection site ( A ) , which therefore act as internal controls . ( E ) Graph displaying the initial recovery rate of DSpd-2-GFP fluorescence at centrosomes bound with antibodies ( as indicated below the graph ) relative to the centrosomes not bound by antibodies ( control ) . The rate of recovery was calculated by measuring the gradient of the initial linear phase of recovery that occurred over the first 60 s after photobleaching . Note how anti-Asl antibodies reduce the rate of DSpd-2-GFP incorporation into the PCM by ∼75% , but that anti-DSas-4 or anti-D-PLP antibodies have little or no effect . Error bars = standard error . See also Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 01210 . 7554/eLife . 03399 . 013Video 3 . The rate of DSpd-2-GFP incorporation into the PCM is reduced in embryos injected with anti-Asl antibodies . ( related to Figure 5 ) . All Videos shown here are maximum intensity projections of image stacks . These videos show the fluorescence recovery of DSpd-2-GFP ( green ) at centrosomes in embryos that were injected with fluorescently labelled antibodies ( red ) against Asl ( A , B ) , DSas-4 ( C ) , or D-PLP ( D ) . Time before and after photobleaching ( t = 0 s ) is shown at the top right of ( D ) . An example of a centrosome in an embryo injected with anti-Asl antibodies that was located far from the injection site and so received a low concentration of the antibody is shown in ( A ) ; these centrosomes acted as internal controls . Panels ( B–D ) show examples of centrosomes located close to the injection sites of antibodies against Asl ( B ) , DSas-4 ( C ) , and D-PLP ( D ) ; these centrosomes received a high concentration of each antibody . Note that DSpd-2GFP fluorescence recovered at a much slower rate at centrosomes that bound anti-Asl , whereas DSpd-2GFP fluorescence recovered at near normal rates at centrosomes that bound anti-DSas-4 or anti-D-PLP antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 01310 . 7554/eLife . 03399 . 014Figure 6 . A yeast-two-hybrid analysis examining the interactions between Asl , DSpd-2 , and Cnn . A schematic summary of the yeast two-hybrid interactions observed between Asl , DSpd-2 , and Cnn . The shades of the lines indicate the strength of the interactions observed: strong ( dark blue ) , medium ( blue ) , and weak ( light blue ) . The characteristics of the lines indicate the number of different positive reporter assays for each interaction: solid line ( 3/3 assays ) , large dashed line ( 2/3 assays ) , and small dashed line ( 1/3 assays ) . Arrows point from bait to prey fragments—double-headed arrows indicate that the interaction scored positive with either fragment as bait or prey . See Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 01410 . 7554/eLife . 03399 . 015Figure 6—source data 1 . A yeast-2-hybrid analysis testing the interactions between various Asl , DSpd-2 , and Cnn fragments . ( Related to Figure 6 ) A yeast-2-hybrid analysis was carried out using various bait and prey fragments of each protein ( as indicated in columns A and B ) . Three different reporters were tested—His ( columns C–F ) , Ade ( columns G and H ) , and LacZ ( column I ) . Interaction levels are indicated as strong , medium , weak , or none . All combinations of baits and preys were tested , but only those that scored positive in at least one assay are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 015 We next addressed how Cnn might be recruited to mother centrioles . We previously showed that anti-DSpd-2 or anti-Asl antibodies , but not anti-D-PLP or anti-Sas-4 antibodies , inhibit the rate of Cnn incorporation into the PCM by ∼75% and ∼55% , respectively , suggesting that both DSpd-2 and Asl may have a role in recruiting Cnn to centrosomes ( Conduit et al . , 2010 ) . We noticed , however , that the distribution of DSpd-2-GFP , but not Asl-GFP , around the mother centriole closely matched the distribution of the initial binding sites for GFP-Cnn , as revealed by our 3D-SIM FRAP experiments ( compare the recovering GFP-Cnn signal at t = 60 s in Figure 4B to the unbleached DSpd-2-GFP signal at t = −30 s in Figure 4A; also compare the average profile of DSpd-2-GFP to the average initial recovery profile of GFP-Cnn , Figure 4D ) . This suggests that DSpd-2 may provide the major centrosomal binding site for Cnn in embryos . In support of this possibility , we found that the amount of Cnn localized to centrosomes in eggs was reduced by ∼80% in the absence of DSpd-2 ( Figure 7 ) consistent with previous reports that the centrosomal localization of Cnn is perturbed in dspd-2 mutant brain cells ( Dix and Raff 2007; Giansanti et al . , 2008 ) . Moreover , our Y2H analysis revealed several strong direct interactions between DSpd-2 and Cnn but only one , much weaker , direct interaction between Cnn and Asl ( Figure 6 , Figure 6—source data 1 ) . Although we cannot be certain that Cnn and DSpd-2 interact directly in vivo , collectively our data indicate that DSpd-2 has an important , and potentially direct , role in recruiting Cnn to mother centrioles in fly embryos , while Asl appears to have a more minor role . 10 . 7554/eLife . 03399 . 016Figure 7 . The centrosomal levels of Cnn are strongly reduced in eggs lacking DSpd-2 . ( A and B ) Images show centrosomes co-stained with DSas-4 antibodies ( red ) and Cnn antibodies ( green ) in fixed eggs that either contained ( A ) or lacked ( B ) endogenous DSpd-2 . As embryos lacking DSpd-2 fail in pronuclear fusion ( Dix and Raff , 2007 ) ( and so fail to develop ) , we induced the de novo formation of centrosomes in WT and DSpd-2 mutant eggs by over-expressing Sak kinase ( Peel et al . , 2007; Rodrigues-Martins et al . , 2007 ) . The centrosomal levels of Cnn are dramatically reduced in the absence of DSpd-2 . ( C ) The graph quantifying the centrosomal levels of Cnn , DSpd-2 , γ-tubulin , and α-tubulin in Sak over-expressing eggs that either contained ( black bars ) or lacked ( white bars ) endogenous DSpd-2 . The centrosomal levels of Cnn , γ-tubulin , and α-tubulin are dramatically reduced in the absence of DSpd-2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 016 To investigate whether Cnn has a role in localizing DSpd-2 to centrosomes , we examined the distribution and dynamics of DSpd-2-GFP in cnn mutant embryos . Although there was a dramatic reduction in the amount of DSpd-2-GFP associated with centrosomes in the absence of Cnn ( Figure 8A , B ) , FRAP experiments revealed that the initial rate of DSpd-2-GFP incorporation was unperturbed ( Figure 8C–E ) . The localization of Asl-GFP ( which appears to recruit DSpd-2 ) was also largely unperturbed in cnn mutant embryos ( Figure 8F , G ) . The residual DSpd-2-GFP appeared to be more tightly associated with the centrioles in the absence of Cnn ( Figure 8B ) , while the PCM fraction of the protein moved rapidly away from the centrioles in small ‘flares’ ( Video 4 ) . These observations suggest that Cnn is not required for the initial incorporation of DSpd-2 into the PCM , but is required for the proper maintenance of DSpd-2 within the PCM . 10 . 7554/eLife . 03399 . 017Figure 8 . Cnn helps maintain DSpd-2 in the PCM . ( A and B ) Images show the localization of DSpd-2-GFP at centrosomes in either WT ( A ) or cnn mutant ( B ) embryos . ( C and D ) Images show the initial dynamic behaviour of DSpd-2-GFP at centrosomes in either WT ( C ) or cnn mutant ( D ) embryos; time before and after photobleaching ( t = 0 s ) is indicated . ( E ) Quantification of DSpd-2-GFP fluorescence recovery at centrosomes in either WT ( black line ) or cnn mutant ( red line ) embryos . The initial rate of DSpd-2-GFP fluorescence recovery is very similar in both WT and cnn mutant embryos , revealing that the initial incorporation of DSpd-2-GFP into the PCM is not dependent on Cnn . ( F and G ) Images show the localization of Asl-GFP at centrosomes in living embryos in the presence ( F ) or absence ( G ) of Cnn . The ability of Asl to localize efficiently in the absence of Cnn presumably explains why DSpd-2 can still be recruited to centrioles at normal rates in the absence of Cnn . Error bars = standard error . See Video 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 01710 . 7554/eLife . 03399 . 018Video 4 . The centrosomal localization of DSpd-2-GFP is perturbed in the absence of Cnn . ( related to Figure 8 ) . All Videos shown here are maximum intensity projections of image stacks . These videos illustrate the dynamic behaviour of DSpd-2-GFP at centrosomes in embryos where Cnn is present ( A ) or where Cnn is absent ( B ) . Note how in the absence of Cnn , DSpd-2-GFP cannot properly spread out through the PCM and a haze of DSpd-2-GFP fluorescence , including small particles of DSpd-2-GFP , appears to be rapidly lost from the centrosomes . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 018 The experiments described above are consistent with the idea that DSpd-2 and Cnn form molecular scaffolds during mitosis that initially assemble around the mother centriole and then move slowly outward . Both proteins have been implicated in PCM recruitment , so we reasoned that such scaffolds might provide a platform on which other PCM proteins assemble during mitosis . To test this idea , we analyzed larval brain cells where centrioles organize almost no PCM or MTs during interphase but mature to organize large amounts of both during mitosis ( Martinez-Campos et al . , 2004; Rogers et al . , 2008; Figure 9A ) . As shown previously , centrosome maturation in mitotic brain cells that lacked either Cnn ( Megraw et al . , 2001; Lucas and Raff , 2007 ) or DSpd-2 ( Dix and Raff , 2007; Giansanti et al . , 2008 ) was perturbed , but not abolished , and mitotic centrosomes organized levels of PCM that were significantly above those observed in interphase cells ( Figure 9B , C , F , G ) . Remarkably , however , centrosome maturation appeared to be abolished in the mitotic brain cells lacking both Cnn and DSpd-2 ( Figure 9D , H ) . This effect was specific to the loss of both Cnn and DSpd-2 , as centrosomes in mitotic cells lacking either Cnn or DSpd-2 that also lacked the abundant PCM protein D-TACC could still partially mature ( Figure 9—figure supplement 1 ) . Thus , most PCM proteins appear to rely on Cnn and DSpd-2 for their recruitment to mitotic centrosomes . 10 . 7554/eLife . 03399 . 019Figure 9 . Cnn and DSpd-2 cooperate to recruit the mitotic PCM . ( A–D ) Graphs show the average fluorescence intensities of interphase ( blue dots ) and mitotic ( black dots ) centrosomes from either WT ( A ) , cnn mutant ( B ) , dspd-2 mutant ( C ) , or cnn;dspd-2 double mutant ( D ) larval brain cells stained for various centrosomal proteins ( as indicated below graphs ) . Each data-point represents the average centrosome value from one brain . The horizontal red bars indicate the average value of all the brains . All the PCM proteins are still partially recruited to centrosomes in the absence of Cnn or DSpd-2 ( with the possible exception of Aurora A , which does not appear to be recruited in the absence of DSpd-2 ) . The mitotic PCM levels do not rise above interphase levels in the absence of both Cnn and DSpd-2 , indicating that centrosome maturation has been abolished . ( E–L ) Images show typical mitotic cells from either WT ( E and I ) , cnn ( F and J ) , dspd-2 ( G and K ) , or cnn;dspd-2 double mutant ( H and L ) larval brain cells stained for the centriole marker Asl ( red ) , mitotic DNA ( phospho-histone H3 , blue ) , and either the PCM marker γ-tubulin ( green , E–H ) or MTs ( I–L , green ) . Error bars = SEM . See also Figure 9—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 01910 . 7554/eLife . 03399 . 020Figure 9—figure supplement 1 . Centrosome maturation is not abolished in cnn;tacc or dspd-2;tacc double mutants . ( A and B ) Images and associated graphs show mitotic and interphase larval brain cells stained for γ-tubulin ( green ) , the centriole marker Asl ( red ) , and mitotic DNA ( phospho-histone H3 , blue ) from either cnn;tacc double mutant ( A ) or dspd-2;tacc double mutant larvae ( B ) . Graphs display the average fluorescent intensities of mitotic and interphase centrosomes ( relative to a WT mitotic value of 1 ) stained for γ-tubulin . Each data-point represents the average centrosome value from one brain . Note how centrosomes still partially mature in each mutant combination , as the levels of γ-tubulin centrosomal fluorescence are significantly higher in mitotic cells than in interphase cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 02010 . 7554/eLife . 03399 . 021Figure 9—figure supplement 2 . Centrosomes in cnn;dspd-2 double mutants fail to organise MTs . Images show selected time-points from videos of either WT ( A ) , cnn ( B ) , dspd-2 ( C ) , or cnn;dspd-2 double mutant ( D ) brain cells expressing the MT marker Jupiter-mCherry ( pseudo-coloured green ) and the centriole marker GFP-PACT ( pseudo-coloured red ) ; time before and after anaphase onset ( t = 0 s ) is indicated . Arrows indicate the position of the centrioles . The centrioles in WT cells , or cnn or dspd-2 single mutant cells can organize MT asters , but no centriole-associated MTs can be detected in the cnn;dspd-2 double mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 021 To test whether centrioles in cells lacking both Cnn and DSpd-2 were really unable to form mitotic MTOCs , we analyzed MT behaviour . In fixed prophase cells ( when centrosomal MT asters are most prominent ) , centrosomal asters were detectable in 100% of WT cells ( Figure 9I; n = 11 ) , 55% of cells lacking Cnn ( Figure 9J; n = 58 ) , 57% of cells lacking DSpd-2 ( Figure 9K; n = 14 ) , but in 0% of cells lacking both proteins ( Figure 9L; n = 15 ) . In live mitotic brain cells co-expressing the centriolar marker GFP-PACT and the MT marker Jupiter-mCherry ( Figure 9—figure supplement 2; Video 5 ) , centrosomes with MT asters were detectable in almost all cells lacking either Cnn ( n = 18/18 ) or DSpd-2 ( n = 24/25 ) , but in almost no cells lacking both proteins ( n = 1/24 ) . We conclude that cells lacking both DSpd-2 and Cnn are unable to assemble centrosomal MTOCs during mitosis . 10 . 7554/eLife . 03399 . 022Video 5 . Centrosomal MTOC activity is only abolished in cells lacking both Cnn and DSpd-2 . ( Related to Figure 9 ) . All Videos shown here are maximum intensity projections of image stacks . These videos show the distribution of the centriole marker GFP-PACT ( pseudo-coloured red ) and the MT marker Jupiter-mCherry ( pseudo-coloured green ) in either WT ( A ) , cnn ( B ) , dspd-2 ( C ) , or cnn;dspd-2 mutant neuroblasts as the cells progress through mitosis . Time before and after anaphase onset ( t = 0 s ) is shown at the top right of each panel . Note how the centrosomal MT asters can be observed in WT ( A ) , cnn ( B ) , and dspd-2 ( C ) mutant neuroblasts , and these MT asters appear to contribute to spindle assembly . No centrosomal MT asters , however , can be observed in cnn;dspd-2 double mutant neuroblasts and the spindle appears to form independently of centrosomes ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 022 Several hundred proteins are recruited to the PCM that expands around the centrioles during centrosome maturation in mitosis , but how so many proteins are organized into a functional mitotic centrosome has remained mysterious . Remarkably , we show here that the assembly of the mitotic PCM in flies appears to depend on just two proteins , Cnn and DSpd-2 . Both proteins appear to form scaffolds that initially assemble around the mother centriole and then spread outward , forming a dynamic platform upon which most , if not all , other PCM proteins ultimately assemble . DSpd-2 and Cnn partially depend on each other for their centrosomal localization , and both proteins are required to ensure robust centrosome maturation . In the absence of one of these proteins , reduced levels of the other protein still localize around the centrioles and can support the partial assembly of the mitotic PCM . In the absence of both the proteins , mitotic PCM assembly appears to be abolished ( Figure 9 ) . How are DSpd-2 and Cnn recruited to mother centrioles ? Our results strongly suggest that in fly embryos Asl initially helps recruit DSpd-2 to centrioles and DSpd-2 then helps to recruit Cnn . Cnn does not appear to be required to recruit either Asl or DSpd-2 to centrosomes , but it is required to properly maintain DSpd-2 within the PCM . We speculate that this interaction between DSpd-2 and Cnn creates a positive feedback loop that drives the dramatic expansion of the PCM scaffold around mother centrioles during mitosis ( Figure 10A ) . Although , we have identified direct interactions between Asl and DSpd-2 and between DSpd-2 and Cnn by Y2H , and the endogenous proteins can all co-immunoprecipitate with one another in fly embryo extracts ( Conduit et al . , 2010; Gopalakrishnan et al . , 2011 ) , we stress that we cannot be certain that these interactions are direct in vivo . 10 . 7554/eLife . 03399 . 023Figure 10 . A model for mitotic PCM assembly in flies . Schematics illustrate a putative pathway of mitotic PCM assembly in a WT cell ( A ) , or in cells lacking either Cnn ( B ) , DSpd-2 ( C ) , or both Cnn and DSpd-2 ( D ) . A top view of the mother centriole is shown surrounded by a layer of Asl ( grey ) ; solid arrows represent recruiting interactions , dotted arrows represent maintaining interactions . Arrow thickness reflects the relative strength of the recruitment or maintenance , and the size of the text reflects the amount of protein localized at centrosomes . In WT cells ( A ) , Asl has an important role in recruiting DSpd-2 to centrosomes , which in turn has an important role in recruiting Cnn; Cnn then has an important role in maintaining DSpd-2 at centrosomes . Thus , a positive feedback loop is generated where increasing amounts of DSpd-2 can recruit increasing amounts of Cnn , which can then maintain increasing amounts of DSpd-2 . DSpd-2 and Cnn both independently recruit other PCM components ( red ) , which themselves help support the PCM structure and can recruit further PCM components . In the absence of Cnn ( B ) , Asl can still recruit DSpd-2 normally , but DSpd-2 cannot efficiently accumulate around the centrioles . The reduced levels of DSpd-2 recruit reduced levels of PCM . In the absence of DSpd-2 ( C ) , an alternative pathway recruits reduced levels of Cnn . This pathway most likely involves Asl ( as indicated here ) , as inhibiting Asl reduces the rate of Cnn incorporation into the PCM ( Conduit et al . , 2010 ) and Asl and Cnn appear to weakly interact in a Y2H analysis ( Figure 6 ) ; other pathways , however , could also be involved . The reduced levels of Cnn recruit reduced levels of PCM . In the absence of Cnn and DSpd-2 ( D ) , no mitotic PCM can be assembled . DOI: http://dx . doi . org/10 . 7554/eLife . 03399 . 023 The requirement for Asl to initiate the mitotic recruitment of DSpd-2 and Cnn probably explains why these proteins are specifically recruited to mother centrioles . We recently showed that although Asl is essential for centriole duplication , it is not incorporated into daughter centrioles until they have passed through mitosis and matured into new mother centrioles ( Novak et al . , 2014 ) , and Asl/Cep152 proteins mainly localize to mother centrioles in several species ( Sir et al . , 2011; Fu and Glover , 2012; Lawo et al . , 2012; Mennella et al . , 2012; Sonnen et al . , 2012 ) . The PCM appears to be preferentially associated with mother centrioles in many systems ( Piel et al . , 2000; Conduit et al . , 2010; Wang et al . , 2011 ) . Our findings provide a potential explanation for why this is so , and raise the intriguing possibility that all mitotic PCM may be organized exclusively by mother centrioles . Although DSpd-2 seems to be the major recruiter of centrosomal Cnn in embryos , there must be an alternative recruiter , as the centrosomal localization of Cnn is not abolished in the absence of DSpd-2 ( Figure 7 , Figure 9C ) . Asl is an attractive candidate as anti-Asl antibodies perturb Cnn recruitment to centrioles ( Conduit et al . , 2010 ) ( although this could be an indirect consequence of their effect on DSpd-2 recruitment ) , and Asl and Cnn interact in our Y2H analysis . Moreover , human Cep152/Asl has a role in the centrosomal recruitment of human Cdk5Rap2/Cnn ( Firat-Karalar et al . , 2014 ) . Interestingly , in flies this alternative pathway appears to be stronger in larval brain cells than in eggs/embryos: in the absence of DSpd-2 , Cnn levels are reduced by only ∼35% in brains ( Figure 9C ) but by ∼80% in eggs ( Figure 7C ) . Thus , the detail of mitotic PCM assembly pathway may vary between different cell types even in the same species . Our data suggest that after DSpd-2 and Cnn have been recruited to centrioles , they rapidly assemble into scaffolds that then move slowly away from the centrioles . For Cnn , there is strong data indicating that scaffold assembly is regulated by phosphorylation . Cnn contains a phospho-regulated multimerization ( PReM ) domain that is phosphorylated by Polo/Plk1 in vitro and at centrosomes during mitosis in vivo ( Dobbelaere et al . , 2008; Conduit et al . , 2014 ) . Mimicking phosphorylation allows the PReM domain to multimerize in vitro and Cnn to spontaneously assemble into cytosolic scaffolds in vivo that can organize MTs . Conversely , ablating phosphorylation does not interfere with Cnn recruitment to centrioles , but inhibits Cnn scaffold assembly ( Conduit et al . , 2014 ) . We speculate that , like Cnn , DSpd-2 can assemble into a scaffold and that this assembly is regulated in vivo so that it only occurs around mother centrioles . It remains unclear , however , whether DSpd-2 itself can form a scaffold , or whether it requires other proteins to do so . It is striking that both DSpd-2 and Cnn exhibit an unusual dynamic behaviour at centrosomes . Both proteins incorporate into the PCM from the inside out , and are in constant flux , as the molecules that move slowly outward away from the centrioles are replaced by newly incorporated molecules close to the centriole surface ( see Figure 1I ) . This inside out assembly is likely to have important consequences , as it means that the events close to the centriole surface , rather than at the periphery of the PCM , can ultimately regulate mitotic PCM assembly . This may be particularly important in cells where centrioles organise centrosomes of different sizes , as is the case in certain asymmetrically dividing stem/progenitor cells ( Lesage et al . , 2010; Nigg and Stearns , 2011; Pelletier and Yamashita , 2012 ) . Fly neural stem cells , for example , use centrosome size asymmetry to ensure robust asymmetric division ( Rebollo et al . , 2007; Rusan and Peifer , 2007; Januschke et al . , 2013 ) , and there is strong evidence that new and old mother centrioles differentially regulate the rate of Cnn incorporation in these cells ( Conduit and Raff , 2010 ) . Moreover , mutations in human Cdk5Rap2/Cnn have been implicated in microcephaly ( Bond et al . , 2005 ) , a pathology linked to a failure in neural progenitor cell proliferation , although the precise reason for this is unclear ( Bond and Woods , 2006; Megraw et al . , 2011 ) . Although DSpd-2 and Cnn have a major role in centrosome maturation , we stress that other PCM components are likely to make important contributions . Pericentrin , for example , has been implicated in PCM recruitment in several systems ( Martinez-Campos et al . , 2004; Zimmerman et al . , 2004; Lee and Rhee , 2011; Lawo et al . , 2012; Mennella et al . , 2012; Kim and Rhee , 2014 ) , and the fly homologue , D-PLP , forms ordered fibrils in cultured S2 cells that extend away from the centriole wall and support PCM assembly in interphase ( Mennella et al . , 2012 ) . These centriolar fibrils , however , cannot explain how centrioles organize such a vastly expanded PCM matrix during mitosis , and D-PLP appears to have an important , but more minor , role in mitotic PCM assembly in vivo ( Martinez-Campos et al . , 2004 ) . Nevertheless , proteins like D-PLP will certainly help recruit other PCM proteins and help form structural links within the PCM , thus strengthening the mitotic PCM matrix . The important distinction is that , in flies at least , most proteins , including the PCM fraction of D-PLP , are recruited into the PCM by an underlying PCM scaffold , whereas DSpd-2 and Cnn appear to form this scaffold . Homologues of Asl , DSpd-2 , and Cnn have been implicated in PCM assembly in many species ( Pelletier et al . , 2004; Gomez-Ferreria et al . , 2007; Zhu et al . , 2008; Haren et al . , 2009; Barr et al . , 2010; Cizmecioglu et al . , 2010; Hatch et al . , 2010; Joukov et al . , 2010; Decker et al . , 2011; Kim and Rhee , 2014 ) , suggesting that the mechanism of mitotic PCM recruitment we identify here in flies may be conserved through evolution . To our knowledge , however , no PCM component has yet been shown to assemble from the inside out and to flux away from the centrioles in any other system . Nevertheless , although the precise molecular details will likely vary from cell type to cell type and from species to species , we suspect that this unusual dynamic behaviour of an underlying mitotic PCM scaffold will prove to be a general feature of mitotic centrosome assembly in many systems . P-element-mediated transformation vectors were made by introducing a full-length DGp71WD or D-PLP cDNA ( Kawaguchi and Zheng , 2004 ) into the Ubq-GFPCT Gateway vector ( Basto et al . , 2008 ) . Transgenic lines were generated either by Genetic Services , Inc . , Cambridge , MA , Bestgene , Inc . , Chino Hills , CA or the Fly Facility in the Department of Genetics , Cambridge , UK . Other GFP , RFP , and mCherry fusions have been described previously: GFP-Cnn ( Lucas and Raff , 2007 ) , DSpd-2-GFP ( Dix and Raff , 2007 ) , Aur-A-GFP ( Lucas and Raff , 2007 ) , RFP-PACT ( Conduit et al . , 2010 ) , Polo-GFP ( Buszczak et al . , 2006 ) , Jupiter-mCherry ( Callan et al . , 2010 ) , γ-tubulin-GFP ( Hallen et al . , 2008 ) , DSas-4-GFP ( Novak et al . , 2014 ) , and Asl-GFP ( Blachon et al . , 2008 ) . To examine the dynamics of DSpd-2-GFP and Asl-GFP at centrosomes , using both standard spinning disc confocal imaging and 3D-SIM imaging , we analyzed embryos from mothers expressing two copies of Ubq-DSpd-2-GFP in a DSpd-2Z35711/DSpd-2Df ( 3L ) st-j7 hemizygous mutant background ( Dix and Raff , 2007; Giansanti et al . , 2008 ) or two copies of Asl-GFP ( expressed from its endogenous promoter ) in an aslmecd mutant background ( Blachon et al . , 2008 ) . For analysing the dynamics of γ-tubulin-GFP , AurA-GFP , DGp71WD-GFP , Polo-GFP , DSas-4-GFP , and D-PLP-GFP at centrosomes , we analyzed embryos from mothers expressing either two copies of γ-tubulin-GFP ( expressed under the ncd promoter ) , Ubq-AurA-GFP , Ubq-DGp71WD-GFP , Polo-GFP ( expressed under its endogenous promoter ) , DSas-4-GFP ( expressed under its endogenous promoter ) or one copy of Ubq-D-PLP-GFP in a WT background . For analysing the dynamics of GFP-Cnn at centrosomes using 3D-SIM imaging , we analyzed embryos from mothers expressing two copies of Ubq-GFP-Cnn in a cnnf04547/cnnHK21 hemizygous mutant background . For comparing the dynamics of DSpd-2-GFP in a WT background to the dynamics of DSpd-2-GFP in the absence of Cnn , we analyzed embryos from mothers expressing one copy of pUbq-DSpd-2-GFP in a WT background and embryos from mothers expressing one copy of pUbq-DSpd-2-GFP in a cnnHK21/cnndf ( 2R ) BSC306 hemizygous mutant background . For comparing the localization of Asl-GFP in a WT background to the localization of Asl-GFP in the absence of Cnn , we analyzed embryos from mothers expressing one copy of pUbq-Asl-GFP in a WT background and embryos from mothers expressing one copy of pUbq-Asl-GFP in a cnnf04547/cnnHK21 hemizygous mutant background . For examining PCM recruitment in larval brain cells lacking Cnn , DSpd-2 or Cnn , and DSpd-2 , we analyzed cnnf04547/cnnHK21 hemizygous mutants , dspd-2Z35711/dspd-2Df ( 3L ) st-j7 hemizygous mutant , or cnnf04547/cnnHK21; dspd-2Z35711/dspd-2Df ( 3L ) st-j7 double hemizygous mutant larval brains , respectively . For examining the behaviour of MTs in living larval brain cells , we analyzed brains expressing one copy of Ubq-GFP-PACT and one copy of Jupiter-mCherry ( expressed under its endogenous promoter ) in either a WT , cnnf04547/cnnHK21 hemizygous mutant , dspd-2Z35711/dspd-2Df ( 3L ) st-j7 hemizygous mutant , or cnnf04547/cnnHK21; dspd-2Z35711/dspd-2Df ( 3L ) st-j7 double hemizygous mutant background . For immunofluorescence analysis , we used the following antibodies: rabbit anti-Cnn ( 1:1000 ) ( Lucas and Raff , 2007 ) , rabbit anti-DSpd-2 ( 1:500 ) ( Dix and Raff , 2007 ) , mouse anti-γ-tubulin ( 1:500; GTU88 , Sigma-Aldrich , St . Louis , MO ) , rabbit anti-D-TACC ( 1:500 ) ( Gergely et al . , 2000 ) , rabbit anti-DGp71WD ( 1:500 ) ( Vérollet et al . , 2006 ) , rabbit anti-AurA ( 1:500 ) ( Barros et al . , 2005 ) , mouse anti-α-tubulin ( 1:1000; DM1α; Sigma-Aldrich ) , guinea-pig anti-Asl ( 1:500 ) ( this study ) , and anti-PhosphoHistoneH3 ( mouse , 1:2000 , AbCam , UK or rabbit , 1:500 , Cell Signalling Technology , Danvers , MA ) . Secondary antibodies were from Molecular Probes ( Invitrogen , Carlsbad , CA ) : Alexa Fluor 488 , 568 , and 647 ( all used at 1:1000 ) . For antibody injection experiments , we used rabbit anti-Asl ( aa665–995 ) , rabbit anti-DSas-4 ( aa1–260 ) , and rabbit anti-D-PLP ( aa683–974 ) affinity purified antibodies . We also tested rabbit anti-D-PLP antibodies raised against aa1805–2137 , which are predicted to also recognise the N-terminus of the short D-PLP isoform ( aa8–350 ) , and found that these antibodies , like the rabbit anti-D-PLP ( aa683–974 ) antibodies , did not significantly perturb the DSpd-2-GFP incorporation ( data not shown ) . FRAP experiments were carried out as described previously ( Conduit et al . , 2014 ) . Photobleaching was carried out in S phase , which is when mitotic PCM is most actively recruited in these rapidly cycling embryos . We used ImageJ to calculate the fluorescence profile of each centrosome at each time-point . We first scaled the images so that each pixel was split into 25 ( 5 × 5 ) pixels in order to increase the resolution of our radial profiling . We then calculated the centre of mass of the centrosome by thresholding the image and running the ‘analyze particles’ ( centre of mass ) macro on the most central Z plane of the centrosome . We then centred concentric rings ( spaced at 0 . 028 μm and spanning across 3 . 02 μm ) on this centre and measured the average fluorescence around each ring ( radial profiling ) . After subtracting the average cytosolic signal and normalising , so the peak intensity of the pre-bleached image was equal to 1 , we mirrored the profiles to show a full symmetric centrosomal profile . For each time-point , an average distribution from at least 10 centrosomes was calculated . For analysing DSpd-2-GFP recovery in the centre and periphery of the PCM , we bleached centrosomes at the start of S-phase and then again 3 minutes later ( still in S-phase ) ; we measured the fluorescence recovery using radial profiling , as described above . The central PCM measurements were calculated as the average fluorescence intensity of 5 measurements taken between 0 . 028 μm and 0 . 14 μm , from the centre of the centrosome . The peripheral PCM measurements were calculated as the average fluorescence intensity of 5 measurements taken between 0 . 62 mm and 0 . 73 μm from the centre of the centrosome . Ten centrosomes from 10 embryos were analyzed; values were averaged to produce each data point . The cytosolic signal was subtracted before plotting the recovery graphs . To examine AurA-GFP and Polo-GFP recovery in the centre and periphery of the PCM , the original AurA-GFP and Polo-GFP FRAP data were re-analyzed by measuring the recovery in the same regions as for the DSpd-2-GFP analysis . We fixed embryos in methanol , homogenized 50 embryos per genotype in 100 μl sample buffer , and ran either 5 μl or 10 μl on NuPAGE 3–8% Tris-acetate pre-cast gels ( Life Technologies , Carlsbad , CA ) . The proteins were transferred onto nitrocellulose membrane and loading was initially checked using Ponceau staining . The membrane was then blocked and probed with antibodies against the protein in question and against the GFP . Living embryos were imaged at 21°C on a DeltaVision OMX V3 Blaze microscope ( GE Healthcare , UK ) equipped with a 60x/1 . 42 oil UPlanSApo objective ( Olympus ) , 405 nm and 488 nm and 593 nm diode lasers and sCMOS cameras ( PCO ) . 3D-SIM image stacks were acquired with 5 phases 3 angles per image plane and 0 . 125μm z-distance between sections . The raw data was computationally reconstructed with SoftWoRx 6 . 0 ( Applied Precision ) using Wiener filter settings 0 . 002 and channel specifically measured optical transfer functions to generate a super-resolution 3D image stack with a lateral ( x-y ) resolution of 100-130 nm ( wavelength-dependent ) and an axial ( z ) resolution of ∼300 nm ( Schermelleh et al , 2008 ) . For two colour images , Images from the different color channels were registered with alignment parameter obtained from calibration measurements with 0 . 2 μm diameter TetraSpeck beads ( Life Technologies ) using the OMX Editor software . Images were processed using SoftWorx software ( GE Healthcare ) . Images shown are maximum intensity projections of several z-slices . When analysing the effect of MT de-polymerisation , embryos were first injected with 1 mM colchicine solution and imaged 20–60 min later . To perform 3D-SIM FRAP , we utilized the software development kit ( SDK ) from GE Healthcare . This allowed us to create a custom acquisition sequence that first acquired a single Z-stack in 3D-SIM , then performed single or multiple spot photobleaching ( using the standard OMX galvo scanner TIRF/photo-kinetics module ) , and then performed time lapse imaging in 3D-SIM mode . The centrosomal profiles were calculated in a similar way to that described above , except that the concentric rings for Asl-GFP , DSpd-2-GFP , and GFP-Cnn were spaced at 0 . 0055 μm , 0 . 011 μm , and 0 . 0109 μm and spanned across 1 . 86 μm , 3 . 28 μm , and 3 . 28 μm , respectively . For generating the average 3D-SIM profiles for Asl-GFP , DSpd-2-GFP , and GFP-Cnn , we averaged profiles from 11 , 24 , and 15 centrosomes , respectively . Affinity-purified antibodies were covalently coupled to Texas Red , as described previously ( Gergely et al . , 2000 ) . Antibodies were injected at the start of a mitotic cycle , and embryos were observed on the Spinning Disk confocal system described above . Centrosomes were bleached in pairs: one centrosome located close to the injection site ( experimental ) and one centrosome located far from the injection site ( control ) . Between 2 and 3 centrosome pairs were bleached per embryo , with an average of 7 embryos injected for each antibody , and the data were collated . The average initial rate of DSpd-2-GFP incorporation at control and experimental centrosomes was compared using a paired Student's t test . Bait and prey fragments were cloned , introduced into yeast , and tested for interactions as described previously ( Conduit et al . , 2014 ) . For the baits , fragments encoding the N-terminal , middle , and C-terminal thirds of the proteins were cloned , along with fragments encoding the N-terminal two-thirds , C-terminal two-thirds , and the full-length protein . For the preys , smaller ∼200 aa fragments and larger combinations of these fragments , including the full-length protein , were cloned . For the analysis of centrosomal fluorescence levels of PCM components , third instar larval brains were dissected and incubated in 100 mM colchicine in Schneider's medium , Sigma-Aldrich for 1 hr at 25°C . Colchicine treatment de-polymerizes the MTs and prevents centrosome ‘rocketing’ in cnn mutants ( Lucas and Raff , 2007 ) , allowing a more accurate quantification of PCM recruitment . The brains were then fixed in paraformaldehyde containing 100 mM PIPES , 1 mM MgSO4 , and 2 mM EGTA pH 6 . 95 for 5 min at room temperature , washed in PBS and then 45% and 60% acetic acid , squashed under a coverslip , post-fixed in methanol , washed in PBT , and then stained with the appropriate antibodies . Images were collected on an Olympus FV1000 scanning confocal microscope using a 60× , 1 . 4 NA oil objective and maximum intensity projections were made . For each brain at least five images containing multiple cells in both mitosis ( as shown by positive Phospho-Histone H3 staining ) and interphase were collected . At least fivebrains were imaged for each mutant and staining combination and an average of 40 mitotic and 37 interphase centrosomes were measured per brain . Centrosome fluorescence was calculated by measuring the total fluorescence in a boxed region around the centrosome and subtracting the local cytoplasmic background fluorescence . The average value of all the centrosomes from a single brain was used for each data point . The average value of these data points for mitotic and interphase cells were compared using a Mann–Whitney test . For the analysis of MT asters , the cells were fixed and stained as above , but were not pre-treated with colchicine . A cell was scored as positive if at least 1 centrosome had detectable astral MTs . Third instar larval brains were dissected and either semi-squashed under a coverslip or mounted whole in Schneider's medium and then imaged on the Perkin Elmer Spinning Disk confocal system described above . Cells were filmed progressing from interphase/prophase through mitosis . A cell was scored as positive if at least 1 centrosome had detectable astral MTs .	Long protein filaments called microtubules perform a range of roles inside cells—for example , they give the cell its shape and help to divide its genetic material during cell division . In animal cells , microtubules emerge from structures called centrosomes . These contain two cylindrical structures called centrioles that are surrounded by a matrix of pericentriolar material made from several hundred different proteins . Problems with centrosomes have been linked to several disorders , including cancer . In the fruit fly Drosophila , it was long thought that the pericentriolar material assembles on an underlying ‘scaffold’ , the composition of which had remained unclear . A protein called Centrosomin was a good candidate molecule , as it is required to maintain the proper structure of the pericentriolar material . In addition , Centrosomin molecules continuously spread away from the centrioles into the matrix providing a clear centriole–matrix connection . However , if Centrosomin is not present in a cell , some protein is still recruited around the centrioles . Conduit et al . therefore suspected that Centrosomin works together with another protein to build the scaffold . Conduit et al . used super-resolution microscopy to observe the behaviour of several proteins , thought most likely to help Centrosomin to form the scaffold . Only one , called DSpd-2 , builds outwards from the centrioles like Centrosomin . Genetic tests showed that both Centrosomin and DSpd-2 are important for the other proteins to localize to the pericentriolar material . If one of either Centrosomin or DSpd-2 is missing from the cell , reduced amounts of protein are recruited around the centrioles but the matrix still partially forms . Without both proteins , however , the matrix does not form at all . Conduit et al . found that a third protein helps to recruit Centrosomin and DSpd-2 to the older of the two centrioles ( also known as the mother centriole ) . DSpd-2 then draws in more Centrosomin . As Centrosomin helps to hold the DSpd-2 proteins in the pericentriolar material , this enables even more Centrosomin to be recruited , and so forms a positive feedback loop that helps the scaffold to continue growing . The findings of Conduit et al . provide a simple mechanism for building the scaffold that supports the formation of the centrosome in the fruit fly Drosophila . Whether a similar mechanism is used to construct centrosomes in other species remains to be investigated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2014	A molecular mechanism of mitotic centrosome assembly in Drosophila
In order to regenerate tissues successfully , stem cells must detect injuries and restore missing cell types through largely unknown mechanisms . Planarian flatworms have an extensive stem cell population responsible for regenerating any organ after amputation . Here , we compare planarian stem cell responses to different injuries by either amputation of a single organ , the pharynx , or removal of tissues from other organs by decapitation . We find that planarian stem cells adopt distinct behaviors depending on what tissue is missing to target progenitor and tissue production towards missing tissues . Loss of non-pharyngeal tissues only increases non-pharyngeal progenitors , while pharynx removal selectively triggers division and expansion of pharynx progenitors . By pharmacologically inhibiting either mitosis or activation of the MAP kinase ERK , we identify a narrow window of time during which stem cell division and ERK signaling produces pharynx progenitors necessary for regeneration . These results indicate that planarian stem cells can tailor their output to match the regenerative needs of the animal . When faced with injury or disease , many animals can repair or even replace damaged tissue . This process of regeneration is observed across animal species , and is often fueled by tissue-resident stem cells ( Bely and Nyberg , 2010; Sánchez Alvarado and Tsonis , 2006; Tanaka and Reddien , 2011 ) . In response to injury , stem cells accelerate the production of specific types of differentiated cells to repair damaged tissues . For example , in adult mammals , injuries to the intestine , skin , or lung induce stem cells to increase proliferation rates and alter their differentiation potential ( Buczacki et al . , 2013; Stabler and Morrisey , 2017; Tetteh et al . , 2015; Tumbar et al . , 2004 ) . These findings suggest that injury can modify the behavior of stem cells to promote repair , but how these changes contribute to tissue regeneration remains unclear . The freshwater planarian Schmidtea mediterranea is an ideal model organism to study the interaction between injury and tissue repair due to their virtually endless ability to regenerate ( Ivankovic et al . , 2019 ) . This ability is driven by an abundant , heterogeneous population of stem cells ( Adler and Sánchez Alvarado , 2015; Reddien , 2018; Zhu and Pearson , 2016 ) . Defined by ubiquitous expression of the argonaute transcript piwi-1 ( Reddien et al . , 2005 ) , the planarian stem cell population consists of pluripotent stem cells capable of reconstituting the entire animal ( Wagner et al . , 2011 ) and likely organ-specific progenitors ( Figure 1A; Scimone et al . , 2014a; van Wolfswinkel et al . , 2014; Zeng et al . , 2018 ) . These progenitors express organ-specific transcription factors required for the maintenance and regeneration of planarian organs , including a pharynx , primitive eyes , muscle , intestine , an excretory system and a central nervous system ( Figure 1A ) , all enveloped in epithelium ( Roberts-Galbraith and Newmark , 2015 ) . Expression of specific progenitor markers in piwi-1+ stem cells ( Fincher et al . , 2018; Plass et al . , 2018; Scimone et al . , 2014a; van Wolfswinkel et al . , 2014; Zeng et al . , 2018 ) provides an opportunity to link the behavior of organ-specific progenitors with injury by tracking stem cell behavior as organ regeneration initiates . During homeostasis , planarian stem cells replenish organs by steady proliferation that drives cellular turnover ( Pellettieri and Sánchez Alvarado , 2007 ) . Within hours of any injury , a general increase in stem cell division occurs , along with vast transcriptional changes ( Baguñà , 1976; Gaviño et al . , 2013; Sandmann et al . , 2011; Wenemoser et al . , 2012; Wenemoser and Reddien , 2010 ) . These changes are only sustained beyond the first day if tissue is removed , in what is referred to as the ‘missing tissue response’ ( Baguñà , 1976; Gaviño et al . , 2013; Wenemoser and Reddien , 2010 ) . Activated by injury , the extracellular signal-regulated kinase ( ERK ) contributes to many of these wound-induced transcriptional changes , in addition to regulating stem cell proliferation , differentiation , and survival ( Owlarn et al . , 2017; Shiroor et al . , 2020; Tasaki et al . , 2011 ) . Because these injury-induced changes have predominantly been characterized by analyzing broad stem cell behaviors , how they regulate the transition from homeostasis to regeneration of particular organs are key issues to resolve . Most planarian organs extend throughout the entire body ( Figure 1A ) , and injuries often cause simultaneous damage to multiple organs ( Elliott and Sánchez Alvarado , 2013 ) . The resulting complex regenerative response has limited our ability to decipher how stem cells respond to damage of particular organs . Unlike most planarian organs , except the eye , the pharynx is anatomically distinct ( Adler and Sánchez Alvarado , 2015; Kreshchenko , 2009 ) . Importantly , it can be completely and selectively removed without perturbing other tissues by brief exposure to sodium azide ( Figure 1B; Adler et al . , 2014; Shiroor et al . , 2018 ) . Because only a single organ is removed , pharynx amputation vastly simplifies the regeneration challenge posed to the animal . Previous work identified the forkhead transcription factor FoxA as an essential regulator of pharynx regeneration ( Adler et al . , 2014; Scimone et al . , 2014a ) . Under homeostatic conditions , FoxA is expressed in the pharynx and a subset of stem cells . Pharynx amputation triggers an increase in FoxA+ stem cells , demonstrating that injury expands the pool of pharynx progenitors . These properties allow us to dissect how stem cells respond to loss of a specific organ and are regulated to restore it . The ability of planarians to replace exactly the tissues that have been damaged or removed by injury remains one of the outstanding questions in regeneration ( Mangel et al . , 2016; Nishimura et al . , 2011 ) . Previous studies have suggested that stem cells selectively increase the output of specific progenitors of depleted organs , implying a targeted mode of regeneration ( Figure 1C; Thi-Kim Vu et al . , 2015 ) . However , others have shown that stem cells respond indiscriminately to tissue removal , incorporating new cells into tissues regardless of whether they have been damaged , suggesting a non-targeted mode of regeneration ( LoCascio et al . , 2017 ) . Based on these findings , the authors proposed that stem cells non-selectively increase production of any nearby progenitors , determined by the size and position of a wound , rather than the identity of missing tissues ( Figure 1C ) . These two seemingly contradictory models introduce uncertainty into our understanding of the relationship between tissue loss and the stem cell behaviors that ultimately contribute to the regeneration of missing tissues . By performing an in-depth analysis of specific populations of stem cells in response to different injuries , we show that stem cells can sense the identity of missing tissues . Inflicting various injuries to both the pharynx and body defines distinct contributions of stem cells to regenerated tissue depending on when they divide relative to injury . Planarian stem cells respond to organ loss by selectively increasing expression of organ-specific transcription factors required for subsequent regeneration . Amputation of non-pharyngeal tissues only amplifies non-pharyngeal progenitors , while removal of pharynx tissue selectively increases pharynx progenitors . This increase in pharynx progenitors , and subsequent pharynx regeneration , depends on stem cell division and the MAP kinase ERK , during defined times after tissue loss . Unlike the pharynx , eye regeneration following selective removal is not dependent on stem cell division or ERK signaling , suggesting that different injuries may require distinct regenerative mechanisms . We propose that , in addition to non-targeted and passive modes of regeneration , stem cell behavior can be altered by the loss of specific tissues , selectively channeling their output towards replacement of missing organs . Planarian stem cells have distinct responses to injury depending on whether or not tissue has been removed . Any injury induces a proliferative response within hours , while tissue removal causes a sustained response for up to 4 days , leading to localized proliferation and differentiation ( Baguñà , 1976; Wenemoser and Reddien , 2010 ) . Therefore , it has been hypothesized that the initial injury response is a ‘general’ mechanism for repair , whereas the later ‘missing-tissue response’ may be tailored to target the replacement of lost tissue . To evaluate the outcome of stem cell proliferation during these specific injury responses , we altered the timing of stem cell labeling relative to different injuries and analyzed the prevalence of labeled cells in mature organs . We labeled stem cells with the thymidine analogue F-ara-EdU ( Neef and Luedtke , 2011 ) for 4 hr either immediately or 1 day after pharynx or head amputation . We then analyzed F-ara-EdU+ cells in the pharynx 7 days after amputation ( Figure 1D ) . When F-ara-EdU was applied immediately after amputation ( D0 ) , we observed increased F-ara-EdU+ cells in the pharynx following either pharynx or head removal , as compared to intact controls ( Figure 1E , F ) . The timing of this pulse , relative to injury , confirms a previous study showing that amputation stimulates general incorporation of newly generated cells into non-injured tissues ( LoCascio et al . , 2017 ) . However , F-ara-EdU administration 1 day after amputation ( D1 ) resulted in a specific increase in F-ara-EdU+ pharynx cells only after removal of the pharynx , but not the head ( Figure 1E , F ) . To determine if increased tissue production requires tissue removal , prior to F-ara-EdU administration we performed incisions anterior to the pharynx , which damaged the body without removing any tissue ( Figure 1F ) . However , the number of F-ara-EdU+ cells in the pharynx were comparable to controls , suggesting that tissue removal strongly stimulates production of new tissue , while injury alone does not . To determine whether other tissues incorporate new stem cells in a time-dependent manner relative to injury , we analyzed the number of newly generated neurons in the brain by combining F-ara-EdU staining with FISH for the neuronal marker ChAT ( Figure 1—figure supplement 1A; Wagner et al . , 2011 ) . In the newly regenerated brain , head amputation increased F-ara-EdU+ ChAT+ neurons after both F-ara-EdU pulse conditions , as compared to intact controls . However , the number of F-ara-EdU+ ChAT+ brain neurons were comparable to controls after either pharynx amputation or incisions , regardless of when the F-ara-EdU pulse was administered ( Figure 1—figure supplement 1B , C ) . Because chemical pharynx removal does not increase production of neural tissue ( Figure 1—figure supplement 1B , C ) , while its surgical removal does ( LoCascio et al . , 2017 ) , non-targeted regenerative mechanisms may require injury to specific types of tissues , such as body-wall muscle , epithelia , or intestine . In fact , amputation-specific transcriptional changes important for regeneration have recently been identified within these tissues ( Witchley et al . , 2013; Lander and Petersen , 2016; Scimone et al . , 2016; Benham-Pyle et al . , 2020 ) . These results suggest that while cells generated immediately after tissue removal can be broadly deployed to all surrounding tissues , those generated 1 day later are targeted toward only those that are missing . Our data so far indicate that injury channels the output of stem cells towards missing tissues . If this targeted model is true , injuries that do not remove pharynx tissue , like head amputations , should not increase pharynx progenitors ( Figure 1C ) . Alternatively , if regeneration is non-targeted , injury should non-selectively increase production of any nearby progenitors ( LoCascio et al . , 2017 ) . If this is the case , head amputation , where pharyngeal tissue is not removed , should also stimulate an increase in pharynx progenitors ( Figure 1C ) . To test the response of organ-specific progenitors ( Figure 1A ) to loss of different tissues , we challenged animals with either head or pharynx amputation and analyzed changes in expression of organ-specific progenitor markers within piwi-1+ stem cells . First , we labeled pharynx progenitors with double fluorescent in situ hybridization ( FISH ) for piwi-1 and the pharynx-specific progenitor marker FoxA , 3 days after pharynx or head amputation . We then quantified pharynx progenitors in the same region , anterior to the pharynx ( Figure 2A ) . As previously reported , we found that pharynx removal caused a significant increase in pharynx progenitors as compared to intact controls ( Adler et al . , 2014; Scimone et al . , 2014a ) . By contrast , head amputation did not influence the number of pharynx progenitors , which were similar to intact animals ( Figure 2A , B ) . To determine when pharynx progenitors emerge and how long they persist , we quantified the number of pharynx progenitors at various times after pharynx amputation and found that they significantly increased 3 days after amputation ( Figure 2C ) . Because injury broadly influences stem cell behavior , we also analyzed the proportion of these pharynx progenitors relative to all other piwi-1+ stem cells at the same times after pharynx and head amputation and found a similar trend ( Figure 2—figure supplement 1A , B ) . These data indicate that pharynx progenitors are selectively produced 3 days after pharynx loss , but not after loss of other tissue types . To determine which types of injuries stimulate an increase in pharynx progenitors , we inflicted various injuries to or around the pharynx ( Figure 2D ) . We then labeled and quantified pharynx progenitors 3 days later . Incisions that damaged the pharynx without removing any tissue failed to stimulate an increase in piwi-1+FoxA + stem cells . However , partial removal of the pharynx ( ~50–80% ) caused a significant increase in pharynx progenitors compared to intact controls . We also performed flank resections , which removed tissue from regions adjacent to the pharynx but did not damage the pharynx itself . Despite being previously shown to increase new cells into the uninjured pharynx with BrdU labeling ( LoCascio et al . , 2017 ) , we observed comparable numbers of pharynx progenitors as in intact controls ( Figure 2D ) . To determine if the increase in pharynx progenitors was localized , we analyzed the same-sized regions in tails , farther from the site of amputation . However , we did not detect an increase in pharynx progenitors in tails after either pharynx or head amputation as compared to intact controls ( Figure 2—figure supplement 1C ) . These data suggest that stimulation of pharynx progenitor production is local and requires recognition of lost pharynx tissue , but not necessarily loss of the entire organ . Based on our finding that pharynx progenitors increase only in response to missing pharynx tissue , we hypothesized that the specific pairing of organ loss and progenitor increase would be true for other organs . Besides the pharynx , the eye is the only other planarian organ that is anatomically restricted and thus can be fully removed without leaving any remaining tissue behind ( Lapan and Reddien , 2012; LoCascio et al . , 2017 ) . Expression of the eye-specific transcription factor ovo is required for eye regeneration ( Figure 1A; Flores et al . , 2016; Lapan and Reddien , 2012; Rouhana et al . , 2013; Scimone et al . , 2011; Scimone et al . , 2017; Scimone et al . , 2014a ) and ovo+ piwi-1+ eye progenitors increase after decapitation ( Lapan and Reddien , 2012 ) . Conversely , following pharynx amputation , this increase did not occur ( Figure 2E , Figure 2—figure supplement 2A ) , indicating that pharynx loss does not stimulate the production of eye progenitors . We also quantified the responses of organ-specific progenitors for muscle ( myoD+ ) , intestine ( gata-4/5/6+ ) , the excretory system ( six-1/2+ ) and the nervous system ( pax6a+ ) ( Figure 1A; Flores et al . , 2016; Scimone et al . , 2011; Scimone et al . , 2017; Scimone et al . , 2014a ) after either pharynx or head removal . With the exception of intestinal progenitors , all others showed a similar behavior , increasing within 3 days after head removal , but not pharynx removal ( Figure 2E , Figure 2—figure supplement 2B–E ) . Analysis of the proportion of progenitors relative to all piwi-1+ stem cells in the same regions yielded comparable outcomes ( Figure 2—figure supplement 1B ) . No changes in any of these organ-specific progenitors were observed in tail regions after pharynx or head amputation , indicating that it is a local response ( Figure 2—figure supplement 1C ) . Together , these data indicate that planarian stem cells sense the loss of missing tissues to initiate their regeneration through the selective expansion of organ-specific progenitors . All these organ-specific transcription factors , with the exception of pax6a , are required for regeneration of their cognate organ ( Adler and Sánchez Alvarado , 2017; Flores et al . , 2016; Lapan and Reddien , 2012; Pineda et al . , 2002; Scimone et al . , 2011; Scimone et al . , 2017 ) . Therefore , to test whether or not these transcription factors regulate pharynx regeneration , we knocked them down with RNAi and assayed feeding ability 7 days after pharynx amputation ( Adler et al . , 2014; Ito et al . , 2001 ) . Unlike FoxA ( RNAi ) , knockdown of other organ-specific progenitor markers did not impact the recovery of feeding ability ( Figure 2—figure supplement 3A ) , despite efficient knockdown and manifestation of known phenotypes ( Figure 2—figure supplement 3B , data not shown ) . Because pax6a is not required for brain regeneration , we performed RNAi of coe , a neural progenitor marker that is required for brain regeneration ( Cowles et al . , 2013 ) . However , coe knockdown did not affect the recovery of feeding ability ( Figure 2—figure supplement 3A ) . Therefore , it is unlikely that any of these other progenitor markers contribute to pharynx regeneration , despite the presence of muscle and neural tissue within the pharynx . The increase in pharynx progenitors following pharynx amputation suggests that stem cells may divide in response to pharynx loss to selectively amplify pharynx progenitors . Because stem cells are the only dividing cells in planarians ( Morita and Best , 1984 ) , we can visualize stem cell division with antibody staining for histone H3Ser10 phosphorylation ( H3P ) . Using this mitotic marker , we verified that stem cell division increased near wounds beginning 1 day after either pharynx or head removal ( Figure 3A; Adler et al . , 2014; Baguñà , 1976 ) . To determine if these dividing stem cells are pharynx progenitors , we combined antibody staining for H3P with FISH for FoxA . One day after pharynx removal , we observed higher numbers of H3P+ pharynx progenitors in regions adjacent to wounds as compared to intact animals ( Figure 3B ) . To determine how soon after amputation these pharynx progenitors initiate division , and how long it persists , we quantified the coincidence of FoxA+ H3P+ cells at various times after pharynx amputation in the pre-pharyngeal region ( Figure 3B ) . Division of pharynx progenitors increased within 6 hr of pharynx amputation , peaked within 2 days , and returned to homeostatic levels by 5 days after amputation ( Figure 3C ) . Despite an overall increase in H3P+ stem cells 1 day after head amputation ( Figure 3A; Baguñà , 1976 ) , numbers of H3P+ pharynx progenitors did not correspondingly increase ( Figure 3B , D ) . Analysis of H3P+ pharynx progenitors relative to all dividing stem cells in the same prepharyngeal region sustained this dichotomy ( Figure 3—figure supplement 1A ) . Interestingly , we were able to detect what appeared to be instances of both symmetric and asymmetric distribution of FoxA in cells undergoing anaphase ( Figure 3—figure supplement 1B ) . Together , these data show that pharynx progenitors are selectively stimulated to divide in response to pharynx loss . In addition to the selective division of excretory system progenitors that occurs after RNAi depletion of excretory tissues ( Thi-Kim Vu et al . , 2015 ) , progenitors in the epidermal lineage have also been shown to divide following head amputation ( van Wolfswinkel et al . , 2014 ) . Therefore , we tested whether non-pharyngeal progenitors are selectively stimulated to divide 1 and 2 days after head amputation . Although the kinetics of each differed slightly , excretory system ( six1/2+ ) , nervous system ( pax6a+ ) , and muscle ( myoD+ ) progenitor division increased 2 days after head amputation , while intestinal ( gata-4/5/6+ ) progenitors did not . Additionally , division of nervous system and muscle progenitors increased only after head but not pharynx amputation ( Figure 3E , Figure 3—figure supplement 2A–D ) . Analysis of these dividing progenitors relative to all H3P+ stem cells recapitulated these results , with the exception of six-1/2+ excretory progenitors , which did not proportionally increase after either amputation ( Figure 3—figure supplement 1C ) . Despite minor differences of each progenitor , the overall trend supports the notion that loss of non-pharyngeal tissues triggers division of stem cells expressing non-pharyngeal progenitor markers while in most cases , pharynx loss does not . This mitotic response appears again to be local , as we did not observe increased division of any of these organ-specific progenitors in tail regions , distant from wounds ( Figure 3—figure supplement 3A–D ) . Even after head amputation , we were unable to detect any dividing eye progenitors ( ovo+H3P+ ) ( Figure 3—figure supplement 4A ) , unless we stalled mitotic exit with the microtubule destabilizing drug nocodazole . Following nocodazole treatment , we detected rare instances of H3P+ eye progenitors; however , they did not increase after head amputation ( Figure 3—figure supplement 4B ) . Therefore , while eye progenitors do divide , their division dynamics do not seem to be affected by injury , similar to intestinal progenitors ( Figure 3E , Figure 3—figure supplement 2D ) . This finding suggests that there may be other eye and intestinal progenitors upstream of those expressing ovo or gata-4/5/6 , or that expansion of these progenitors may occur via transcriptional upregulation . In planarians , pulse-chase experiments using thymidine analogs have shown that cell division contributes to the production of regenerated tissues ( Cowles et al . , 2013; Eisenhoffer et al . , 2008; Forsthoefel et al . , 2011; LoCascio et al . , 2017; Newmark and Sánchez Alvarado , 2000; Wagner et al . , 2011 ) . Our results above identified an elevation in pharynx progenitor division within 2 days after pharynx removal ( Figure 3C ) that correlates with cell cycle entry of stem cells destined for missing pharynx tissue 1 day after amputation ( Figure 1E , F ) . Together , these results define a window of 1–2 days after amputation in which pharynx progenitor division may selectively contribute to pharynx regeneration . Because this timeframe directly precedes the expansion of pharynx progenitors 3 days after pharynx amputation ( Figure 2C ) , we hypothesized that stem cell division increases to specifically generate the progenitors that are necessary for pharynx regeneration . To test this possibility , we blocked stem cell division with nocodazole , which induces a metaphase arrest with as little as 24 hr of exposure , resulting in the accumulation of mitotic ( H3P+ ) stem cells ( Figure 4A; Grohme et al . , 2018; Molinaro et al . , 2021; van Wolfswinkel et al . , 2014 ) . To specifically inhibit mitosis 1–2 days after pharynx amputation , we soaked animals in nocodazole for 24 hr , beginning 1 day after pharynx amputation ( Figure 4B ) . We then assayed pharynx regeneration via recovery of feeding behavior . Animals treated with nocodazole for 24 hr had drastic delays in recovery of feeding , compared to DMSO-treated controls , with only 50% of worms regaining the ability to eat within 20 days , and 100% within 32 days of amputation ( Figure 4C ) . To verify that nocodazole treatment under these conditions delayed pharynx regeneration , we examined pharynx anatomy with the marker laminin , which is strongly expressed in the mouth and pharynx , and weakly expressed in the body where the pharynx attaches ( Adler et al . , 2014; Cebrià and Newmark , 2007 ) . As expected , while residual laminin expression was retained within the body , animals treated with nocodazole 1–2 days after pharynx amputation completely lacked a pharynx with its characteristic layered structure 7 days after amputation and sustained severe defects even up to 14 days ( Figure 4D ) . Treatment with nocodazole for a full 48 hr , beginning immediately after amputation ( Figure 4B ) , did not exacerbate the delay in feeding ability or defects in pharynx anatomy ( Figure 4C , D ) . These findings suggest that stem cell division outside the 1–2 day window has a minor contribution to pharynx regeneration . To verify this , we soaked animals in nocodazole for 24 hr increments surrounding this 1–2 day window , beginning either immediately , or 2 days after pharynx amputation , which recovered feeding behavior at a similar rate as controls ( Figure 4—figure supplement 1A , B ) . Further , animals treated from 0 to 1 days after amputation had only minor defects in pharynx anatomy , while those treated 2–3 days after amputation were normal ( Figure 4—figure supplement 1C ) . Therefore , we conclude that stem cell division within a critical window of 1–2 days after amputation fuels the majority of pharynx regeneration . To test whether stem cell division within this critical window generates pharynx progenitors , we exposed animals to nocodazole 1–2 days after amputation , and analyzed the impact on FoxA+ stem cells . First , we verified that this treatment caused an extensive increase in H3P+ pharynx progenitors 2 days after amputation ( Figure 4—figure supplement 2A , B ) , illustrating that they were arrested in mitosis . Second , we performed double FISH for FoxA and piwi-1 3 days after pharynx removal and found that nocodazole treatment caused a dramatic decrease in pharynx progenitors compared to controls ( Figure 4E , F ) . Importantly , intact animals treated similarly with nocodazole showed no difference in the abundance of pharynx progenitors compared to controls ( Figure 4—figure supplement 2C , D ) . Therefore , perturbing stem cell division during this brief window specifically impacts the production of pharynx progenitors during regeneration . To determine if stem cell division during other times contributed to pharynx progenitor production , we again exposed animals to nocodazole for 0–1 and 2–3 days after pharynx amputation . While we observed some defects in the production of pharynx progenitors , they were more subtle than those in animals treated for 1–2 days after amputation ( Figure 4—figure supplement 2E ) . Together , our data show that stem cell division in a critical window of 1–2 days after amputation produces pharynx progenitors that are likely essential for pharynx regeneration . The mitogen activated protein ( MAP ) kinase pathway drives proliferation and differentiation during development and regeneration in many organisms ( Ghilardi et al . , 2020; Patel and Shvartsman , 2018 ) . In planarians , phosphorylation of the MAP kinase ERK is the earliest known injury-induced signal required for regeneration . ERK activity regulates broad stem cell proliferation and is required for transcriptional changes that drive axial repatterning ( Owlarn et al . , 2017; Tasaki et al . , 2011 ) . To determine whether pharynx loss also induces ERK phosphorylation , we performed a western blot with an antibody against phosphorylated ERK ( pERK ) . pERK increased 15 min after pharynx amputation and returned to baseline levels within 6 hr ( Figure 5A ) . Therefore , similar to other injuries ( Owlarn et al . , 2017 ) , pharynx amputation also activates ERK by phosphorylation soon after injury . Exposing animals to PD0325901 ( PD ) , an inhibitor of the upstream ERK-activating MEK kinase , blocks ERK phosphorylation and permanently inhibits regeneration following substantial anterior tissue removal ( Owlarn et al . , 2017 ) . To determine whether ERK is also required for pharynx regeneration , we exposed animals to PD for 5 days immediately after pharynx amputation and then assayed feeding behavior ( Figure 5B , C ) . While DMSO-treated control animals regained the ability to feed within 7 days , animals treated with PD from 0 to 5 days after amputation had substantial delays in feeding , with 50% of worms feeding by day 13 and all worms feeding by day 29 ( Figure 5C ) . Depending on the timing , delaying administration of MEK inhibitors relative to amputation partially or completely rescues anterior regeneration , and suggests that ERK acts within the first day of regeneration ( Owlarn et al . , 2017 ) . To pinpoint when ERK signaling is important for pharynx regeneration , we delayed PD exposure for 1 or 2 days after pharynx amputation , and again assayed feeding ( Figure 5B , C ) . Animals exposed 1 day after pharynx amputation ( 1–6 days ) had delayed feeding ability , similar to those treated immediately after amputation ( 0–5 days ) , suggesting that ERK activity within the first day of amputation is dispensable for pharynx regeneration . Animals exposed 2 days after amputation ( 2–7 days ) regained the ability to feed at rates similar to controls ( Figure 5C ) , indicating that ERK is essential for regeneration within the first 2 days after pharynx amputation . Therefore , ERK likely acts primarily between 1 and 2 days after pharynx amputation . This timing occurs after the increase in pERK following injury has already subsided ( Figure 5A ) , suggesting that pharynx regeneration may be facilitated by homeostatic levels of ERK signaling instead of its injury-induced high level activation . To verify that the inability of ERK-inhibited animals to feed was caused by defects in regeneration , we analyzed pharynx anatomy 7 days after pharynx amputation with FISH for laminin . Animals exposed to PD for 5 days , beginning 0 or 1 day after amputation , lacked a pharynx , while pharynges in animals exposed beginning 2 days after amputation were comparable to controls ( Figure 5D ) , mirroring the results of our feeding assay ( Figure 5C ) . PD-exposed animals eventually regenerated a pharynx within about 2 weeks ( Figure 5C , Figure 5—figure supplement 1A ) , suggesting that ERK inhibition does not permanently block pharynx regeneration . Therefore , we confirmed the effects of ERK inhibition by repeating pharynx regeneration experiments with UO126 ( UO ) , another potent , but structurally independent , MEK inhibitor and observed the same outcomes ( Figure 5—figure supplement 1B–D ) . Furthermore , exposure to PD or UO eliminated pERK after pharynx amputation on a western blot ( Figure 5—figure supplement 1E ) and also permanently blocked regeneration after extensive anterior tissue removal , for up to 70 days ( Figure 5—figure supplement 1F ) . Together , these results define a window 1–2 days after amputation in which ERK activity is required for pharynx regeneration . Soon after amputation , ERK signaling is required for upregulation of several genes including follistatin ( fst ) ( Owlarn et al . , 2017 ) , which accelerates regeneration by inhibiting activin-1 and -2 ( Gaviño et al . , 2013; Roberts-Galbraith and Newmark , 2013; Tewari et al . , 2018 ) . Like ERK inhibition , fst ( RNAi ) prevents regeneration following substantial anterior tissue removal ( Owlarn et al . , 2017; Tewari et al . , 2018 ) . However , if less tissue is removed , the requirement for fst diminishes ( Tewari et al . , 2018 ) . Likewise , when we amputated animals pre-pharyngeally and maintained them to PD for 5 days , head regeneration was initially delayed in 100% of animals 7 days after amputation , but eventually occurred in 88% of animals within 2 weeks ( Figure 5—figure supplement 1G ) , on a similar timeline as pharynges ( Figure 5C , D , Figure 5—figure supplement 1A ) . Therefore , ERK may mediate pharynx regeneration via fst expression , resulting in a short-lived block of pharynx regeneration . To test this , we analyzed fst expression , which increases within 6 hr of head amputation ( Gaviño et al . , 2013 ) , but not until 24 hr after pharynx amputation ( Figure 5—figure supplement 2A ) , suggesting that injury-induced ERK activation ( Figure 5A ) may not always coincide with fst upregulation . Despite upregulation of fst expression after pharynx amputation , fst ( RNAi ) animals regained the ability to feed at a normal rate ( Figure 5—figure supplement 2B and C ) , indicating that fst is not required to accelerate pharynx regeneration . Therefore , although pharynx loss eventually induces fst expression , regulation of pharynx regeneration via ERK is independent of fst . Our data suggests that ERK activity is required 1–2 days after pharynx amputation , just prior to the emergence of pharynx progenitors 3 days after amputation . We hypothesized that ERK may promote the production of these progenitors , which we tested by maintaining animals in PD for 3 days following pharynx amputation ( Figure 5E ) . PD exposure significantly inhibited the increase in FoxA+ piwi-1+ cells typically observed after pharynx amputation ( Figure 5E , F ) . Importantly , intact animals treated with PD for 3 days showed no difference in the abundance of pharynx progenitors as compared to controls ( Figure 5—figure supplement 3A ) , indicating that ERK activity for this duration is not necessary for maintaining pharynx progenitors during homeostasis . Therefore , the decrease in pharynx progenitors following pharynx amputation and PD treatment is likely due to their reduced production , rather than altered survival . To determine when ERK promotes pharynx progenitor production , we exposed animals to PD in 24 hr increments during this 3-day window ( Figure 5F ) . While PD treatment starting 0 or 2 days after pharynx amputation had no effect on pharynx progenitors , treatment starting 1 day after amputation significantly inhibited pharynx progenitor increase , similar to those treated for three full days ( Figure 5F ) . Similar experiments with UO yielded the same outcomes ( Figure 5—figure supplement 3A , B ) , demonstrating that ERK activity 1–2 days after pharynx amputation promotes the production of pharynx progenitors during regeneration . To test whether PD-mediated inhibition of pharynx progenitor production following pharynx loss has long-term consequences on pharynx regeneration , we exposed animals to PD for 0–3 days and 1–2 days after amputation , and assayed feeding behavior . These animals had delays in feeding , although less severe than those treated for five full days ( Figure 5—figure supplement 3C ) , presumably because animals start recovering after drug washout . Meanwhile , those treated for 0–1 day after pharynx amputation recovered feeding ability at normal rates ( Figure 5—figure supplement 3C ) , illustrating that PD exposure that does not affect the production of pharynx progenitors does not delay regeneration . Together , these results indicate that ERK activity 1–2 days after amputation promotes pharynx regeneration by contributing to the increase in pharynx progenitors 3 days after amputation . Because ERK signaling regulates broad stem cell division associated with missing tissue in planaria ( Owlarn et al . , 2017 ) , the decrease in pharynx progenitor production after drug exposure could be a result of reduced stem cell division . To determine if amputation-induced pharynx progenitor division depends on ERK activity , we exposed animals to MEK inhibitors 1–2 days after pharynx amputation , and analyzed FoxA+ H3P+ stem cells . Neither PD or UO exposure substantially impacted the number of dividing pharynx progenitors as compared to controls ( Figure 5G , H , Figure 5—figure supplement 3D ) , despite an overall decrease in H3P+ stem cells in the same animals ( Figure 5—figure supplement 3E ) . These data indicate that ERK signaling is not required for the selective increase in pharynx progenitor division induced by pharynx loss . Because the timing of ERK’s requirement for increasing pharynx progenitors overlaps with when stem cell division is required ( Figure 4C–F ) 1–2 days after pharynx removal , ERK signaling is unlikely to regulate initiation of pharynx progenitor division . Instead , ERK likely promotes pharynx progenitor production and regeneration by regulating FoxA expression and stem cell differentiation . Unlike pharynx removal , selective removal of the eye does not increase stem cell division or expansion of eye-specific progenitors . Instead , following resection , the eye regenerates by passive homeostatic turnover that is not regulated by the presence or absence of the eye ( LoCascio et al . , 2017 ) . Therefore , we speculated that eye regeneration , following specific removal , may not have the same requirements as the pharynx . To test whether eye regeneration relies on stem cell division and ERK signaling , we performed eye-specific resections and immediately exposed animals to either nocodazole or MEK inhibitors ( Figure 6A ) . We monitored eye regeneration with FISH for ovo and the eye-specific marker opsin Sánchez Alvarado and Newmark , 1999 . We confirmed that eye tissue was successfully removed by the absence of ovo+ opsin + photoreceptors immediately after surgery ( Figure 6B ) . Exposure to either nocodazole for 2 days , or MEK inhibitors for 5 days , did not impact eye regeneration ( Figure 6B , C ) , despite a complete block of pharynx regeneration under these conditions ( Figure 4D , Figure 5D , Figure 5—figure supplement 1D ) . Our results show that , unlike pharynx regeneration , eye regeneration does not require stem cell division or ERK activity after selective removal . Instead , much like the maintenance of pharynx progenitors in intact animals ( Figure 4—figure supplement 2D , Figure 5—figure supplement 3A ) , eye regeneration is unaffected by drug treatments . Therefore , it is possible that pre-existing eye progenitors , or more likely , the continued production of homeostatic levels of eye progenitors ( LoCascio et al . , 2017 ) is sufficient for eye regeneration following selective removal . Previous studies have shown that eye regeneration may occur in alternative ways depending on the context of the wound . While eye-specific resections do not invoke typical injury responses , more severe injuries stimulate the expansion of ovo+ eye progenitors ( Lapan and Reddien , 2012; LoCascio et al . , 2017 ) . Because wounds generated from eye resection do not stimulate the same response as more severe injuries , their repair may not depend on the same mechanisms facilitated by proliferation and ERK signaling . Therefore , we tested whether stem cell division and ERK activity are required for eye regeneration after head amputation ( Figure 6A ) . In controls , ovo+ opsin + photoreceptors re-emerged 7 days after amputation ( Figure 6D , E; Lapan and Reddien , 2011 ) . By contrast , animals treated with either nocodazole for 2 days or ERK inhibitors for 5 days failed to regenerate eyes , despite the presence of ovo+ eye progenitors ( Figure 6D , E ) . We conclude that unlike pharynx regeneration , eye regeneration only requires stem cell division and ERK signaling in the context of more severe injuries . Previous work has suggested two potential models underlying planarian regeneration . One group proposed a non-targeted model , in which stem cells broadly incorporate into both damaged and undamaged tissue , dependent on indiscriminate amplification of progenitors triggered by nearby wounds ( LoCascio et al . , 2017 ) . On the other hand , a targeted model suggests that stem cells amplify organ-specific progenitors in response to perturbations to organ tissues ( Thi-Kim Vu et al . , 2015 ) . By labeling proliferating stem cells with F-ara-EdU at different times after amputation , we have uncoupled the contribution of these two mechanisms to the production of regenerated tissues . Depending on the type of injury , cells generated soon after amputation are channeled into all nearby tissues , even undamaged ones . However , those generated 1 day after amputation are selectively targeted for missing tissues . Importantly , the timing of this targeted mechanism overlaps with an essential peak in pharynx progenitor division , revealing that this mechanism generates organ progenitors necessary for the replacement of missing tissues . The cells that regenerate the pharynx are primarily generated through this targeted mechanism , mediated by stem cell division and ERK signaling between 1and 2 days after amputation ( Figure 7 ) . Smaller lesions , such as eye resections , do not stimulate a proliferative wound response ( LoCascio et al . , 2017 ) , nor do they depend on stem cell division and ERK signaling for subsequent regeneration . A previous study proposed that regeneration initiation requires signals regulated by both injury and tissue loss ( Owlarn et al . , 2017 ) . If eye removal is not detected as an injury , and is insufficient to trigger a ‘missing tissue signal’ , this could explain why eye regeneration , following resection , happens passively via homeostatic cellular turnover in the regenerating eye ( LoCascio et al . , 2017 ) . Further , while surgical removal of the pharynx does increase production of neural tissue ( LoCascio et al . , 2017 ) , selective chemical pharynx removal does not , suggesting that non-targeted regeneration may require injury to specific types of tissues , such as body wall muscle or epithelia , in addition to tissue loss . Therefore , multiple avenues lead to regeneration: passive homeostatic turnover in the absence of typical injury responses , and both targeted and non-targeted mechanisms that increase cellular production when tissue is lost . Our work highlights several key features that are important to consider for future analysis of regeneration . In particular , depending on the size and severity of the wound , initial injury signals may vary , triggering a differential requirement for stem cell division and regulatory signaling pathways . Our results underscore the distinct requirements for proliferation and ERK activity in regeneration of eyes and the pharynx . Also , the timing of experimentation and analysis should be deliberate , as we show that F-ara-EdU-labeled stem cells are differentially incorporated into regenerating and non-regenerating tissues depending on when they are labeled relative to injury . Another key consideration is the intrinsic heterogeneity of cell populations that are analyzed . Stem cells identified by piwi-1 and H3P staining encompass cells with different potencies and differentiation states ( van Wolfswinkel et al . , 2014; Zeng et al . , 2018 ) , which may respond uniquely to the changing environment of a regenerating animal . Therefore , broad analysis of stem cells lacks the resolution required to tease apart the intricacies involved in coordinating regeneration . Restricting our analysis to organ progenitors and even further to those that are actively dividing narrows this focus . However , our observation of the asymmetric and symmetric segregation of FoxA in dividing cells suggests that heterogeneity exists even within these subsets of stem cells . Resolving the specific stem cells responsible for driving different modes of regeneration , and when cell fate is established in them , will be an exciting area for future work . Phosphorylation of ERK promotes regeneration in many animals ( DuBuc et al . , 2014; Wan et al . , 2012; Yun et al . , 2014 ) . In planaria , ERK has been implicated as a regeneration trigger , as it is briefly activated by phosphorylation within minutes of injury and is required for wound-induced transcription , stem cell differentiation and broad stem cell proliferation ( Owlarn et al . , 2017; Tasaki et al . , 2011 ) . ERK also functions to re-establish axial patterning during regeneration ( Owlarn et al . , 2017; Umesono et al . , 2013 ) , which depends on a network of positional cues that are expressed in muscle cells throughout the body ( Lander and Petersen , 2016; Scimone et al . , 2016; Witchley et al . , 2013 ) . Because tissue removal from the body requires re-establishment of these positional cues for regeneration to proceed ( Rink , 2018 ) , it has been difficult to distinguish ERK’s roles in organ regeneration . Unlike amputations to the body , pharynx removal does not broadly disrupt positional cues , which has allowed us to pinpoint a distinct role for ERK in organ regeneration . Both ERK activity and stem cell division act simultaneously , but independently , 1–2 days after pharynx loss , to drive the expansion of pharynx progenitors 3 days after amputation . These events occur after the injury-induced increase in pERK has subsided . Also , ERK activity is dispensable for pharynx progenitor division . Taken together , these results suggest that an ERK-independent signal triggers division of FoxA+ stem cells , and that ERK acts later during organ regeneration to facilitate stem cell differentiation or maintain cell fate . Among the ERK-dependent wound-induced genes is follistatin ( fst ) ( Owlarn et al . , 2017 ) , which promotes regeneration in ways similar to ERK ( Tewari et al . , 2018 ) . Interestingly , both ERK and fst are absolutely essential for regeneration following substantial anterior tissue removal , but become less so if smaller amounts of tissue are removed . This variability may be due to the extent of axial patterning disruption induced by different amputations . In fact , the inability of fst ( RNAi ) animals to regenerate is entirely dependent on their failure to reset positional information , as inhibition of wnt signals that restrict head formation rescue these regeneration defects ( Tewari et al . , 2018 ) . Therefore , it is likely that injury-induced fst expression and ERK phosphorylation primarily regulate regeneration initiation by establishing repatterning rather than triggering stem cell behaviors that directly contribute to organ regeneration . ERK inhibition reduces broad stem cell division after tissue removal ( Owlarn et al . , 2017 ) , but not the specific increase in division of pharynx progenitors that accompanies pharynx loss , suggesting that not all injury-induced stem cell behaviors may be critical for regeneration . Further , rescuing axial repatterning , and thus regeneration after fst knockdown , does not rescue defects in missing tissue-induced stem cell proliferation and apoptosis ( Tewari et al . , 2018 ) . Therefore , the ‘missing-tissue response’ may encompass multiple events including patterning , broad stem cell division , and the generation of organ-specific progenitors that contribute independently to different aspects of regeneration . Receptor tyrosine kinases such as the epidermal growth factor receptor ( EGFR ) and the fibroblast growth factor receptor ( FGFR ) have been shown to play critical roles in signaling upstream of ERK in many organisms ( Patel and Shvartsman , 2018 ) , making them intriguing candidates for potential regulators of regeneration . In planarians , egfr-3 is required to activate ERK during regeneration ( Fraguas et al . , 2017 ) and is also involved in stem cell differentiation ( Fraguas et al . , 2011; Lei et al . , 2016 ) . Other studies have highlighted roles for the ligand egf-4 and the receptors egfr-1 and egfr-5 in the differentiation of stem cells into brain , intestinal and excretory tissues , respectively ( Barberán et al . , 2016; Fraguas et al . , 2014; Rink et al . , 2011 ) . Further , some FGFRL-Wnt circuits restrict pharynx formation to the trunk region , possibly through regulation of FoxA expression ( Lander and Petersen , 2016; Scimone et al . , 2016 ) . Whether any of the planarian EGF or FGF ligands or receptors similarly regulate the production of pharyngeal progenitors remains to be determined ( Cebrià et al . , 2002; Ogawa et al . , 2002 ) . By studying the dynamics of FoxA expression in stem cells after pharynx or head removal , we have uncovered shifts in stem cell heterogeneity that depend on the presence or absence of a particular organ . Intriguingly , we find that both the overall number of FoxA+ pharynx progenitors , as well as those that are actively dividing , increase only after pharynx removal . Therefore , stem cells sense the absence of the pharynx and channel their proliferative output toward the population of stem cells required to regenerate it . Combined with our analysis of non-pharyngeal progenitor dynamics after different amputations , our results suggest that the heterogeneity of the stem cell population can be differentially deployed depending on the severity of the injury and the particular tissues that need repair . Interestingly , removal of non-pharyngeal tissues like the head , does not increase pharynx progenitor proliferation but nevertheless results in increased F-ara EdU+ cells within the pharynx , raising a conundrum regarding the source of these cells . One possibility is that other , non-FoxA+ progenitors which have yet to be identified , may contribute to pharynx regeneration . However , no other tissue-specific transcription factors or proposed pharynx progenitor markers ( meis , twist , or dd_554 ) appear to be required for pharynx regeneration ( Cowles et al . , 2013; Scimone et al . , 2014a; Zhu et al . , 2015 ) . Alternatively , a recent study has suggested that planarian stem cells , even those expressing organ-specific transcription factors , may harbor a large degree of plasticity that allows fate switching between stem cell and progenitor types ( Raz et al . , 2021 ) . While this hypothesis has not been tested in the context of injury , it is possible that stem cells generated soon after tissue loss could adopt a pharynx progenitor fate at various times over the course of regeneration , which would not necessarily generate a detectable increase in pharynx progenitors at any one time . It will be interesting to explore the potential of these cells in more detail when true lineage-tracing becomes possible in planarians . Surprisingly , stem cell division in a narrow window , 1–2 days following pharynx amputation , is absolutely essential for pharynx regeneration , and coincides with the elevation of pharynx progenitor division that directly precedes their increase 3 days after pharynx amputation . The requirement for division in this brief moment after amputation suggests that stem cells detect tissue loss through transient signals regulated by injury . Indeed , a recent study has identified a population of potentially slow-cycling stem cells , reminiscent of reserve stem cells in mammals , that may be specifically induced to enter the cell cycle by tissue loss ( Bankaitis et al . , 2018; Molinaro et al . , 2021 ) . Whether this distinct population of stem cells contributes to the regeneration of particular organs is not known . Regulatory signals could either be produced upon injury to promote regeneration , or released from inhibitory cues that might emanate from organs when they are present ( Rink , 2018; Ziller-Sengel , 1967; Ziller-Sengel , 1965 ) . Intriguingly , a recent study characterizing transient amputation-induced transcriptional changes revealed that the majority of these changes occur within differentiated cell types ( Benham-Pyle et al . , 2020 ) . The possibility of transient signals customized to particular organs and the ability of stem cells to readily sense them may explain how planarians exhibit such rapid and robust regeneration of all organs . Cell fate acquisition can occur throughout the cell cycle ( Fichelson et al . , 2005; Pauklin and Vallier , 2014; Soufi and Dalton , 2016 ) . The increase in FoxA expression in both actively dividing stem cells , and those outside of M-phase , does not pinpoint a particular time in the cell cycle where fate acquisition during regeneration might occur . Tissue loss could generate fleeting signals sensed by stem cells that influence them to adopt a specific cell fate during division to compensate for missing tissue . Alternatively , stem cells already expressing organ-specific markers may be poised to divide upon receiving such a signal , allowing them to quickly initiate regeneration upon exit of the cell cycle . Indeed , studies in human hepatoma cell lines have shown that FoxA1 remains attached to chromatin during mitosis , contributing to rapid activation of downstream targets following mitosis during liver differentiation ( Caravaca et al . , 2013 ) . It will be interesting to explore these possibilities in future studies . Mammalian homologs of FoxA were the first identified ‘pioneer’ transcription factors , characterized by their ability to engage closed chromatin and drive organogenesis ( Hsu et al . , 2015; Iwafuchi-Doi and Zaret , 2016; Lam et al . , 2013; Zaret and Mango , 2016 ) . This raises the possibility that pioneer factors may be viable in vivo targets for achieving regeneration of entire organs . In fact , overexpression of a related mammalian transcription factor , FoxN , is sufficient to drive regeneration of the thymus in mice ( Bredenkamp et al . , 2014 ) . The increased proliferation of stem cells expressing FoxA after pharynx removal suggests that activation of pioneer factors may also drive organ regeneration in planaria . Other pioneer factors , including gata-4/5/6 , soxB1-2 and FoxD , are also expressed in planarian stem cells and are required for regeneration of the intestine ( Flores et al . , 2016; González-Sastre et al . , 2017 ) , sensory neurons ( Ross et al . , 2018 ) , and anterior pole ( Scimone et al . , 2014b; Vogg et al . , 2014 ) , respectively . Therefore , upregulation of pioneer factors in stem cells may be a general strategy used to initiate organ regeneration . Identifying the regulatory mechanisms responsible for the selective activation of pioneer factors in stem cells may be an ideal approach to understanding how organisms initiate regeneration of targeted organs in vivo . In conclusion , our work sheds light on the flexibility and dynamic responses of stem cells to different injuries , and highlights potential mechanisms to activate organ-specific transcriptional programs required for regeneration . Animals of Schmidtea mediterranea asexual clonal line CIW4 were maintained in a recirculating water system ( Arnold et al . , 2016; Merryman et al . , 2018 ) containing Montjuïc salts ( planaria water ) ( Cebrià and Newmark , 2005 ) . Prior to experiments , animals were transferred to static culture and maintained in planaria water supplemented with 50 µg/mL gentamicin sulfate . Animals used for experiments were between 2 and 3 mm in length and starved for approximately 5–7 days . Pharynx removal was performed by chemical amputation as previously described ( Adler et al . , 2014; Shiroor et al . , 2018 ) . Planarians ( 2–3 mm in size ) were placed in 100 mM sodium azide diluted in planaria water . After 4–7 min , the pharynx extended out of the body and was plucked off using fine forceps ( #72700-D; Electron Microscopy Sciences ) . Animals were kept in sodium azide for no longer than 10 min , rinsed three times , and then transferred into a fresh dish . For pharynx incisions and partial amputations , animals were soaked in tricaine solution ( 4 g/L in 21 mM Tris pH 7 . 5 ) diluted 1:3 in planaria water which causes the pharynx to extend but not detach . Pharynx incisions were created by using forceps to snip along the length of the pharynx . For partial pharynx amputations , the proximal end of the pharynx was snipped off with forceps or trimmed with a scalpel . To resect eyes , animals were immobilized on moist filter paper , and eyes were scraped out using the tips of fine forceps . All other amputations and injuries were performed with a micro feather scalpel ( #72046–15 or #72045–45; Electron Microscopy Sciences ) . For direct comparisons to pharynx-amputated animals , head-amputated and intact animals were soaked in sodium azide for 2–3 min . Animals were soaked in 0 . 5 mg/mL F-ara-EdU ( Sigma T511293 ) in planaria water containing 3% DMSO for 4 hr either immediately or 24 hr after amputation and fixed 7 days after amputation . Animals were fixed as previously described ( Pearson et al . , 2009 ) with minor modifications . Briefly , animals were killed in 7 . 5% N-acetyl-cysteine in PBS for 7 . 5 min and fixed in 4% paraformaldehyde in PBSTx ( PBS + 0 . 3% Triton X-100 ) for 30 min . Worms were then rinsed twice with PBSTx and incubated in pre-warmed reduction solution ( PBS + 1% NP-40 + 50 mM DTT + 0 . 5% SDS ) at 37°C for 10 min . Worms were rinsed twice more with PBSTx , dehydrated in a methanol series and stored at −20°C . For F-ara-EdU detection , following fixation , animals were rehydrated and bleached in 6% H2O2 overnight . Animals were then treated with proteinase K ( 10 µg/mL proteinase K and 0 . 1% SDS in PBSTx ) for 15 min , and post-fixed in 4% formaldehyde in PBSTx for 10 min . A F-ara-EdU development solution was made containing PBS + 1 mM CuS04 and 100 µM Oregon Green 488 azide ( Thermo Fisher O10180 ) . Freshly made 100 mM ascorbic acid was added to this solution immediately before administering it to samples , which were then incubated for 30 min in the dark . Following a few rinses with PBSTx , animals were post-fixed , rinsed 2x in PBSTx , and put through in situ ( see below ) . Following in situ , animals were placed in K block ( 5% inactivated horse serum , 0 . 45% fish gelatin , 0 . 3% Triton-X and 0 . 05% Tween-20 diluted in PBS ) at room temperature for 4 hr or 4°C overnight . To detect F-ara-EdU , animals were incubated with 1:1000 anti-Oregon Green-HRP ( Thermo Fisher A21253 ) and counterstained with DAPI in K block at 4°C overnight . Antibodies were washed off in PBSTx , pre-incubated with tyramide ( 1:2000 FAM ) for 10 min and developed for 15 min . Colorimetric in situ hybridizations were performed as described in Pearson et al . , 2009 using anti-DIG-AP ( Roche 11093274910 ) at 1:3000 . Fluorescent in situ hybridizations were performed as in King and Newmark , 2013 with minor modifications . Briefly , animals were rehydrated and bleached ( 5% formamide , 1 . 2% H2O2 in 0 . 5x SSC ) for 2 hr , then treated with proteinase K ( 4 µg/mL in PBSTx , Thermo Fisher 25530049 ) . Following overnight hybridizations at 56°C , samples were washed 2x each in wash hybe ( 5 min ) , 1:1 wash hyb:2X SSC-0 . 1% Tween 20 ( 10 min ) , and 2X SSC-0 . 1% Tween 20 ( 30 min ) , 0 . 2X SSC-0 . 1% Tween 20 ( 30 min ) at 56°C followed by 3 × 10 min PBSTx washes at room temperature . Subsequently , animals were placed in blocking solution ( 0 . 5% Roche Western Blocking Reagent and 5% inactivated horse serum diluted in PBSTx ) . Animals were then incubated with an appropriate antibody: 1:1000 anti-DIG-POD ( Roche 11207733910 ) or 1:1000 anti-FITC-POD ( Roche 11426346910 ) in blocking solution at 4°C overnight followed by several washes with PBSTx . For development with FAM ( 1:2000 ) or Cy3 ( 1:7500 ) , animals were preincubated with tyramide in borate buffer for 30 min and then developed with 0 . 005% H2O2 in borate buffer for 45 min . For development with rhodamine , animals were pre-incubated with tyramide ( 1:5000 ) for 10 min and developed for 15 min . To inactivate peroxidases , animals were treated with 200 mM sodium azide or 4% H2O2 in PBSTx for 1 hr , then rinsed with PBSTx >6 times before application of the next antibody . For H3P detection , following in situ , animals were incubated with anti-phosphohistone H3 ( Ser10 ) antibody ( Abcam , Cambridge , MA Ab32107 ) diluted 1:1000 in blocking solution ( 0 . 5% Roche Western Blocking Reagent and 5% inactivated horse serum in PBSTx ) for 2 days at 4°C . Primary was washed off with PBSTx followed by incubation with goat anti-rabbit-HRP ( Thermo Fisher 31460 ) diluted 1:2000 in PBSTx overnight at 4°C . Antibody was washed off with PBSTx and samples were pre-incubated and developed with rhodamine tyramide as described above . For all in situ and immunostaining experiments , DAPI [5 µg/mL] ( 1:5000 dilution; Thermo Scientific ) was added along with the last antibody ( except for colorimetric in situ ) . After the final development , animals were soaked in ScaleA2 ( 4M urea , 20% glycerol , 0 . 1% Triton X-100 , 2 . 5% DABCO ) ( Hama et al . , 2011 ) for at least 3 days . Animals were mounted ventral side up except for those stained for ovo , which were dorsal side up , and embedded in Aqua-Polymount ( Polysciences Inc 18606 ) . To maintain consistent sample thickness , animals were mounted in wells cut from a double layer of double stick tape ( Scor-Pal 6’ wide Scor-Tape 209 ) . Ten animals per condition were snap frozen in lysis buffer ( 50 mM Hepes pH 7 . 5 , 1 mM EGTA , 1 mM MgCl2 , 100 mM KCl , 10% glycerol , 0 . 05% NP40 , and 0 . 5 mM DTT ) ( Zanin et al . , 2011 ) containing Pierce protease and phosphatase inhibitors ( Thermo Fisher A32965 and A32957 ) . A cup horn sonicator ( Branson Ultrasonics Corporation , Danbury , CT ) chilled to 4°C was used to generate extracts by sonication for a total of 2 min with 1 s pulses at 90% amplitude . Total protein was quantified using a NanoDrop OneC ( Thermo Fisher ) . After quantification , Bolt LDS sample buffer ( Life Technologies B0007 ) was added to the extracts and 100 µg of each sample was run on a polyacrylamide gel ( Bolt 4–12% Bis-Tris , Invitrogen NW04125BOX ) . The gel was transferred onto a PVDF Immobilon membrane ( Merck Millipore IPFL00010 ) using the Pierce Power Blot Cassette system ( Thermo Scientific ) , then treated with Odyssey blocking buffer ( LI-COR 927–40000 ) for 1 hr at RT . Membranes were incubated overnight at 4°C with mouse anti-tubulin ( Sigma/Millipore T5168 ) and rabbit anti-Phospho-p44/42 MAPK ( Erk1/2 ) ( Cell Signaling Technologies 4370S ) , diluted 1:1000 in blocking buffer . The blot was washed 3 × 10 min in TBST ( TBS +10% Tween 20 ) and incubated for 1 hr at RT with goat anti-mouse Alexa Flour 488 ( Thermo Fisher A11029 ) and goat anti-rabbit IRDye 800CW ( LI-COR 926–32211 ) secondary antibodies ( diluted 1:4000 and 1:20 , 000 respectively in Odyssey blocking buffer ) . Membranes were then washed as before and imaged using a Bio-Rad ChemiDoc MP . Western blots were repeated at least twice with comparable results . Nocodazole ( Sigma M1404 ) was administered in 24 or 48 hr increments at 50 ng/mL . PD0325901 ( EMD Millipore/Calbiochem 4449685 MG ) and UO126 ( Cell Signaling Technologies 9903S ) were administered at 10 µM and 25 µM , respectively . Drugs were diluted in planaria water containing 0 . 05% DMSO . Animals were rinsed three times after treatment and either fixed immediately , or transferred to a new dish and rinsed daily until further experimentation . Animals were fed 20 µL of colored food ( 4:1 liver:milliQ water with 2% red food coloring ) in a petri dish . After approximately 30 min , the number of animals with red intestines were scored . For time courses , feeding assays started at 4 days post-amputation and any animals that ate were removed from the dish . Feeding assay time courses were repeated at least three times with ~20 animals assayed per experiment . RNAi was performed as previously described ( Rouhana et al . , 2013 ) , with in vitro-synthesized double-stranded RNA ( dsRNA ) . dsRNA was diluted to a final concentration of 400 ng/µL in colored food . RNAi food was administered every 3 days , six times in total , except for gata-4/5/6 and six-1/2 , which caused phenotypes after 1–2 feeds . C . elegans unc22 dsRNA was used as a control . Amputations were carried out 5–7 days after the last feed . All RNAi experiments were repeated at least twice with ~10 animals per experimental group . Whole-mount colorimetric in situ hybridizations and live worms were imaged on a Leica M165F . Fluorescent in situ hybridizations were imaged on a Zeiss 710 confocal microscope using a 25x objective with 2 . 28 µm z-sections . ImageJ software was used for processing and quantification ( Schindelin et al . , 2012 ) . All samples were quantified without blinding by manual examination of optical sections of overlaid fluorescence channels in pre-defined regions of animals as indicated in figures . Cells were identified as positive for markers if fluorescence coincided with DAPI signal and was easily distinguishable from background levels , as demonstrated in figures with images . Quantification was performed in a minimum of 6 animals per experimental group . For piwi-1+ progenitor analysis , a 6000 μm2 region in the same location of the pre-pharyngeal area was captured at 1 . 3x zoom . Quantification included 20 z-sections ( 45 . 6 µm ) beginning at the first piwi-1+ cell . For H3P analysis , the entire pre-pharyngeal region was imaged , captured at 0 . 6x zoom . Quantification included 30 z-sections ( 68 . 4 µm ) beginning at the first H3P+ cell and was normalized to area . Representative confocal images are partial projections of ~5 z-sections from regions that were used for quantification , re-imaged at 4x zoom . For F-ara-EdU quantification , images were captured at 1x zoom . In most cases , the pharynges and brains of each experimental group were imaged from the same animal . All visible F-ara-EdU+ cells in the pharynx and all F-ara-EdU+ ChAT+ cells in the brain were quantified throughout the entire organ . Representative F-ara-EdU images are projections of the entire analyzed region . In bar graphs , symbols represent individual animals , and shapes distinguish biological replicates . Statistical analysis was performed using PRISM-Graphpad version nine or GraphPad Quick Calcs to perform one-way ANOVA with Tukey test , unpaired t-test or Fisher’s Exact test as indicated in figure legends . p≤0 . 05; p≤0 . 01; p≤0 . 001; and **p≤0 . 0001 . Sequences for all transcripts used in this study were cloned using the following primers: GeneSmed IDForward primerReverse primerpiwi-1dd_Smed_v6_659_0_1gaccaagaagaggaggtctccgcgttcgcgaattctgtcattFoxAdd_Smed_v6_10718_0_4aacgacctcaacggaatgtttcatgcgccaaagttaaggataovodd_Smed_v6_48430_0_1aatgcccacagatttgtccataaagtgaattcgggtgmyoDdd_Smed_v6_12634_0_1ctattccggtccatactcagcactcttgatcaactttcctcggata-4/5/6dd_Smed_v6_4075_0_1gtccgtaagatccacgatccgtgattgaggaatagggcttcgsix-1/2dd_Smed_v6_9774_0_1ccttgtcagggatctaatccggtgaggatgataagttgggpax6add_Smed_v6_17726_0_1ctgggcataaatcaaaccgccttgggggataaactgatcccoedd_Smed_v6_9893_0_1cgaagagcagacaacagcacttttaccaacacccgattgclaminindd_Smed_v6_8356_0_1agtcgctggcaaagtgcatctaatgatgcgtggtatccacagfstdd_Smed_v6_9584_0_1cagtggtgtgcaatttagcgagttcgcaggtattcttggtttcgtaattcgChATdd_Smed_v6_6208_0_1tcggttgctgaaggtattgcaggcatatagcattctacacgg	Many animals can repair and regrow body parts through a process called regeneration . Tiny flatworms called planaria have some of the greatest regenerative abilities and can regrow their whole bodies from just a small part . They can do this because around a fifth of their body is made of stem cells , which are cells that continuously produce new cells and turn into other cell types through a process called differentiation . Measuring the gene activity in stem cells from planaria shows that these cells are not all the same . Different groups of stem cells have specific genes turned on which are needed to regrow certain body parts . It is unclear whether all stem cells respond to injuries in the same way , or whether the stem cells that respond are specific to the type of injury . For example , stem cells needed to repair the gut may respond more specifically to gut injuries than to other damage . Bohr et al . studied how stem cells in planaria respond to different injuries , by comparing an injury to a specific organ to a more serious injury involving several organs . The specific injury was the loss of the pharynx , the feeding organ of the flatworm , while the more serious injury was the loss of the entire head . Within hours of removing the pharynx , stem cells that were poised to develop into pharyngeal cells became much more active than other stem cell types . When the head was removed , however , a wide range of stem cells became active to make the different cell types required to build a head . This suggests that stem cells monitor all body parts and respond rapidly and specifically to injuries . These findings add to the understanding of regeneration in animal species , which is of great interest for medicine given humans’ limited ability to heal . Many of the genetic systems that control regeneration in planaria also exist in humans , but are only active before birth . In the long-term , understanding the key genes in these processes and how they are controlled could allow regeneration to be used to treat human injuries .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "developmental", "biology" ]	2021	Planarian stem cells sense the identity of the missing pharynx to launch its targeted regeneration
Glioblastoma multiforme ( GBM ) is the most aggressive human primary brain cancer . Using a Trp53-deficient mouse model of GBM , we show that genetic inactivation of the Atm cofactor Atmin , which is dispensable for embryonic and adult neural development , strongly suppresses GBM formation . Mechanistically , expression of several GBM-associated genes , including Pdgfra , was normalized by Atmin deletion in the Trp53-null background . Pharmacological ATM inhibition also reduced Pdgfra expression , and reduced the proliferation of Trp53-deficient primary glioma cells from murine and human tumors , while normal neural stem cells were unaffected . Analysis of GBM datasets showed that PDGFRA expression is also significantly increased in human TP53-mutant compared with TP53-wild-type tumors . Moreover , combined treatment with ATM and PDGFRA inhibitors efficiently killed TP53-mutant primary human GBM cells , but not untransformed neural stem cells . These results reveal a new requirement for ATMIN-dependent ATM signaling in TP53-deficient GBM , indicating a pro-tumorigenic role for ATM in the context of these tumors . Glioblastoma multiforme ( GBM ) is the most common and aggressive form of primary brain cancer in adults . Despite improvements in clinical care , a rapid disease progression and insufficient understanding of the etiology of these tumors results in very poor survival prognosis . The standard treatment for GBM patients involves a combination of the DNA damaging agent temozolomide together with radiotherapy ( Stupp et al . , 2005 ) . However , the benefits of current treatment regimes come with severe side effects for patients , as well as drug resistance and inevitable recurrence of the tumor ( Wen and Kesari , 2008 ) . Over recent years , great efforts by individual research groups and consortia have shed light on the key genetic events that lie at the heart of human GBM formation , raising the possibility of a more targeted approach to therapy . This has identified activation of growth factor receptor signaling ( for example by amplification or overexpression of the PDGF receptor ) and direct or indirect inactivation of the TP53 and retinoblastoma tumor suppressors as core deregulated pathways in human GBM ( Cancer Genome Atlas Research Network , 2008 ) . Unfortunately , many of these pathways are difficult to target pharmacologically in vivo , and for others , like PDGFR signaling , available inhibitors have so far lacked clinical impact ( Wen et al . , 2006; Rich et al . , 2004 ) . Hence , an important goal of GBM research is to identify novel , more effective therapies to generate better outcomes . One of the central determinants of tumor progression and the response to therapy is the DNA damage response ( Lord and Ashworth , 2012 ) . The DNA damage kinase ATM is known primarily as a tumor suppressor , through its role in the response to DNA double-strand breaks ( Shiloh and Ziv , 2013 ) , and systemic loss of ATM has previously been shown to accelerate glioblastoma progression ( Squatrito et al . , 2010 ) . This tumor suppressive role is in line with the activation of the DNA damage response in precancerous lesions as a barrier to tumorigenesis ( Bartkova et al . , 2005 ) . A recent study has , however , demonstrated loss of ATM signaling to inhibit the growth of TP53-null tumor xenografts , via stabilization of p14ARF ( Velimezi et al . , 2013 ) , suggesting that ATM function in cancer is highly context-dependent . As well as its canonical activation at break sites , ATM signaling also occurs in response to other cellular stresses ( Bakkenist and Kastan , 2003 ) , and this mode of ATM signaling requires the ATM INteractor ATMIN ( Kanu and Behrens , 2007 ) . ATMIN interacts with the ATM kinase in basal conditions and disassociates from ATM in response to ionizing radiation , to allow ATM to interact with the MRN complex at double-strand break sites ( Zhang et al . , 2014 ) . ATMIN also has ATM-independent functions , most notably the transcriptional activation of Dynll1 , a motor protein involved with ciliogenesis and crucial for lung development ( Jurado et al . , 2012a; Goggolidou et al . , 2014 ) . In addition , ATMIN has been shown to counteract oxidative damage in the brain ( Kanu et al . , 2010 ) and to protect against B cell lymphomagenesis ( Loizou et al . , 2011 ) , but its role in other cancer types has not yet been determined . A key downstream mediator of the DNA damage response pathway , activated by both the double-strand break and ATMIN-dependent responses , is the tumor suppressor TP53 . Precancerous lesions in which the DNA damage response is activated are under selective pressure to lose or mutate TP53 , and loss of TP53 is known to cooperate with several genes to accelerate tumorigenesis ( as summarized in the IARC database [Petitjean et al . , 2007] ) . Among these cooperating changes is loss of ATM , which induces rapid T-cell lymphoma development ( Westphal et al . , 1997 ) . TP53 is also one of the most commonly mutated genes in human GBM . Consequently , the majority of available mouse models of GBM use deletion of one or both copies of the Trp53 gene in combination with other mutations ( Chen et al . , 2012 ) . Overexpression of the Pdgf receptor ligand Pdgfrb in adult Nestin-positive neural stem cells , for instance , results in glioma formation , which is accelerated in a Trp53-mutant background ( Squatrito et al . , 2010 ) . In this study , we demonstrate that congenital loss of Trp53 in the mouse brain is sufficient to precipitate spontaneous glioblastoma formation , and that this correlates with upregulation of Pdgfra . Further , we show that ATMIN plays a critical role in GBM formation , promoting Pdgfra protein and gene expression in a Trp53-deficient background , using an in vivo glioma model as well as neural stem cell and primary tumor cell cultures . Importantly , we find that these results are translatable to therapeutic ATM inhibition in human patient-derived GBM stem cells , and that combining ATM inhibition with PDGFRA inhibition results in synergistic tumor cell killing with minimal effects on untransformed cells . Loss of TP53 is one of the earliest occurring events in human GBM initiation ( Maher et al . , 2001; Wang et al . , 2009; Ohgaki et al . , 2004; Mazor et al . , 2015; Johnson et al . , 2014 ) . This inevitably results in the accumulation of a plethora of secondary hits , which , after a long latency period , leads to tumor formation . To recapitulate this chain of events in mice , we deleted Trp53 as an initial driver during neural development ( using p53f/f; Nestin-Cre ( p53ΔN ) mice ) and monitored brain tumor formation in late adult life . After 8 months , brain tumors arose in p53ΔN mice with high penetrance ( Figure 1A and B ) , similar to previous observations using an hGFAP-Cre model ( Wang et al . , 2009 ) . Animals began to show neurological symptoms including reduced movements and tremor at an average age of 263 days . When examined histologically , many p53ΔN tumors had features consistent with WHO classification criteria for Grade IV GBM ( Figure 1C ) . Out of 17 animals , 14 ( 82% ) showed diffusely infiltrative astrocytic brain tumors , the majority of which were classified as GBM by independent analysis ( 57% Grade IV , glioblastoma; 29% Grade III , anaplastic astrocytoma; 14% Grade II , fibrillary astrocytoma ) , while one displayed an osteosarcoma ( Figure 1B and D , Figure 1—figure supplement 1 and Supplementary file 1 ) . p53ΔN GBMs displayed pseudopalisading necrosis ( i ) , microvascular proliferation with endothelial hyperplasia ( ii ) and occasional endovascular thrombosis ( iii ) , hallmarks of human GBM tumors ( Figure 1E [i–iii] ) . The markers Gfap ( iv ) , Nestin ( v ) , and Olig2 ( vi ) were expressed in all tumors , while they were negative for the neuronal marker NeuN , supporting the diagnosis of glioma ( Figure 1E ( iv–vi ) and Figure 1—figure supplement 2 ) . Intertumoral and intratumoral heterogeneity , a classic hallmark of high-grade gliomas , was also frequently observed ( Figure 1F ) . These data indicate that early embryonic loss of Trp53 in the brain is sufficient to promote GBM formation . The long latency of these tumors , together with the genome instability and transcriptional changes known to be induced by loss of Trp53 , indicated that these gliomas almost certainly develop as a result of secondary mutations arising in the Trp53-null brain . In line with this notion , mouse SNP array data revealed large chromosomal gains and losses ( typical tumorigenic genetic changes ) in four out of five analyzed p53ΔNgliomas , whereas no changes were detected in one wt and three Trp53-deficient non-tumorigenic control NS cells ( Figure 1—figure supplement 3 ) . These secondary hits are not congenitally predefined and thus possibly allow a better representation of the genetic diversity of the disease in patients . In addition , the high frequency of brain tumor formation , proportion of high-grade tumors , and consistency of tumor latency on a mixed genetic background make this a valuable GBM model . 10 . 7554/eLife . 08711 . 003Figure 1 . Loss of Trp53 is sufficient to induce GBM with high penetrance . ( A ) Kaplan-Meier curves showing tumor-free survival in p53f/f; Nestin-Cre ( p53ΔN ) and control Nestin-Cre mice . ( B ) Status of mouse cohorts at 450 days , showing tumor incidence . ( C ) H&E-stained p53ΔN brain tumor sections showing histological features of malignant GBM . Arrowheads indicate ( i ) a mitotic cell , ( ii ) neo-vascularization , ( iii ) rosetta formation , ( iv ) mitotic catastrophe , ( v ) a multinucleated giant cell , and ( vi ) large areas of necrosis . ( D ) Grades of gliomas for p53ΔN mice . ( E ) Examples of human GBM hallmarks observed in p53ΔN tumors: pseudopalisading necrosis ( i ) , microvascular proliferation with endothelial hyperplasia ( ii ) , and endovascular thrombosis ( iii ) ( all H&E ) . Immunohistochemistry shows expression of the glial markers GFAP ( iv ) , NESTIN ( v ) , and OLIG2 ( vi ) in p53ΔN tumors . ( F ) p53ΔN glioblastomas display high inter- and intra-tumoral heterogeneity . H&E images i-ii , iii-iv , and v-vi represent different regions of the same tumors #1 , #2 , and #3 respectively . GBM , Glioblastoma multiforme . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 00310 . 7554/eLife . 08711 . 004Figure 1—figure supplement 1 . Histological features of lower grade tumors observed in p53ΔN animals . Hematoxylin and Eosin ( H&E ) and antibody-stained sections of p53ΔN brains showing representative examples of low-grade fibrillary astrocytoma ( WHO Grade II ) and anaplastic astrocytoma ( WHO Grade III ) . Ki67 immunolabeling indicates progressively increased proliferation of atypical glial cells with increased tumor gradeDOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 00410 . 7554/eLife . 08711 . 005Figure 1—figure supplement 2 . Histological features of glioblastomas observed in p53ΔN animals . H&E and immunohistochemical staining of sections from p53ΔN control brain tissue ( top row ) and gliomas ( lower row ) . 'N' denotes the normal brain region adjacent to the tumor mass . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 00510 . 7554/eLife . 08711 . 006Figure 1—figure supplement 3 . SNP array analysis on primary cells derived from GBM bearing p53ΔN , non-tumor bearing p53ΔN , and wild-type ( wt ) animals . Chromosomal copy number variations are found in primary tumor cells derived from p53ΔN gliomas ( Stupp et al . , 2005; Wen and Kesari , 2008; Cancer Genome Atlas Research Network , 2008; Wen et al . , 2006; Rich et al . , 2004 ) but not in non-tumorous adult p53ΔN NSCs ( Stupp et al . , 2005; Wen and Kesari , 2008; Cancer Genome Atlas Research Network , 2008 ) or adult wt NSCs . Log R Ratio plots show copy number states for all mouse autosomes for tumor , wt , and p53ΔN cells . Grey points represent probes in a normal copy number state . Green and light green represents gain and amplification calls , respectively , while red and light red points represent hemi- and homozygous losses , respectively . NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 00610 . 7554/eLife . 08711 . 007Figure 1—figure supplement 4 . Table representing CGHcall output for SNP array analysis compressed to comparable regions between arrays in the series . Genomic position and corresponding CGHcall copy number state of samples ( 0 – homozygous loss , 1 – hemizygous loss , 2 – normal state , 3 – gain , 4 – amplification ) are included in the table . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 007 As an integral part of the response to genomic stress , activation of the DNA damage response pathway typically represents a barrier to tumorigenesis . Consistent with this , it has previously been shown that loss of the DNA damage kinase Atm accelerates tumorigenesis in a Pdgf-driven glioblastoma model ( Squatrito et al . , 2010 ) . In addition , deletion of the Atm cofactor Atmin promotes B cell lymphomagenesis ( Loizou et al . , 2011 ) . Whether loss of Atmin affects tumorigenesis in other cancer types is so far unknown . To investigate the potential role of Atmin in glioma formation , we crossed AtminΔN mice ( Kanu et al . , 2010 ) with p53ΔN mice . Strikingly , AtminΔN; p53ΔN mice showed significantly longer tumor-free survival than p53ΔNanimals ( p<0 . 0002 ) ( Figure 2A ) . The largest proportion of AtminΔN; p53ΔNmice ( 43% ) did not develop any signs of sickness and were still alive after 450 days; a further 36% succumbed either to non-tumor-related illness or , in one case , a pancreatic tumor ( Figure 2B and Supplementary file 1 ) . Only three animals out of a cohort of 14 AtminΔN; p53ΔNdouble mutants ( 21% ) initiated glial tumors ( Figure 2B ) , compared with 82% of p53ΔNmice over the same period . The three GBMs arising in AtminΔN; p53ΔNdouble-mutant mice arose later but were histologically similar to those observed in p53ΔN mice ( Figure 2—figure supplement 1 ) , and did not escape Atmin deletion , suggesting that the requirement for Atmin can eventually be overcome , but only in a small proportion of cases . Importantly , central nervous system deletion of Atmin alone ( AtminΔN ) did not affect brain morphology or histology ( Kanu et al . , 2010 ) and the mice remained tumor free ( Figure 2A ) . Thus , Atmin deletion strongly suppresses GBM formation in p53ΔN animals . 10 . 7554/eLife . 08711 . 008Figure 2 . Loss of Atmin rescues GBM formation in p53ΔN brains . ( A ) Kaplan-Meier curves showing tumor-free survival in AtminΔN and AtminΔN; p53ΔN mice . p53ΔN curve from Figure 1A ( same experiment ) is shown for comparison . ( B ) Status of mouse cohorts at 450 days , showing tumor incidence . ( C ) Atmin loss rescues the increased proliferation of p53ΔN NSCs . ( D ) Mean percentage of BrdU-positive NSCs from the indicated genotypes , assessed by FACS following a 2-hr BrdU pulse . ( E ) Mean number of DAPI-permeable ( non-viable ) cells after IR or ( F ) hypoxia , showing sensitivity of AtminΔN; p53ΔN NSCs to hypoxia but not IR . n . s . , not significant; * p<0 . 05 , ** p<0 . 01 . ( G ) Scheme of orthotopic NSC transplant experiment ( left ) and Kaplan-Meier curves indicating survival of NOD/SCID mice orthotopically transplanted with p53ΔN and AtminΔN; p53ΔN NSCs ( right ) . Error bars represent the SEM of three biological repeats , and two biological repeats for ( E ) . IR , Ionizing radiation; NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 00810 . 7554/eLife . 08711 . 009Figure 2—figure supplement 1 . Histological features of glioblastomas observed in p53ΔN and AtminΔN; p53ΔN animals . H&E and immunohistochemical staining of sections from gliomas derived from AtminΔN; p53ΔN brains ( top row ) showing expression of Nestin and Olig2 in the tumor cells as well as scattered expression of Gfap , similar to Grade IV GBMs from p53ΔN animals ( lower row ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 00910 . 7554/eLife . 08711 . 010Figure 2—figure supplement 2 . Examples of murine neural stem cells in culture . Neural stem cells ( NSCs ) were isolated according to published protocols and cultured first as neurospheres ( left ) , then maintained in adherent culture ( center ) . They maintain the ability to differentiate , as shown by Map2 staining after 5 days ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01010 . 7554/eLife . 08711 . 011Figure 2—figure supplement 3 . AtminΔN NSCs proliferate at a similar rate as wild-type NSCs . Error bars represent the SEM of two biological repeats . NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01110 . 7554/eLife . 08711 . 012Figure 2—figure supplement 4 . Gross genome stability is unaffected in Atmin/Trp53 double mutant NSCs . ( A ) Percentages of cells in G1 , S and G2/M cell cycle phases , as measured by FACS . ( B ) AtminΔN; p53ΔNand p53ΔN NSCs have similar proportions of cells with >4n DNA content as wild-type NSCs , as judged by FACS of propidium iodide ( PI ) -stained cells . Representative images of two biological repeats per genotype are shown for ( A ) and ( B ) . ( C ) Metaphase spread analysis of chromosome number from NSCs of the indicated genotypes . At least 15 cells were analyzed for each indicated genotype . NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01210 . 7554/eLife . 08711 . 013Figure 2—figure supplement 5 . Radiation-induced arrest is similarly impaired in Trp53-mutant and Atmin/Trp53 double mutant NSCs . Mean average ratio of BrdU-positive IR-treated NSCs relative to untreated NSCs , showing similar impaired arrest in AtminΔN; p53ΔNand p53ΔN NSCs . Error bars represent the SEM of two biological repeats . IR , Ionizing radiation; NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01310 . 7554/eLife . 08711 . 014Figure 2—figure supplement 6 . Tumors arising from orthotopic injection of NSCs are Nestin and Ki67-positive . Adjacent sections from a representative tumor arising from injected p53ΔN NSCs stained with Nestin and Ki67 . Control , uninjected brain . NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01410 . 7554/eLife . 08711 . 015Figure 2—figure supplement 7 . Asymptomatic AtminΔN; p53ΔN-injected animals show persistent GFP-positive cells . Bright field and fluorescent images of asymptomatic mouse brains 155 days after injection with wt or AtminΔN; p53ΔN NSCs , showing persistence of injected cells . NSC , Neural stem cellDOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 015 Loss of Trp53 affects diverse fundamental cellular processes including proliferation , genome stability , cell cycle arrest , and apoptosis ( Vousden and Lane , 2007 ) . To understand the involvement of Atmin in the suppression of the Trp53-null phenotype , we isolated neural stem cells ( NSCs ) ( Figure 2—figure supplement 2 ) from p53ΔNembryos and tested the effect of Atmin loss on these processes . p53ΔNNSCs proliferated more rapidly than wild-type ( wt ) controls . This proliferative advantage , however , was rescued in AtminΔN; p53ΔN NSCs ( Figure 2C and D ) , indicating that Atmin is required for the hyperproliferation of Trp53-deficient NSCs . Loss of Atmin alone did not alter proliferation of NSCs ( Figure 2—figure supplement 3 ) . FACS profiling of p53ΔN and AtminΔN; p53ΔN NSCs revealed a decrease in the proportion of AtminΔN; p53ΔN cells in S phase compared with p53ΔN NSCs , in agreement with the reduced proliferation rate , but no change in cells with a >4n DNA content , and metaphase spreads at passage 10 revealed no obvious difference in ploidy ( Figure 2—figure supplement 4 ) . We analyzed cell cycle arrest and cell death in response to ionizing radiation ( IR ) by quantifying the percentage of BrdU-incorporating and DAPI-permeable NSCs respectively . As expected , p53ΔN NSCs arrested less efficiently than wt NSCs and cell death was reduced; but cell cycle arrest and cell death were not rescued in AtminΔN; p53ΔN NSCs ( Figure 2E and Figure 2—figure supplement 5 ) . In addition to IR-induced cell death , p53ΔN NSCs were more resilient to hypoxia-induced death , consistent with previous studies ( Liu et al . , 2007; Graeber et al . , 1996 ) . Atmin deletion in these cells re-sensitized them to hypoxia , to similar levels as wt NSCs ( Figure 2F ) . Thus , loss of Atmin is able to rescue some of the phenotypes of Trp53 loss , such as hyperproliferation and hypoxia induced death , but not others , such as IR-induced cell cycle arrest and cell death . As increased proliferation and hypoxia resistance are attributes commonly found in tumor initiating cells of solid tumors , including glioma ( Graeber et al . , 1996; Gilbertson and Rich , 2007 ) , we evaluated the tumorigenic potential of p53ΔN and AtminΔN; p53ΔN NSCs in vivo . We performed intracranial injections of NSCs isolated from wt , p53ΔN , and AtminΔN; p53ΔNembryos ( Figure 2G ) . Five out of five animals injected with p53ΔN NSCs died within 93 days post-injection , while four out of five wt and three out of four p53ΔN; AtminΔNNSC-injected animals were still alive even at 155 days , the endpoint of the experiment ( Figure 2G ) . Histologically , injected cells appeared as atypical glial cells expressing Nestin and Ki67 , diffusely infiltrating the host brain ( Figure 2—figure supplement 6 ) , similar to previous observations with human GBM xenografts ( Stricker et al . , 2013 ) . Fluorescent imaging of asymptomatic mice at the experimental endpoint readily detected injected AtminΔN; p53ΔN and wt NSCs ( Figure 2—figure supplement 7 ) , suggesting that these cells survived in the host brain , but did not induce lethality . Thus , genetic inactivation of Atmin greatly impairs the tumorigenic potential of p53ΔNcells . To understand the molecular basis of the observed attenuation of tumorigenic potential , we performed gene expression profiling on wt , p53ΔN , and AtminΔN; p53ΔNembryonic NSCs . Compared with wt NSCs , 145 genes were downregulated more than 1 . 5-fold in p53ΔN , whereas 77 were overexpressed ( Figure 3A ) . Many of the canonical Trp53 target genes , such as Bax , Puma , and Cdkn1a , were downregulated to a similar extent in AtminΔN; p53ΔN compared with p53ΔNNSCs . However , 27% of genes deregulated in p53ΔNNSCs returned to wt expression levels when Atmin was also deleted ( 36/145 of the decreased , and 24/77 of the increased genes; examples in Figure 3B ) . When this subset of genes ‘rescued’ in double mutant NSCs was queried against the Cancer Genome Atlas ( TCGA ) database , in several cases abnormal levels of these transcripts were found to be associated with human GBM ( Cancer Genome Atlas Research Network , 2008; Suvasini et al . , 2011 ) . Particularly notable among this list was the platelet-derived growth factor receptor alpha ( Pdgfra ) ( Figure 3C ) . Elevated levels of PDGFRA have been observed in human gliomas of various malignancy grades ( Engström et al . , 2012 ) , and increased PDGFR signaling has been shown to induce glioma-like growths in vivo ( Jackson et al . , 2006 ) . Increased PDGFRA levels are also a characteristic hallmark of the proneural GBM subtype as classified in ( Brennan et al . , 2013; Verhaak et al . , 2010 ) . Interestingly , we found that p53ΔNNSCs were associated most closely with the proneural subtype when comparing the microarray expression profile from the murine NSCs to that of publically available human GBM samples from TCGA ( Verhaak et al . , 2010 ) ( Figure 3—figure supplement 1 ) , which is in agreement with elevated Pdgfra mRNA as well as protein levels ( Figure 3D ) . This increased protein expression was also apparent in high-grade GBMs from endstage p53ΔNanimals ( Figure 3E and Figure 3—figure supplement 2 ) . Other growth factor receptors , like Egfr , and proteins commonly deregulated in glioma including Rb , Pten , Cyclin D2 , Junc , and Cdk4 ( Wang et al . , 2009 ) remained unaltered in p53ΔNcells and tumors ( Figure 3—figure supplements 2 and 3 ) , as did the closely related Pdgf receptor Pdgfrb and the receptor ligands Pdgfa and Pdgfb ( Figure 3—figure supplement 4 ) . 10 . 7554/eLife . 08711 . 016Figure 3 . Atmin regulates Pdgfra expression in a mutant Trp53 background . ( A ) Heatmap showing genes deregulated more than 1 . 5-fold in Trp53 null NSCs and their corresponding expression in AtminΔN; p53ΔN double null NSCs . ( B , C ) qRT-PCR validation of GBM implicated genes that show deregulated expression in p53ΔN but not AtminΔN; p53ΔN NSCs . ( D ) Elevated Pdgfra protein expression in p53ΔN but not AtminΔN; p53ΔN NSCs . Corresponding p-Akt and total Akt levels were probed to analyze pathway activation . Actin was used as loading control . ( E ) Tumor section from a GBM arising from a p53ΔN animal showing elevated Pdgfra expression . 'N' denotes the normal brain region adjacent to the tumor mass . Dotted lines indicate tumor border . ( F ) Myc-Atmin overexpression increases expression of Pdgfra mRNA in AtminΔN; p53ΔN but not wt cells . ( G ) Myc-Atmin overexpression increases Pdgfra protein levels in AtminΔN; p53ΔN NSCs . Tubulin was used as loading control . ( H ) Atmin silencing using two independent shRNA constructs reduces Pdgfra expression , assessed by qRT-PCR . ( I ) Stable Pdgfra silencing using two independent shRNA constructs reduces primary tumor cell proliferation , measured using IncuCyte timelapse microscopy . Error bars represent the SEM of at least three biological repeats and two biological repeats in ( I ) . * p<0 . 05 , p<0 . 01 , * p<0 . 001 . NSC , Neural stem cellDOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01610 . 7554/eLife . 08711 . 017Figure 3—figure supplement 1 . The transcriptional profile of p53ΔN NSCs is most closely related to the human proneural GBM subtype , relative to expected distances . Relative to expected distance away from the centroids , p53ΔNis closest to the proneural category ( p=0 . 036 , simulation test ) and the p53ΔN:proneural pairing is the only one that is significant – the grey shading represents the lower 5% tail of the simulated distances . NSC , Neural stem cellDOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01710 . 7554/eLife . 08711 . 018Figure 3—figure supplement 2 . Pdgfra expression is elevated in p53ΔNtumors , but Egfr is not . Immunohistochemistry of GBM sections showing elevated Pdgfra , but not Egfr , expression in a p53ΔN tumor . Esophagus section is shown as positive control for Egfr staining . GBM , Glioblastoma multiformeDOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01810 . 7554/eLife . 08711 . 019Figure 3—figure supplement 3 . Some markers frequently altered in GBM are unaltered in p53ΔN NSCs . Western blots of NSCs of the indicated genotypes showing levels of proteins commonly altered in glioma but unaffected by Atmin/Atm deletion . GBM , Glioblastoma multiforme; NSC , Neural stem cellDOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 01910 . 7554/eLife . 08711 . 020Figure 3—figure supplement 4 . Pdgf ligand expression is not significantly altered in p53ΔN NSCs . q-PCR showing similar expression levels of the Pdgfr ligands Pdgfa and Pdgfb in p53ΔN and AtminΔN; p53ΔN cells . Pdgfrb was not detected . Error bars represent the SEM of at least three biological repeats . NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 02010 . 7554/eLife . 08711 . 021Figure 3—figure supplement 5 . Two independent shRNAs cause Pdgfra knockdown . ( A ) Isolation and culture of p53ΔN primary tumor cells from affected mice . ( B–C ) Western blot ( B ) and qPCR ( C ) showing knockdown of Pdgfra using two independent shRNA constructs . Error bars represent the SEM of at least two biological repeats . * p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 02110 . 7554/eLife . 08711 . 022Figure 3—figure supplement 6 . miR34a expression is reduced in p53ΔN NSCs . Error bars represent the SEM of at least three biological repeats . * p<0 . 001DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 022 Importantly , deletion of Atmin together with Trp53 completely rescued the elevated Pdgfra expression , reducing it to wild-type levels ( Figure 3C and D ) . Overexpression of tagged Atmin in double mutant NSCs restored elevated Pdgfra expression at both the mRNA and protein levels ( Figure 3F and G ) , while acute silencing of Atmin in p53ΔNNSCs , using two independent shRNA constructs , reduced Pdgfra levels ( Figure 3H ) , implying that sustained Pdgfra upregulation in p53ΔNNSCs constantly requires Atmin function to maintain increased Pdgfra expression . Thus , interference with Atmin function is sufficient to reduce Pdgfra overexpression in Trp53-deficient cells . To assess whether Trp53 null tumors require sustained Pdgfra overexpression , we isolated primary tumor cells from GBM-bearing p53ΔNmice and induced stable knockdown of Pdgfra using shRNA ( Figure 3—figure supplement 5 ) . This resulted in a significant decrease in the proliferation rate of these primary tumor cells ( Figure 3I ) , emphasizing the importance of Atmin function in supporting elevated Pdgfra expression in p53ΔNgliomas . Atmin has two known functions: an Atm-dependent function , in which Atmin interacts with Atm and is required for Atm signaling in several stress contexts , and an Atm-independent function , in which Atmin is required for transcription of the dynein light chain Dynll1 . To determine whether acute silencing of Atmin in tumor cells might reduce Pdgfra expression by disrupting Atm signaling , we tested the effects of pharmacological Atm inhibition in p53ΔNprimary tumor cells . Treatment with an Atm inhibitor ( ATMi ) significantly reduced Pdgfra transcript and protein levels ( Figure 4A and B ) and efficiently reduced proliferation of p53∆Nprimary tumor cells ( Figure 4C ) , similar to our observations after Pdgfra silencing . ATMi treatment also reduced Pdgfra expression in p53ΔNNSCs ( Figure 4D and Figure 4—figure supplement 1 ) . Consistent with these results , Atm-/-; p53ΔNdouble-mutant NSCs had dramatically lower Pdgfra expression levels than p53ΔNNSCs , comparable to AtminΔN; p53ΔNand wt NSCs ( Figure 4E and F ) . Furthermore , in vitro proliferation assays showed that genetic loss of Atm reduced the proliferation of p53ΔNNSCs to wt levels ( Figure 4G ) . Importantly , the reduction in cell proliferation in vitro correlated with reduced tumorigenic potential of the NSCs when orthotopically injected into the brains of NOD/SCID mice . Mice injected with Atm-/-; p53ΔNNSCs survived up to 50% longer than those receiving p53ΔNNSCs ( Figure 4H ) . However , Atmin deletion ( Figure 2G ) was more effective than deletion of Atm ( Figure 4H ) . Atmin can also act as a transcription factor ( Jurado et al . , 2012a; 2012b; Goggolidou et al . , 2014 ) , and it is conceivable that this Atm-independent function of Atmin contributes to the suppression on GBM . The similar effect of Atmin deletion and Atm inhibition on Pdgfra expression and proliferation of NSCs and primary tumor cells supports the hypothesis that Atmin functions in these processes via its modulation of Atm signaling . Given the extensively characterized role of PDGFRA in cell proliferation and in particular in glioma , it is reasonable to propose that the reduction in Pdgfra expression upon Atm loss contributes to the reduced tumorigenic potential observed . 10 . 7554/eLife . 08711 . 023Figure 4 . Atm inhibition reduces Pdgfra expression and reduces tumorigenic potential in murine p53ΔN primary tumor cells and NSCs . ( A–B ) Atm inhibitor ( ATMi ) treatment reduces Pdgfra protein levels ( A ) and Pdgfra expression ( B ) in p53ΔN primary tumor cells . ( C ) ATMi treatment reduces in vitro proliferation of p53ΔN primary tumor cells . Error bars represent the SEM of two biological repeats . ( D ) ATMi reduces Pdgfra expression in p53ΔN NSCs , assayed by qPCR . ( E–F ) Pdgfra expression ( E ) and Pdgfra protein levels ( F ) are reduced in Atm-/-; p53ΔN and AtminΔN; p53ΔN compared with p53ΔN NSCs . cDNA is normalized to Actin levels . Error bars represent the SEM of at least three biological repeats . s . e . and l . e . denote short and long exposures of the same blot , respectively . ( G ) Genetic loss of Atm reduces in vitro proliferation of p53ΔN NSCs . Error bars represent the SEM of two biological repeats . ( H ) Kaplan-Meier curves indicating increased survival of NOD/SCID mice orthotopically transplanted with Atm-/-; p53ΔN NSCs compared with p53ΔN NSCs . n . s . , not significant; * p<0 . 05 , p<0 . 01 , * p<0 . 001 . NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 02310 . 7554/eLife . 08711 . 024Figure 4—figure supplement 1 . ATM inhibitor ( ATMi ) treatment reduces Pdgfra protein levels in p53ΔN NSCs . NSCs , Neural stem cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 024 Since Atm is best known as a DNA damage signaling kinase , we examined whether the effect of Atm on Pdgfra expression and tumorigenicity was related to altered DNA damage signaling in p53ΔNglioma cells . Neither p53ΔNnor AtminΔN; p53ΔNNSCs showed increased γh2ax or 53bp1 foci in basal conditions ( Figure 5A and B ) , suggesting that the different proliferation rates are not caused by changes in endogenous DNA damage . Pdgfra expression was not affected by IR or hydroxyurea ( HU ) , two well-described inducers of ATM signaling , either in wt or in p53ΔNNSCs ( Figure 5C ) and no increased Atm substrate phosphorylation could be detected in untreated p53ΔN NSCs ( Figure 5D ) , suggesting that Pdgfra upregulation is not a consequence of stimulation of DNA damage signaling . Atm substrate phosphorylation was also comparable in IR- and HU-treated wt , p53ΔNand AtminΔN; p53ΔNNSCs ( Figure 5D ) , suggesting that the response to DNA damage stimuli is comparable in these cells . This suggests that while Atm is required for the increased Pdgfra expression in p53ΔN cells , this does not involve alterations in IR or HU-induced Atm signaling , but at this point does not exclude the possibility that increased ROS ( reactive oxygen species ) levels in p53∆N GBM cells might contribute to Atm pathway activation and subsequent Pdgfra induction and GBM development . 10 . 7554/eLife . 08711 . 025Figure 5 . Atm signaling in response to DNA damaging agents is remarkably unaffected in p53ΔN and AtminΔN; p53ΔN cells and DNA damaging treatments do not affect Pdgfra expression . ( A ) p53ΔN and AtminΔN; p53ΔN NSCs do not show elevated γh2ax and 53bp1 foci in untreated conditions , suggesting low endogenous damage . Cells treated with 5Gy IR ( wt +IR ) are shown as a positive control . ( B ) Quantification of γh2ax and 53bp1 foci in ( A ) . Error bars indicate 95% confidence intervals . ( C ) qPCR of Pdgfra in wt and p53ΔN NSCs in untreated conditions and after DNA damage-inducing stimuli . cDNA is normalized to Actin levels . Error bars represent SEM of three biological repeats . ( D ) Western blots depicting Atm substrate phosphorylation after the indicated stimuli in NSCs of different genotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 025 To investigate the relevance of ATM signaling for PDGFRA upregulation in human GBM , we took advantage of patient derived human GBM neural stem cells ( GNSCs ) ( Stricker et al . , 2013 ) that have documented TP53 mutations ( lines G26 , G166 , and G179 , Figure 6A ) . We confirmed that cell extracts from each of these human tumors were unable to activate a luciferase construct containing multiple TP53-binding elements ( Figure 6—figure supplement 1 ) . PDGFRA expression was significantly increased in two out of three human GNSC lines compared with untransformed human neural stem cells ( cb660; ( Engström et al . , 2012; Pollard et al . , 2009 ) ( Figure 6B ) . To assess whether this observation is representative of the wider human GBM spectrum , we analyzed the gene expression patterns in the 'Glioblastoma Multiforme ( TCGA , provisional ) ' dataset from The Cancer Genome Atlas . We analyzed 153 RNA-seq and 500 microarray datasets for 518 glioblastoma samples and completed these with the clinical subgroup information from ( Brennan et al . , 2013 ) . The gene expression datasets contained z-scores representing the differences in expression levels of PDGFRA between cancer and control samples ( Figure 6—figure supplements 2 and 3 ) . The RNA-seq and microarray data were handled separately; for each , we observed higher average PDGFRA expression in TP53-mutant glioblastoma samples compared with TP53 wild-type samples ( Figure 6C , D ) . We also observed higher PDGFRA expression levels in the Proneural and GCiMP GBM subtypes when compared to the Classical , Neural , and Mesenchymal ( Figure 6E and Figure 6—figure supplement 4 ) . 10 . 7554/eLife . 08711 . 026Figure 6 . ATM inhibitor treatment reduces elevated PDGFRA expression and decreases proliferation in human GBM tumor cells . ( A ) Mutational status of TP53 , PTEN , IDH1 , and EGFR in the indicated glioma neural stem stell ( GNSC ) lines . n . e . , not expressed . wt , wild type . ( B ) Western blot depicting PDGFRA , TP53 , PTEN , and ACTIN expression in three GNSC lines ( G26 , G166 , and G179 ) and cb660 control cells . ( C , D ) Box plots showing PDGFRA expression in TP53-wt and TP53-mutant TCGA human glioblastoma datasets , measured by microarray ( n=153 ) . ( C ) or RNA-Seq ( n=500 ) . ( D ) Data from 518 patient samples in total with an overlap of 135 patients . ( E ) Box plots showing PDGFRA expression z-scores in TCGA human glioblastoma subtypes , measured by RNA-Seq ( n=150 ) . p-Values in ( C–E ) calculated using Wilcoxon’s test . ( F ) qRT-PCR showing decreased PDGFRA expression in TP53-mutant GNSC lines after ATM inhibitor ( ATMi ) treatment . cDNA is normalized to GAPDH levels . Error bars represent the STDEV of two biological repeats . ( G ) Western blot showing knockdown of PDGFRA using a doxycycline ( Dox ) -inducible shRNA construct . ( H ) PDGFRA knockdown using doxycycline ( Dox ) -inducible shRNA reduces proliferation of TP53-mutant G179 GNSCs . Cell confluence measured by IncuCyte timelapse microscopy . Error bars represent the SEM of three biological repeats . * p<0 . 05 , *** p<0 . 001 , n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 02610 . 7554/eLife . 08711 . 027Figure 6—figure supplement 1 . Luciferase reporter assay for TP53 activity using the p53-550RE construct in human GNSC ( G26 , G166 , G179 ) and control NSC ( cb660 ) lines . NSC , Neural stem cell . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 02710 . 7554/eLife . 08711 . 028Figure 6—figure supplement 2 . Correlation of PDGFRA expression levels in TCGA glioblastoma samples represented in both microarray and RNASeq datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 02810 . 7554/eLife . 08711 . 029Figure 6—figure supplement 3 . Comparison of the distribution of PDGFRA expression levels in human glioblastoma samples measured by RNASeq and by microarray . Data from TCGA . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 02910 . 7554/eLife . 08711 . 030Figure 6—figure supplement 4 . Box plots showing PDGFRA expression z-scores in TCGA human glioblastoma subtypes measured by microarray ( n=487 ) . p-Value is calculated using Wilcoxon’s test , combining GCiMP and Proneural versus Neural , Classical , and Mesenchymal . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 03010 . 7554/eLife . 08711 . 031Figure 6—figure supplement 5 . Control for Figure 6G and H showing efficient knockdown of PDGFRA after doxycycline administration . q-PCR showing knockdown of PDGFRA in GNSCs using a doxycycline ( Dox ) -inducible shRNA construct . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 031 In line with the data from our mouse model , treatment with an ATMi strongly reduced PDGFRA expression in GNSCs , whereas no significant effect on PDGFRA expression was detected in cb660 controls ( Figure 6F ) . To determine whether a reduction in PDGFRA levels would be sufficient to inhibit the proliferation of primary GNSCs from human tumors , G179 cells were infected with doxycycline ( dox ) -inducible shRNA constructs targeting PDGFRA ( Figure 6G and Figure 6—figure supplement 5 ) , and their proliferation was monitored over a period of 10 days . In agreement with our results in the mouse , depletion of PDGFRA led to a significant decrease in the proliferation rate of human GNSCs ( Figure 6H ) . Thus , both murine and human TP53-mutant glioma cells are sensitive to loss of PDGFRA . To analyze the physiological effects of ATM inhibition on GNSCs , proliferation was monitored over a period of 7 days . Similar to our observations in murine NSCs , ATMi treatment reduced the proliferation rate of human TP53 mutant GNSCs , but not cb660 control cells ( Figure 7A ) . PDGFR inhibitors ( PDGFRi ) have previously been shown in vitro and in vivo to reduce tumor growth in a glioma model ( Kilic et al . , 2000 ) . These advances , however , have so far failed to transition into clinical practice ( Wen et al . , 2006; Rich et al . , 2004 ) . We reasoned that ATMi-mediated reduction in PDGFRA protein levels could potentiate the effects of PDGFR inhibition . Consistent with this , we observed that co-treatment of GNSCs , but not untransformed NSCs ( e . g . cb660 ) , with relevant ATM and PDGFR inhibitor concentrations further reduced cell proliferation ( Figure 7A and Figure 7—figure supplement 1 ) . More detailed analysis revealed that this treatment combination induced substantial cell death , while untransformed NSCs remained largely unaffected ( Figure 7B , C ) . The proportion of apoptotic GNSCs receiving the combination treatment was quantifiable by microscopic detection of a caspase-cleaved fluorescent substrate after up to four days of drug treatment; thereafter , dead cells detached from the plate , precluding accurate quantification . Essentially , all GNSCs were eventually killed by the combination of ATM and PDGFR inhibitors , as judged by cell detachment . In contrast , PDGFRi treatment by itself had little effect on cell death and proliferation ( Figure 7 A–C ) . Seeding GNSCs at higher confluence gave similar results ( Figure 7—figure supplement 2 ) , as did a second , independent PDGFRi in combination with the ATMi ( Figure 7—figure supplement 3 ) , indicating that cell death was not an indirect consequence of slower proliferation induced by the ATMi or a side effect of a particular drug . Hence , ATM is required for PDGFRA overexpression in both murine and human GBM cells , and combined inhibition of ATM and PDGFR induces lethality in TP53-mutant glioma cells , promising new opportunities for future GBM treatment . 10 . 7554/eLife . 08711 . 032Figure 7 . Combinatorial treatment with both ATM and PDGFR inhibitors induces apoptosis of human glioma stem cells . ( A ) Growth curves showing the proliferation of wild-type control human neural stem cells ( cb660 ) and two independent cell lines isolated from TP53-deficient human gliomas ( G166 & G179 ) over 7 days in the presence of vehicle control ( DMSO ) , ATM inhibitor ( ATMi ) , PDGFR inhibitor ( PDGFRi ) , or both inhibitors . Error bars depict STDEV from two biological repeats . ( B ) Representative bright-field/fluorescent images of the same cell types treated as above after 4 days . Cells undergoing apoptosis are labeled by emitting GFP . ( C ) Quantification of apoptotic cells after the treatments indicated above , represented as % apoptotic cells from total live cells . Error bars depict STDEV from three biological repeats . After four days , it became impossible to quantify apoptotic GNSCs receiving the combination treatment , since they detached completely from the culture plate . These percentages thus represent an underestimate of the total extent of G166 and G179 cell death with ATMi + PDGFRi . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 03210 . 7554/eLife . 08711 . 033Figure 7—figure supplement 1 . ATM and PDGFR inhibitors were used at minimally required concentration to achieve pathway inhibition . ( A ) G179 cells were treated with DMSO or 400 nM , 2 μM or 10 μM of ATMi for 16 hrs , then subjected to 5Gy of ionizing irradiation , harvested 30 min later and analyzed by Western blotting . Pre-treatment of cells with the ATMi at 10 μM clearly prevented IR-induced ATM auto-phosphorylation at Serine 1981 and phosphorylation of its downstream substrates: KAP1 and CHK2 , at Serine 824 and Threonine 68 , respectively . ( B ) G179 cells were treated with DMSO , 200 nM or 1 μM of PDGFRA inhibitor III for 2 hrs , harvested and analyzed by Western blotting . 1 μM inhibitor prevented the phosphorylation of AKT ( downstream substrate of PDGFRA ) at Serine 473 . ( C ) P-PDGFRA levels were below detection , therefore total PDGFRA was immunoprecipitated from cell lysates prior to Western blotting . Treatment of cells with the PDGFRA inhibitor III at 1 μM clearly reduced phosphorylation of PDGFRA at Tyrosine 754 . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 03310 . 7554/eLife . 08711 . 034Figure 7—figure supplement 2 . Bright-field images of timelapse microscopy using cells seeded at high confluency . Representative bright-field images of cb660 , G166 , and G179 GNSC lines seeded at high confluency and treated over 7 days with vehicle control ( DMSO ) , ATM inhibitor ( ATMi ) , PDGFR inhibitors III or V , or both ATM and PDGFR inhibitors . Values indicate% confluence at the indicated point in the time course as measured using IncuCyte software . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 03410 . 7554/eLife . 08711 . 035Figure 7—figure supplement 3 . Apoptosis quantification of a second PDGFR inhibitor . ( A ) Representative bright-field/fluorescent images of the same cell types treated over 4 days with vehicle control ( DMSO ) , PDGFR inhibitor V ( PDGFRi V ) , or ATM inhibitor in combination with PDGFR inhibitor V ( ATMi + PDGFRi V ) . Cells undergoing apoptosis are labelled by emitting GFP . ( B ) Quantification of apoptotic cells four days after the treatments indicated above , represented as % apoptotic cells from total live cells . After four days , it became impossible to quantify apoptotic GNSCs receiving the combination treatment , since they detached completely from the culture plate . These percentages thus represent a large underestimate of the extent of G166 and G179 cell death with ATMi + PDGFRi V . DOI: http://dx . doi . org/10 . 7554/eLife . 08711 . 035 Our data indicate that disruption of Atm/Atmin function not only inhibits GBM initiation in Trp53-deficient animals , but also reduces the tumorigenic potential of established human glioma cells , suggesting that ATM inhibitors might be valuable tools in GBM therapy . An important goal of personalized cancer medicine is to identify vulnerabilities of specific tumor genotypes . Since TP53 signaling is the most common genetically altered pathway in human gliomas ( Cancer Genome Atlas Research Network , 2008 ) , it is desirable for future cancer therapies to identify new molecular targets that affect TP53-deficient cells . Here , we have shown that genetic loss of Atmin or Atm reduces GBM formation initiated by deletion of Trp53 . This finding indicates that Atm/Atmin is crucial for Trp53-deficient GBM development , and suggests the use of ATM inhibitors , which have already been developed , for GBM therapy ( Basu et al . , 2012; Golding et al . , 2009; 2012; Batey et al . , 2013 ) . Although other groups have previously observed that loss of Trp53 is sufficient to provoke GBM formation ( Wang et al . , 2009; Zheng et al . , 2008 ) , the potential advantages of this model system have largely been overlooked . Even on a mixed genetic background , mice harboring a Nestin-driven Trp53 deletion show strong consistency in tumor latency , high penetrance of brain tumors and high-grade GBM formation . Such a model , incorporating the stochastic evolution and genetic heterogeneity of human glioblastoma patients , complements existing genetically defined mouse models that show an accelerated disease course ( Chen et al . , 2012 ) . Atm is a rare example of a protein required for formation of a Trp53-deficient tumor that is amenable to pharmacological inhibition , with potential direct therapeutic implications . This pro-tumorigenic function for Atmin and Atm signaling in brain cancer is unexpected , as systemic Atm deletion has previously been found to accelerate Pdgf ligand-induced gliomagenesis ( Squatrito et al . , 2010 ) . However , the previous study was carried out in Trp53 wild-type animals . It was recently shown that TP53 status is crucial in determining the cellular response to ATM inhibition in human glioma cell lines ( Biddlestone-Thorpe et al . , 2013 ) , so this difference might explain the apparently opposite function of Atm in our model . TCGA data indicates that ATM mutation/deletion is rare in TP53-mutant GBM , supporting the hypothesis that ATM loss is not a tumor driver and may even inhibit tumorigenesis in this context . Because the pleiotropic effects of ATM signaling are highly context dependent , factors other than TP53 status must also influence the outcome in vivo . For example , loss of Atm cooperates with Trp53 deficiency and accelerates the progression of T-cell lymphoma ( Westphal et al . , 1997 ) . Thus , the relationship between TP53 and ATM signaling in tumorigenesis may be tissue specific , ranging from cooperation to antagonism . Adding to this context dependency , ATM’s function in the DNA damage response inhibits tumorigenesis in precancerous lesions but can also promote resistance to DNA damaging therapies in established tumors . Indeed , recently published work has shown that ATM inhibition preferentially radiosensitizes TP53 mutant GBM cell lines ( Biddlestone-Thorpe et al . , 2013 ) . Our data reveal an additional pro-tumorigenic effect of Atm , which acts in the absence of ionizing radiation and involves non-canonical , Atmin-dependent signaling . It is possible that the effect of Atm inhibition in our model would also be enhanced by ionizing radiation , since the radiosensitizing effect would be predicted to be largely independent of Atmin ( Kanu and Behrens , 2007 ) . Thus , combination treatment with ATM and PDGFR inhibitors together with radiotherapy may represent a therapeutic opportunity for GBM . Our results indicate that the requirement for the Atm/Atmin pathway in p53∆N GBM lies in the initiation of a program of altered gene expression ( Figure 3A ) and does not directly involve the DNA damage response to double-strand breaks induced by IR or HU ( Figure 5B and C ) . Although Atmin has been implicated directly as a transcription factor for Dynll1 ( Jurado et al . , 2012b ) , a broader role for Atmin in transcriptional regulation had not yet been determined . We find here that Atmin loss leads to widespread rescue of glioma-associated changes in gene expression in the p53∆N background . Several of these genes could be collectively responsible for suppressing GBM formation in AtminΔN; p53∆N animals . The most compelling single candidate , however , is the proto-oncogene Pdgfra . It has been shown that stimulating Pdgf signaling in vivo is sufficient to induce glioma-like growths ( Jackson et al . , 2006 ) . Moreover , PDGFRA is commonly overexpressed in human glioma ( Engström et al . , 2012 ) , but only a subset of gliomas display PDGFRA locus amplification ( Furnari et al . , 2007 ) ( Cancer Genome Atlas Research Network , 2008 ) , indicating that additional mechanisms drive its increased expression . PDGFRA overexpression frequently correlates with LOH on chromosome 17p , where the human TP53 gene is located ( Hermanson et al . , 1996 ) , supporting the view that the oncogenic misexpression of PDGFRA might be a direct consequence of TP53 inactivation in human GBM . In our murine and human glioma lines , Pdgfra levels could be dynamically modulated by depleting Atmin or inhibiting Atm . At present , how Atm/Atmin controls Pdgfra expression in this background is unclear . Chromatin immunoprecipitation experiments in murine NSCs indicated that Atmin does not bind the Pdgfra gene ( unpublished observations ) , making the regulation unlikely to be via direct transactivation , most likely involving currently unidentified intermediate transcriptional regulators of Pdgfra expression . We speculate that Trp53 controls the expression of a negative regulator of Pdgfra , for example , a repressor or a miR RNA , and that Atm signalling is required for the expression or function of this factor . As these intermediate factors could be promising targets for pharmacological inhibition , further investigation into their identity is of therapeutic interest . Although the oncogenic role of PDGFR signaling in glioma is well established , the clinical efficacy of PDGFR inhibitors has so far been disappointing ( Wen et al . , 2006; Rich et al . , 2004 ) . Here , we demonstrate that the effect of these drugs on the proliferation and survival of human glioma cells can be greatly potentiated by combining PDGFR inhibitor treatment with ATM inhibitors . Importantly , the strongest effects of combinatorial treatment on cell proliferation and survival are specific to tumor cells that have lost TP53 function , suggesting that systemic therapy should target TP53-deficient tumors with minimal impairment to normal tissue . A second encouraging result of combining ATM and PDGFRA inhibitors is that they not only reduce tumor cell proliferation , but also potently induce cell death . While our data support cooperation based on simultaneous inhibition of PDGFR signaling at the transcriptional and protein kinase levels , it is likely that this synergy also encompasses broader effects of ATM and PDGFRA inhibition . In addition , current PDGFR inhibitors are not completely specific and also inhibit EGFR , VEGFR , FLT3 , and c-KIT kinases amongst others ( Andrae et al . , 2008; Homsi and Daud , 2007 ) . Similarly , ATM inhibitors have also been shown to display broader kinase inhibitor activity , especially at higher concentrations ( Hickson et al . , 2004 ) . Thus , PDGFR-independent mechanisms could contribute to the observed treatment synergy . In summary , we have identified a novel protumorigenic function of ATM signaling in GBM . Our results present a rationale for expanding the investigation of ATM inhibitors from radiosensitizers to potential therapies in their own right , and point toward improvements in the efficacy of PDGFRA inhibition using combination treatment . Experiments in mice were carried out with the approval of the Crick Institute’s Ethical Review Committee and under the guidance of the Biological Resources Unit . Atminf/f , p53f/f , Atm-/- , and Nestin-Cre mice have been described previously ( Kanu et al . , 2010; Jonkers et al . , 2001; Barlow et al . , 1996; Tronche et al . , 1999 ) . Immunocompromised NOD/SCID mice were maintained in-house . Mice were maintained and bred on a mixed background in pathogen-free conditions , monitored for signs of ill health and culled when moribund . Strain background had no significant effect on the latency or development of gliomas in p53ΔNmice . NSCs were isolated as spheres from fore and midbrains of mouse E13 . 5 embryos . Cells were initially cultured as spheres under self-renewal conditions , as previously described . Adherent NSC cultures were derived as previously described ( Pollard et al . , 2006; Conti et al . , 2005 ) with minor modifications ( Sancho et al . , 2013 ) . Briefly , primary spheres were plated in Neurobasal Medium ( Invitrogen , Grand Island , NY , USA ) supplemented with 1% Penicillin/Streptomycin ( Invitrogen ) , 1% L-glutamine ( Invitrogen ) , 2% B27 supplement ( Invitrogen ) , 1% N-2 supplement ( Invitrogen ) , 20 ng/ml EGF ( PeproTech ) , 20 ng/ml FGF-basic ( PeproTech ) , and 1 μg/ml laminin ( Sigma ) . All experiments were performed using undifferentiated adherent NSCs ( see Figure 2—figure supplement 2 ) . Primary tumor cells were generated from brain tumor samples of symptomatic mice ( Figure 3—figure supplement 5 ) . Tissue was subjected to mechanical and enzymatic dissociation , and single cells initially cultured to form spheres . Tumor spheres formed were then maintained as adherent tumor cell cultures in NSC media as described above . pCMV6-myc-ATMIN was generated by cloning mouse Atmin into a pCMV6 backbone . Silencing and mismatch constructs for Atmin have been described previously ( Kanu and Behrens , 2007 ) . The luciferase p53 response element ( pGL3-550RE ) was kindly provided by Karen Vousden . The ATM inhibitor Ku55993 ( Merck ) was used at 10 µM and replenished every 24 hr . For DNA damage induction , NSCs of the indicated genotypes were either left untreated or subjected to 0 . 8 Gy irradiation or 2 mM hydroxyurea ( HU ) and harvested after 1 hr and 3 hr respectively for RNA extraction or protein lysates . For transfection , NSCs and GNSCs were plated at subconfluence and transfected with Lipofectamine 2000 according to the manufacturer’s protocol ( Invitrogen ) . For luciferase assays , samples were transiently transfected with Firefly and Renilla luciferase reporters and luciferase activity was measured using the Dual-Luciferase Reporter Assay System ( Promega ) , 36-hr post-transfection . Data are expressed as fold induction of luciferase activity after being normalized to expression of thymidine kinase-renilla luciferase ( TK-renilla ) . For Western Blots ( WB ) , NSCs were extracted in RIPA lysis buffer ( NEB ) supplemented with protease inhibitors ( Sigma ) . For immunoprecipitation ( IP ) cells were lysed in IP buffer ( 20mM sodium phosphate buffer , 1 mM EDTA , 0 . 2% NP40 , 150 mM NaCl supplemented with Na-orthovanadate , PMSF , NaF and protease inhibitor mixture [Sigma] ) . After sonication and centrifugation , supernatant was incubated overnight at 4°C with PDGFRA antibody , followed by 2-hr incubation with Dynabeads M-280 ( Invitrogen ) , washed with IP buffer ( 0 mM NaCl , 150 mM NaCl and 1 M NaCl ) and eluted in Laemmli sample buffer . All primary antibodies were used at 1:1000 dilution and secondary antibodies at 1:10000 . The following antibodies were used: p53 , Pdgfra , p-Pdgfra , p-Chk2 , p-Akt , Akt ( all Cell Signaling ) ; Myc-9E10 ( CRUK ) ; Smc1 ( Abcam ) ; p-Smc1 , Chk2 , Tubulin ( all Merck ) ; p-Kap1 , Kap1 ( both Bethyl Labs ) ; Atm ( Santa Cruz ) , p-Atm ( Epitomics ) , p-p53 , Actin , GAPDH , HRP-conjugated goat anti-mouse/rabbit IgG ( all Sigma ) . For qRT-PCR analysis , mRNA was isolated from NSCs using the RNeasy mini kit ( Qiagen ) . Results ( normalized to Actin expression ) are presented as fold change relative to control . Sequences of primers used are listed in Supplementary file 2 . For cell proliferation assays , NSCs were plated in duplicate in 12-well plates and cell number measured every day using a ViCell cell counter . The initial number of cells seeded was used to normalize to 100% . Mean averages were taken from at least four independent lines per genotype taken from different embryos and crosses . For BrdU profiling , cells pulsed with BrdU for 2 hr were fixed in 70% ethanol , stained with BrdU antibody and propidium iodide ( PI ) and analyzed by flow cytometry . For hypoxia assay , NSCs were seeded in duplicate into 6-well plates and next day placed into a hypoxic chamber at 0 . 1% O2 for 72 hr . ATM inhibitor was replenished after 24 hr and 48 hr . For cell death analysis , control-treated and hypoxia-treated cells were subjected to DAPI staining and analyzed via flow cytometry . Mean averages were taken from at least three independent experiments . To determine IR-induced death , NSCs were irradiated at 5Gy and after 24 hr incubated with 4'-6-Diamidino-2-phenylindole ( DAPI; Sigma ) and assessed for apoptosis via flow cytometry . G1/S phase arrest was assessed by irradiating NSCs at 5Gy and after 18 hr pulsing with BrdU for 1 . 5 hr . Cells were prepared for flow cytometry as described above and gated on BrdU-positive cells . Mean averages were taken from at least three independent experiments . 1x105 early passage ( p . 4 ) GFP labeled NSCs of the indicated genotypes were injected using a stereotaxic frame into the striatum of 6- to 8-week-old mice ( NOD/SCID strain ) , following administration of general anaesthesia as previously described . Animals were monitored daily and culled when moribund . Fetal NS cell lines and GNS lines derived from human glioma samples have been described previously ( Stricker et al . , 2013; Pollard et al . , 2009 ) . The same culture conditions apply as previously described for mouse NSCs . Cell proliferation was measured over a period of four or seven days using an incubator microscope system for live cell imaging and measurement of cell confluence over several days ( IncuCyte ) and cell death was quantified using CellPlayer Caspase 3/7 Reagent from Essen Bioscience . Cells were seeded at 1 . 2x104 or 2 . 4x104 into a 48-well plate and images taken every 30 min . The ATM inhibitor KU55993 was added at 10 μM . PDGFR inhibitors III and V ( CAS 205254-94-0 and CAS 347155-76-4 respectively , both Merck ) were used at 1μM and replenished every 24 hr over a period of four or seven days as indicated . Stable shPDGFRA GNSC lines were generated by viral infection of GNSCs with pTRIPZ-shPDGFRA targeting and control vectors ( Thermo Fisher ) . Doxycycline was added at a final concentration of 0 . 5 μg/ml and replenished every day . All mice were euthanized in a CO2 chamber to preserve brain tissue . Brain sections were cut at 4 μm for Hematoxylin & Eosin ( H&E ) staining and all antibody staining . The following antibodies were used for IHC: Nestin ( BD Bioscience ) , Gfap ( Dako ) , Olig2 , Neun ( both Merck ) , Synaptophysin ( Sigma ) , Pdgfra ( Cell Signaling ) , Ki67 ( Abcam ) . Tumor grading was determined on the basis of the WHO grading system for malignant astrocytoma . Three independent NSC lines from each indicated genotype were pooled and submitted for genome wide gene expression profiling using the Illumina Mouse ref 8 v3 . 0 expression bead chip . Raw data were processed using the 'lumi' package ( Du et al . , 2008 ) within Bioconductor ( Gentleman et al . , 2004 ) first by applying the variance-stabilizing transform , and then carrying out quantile normalization . Following this , genes were selected on the basis of a 1 . 5-fold threshold on the mean of the technical duplicates ( technical correlation being higher than 0 . 996 in all cases ) . The microarray data depicted in Figure 3A has been uploaded to the GEO database under the accession number GSE76296 . Using our normalised microarray data , we calculated the average log fold change between each condition and the wild-type samples . We then calculated distance ( following the procedure of [Dabney , 2006] ) of these expression profiles to the centroids of the four GBM subtypes identified in ( Verhaak et al . , 2010 ) . We mapped the expression profile from mouse to human by matching up gene names between the two species , resulting in a correspondence for 674 of the 840 genes in the GBM signature . To assess informally the specificity of the subtypes , and to gain some insight into the strength of the similarities to GBM subtypes , we bootstrapped the log fold changes within samples 1000 times , recording the pairwise distances between bootstrapped samples and fixed GBM subtype-centroids . Total genomic DNA was extracted and purified using DNeasy Blood & Tissue Kit ( Qiagen ) . Copy number variation ( CNV ) detection was performed using the Affymetrix Mouse Diversity Genotyping Array ( MDGA ) at AROS Applied Biotechnology . Genotypes were called from the CEL files using the BRLMM-P algorithm in Affymetrix Power Tools ( APT ) ( available from: http://www . affymetrix . com/estore/partners_programs/programs/developer/tools/powertools . affx ) using default settings , including quantile normalization . PennAffy ( available from: http://penncnv . openbioinformatics . org/en/latest/user-guide/download/ ) was implemented to generate Log R Ratio ( LRR ) and B-allele frequency ( BAF ) using canonical genotype clustering and population frequency of the B-allele ( PFB ) files kindly provided by Locke et al . ( Locke et al . , 2015 ) . Markers that do not probe allelic balance ( exon markers in the MDGA ) were removed . GC content was calculated for 1 Mb windows centred by each marker using BEDTools ( Quinlan and Hall , 2010 ) . LRR was then corrected for genomic waviness by subtracting the median normalised GC content multiplied by a coefficient optimised to generate the minimum variance when subtracted from the LRR . Copy number states were then called for mouse autosomes using the ‘CGHcall’ R package ( van de Wiel et al . , 2007 ) in Bioconductor ( Gentleman et al . , 2004 ) using the 'sdundo' option for the undo . splits parameter for segmentation . Only CNVs larger than 1 Mb were reported . Gene expression measurements were obtained by downloading the 'Glioblastoma Multiforme ( TCGA , provisional ) ' dataset using the cBioPortal ( http://www . cbioportal . org/ , version 05/09/2014 ) for The Cancer Genome Atlas ( http://cancergenome . nih . gov/ ) . The dataset contained 500 microarrays and 153 RNA-seq measurements for 518 glioblastoma patients . Glioblastoma expression subgroup information was obtained from ( Brennan et al . , 2013 ) , and gene expression levels were represented as z-scores . The RNA-seq and microarray data were analyzed separately but the z-scores for PDGFRA are correlated across 135 samples for which both data types are available ( Spearman correlation coefficient = 0 . 85 for 135 samples with both RNA-seq and microarray data , Figure 6—figure supplement 2 ) , indicating that the two measurement types are comparable . Differences in expression z-scores between TP53 mutant and TP53 wt status , as well as glioblastoma subgroups ( combining GCiMP and Proneural versus Neural , Classical , and Mesenchymal ) were tested using the Wilcoxon’s test . Three outlier points from each of the TP53 wt and mutant groups in Figure 6D , as well as six outlier points ( 5 Proneural and 1 Mesenchymal ) in Figure 6E are excluded from the plot for clarity . Tumor-free survival was analyzed using GraphPad Prism 6 and statistical analyses to determine tumor-free survival were performed using the Mantel-Cox test . For TCGA data analysis p values were determined using the Wilcoxon’s test . For all other experiments with error bars , the unpaired Student’s t-test was performed to determine statistical significance .	Glioblastomas are the most common and aggressive brain cancers in adults , and currently lack efficient treatment options . Glioblastoma cells contain genetic mutations that enable them to grow and divide more quickly than they would under normal conditions . The occurrence of these mutations often leads to a functional impairment in so-called 'tumor suppressor' proteins that may have a range of roles , including repairing genetic damage or controlling the rate of cell division . Blake et al . have now studied how some of these tumor suppressor proteins interact . Deleting a prominent tumor suppressor called TP53 from the brain of mice caused these animals to develop glioblastomas . If , however , both TP53 and another tumor suppressor called ATMIN were deleted at the same time , the majority of mice did not develop any brain tumors . Further in-depth profiling of these brain tumor cells revealed that TP53-deleted cells had very high levels of the oncogene PDGFRA , which causes cells to divide more rapidly . These high PDGFRA levels were brought back to normal conditions upon deletion of ATMIN . Blake et al . then studied primary human glioblastoma cells that lack TP53 and found that these cells could be efficiently killed by a combination of drugs that block the activity of PDGFRA and the protein ATM , which is known to work in concert with ATMIN . Importantly , this combination of drugs did not adversely affect healthy brain cells , opening up new strategies and potential treatment options for glioblastoma patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2016	Inactivation of the ATMIN/ATM pathway protects against glioblastoma formation
Focused Ion Beam Scanning Electron Microscopy ( FIB-SEM ) can automatically generate 3D images with superior z-axis resolution , yielding data that needs minimal image registration and related post-processing . Obstacles blocking wider adoption of FIB-SEM include slow imaging speed and lack of long-term system stability , which caps the maximum possible acquisition volume . Here , we present techniques that accelerate image acquisition while greatly improving FIB-SEM reliability , allowing the system to operate for months and generating continuously imaged volumes > 106 µm3 . These volumes are large enough for connectomics , where the excellent z resolution can help in tracing of small neuronal processes and accelerate the tedious and time-consuming human proofreading effort . Even higher resolution can be achieved on smaller volumes . We present example data sets from mammalian neural tissue , Drosophila brain , and Chlamydomonas reinhardtii to illustrate the power of this novel high-resolution technique to address questions in both connectomics and cell biology . Many modalities of electron microscopy ( EM ) can probe cellular structure at the nanometer scale . However , despite considerable progress over the past decade in developing high-resolution 3D imaging , there remain important limitations , reflecting an inherent trade-off between resolution and the size of the 3D volume . Different currently available EM methods , and their advantages and disadvantages have been reviewed recently ( Briggman and Bock , 2012; Titze and Genoud , 2016 ) . For demanding applications such as tracing neuronal processes in three dimensions , high resolution in the z axis , in addition to the xy plane , is critical ( Lichtman and Denk , 2011; Meinertzhagen , 2016 ) . FIB-SEM offers exactly this capability , with x , y , and z resolution all <10 nm ( Knott et al . , 2008; Xu and Hess , 2011 ) , in addition to other significant advantages , such as superior registration and fully automated operation . However , to date the FIB-SEM approach has seldom been used in neuroscience because of its severe volume limitation ( Briggman and Bock , 2012; Knott et al . , 2008; Helmstaedter , 2013; Denk et al . , 2012 ) , typically less than the extent of a single neuron . FIB-SEM was originally developed for semiconductor and material research applications without requirements for imaging large volumes; only in the past decade has it been explored as a tool for 3D biological imaging ( Knott et al . , 2008; Xu and Hess , 2011; Narayan and Subramaniam , 2015; Wei et al . , 2012 ) . The typical 3D FIB-SEM procedure uses a focused ion beam to ablate a few nanometer layer from the specimen block-face , followed by SEM imaging of the freshly exposed surface . These steps cycle continually until the entire 3D volume is ablated and imaged . The distinctive advantage of FIB-SEM is the fine z thickness removed with each step , which gives z resolution down to a few nanometers . In contrast , the current state-of-the-art technologies based on diamond knife sectioning ( Hayworth et al . , 2006 ) or diamond knife block-face removal ( Denk and Horstmann , 2004; Wanner et al . , 2015 ) lose consistency when attempting z steps between adjacent images below 20 nm . Deconvolution based on multiple images with varying landing energies can improve z resolution on thin sections ( Boughorbel et al . , 2012; FEI Teneo VS Technology: http://www . fei . com/teneo-for-life-sciences/ ) , but only to a limited extent . Electron tomography , based on tilting thin sections in TEM , provides excellent z resolution but becomes impractical for reconstructing thick samples due to the tedious stitching requirements of a long series of these tomograms from sequential sections ( Soto et al . , 1994 ) . Most of these EM techniques yield very different resolutions in the x , y , and z directions , and reduced resolution in any one axis can introduce significant burden to subsequent image processing and analysis such as segmentation and proofreading in connectomics studies . Figure 2 and Video 1 illustrate this shortcoming by comparing experimental isotropic 3D data and simulated 3D data with a 40 nm sampling interval in z . The anisotropic image stack was generated by averaging 10 frames of the nearly isotropic 4 nm voxel data set along the z-axis to emulate data from a 40 nm section . This emulation is not identical to a TEM image stack collected from 40-nm-thick sections due to different contrast mechanisms provided by backscattered and transmitted electrons . Nevertheless , it offers a first-order comparison between the two imaging modalities , showing the limitations of TEM’s anisotropic data , where details become poorly resolved in re-sliced non-imaging planes . For example , the xz and zy planes show much degraded resolution compared to the xy image planes , due to poor z resolution . This is particularly troublesome in connectomics ( the study of neural connectivity ) , which needs to resolve fine neural processes parallel to the xy imaging plane . In an isotropic data set , there will be no degradation of resolution at any re-sliced planes at random angles . This feature has substantial benefits for tracing neuronal circuitry , where one needs to follow fine processes that are orientated at random angles . Accordingly , we use a unifying resolution metric , where resolution is defined by the worst case in the x , y , or z direction . To elucidate biological structure , it is helpful to render 3D images , without any axial bias , with isotropic resolution . The worst-case axial resolution then dictates the appropriate minimal isotropic voxel size for sampling and rendering . 10 . 7554/eLife . 25916 . 003Video 1 . Three-dimensional x , y , z data showing 4 nm voxels over 600 nm range of Drosophila neuropil with isotropic resolution ( top row ) , and a section where the data is binned together in z to form 4 x 4 x 40 nm3 voxels , to emulate standard TEM sections . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 003 The graphical summary ( Figure 1 ) , which shows the operating regimes of the different EM methods in terms of sample volume and minimum isotropic resolution , identifies an important region of resolution-volume space that remains inaccessible with current techniques . FIB-SEM provides a logical probe for this region , but until now , technical obstacles have blocked its use . The most prominent such obstacle is the volume limitation , dictated by the limited imaging speed and the limited duration of smooth and consistent ablation . Because the process is destructive , there is little room for error in the ablation-imaging cycle , which requires virtually perfect continuity and consistency . Here , we describe a series of measures that address these limitations , thus transforming FIB-SEM into a tool capable of probing this ‘dark’ region of resolution-volume space . We also provide examples to illustrate the potential of large volume FIB-SEM for both neurobiology and cell biology . 10 . 7554/eLife . 25916 . 004Figure 1 . A comparison of various 3D imaging technologies in the application space defined by resolution and total volume . The resolution value indicated by the bottom boundary for each technology regime represents the minimal isotropic voxel it can achieve , while the size value indicated by the right boundary is the corresponding limit in total volume . An expansion in total volume and improvement in resolution of FIB-SEM would fulfill a desired space at the lower right corner , not yet accessible with any existing technology . The three red diagonal constant imaging time contours indicate the general trade-off between resolution and total volume during FIB-SEM operations of 3 days , 3 months , and 8 years , respectively , using a single FIB-SEM system . These contours are sensitive to staining quality and contrast . The yellow star indicates the intercept between the extrapolated 8-year contour and 1 mm3 volume . Considering the hot-knife overhead and machine maintenance downtime , a more realistic estimate would be ~3 years using 4 FIB-SEM systems . The boundaries of the different imaging technologies outline the regimes where they have a preferential advantage , though in practice there is considerable overlap and only a fuzzy boundary . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 004 With connectomics in mind , we designed a customized FIB-SEM system to address the prevailing deficiencies in imaging speed and duration . Our new system incorporates many prior improvements in ion beam and electron microscopy , and new advances that represent key enabling features for large-volume 3D imaging . The two most important technological advances are imaging speed improvement and error detection followed by seamless recovery . Negative sample biasing , an established procedure in scanning electron microscopy , is typically used to improve resolution ( Bouwer et al . , 2016 ) . However , we have found that a moderate positive bias provides a simple and straightforward way to filter out secondary electrons . This scheme transforms a traditional in-column ( InLens ) detector into an effective backscattered electron detector that captures a larger fraction of the backscattered electrons , resulting a ~ 10x improvement of imaging speed without contrast degradation compared to a traditional energy-selective backscattered ( EsB ) detector alone ( see Figure 11j ) . Considerable engineering effort is required to gain major improvements in system reliability . Most of the individual components , especially those achieved via software controls , are not themselves innovative , but their combined effect is transformative . First , multiple layers of error and disturbance protection , including refinements and additions in hardware , software , and utilities , were introduced to prevent catastrophic failures such as an uncontrolled sample ablation . Detailed descriptions of these strategies can be found in the Technology and Methods section . Second , an extension of the closed-loop control of the ion beam ( first introduced by Denk and co-workers [Boergens and Denk , 2013] ) maintained stability , and enabled a seamless restart of the imaging cycle after interruptions . Third , the FIB column was repositioned to be 90 degrees from the SEM column instead of the standard 52–55 degrees; this enabled a shorter working distance , which enhanced the imaging quality . Together these modifications provide a speedy system with overall virtual reliability exceeding that of its individual components . Biological volumes as large as 1 million µm3 containing biologically meaningful neuronal elements , such as individual modules in a Drosophila brain , can now be routinely acquired in a few months . Furthermore , with a previously reported ultrathick partitioning technique , even larger volumes ( e . g . an entire Drosophila brain ) could potentially be subdivided into small pieces , then imaged with multiple FIB-SEM systems running in parallel ( Hayworth et al . , 2015 ) . The capacity for sustained operation also opens an opportunity for cell biologists to explore research topics dependent upon the ability to resolve fine ( <5 nm ) features in 3D . In such applications , a slower high-resolution imaging modality can acquire volumes up to tens of micrometers . By providing straightforward generation of large high-resolution isotropic FIB-SEM datasets , the strategies outlined here can provide clear visualization of complex fine-grained biological structures , permitting exploration of novel elements of cellular architecture . To assess the value of FIB-SEM’s superior z-axis resolution for connectomics research , a portion of a Drosophila optic lobe ( Takemura et al . , 2015 ) containing seven medulla columns was imaged at an isotropic resolution of 10 × 10 × 10 nm3 voxels . The entire volume of 30 × 30 × 60 µm3 was acquired over 2 weeks , and then segmented , annotated , and proofread . We compared these data with those from a previous study of equivalent material studied with classical serial-section TEM performed on 40 nm-thick sections ( Takemura et al . , 2013 ) . While TEM of sections can take beautiful images of dendrites and synapses , many important details will be obscured if they are oriented in the wrong direction , for example a fine dendrite with its axis running parallel to the image plane of a section is easily lost ( Figure 2 ) . Accordingly , 50% more synaptic connections were detected within a single medulla column in the FIB-SEM data set than in the TEM image stack ( Takemura et al . , 2015 ) . These improvements in accuracy provided by FIB-SEM data analysis represent a gold standard , useful for understanding the level of completeness of a connectome derived from TEM sections . Furthermore , by imaging the intact block-face , registration and alignment is easy , unlike in serial section TEM where section tears , scratches , and distortions require complex corrections . Finally , the rate of volume reconstruction , which includes synapse identification , segmentation , and proofreading was ~3–5x faster for the FIB-SEM data ( Plaza , 2014 ) , thanks to finer z resolution and better image registration . 10 . 7554/eLife . 25916 . 005Figure 2 . Three orthogonal views of a ( 600 nm ) 3 block of Drosophila neuropil with isotropic 4 nm voxels and anisotropic ( 4 x 4 x 40 nm3 ) voxels derived from the isotropic data to emulate 40-nm section data . Video 1 corresponds to this Figure . Scale bar , 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 005 The same FIB-SEM system has been used to image other parts of the Drosophila brain of comparable size , including the antenna lobe and mushroom body ( Takemura et al . , submitted ) , with comparable improvements in both speed and accuracy . To evaluate the suitability of the customized FIB-SEM system for imaging even larger volumes , the Drosophila optic lobe , including medulla , lobula , and lobula plate , was imaged at 8 × 8 × 8 nm3 voxel resolution over a 100 day period , yielding a final four terabyte image volume of about 180 × 100 × 50 µm3 ( 50 µm in the direction of the FIB beam ) , shown in Figure 3 . We encountered multiple unplanned system failures ranging from replacement of the SEM field emitter tip to complete pump failure , in addition to more than 20 planned interruptions reflecting the need to replenish the FIB source every 4–6 days . With a standard FIB-SEM system , these interruptions would have led to multiple gaps and other defects in the final image stack that would make it virtually impossible to perform any large-scale reconstructions of connectivity . However , our customized FIB-SEM system was designed to pause the system promptly when interruptions occurred , and to resume seamlessly after the system returned to normal operation . Only one noticeable imperfection associated with system shutdown and source replacement was passed down to the final 3D volume images . Data from this run are shown in Figure 3b , c , which render x-z and y-z re-sliced views of a large volume that includes medulla , lobula , and lobula plate . Because the voxels were isotropic , re-slicing did not degrade image resolution . Plasma membranes , presynaptic T-bars , and postsynaptic densities were clearly visible in both renderings . Importantly , there were no visible discontinuities due to system interrupts other than the one noted above ( Video 2 ) . The robust handling of both scheduled and random interruptions enables long-term operations spanning months of imaging . 10 . 7554/eLife . 25916 . 006Figure 3 . FIB-SEM images of Drosophila optic lobe . ( a ) A cropped FIB-SEM volume showing medulla ( M ) , lobula ( L ) , lobula plate ( LP ) , and chiasm ( C ) . ( b ) An enlargement of the blue cross-section in ( a ) showing a re-sliced y-z plane . Scale bar , 10 µm . ( c ) An enlargement of the green box in ( a ) showing a re-sliced x-z plane where very fine neural processes are visible . Red arrows indicate synaptic structures . Scale bar , 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 00610 . 7554/eLife . 25916 . 007Video 2 . Re-sliced view of Drosophila optic lobe showing medulla ( left ) , lobula ( upper right ) , lobular plate ( lower right ) . 150 × 64 × 40 µm3 region with 10x zoom , 8 × 8 × 8 nm3 voxel . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 007 This image volume can be segmented and proofread in its entirety to create dense reconstructed data sets where > 90% of the synaptic contacts are identified and assigned to reconstructed neurons . Figure 4 illustrates this for two compartments of the fly brain: the medulla and mushroom body . In the cross-sections of Figure 4b , c , segmented regions assigned to specific neurons that are contained in the imaged volume are colorized in green , while processes traced to neurites that extend tangentially outside the imaged volume are colorized in yellow . Only a small fraction of the area remains red , corresponding to ‘orphaned’ incompletely reconstructed neuron fragments . Note that in the rendering of Figure 4c all cells in the mushroom body have been identified . The few examples in the literature that have achieved comparable levels of volume completeness all required multi-year proofreading or tracing efforts by large teams ( Meinertzhagen , 2016 ) . These dramatic improvements in both the overall efficiency of the reconstruction effort and the degree of completeness of the connectome are important benefits of isotropic block-face FIB-SEM data . 10 . 7554/eLife . 25916 . 008Figure 4 . Examples of densely reconstructed data . ( a ) A 3D rendering of seven columns of the medulla of the Drosophila optic lobe from FIB-SEM showing reconstructed neurons from a ~ 30 , 000 µm3 volume . This reconstruction , which required over five man-years of effort , was ~3–5x faster than a comparable optic lobe reconstruction using an image stack from serial-section TEM , for which we have reconstructed a single medulla column ( Takemura et al . , 2013 ) . Scale bar , 10 µm . ( b ) A cross-section of the neuropil of the medulla in the optic lobe of Drosophila , showing the high degree of reconstruction completeness that is possible with FIB-SEM data . The hexagonal periodicity reflects the hexagonal pattern of the ommatidia of the fly’s retina . The colors illustrate how all neural processes have been assigned . Green indicates various identified columnar input neurons contained within this volume , and yellow indicates axons and arbors of various medulla neurons that branch into or out of this volume . The small remainder ( shown in red ) highlights the ‘left over’ parts , including unidentified and orphaned fragments of neurons and glial processes . Well over 90% of the neuropil volume could be reconstructed and assigned to specific neurons . Scale bar , 1 µm . ( c ) Cross-section of the neuropil of the mushroom body of Drosophila . Notice that virtually all processes in this section have been identified and colorized green ( to denote Kenyon cells ) or yellow ( for other identified mushroom body neurons ) . The only ‘left over’ uncoded processes are a few thin fragments dispersed within the mushroom body boundary that could not be confidently assigned to a specific cell . The mushroom body volume was comparable to the seven-column medulla volume and required a comparable reconstruction effort . Scale bar , 10 µm . Image process , segmentation , and 3D rendering provided by the Janelia FLYEM team , see Acknowledgements . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 008 For even larger samples with the typical electron dose of our standard SEM imaging condition , when dimensions in the direction of the FIB beam exceed 60–100 µm a FIB milling instability emerges , producing curtains and waves of non-uniform material removal that limit data quality ( Lemmens et al . , 2011 ) . We have addressed this by a hot knife partitioning method ( Hayworth et al . , 2015 ) that subdivides a larger sample into 20 µm-thick slabs . The method preserves sample quality to within ~20 nm of the cut surface , enabling effective stitching of 3D connectomic data across the cut slabs . An overview of hot knife partitioning and imaging results are shown in Figure 5 . We have imaged nine slabs of 20 × 250 × 250 µm3 with two FIB-SEM systems in parallel , representing a total volume of over 10 × 106 µm3 that spans the key central complex components of the fly brain . One of these slabs is shown in Video 3 . A complete fly brain of 30 slabs could be imaged in ~5 FIB-SEM-years of acquisition time; because the approach is scalable , multiple FIB-SEM machines could be used to reduce the total time required . Though substantial , this acquisition duration is dwarfed by the overriding component of the dense connectome pipeline: tracing and proofreading of the segmented data , which can take two to three orders of magnitude more man-years ( Plaza , 2014 ) ! Efficient proofreading thus requires teams of hundreds to thousands of people to keep up with the rate of data generation . The superior z resolution of the FIB-SEM data provides a more balanced pipeline for complete reconstruction of the fly brain with better-matched investment between acquisition and proofreading . 10 . 7554/eLife . 25916 . 009Figure 5 . Overview of ultrathick partitioning and imaging results . ( a ) X-ray micro-CT of Drosophila brain cross-section shows central complex structures ( the doughnut-shaped structure at center is the ellipsoid body ) . Yellow planes show locations of hot knife cuts at 20 µm intervals . Red highlighted area shows FIB-SEM imaged volume covering nine hot knife sections ( #22-30 in our notation ) . Example light micrographs of Section #26 and #27 are shown ( dashed box shows FIB-SEM imaged volume in each ) . ( b ) Each hot knife section is flat embedded against a PET laminate , individually mounted on a metal stud , and laser trimmed to dimensions suitable for efficient FIB-SEM imaging ( Hayworth et al . , 2015 ) . ( c ) X-ray micro-CT of individually-mounted hot knife section showing laminar structure . All sections are micro-CT imaged as a quality control prior to FIB-SEM imaging . ( d ) Z-Reslice through FIB-SEM imaged volume of section #26 . Blue box shows location of volume stitch test in protocerebral bridge region . Scale bar , 40 µm . ( e ) Result of volume stitch test in protocerebral bridge region . The FIB-SEM volumes of corresponding regions of adjacent hot knife sections #26 and #27 were computationally flattened and stitched to produce a single FIB-SEM volume suitable for tracing ( Hayworth et al . , 2015 ) . Red dashed line shows stitch line . This stitched volume is available as Video 8 . Scale bar , 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 00910 . 7554/eLife . 25916 . 010Video 3 . Re-sliced views of a hot knife slab containing the Drosophila central complex at various zoom levels . The left panel shows the entire slab at 512 × 512 × 64 nm3 voxel . The center panel shows a cropped region at the bottom of fan shape body ( FB ) with 64 × 64 × 64 nm3 voxel . The right panel shows a cropped region in FB with 16 × 16 × 16 nm3 voxel . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 010 This new capability for long-term operation opens up a whole new application space for high-resolution ( ~4 nm isotropic ) 3D imaging . Exemplary data sets from Drosophila neuropil and Chlamydomonas reinhardtii illustrate the difference in resolution for high-volume vs . high-resolution acquisition ( Figure 6 ) . Considering the large processes found in mammalian neuropil , this may not be required for mammalian connectomics , but the additional resolution can be very useful to decipher the extremely fine processes in Drosophila neuropil ( Meinertzhagen , 2016 ) . The ability to explore synaptic motifs and other details of neuropil at high resolution can be very helpful in interpreting the larger volume but poorer resolution data sets required to generate a full connectome . The operating conditions needed , including lower current to reduce chromatic and spherical aberrations , and lower electron landing energy to reduce point spread function size along z-axis , require lower acquisition rates , implying smaller sampled volumes . The typical trade-off between resolution and volume for a given time constraint is illustrated by constant time contours in Figure 1 ( assuming our baseline Drosophila samples are used ) . The exact placement of these contours depends on specific features of the sample . For example , images can be acquired faster from mammalian neural tissue than from Drosophila , both because its stronger contrast and fewer small processes . 10 . 7554/eLife . 25916 . 011Figure 6 . Improved FIB-SEM resolution reveals more detailed cellular structures in biological samples . Typical images of ( a ) Drosophila central complex and ( b ) Chlamydomonas reinhardtii , using standard 8 × 8 × 8 nm3 voxel imaging condition are shown in the top panels . The bottom panels show the corresponding high-resolution images at 4 × 4 × 4 nm3 voxel . Scale bar , 1 µm . Inset scale bar , 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 011 An 8 × 8 × 8 µm3 high-resolution data set from the Drosophila central complex acquired over 10 days illustrates clearly delineated processes , and synapses with well-defined postsynaptic densities ( see Figure 7 and Video 4 ) . This high-quality data provides significantly more accurate estimates of synaptic connectivity than possible for lower-resolution stacks . Moreover , data of this quality will permit highly automated reconstruction , thus greatly reducing the time required for manual proofreading . In contrast , it would take only ~2 hr to acquire a comparable dataset using the high-throughput mode of acquisition , but it would require hundreds of man-hours of proofreading to correct the dataset because of the small neurites in Drosophila . High-resolution data also provides an accurate ‘gold standard’ for the higher throughput data , helping to interpret the larger dataset and perhaps also serving as a reference for machine learning . The actual resolution of this high-resolution mode can be objectively quantified by intracellular structures of known dimensions . For example , in Figure 7b , the resolved hollow core of the 25 nm outer-diameter/17 nm inner-diameter microtubule confirms a resolution of <3 . 5 nm ( referenced to a 25–75% step edge rise resolution criteria , consistent with an alternative definition for resolution of ( spatial period = 21 nm/2π ) . 10 . 7554/eLife . 25916 . 012Figure 7 . A high-resolution image ( 4 × 4 × 4 nm3 ) of a Drosophila protocerebral bridge ( in the central complex ) reveals fine details of various organelles . ( a ) an 8 × 8 µm2 area overview; ( b ) end-on and side views of microtubule , indicated by green arrows; ( c ) polyribosomes attached to the endoplasmic reticulum , indicated by blue arrows; and ( d ) synaptic vesicles , presynaptic T-bar , and postsynaptic density , shown in two different z planes . Video 4 shows the corresponding full z stack . Scale bar , 1 µm in ( a ) and 500 nm in ( b ) - ( d ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 01210 . 7554/eLife . 25916 . 013Video 4 . Detail of synapse in Drosophila protocerebral bridge showing multiple post synaptic contacts . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 013 Further improvement in resolution/volume for FIB-SEM should pay substantial dividends for cell biology , just as previous two-fold resolution improvements in fluorescence microscopy have enabled important scientific advances by rendering finer details . Here , we illustrate the potential of this approach for cellular neurobiology with data from the nucleus accumbens , a region of the mammalian forebrain involved in reward processing ( see Figure 8a and Video 5 ) . This dataset encompasses much of the soma of one neuron , along with the surrounding neuropil . The endoplasmic reticulum ( ER ) is well resolved and easily segmented , allowing its full 3D structure to be extracted; one is no longer relying on sampling from 2D EM sections to infer its 3D organization . For example , the frequency of ER-to-plasma membrane and ER-to-mitochondrion contacts can be quantified across a whole cell ( Wu et al . , 2017 ) , providing new insight into contact-dependent processes such as lipid transfer . Dendrites with readily-segmentable organelles , and synapses with all vesicles countable are plentiful and could be mined for statistics , for example , quantitative comparisons among synapses from the same axon , or onto the same dendrite . 10 . 7554/eLife . 25916 . 014Figure 8 . Isotropic high-resolution data offer easy visualization of 3D structures through arbitrary slices . ( a ) Two orthoslices ( x–y and x–z ) from nucleus accumbens in a sample of adult mouse brain . ( b ) Specific slices of the same volume as in ( a ) provide easy viewing of Golgi in the context of other nearby organelles . Matching arrows in ( a ) and ( b ) indicate the same ROI’s through different slice views , ( c ) Polyribosomes and nuclear pores ( arrow ) at the nuclear envelope of a Chlamydomonas reinhardtii . The 3D rendering was generated by thresholding a maximum intensity projection where brighter yellow ( polyribosomes ) indicates higher intensity of backscattered electrons due to stronger staining than darker yellow ( nuclear pores ) . Scale bar 1 µm and 100 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 01410 . 7554/eLife . 25916 . 015Video 5 . Nucleus accumbens of a mouse brain . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 015 The ability to reorient the isotropic 3D data set provided by FIB-SEM permits high-resolution examination of arbitrary slices , thus offering new insights . For example , the nucleus accumbens volume reproduced here shows the edge of a soma , a partial nucleus , and the adjoining Golgi apparatus with a barely visible grey region associated ( Figure 8a ) . By rotating that data block into a more informative plane ( Figure 8b ) one can see that this Golgi is corralled by a grey fibrous arc , a structure perhaps formed from rootletin ( Chen et al . , 2015 ) , as suggested by the observed 120 nm periodicity . Remarkably , by following these fibers in 3D through multiple rendered image planes , we see that they connect and terminate onto the basal body of a cilium . In another demonstration of the potential of FIB-SEM for cell biology discovery , we imaged a group of the single-cell green alga Chlamydomonas reinhardtii . C . reinhardtii is widely used as model organism in the study of cilia/flagella structure and function , advancing our understanding of how defects in these organelles cause human diseases . It is also a broadly used model system for eukaryotic photosynthesis , chloroplast biogenesis , light perception , cell-cell recognition and cell cycle control ( Harris , 2001 ) . One cell cropped out of the data volume is shown in Figure 6b . Many details in the structure of the nucleus , mitochondria , ER , and Golgi are visible . Of particular interest is the large cup-shaped chloroplast and the associated light-sensing eyespot and the ‘pyrenoid , ’ a Rubisco-rich structure involved in the first major step of carbon fixation . All these structures are clearly distinguishable; their overall organization and interplay , in various functional conditions and in light-sensitive mutants , would provide new light on mechanisms of photosynthesis . C . reinhardtii flagella have also been extensively studied to understand cell motility . Zooming into the flagellar base ( see inset Figure 6b and Videos 6 and 7 ) , the nine-fold doublet microtubule structure becomes clearly visible , and details of the mature basal body pair and two probasal bodies are revealed . The latter form during basal body replication , at a very early stage of cell division in C . reinhardtii . After cytokinesis , the daughter cell will contain one mature basal body and one newly-formed one from which the flagella pair will grow ( Silflow and Lefebvre , 2001; Preble et al . , 2000 ) . Our imaging of this stage illustrates how recording a large data set that includes a population of cells allows the researcher to capture a variety of details through the cell cycle , as well as being more statistically meaningful than a single tomogram of a thin section of a single cell could ever be . 10 . 7554/eLife . 25916 . 016Video 6 . Whole Chlamydomonas reinhardtii . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 01610 . 7554/eLife . 25916 . 017Video 7 . Flagella structure of a Chlamydomonas reinhardtii . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 01710 . 7554/eLife . 25916 . 018Video 8 . Result of volume stitch test in Drosophila protocerebral bridge region between hot knife sections #26 and #27 . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 018 The wealth of structural data yielded by this approach merits the application of data mining tools . As an illustration of how further cellular details can be extracted from our FIB-SEM data set , we cropped out a thin spherical shell that starts at the nuclear envelope , extending 50 nm outward . By masking out backgrounds from inside the nucleus and those beyond 50 nm outside the nuclear envelope , we can restrict visualization to structures at or very close to the nuclear surface . This shell ( rendered as a 3D section of a sphere in Figure 8c ) reveals all the polyribosomes that decorate the exterior surface of the nucleus . Similar spirals of paired ribosome necklaces were seen previously in EM sections that happened to intersect the polyribosomes at just the right angle ( Christensen et al . , 1987 ) . Here , by virtue of the isotropic character of the 3D data , one is no longer sampling a fortuitous section of the cell generated by the geometry of sectioning . Consequently , it is now possible to count all the polyribosomes on the whole nucleus . The insert of Figure 8c shows the nuclear pores , with their eightfold symmetry , in relation to these polyribosomes . Similar data can be extracted for ER-bound ribosomes throughout the cell . Most FIB-SEM images are assigned an arbitrary grey scale . This limitation deprives one of opportunities 1 ) to understand the true limits of SEM performance , 2 ) to optimize acquisition and identify inefficiencies , 3 ) to quantify absolute staining levels in samples , 4 ) to distinguish instrument vs sample factors , and 5 ) to cross-compare performance across different labs , samples , and FIB-SEM instruments . Initially , motivated by our desire to improve the speed of traditional SEM imaging , we needed to get a better understanding of the mechanisms by which backscattered electrons generate contrast , and to better define the ultimate limits of collection/detection efficiency . By comparing experimental results from specimens of known chemical compositions ( gold , epoxy resin , and metal-organic compounds ) with theoretical simulations using Monte Carlo methods for electron scattering ( Figures 13 , 15 and 17 ) and SIMION for electron optics ( Figure 14 ) , we characterized baselines under different sample biasing conditions ( Figures 11 and 12 ) . The general agreement between simulation and experimental results guided us to optimize SEM imaging with minimal artifacts ( Figure 11 ) . As a foundation for the experiments reported here , we also established two independent methods of quantifying the signal in terms of electrons detected . These explicit electron counts can be compared against models of electron scattering and also to reference standards , to establish best operating conditions ( the results are detailed in the Technology and methods section ) . A current milestone in connectomics is to image a 1 mm3 volume . This is a daunting task for FIB-SEM , given its slower imaging speed compared to other competing methods . With the current throughput used for Drosophila brain , it would take approximately 100 years to acquire 1 mm3 volume at 8 × 8 × 8 nm3 voxel resolution with a single system . However , the task might not be as hopeless as it seems . First , we have seen less demand for resolution in applications requiring large volumes . For example , mammalian brains have relatively larger processes and synapses compared to those of Drosophila . Mouse neuronal circuits should be traceable with minimal isotropic voxel of 16 nm ( Mikula and Denk , 2015 ) . Secondly , protocols for mammalian sample preparation produce higher contrast than those for insects , which allows shorter imaging time . Being able to image with larger voxels benefits FIB-SEM volume throughput in two ways: fewer voxels ( to the third power ) and the ability to use larger imaging current , which allows shorter scanning dwell time to achieve the same shot noise , though the FIB milling overhead is increased . Based on limited data comparing Drosophila and mouse cortex samples acquired on the same FIB-SEM system , we estimate an 8x improvement on volume imaging speed with mammalian brain tissue . As illustrated in Figure 1 , one could reach 1 mm3 with 16 × 16 × 16 nm3 voxels using a single FIB-SEM system in 8 years . Given that we can acquire samples from multiple systems running in parallel , one could expect a more feasible timeline: with our current capacity of four production systems , we estimate that a 1 mm3 volume could be imaged in a total of 3 years , including hot-knife overhead and machine maintenance . The technical developments reported here now make it feasible to extend the intrinsic advantages of FIB-SEM , including excellent z resolution , isotropic voxels , and easy 3D data acquisition to larger volumes , by allowing long-term imaging for weeks , months , or even years . These enhancements can be adopted by other labs or on commercial systems to transform FIB-SEM into an effective tool for connectomics , which demands both high data quality and large data sets . The higher resolution mode of SEM imaging ( ~4 nm isotropic ) can also be harnessed to study volumes of 5–50 µm linear dimension , providing a more detailed view of neuropil to guide the connectomics effort . For cell biology , the access to fine resolution and complete eukaryotic cell-sized volumes make this a practical alternative to the difficult and tedious stitched serial section tomographic TEM approaches now available . Thus , the technical advances reported here open new vistas for the study of biological structures . SEM imaging is usually slower than that of transmission electron microscopy ( TEM ) for several reasons . First , SEM acquires images pixel-by-pixel in series , whereas TEM acquires all pixels of the image in parallel , with orders of magnitude larger imaging current . Second , SEM detects backscattered or secondary electrons , emitted at a much smaller flux than the transmitted electrons measured by TEM . Third , the low SEM landing energy of <2 keV typically used to reduce the electron penetration depth into the block-face ( thus increasing z-axis resolution ) can reduce contrast , especially below 800 eV ( Figure 15 ) . Fourth , the signal-to-noise ratio ( SNR ) of a pixel depends on the number of primary electrons devoted to it , which in turn is determined by the beam current . For resolution of <10 nm at low landing energy , the incoming electron beam current must be limited to <10 nA , since higher current beams require larger apertures , which are subject to greater spherical and chromatic aberrations as well as Coulomb repulsion , creating unacceptable blur in the beam spot . Finally , post-staining is commonly used in TEM and ATUM-SEM ( Hayworth et al . , 2006 ) sections to enhance contrast , whereas FIB-SEM must image the block surface without benefit of the extra contrast from post-staining . As a result of all these factors , FIB-SEM and other block-face scanning methods ( e . g . serial block-face scanning electron microscopy ) require a lower image acquisition rate than TEM to achieve the same SNR . Along with slow throughput , the limited duration of continuous FIB-SEM data acquisition constrains the useable volume . FIB-SEM is destructive and does not allow re-imaging , imposing formidable requirements on system reliability . Interrupts have a direct impact on the total contiguous imaging volume . System drift , routine maintenance , facilities interrupts , or system failures can all terminate a 3D FIB-SEM operation . Focus or beam stigmation of SEM can drift from its optimal settings within 1–2 days , due to environmental or sample stage instability . A pause in the milling/imaging operation is normally needed to correct these drifts and restore image quality . Moreover , a FIB gallium source has a lifetime limited to 3–4 months of continuous operation , and requires reheat or ‘flashing’ every 3 to 5 days . Even without facility or system failures , these regular maintenance events impose hard limits on continuous data acquisition , and thereby the size of contiguous high-quality data sets that can be collected with standard FIB-SEM systems . The ability to operate the system for long periods is crucial for imaging large volumes . Unfortunately , there are many potential interruptions to a FIB-SEM system , with intervals ranging from a few hours to a year ( Figure 9 ) . Some of them relate to system reliability , which is challenging to improve . Others are regular maintenance items that are impossible to eliminate , such as FIB source reheat ( 3 days ) and replacement ( 3 months ) . Since FIB-SEM image acquisition is destructive , any interrupt could be detrimental . For example , a spike in room temperature may cause the SEM focus to drift , and the FIB beam-pointing position relative to the specimen to change , potentially damaging the sample and sabotaging the continuity required for neural tracing of fine processes across the brain . To address these frequent interrupts , we have developed a system that immediately pauses to prevent damage , and resumes seamlessly after restoration of normal operation . By providing high virtual reliability , this system greatly expands the total imaging volume possible . 10 . 7554/eLife . 25916 . 019Figure 9 . Major challenges to long-term reliability and stability of FIB-SEM systems . ( a ) System failure modes with different frequencies of occurrence and resume challenges . ( b ) Corresponding customized solutions . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 019 To gain more insight into mechanisms underlying the detection efficiency , its limits and imaging resolution , we modeled the physics of the electron/sample interaction and the scattered electron detection efficiency , validating the models by comparison with known ‘calibration’ samples . Only a small fraction of the electrons from the incoming primary electron beam will be scattered back and collected by the detector , and an even smaller variation of that signal corresponds to the contrast generated by the stained and unstained parts of the sample . Monte Carlo simulations can provide a useful perspective . The approach described by David Joy ( Joy , 1991 ) was adopted , using the Rutherford energy loss formula between scattering events , modified to use the angle- and energy-dependent elastic scattering cross-sections for the key elements H , C , N , O , P and Os . All these scattering cross-sections for all energies were obtained from a NIST database ( Jablonski A et al . , 2016 ) . Conventional biological sample preparations optimized for serial sectioning can be directly applied to 3D FIB-SEM with minor modifications . Both chemical fixation with mixed aldehydes and high-pressure freezing followed by freeze substitution yielded successful results on FIB-SEM . Images were acquired on a Zeiss NVision40 system using 1 . 1 keV electron beam energy with a sample bias voltage of 0 . 4 kV , which resulted in a landing energy of 1 . 5 keV . The probe current was set at ~3 nA and working distance at ~4 . 5 mm with imaging speed of 1 . 25 MHz . Images acquired on a Zeiss Merlin system ( hybrid with 90 degree mounted FEI Magnum FIB ) used slightly different conditions: landing energy was 1 . 2 keV with 0 . 6 keV electron beam energy with a sample bias of 0 . 6 kV . The probe current was set at ~4 nA and working distance at ~3 mm with imaging speed of 4 MHz . The x-y pixel size was 8 nm for both systems unless noted . SEM images were acquired for every 2 nm of material removal . After the final image series were registered using IMOD ( Kremer et al . , 1996 ) or SIFT plug-in of Fiji ( Schindelin et al . , 2012 ) , every four consecutive images were binned down to one , forming an image stack with isotropic voxels of 8 × 8 × 8 nm3 . The gallium FIB column was operated at 30 keV . A 7-nA probe current was selected for milling with the FEI Magnum column , while a 13- or 27-nA probe current was used in a Zeiss NVision40 . A repeated line scan of 0 . 1 µm pixel and 1-MHz frequency was applied . A 10 Hz PID closed-loop algorithm written in LabVIEW ( National Instruments ) controlled the line scan position relative to the specimen block-face . To minimize overhead , the milling time was typically set to be less than 20% of SEM imaging time . With a 200 × 200 µm2 block-face specimen , the milling time was around 10 s or less for every frame ( 2 nm z-step ) using a 27-nA FIB probe current . A customized hardware , control , and software package was developed to enable long-term acquisition on a FIB-SEM system . Major hardware components included a National Instrument signal generation and acquisition system , temperature sensors , a high-voltage isolation current amplifier , and a home-built computer with RAID6 storage . A National Instrument PXIe-1082 chassis , equipped with two PXI-5421 , one PXIe-5122 , one PXIe-6124 , and one PXIe-6259 , was connected to the RAID6 computer through a PXIe-PCIe8371 bridge card . Collectively , they provided scan signals for SEM imaging and FIB imaging/milling , as well as SEM image collection and storage . This system was able to acquire data from two SEM channels up to 32k x 32k pixel at 100 MHz . It also recorded machine vital signs , such as ambient temperature , specimen current , and Faraday cup currents , at 10 Hz . Software written in LabVIEW was used to control the entire FIB-SEM operation . Voxel size is reciprocally related to imaging speed; the best compromise must be determined for each study . Smaller voxel size and lower imaging speed generate higher quality data sets , but can be impractically slow , depending on the specimen and the biologically relevant sample sizes . The optimal conditions also depend on the specimen preparation , especially the staining conditions and the resulting electron contrast . We performed experiments to determine the optimal-throughput balance between imaging and subsequent analysis for complete circuit reconstruction of Drosophila brain . First , we collected a high-quality image stack of Drosophila medulla at 5 × 5 × 5 nm3 voxel size , with small current and slow scanning speed . Software binning and shot noise were then added to simulate datasets collected with larger voxels and at faster scanning speed ( Figure 19 ) . Based on the automated segmentation and human proofreading speed and error rate , we determined that a voxel sampling of 8 nm was optimal for our Drosophila connectomics studies . This is at least 50% smaller then Nyquist sampling distance for beam blur associated with currents of up to 4 nA characterized by the edge signal intensity drop-off of gold nanoparticles on a carbon surface ( Electron Microscopy Sciences , P . O . Box 550 , 1560 Industry Road , Hatfield , PA 19440 , USA , Part #79511–01 ) . Dwell times of 0 . 8 to 3 µs with 12 , 000 to 48 , 000 primary beam electrons per pixel gave a useable SNR of 5 or better . 10 . 7554/eLife . 25916 . 030Figure 19 . Imaging speed and voxel size study to determine optimal condition for neuronal circuit reconstruction . A high-resolution ( 5 nm ) image stack was first acquired at low imaging speed ( 38 kHz ) . Corresponding larger voxel and more rapidly acquired images were emulated by binning and adding shot noise through software . A condition of around 8 nm and 300 kHz was found to optimize traceability and throughput . Scale bar , 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 25916 . 030 Continuous long-term operation opens up opportunities for studies that require higher resolution . For low-voltage ( ~1 keV ) SEM’s operating with 0 . 1–10 . 0 nA on the primary beam , the resolution is a function of current ( Reimer , 1993 ) . Spherical and chromatic aberration are prime contributors to beam blur , so smaller numerical apertures with smaller beam currents typically improve resolution . Unfortunately , the downside of better resolution is a longer image acquisition time; conversely , a reduced volume size will be accessible in a fixed time . As an example , we have 2 . 7 nm resolution ( measured as 50% rise distance at a step edge ) at 0 . 08 nA vs 5 . 5 nm at 4 . 0 nA . To keep the same sampling-to-resolution ratio , there will be a 2x shorter sampling distance or 8x more voxels for 2x higher resolution in a given volume . Furthermore , since the electron dose per voxel needs to be the same for constant signal-to-noise ( assuming mainly shot noise ) , we need to integrate the current on each smaller voxel for ~50x longer . Together this means that doubling the resolution requires a 400x slower volume acquisition rate ! Thus , instead of ( 100 µm ) 3 , only a ( 14 µm ) 3 cube can be acquired in 100 days . In general for constant signal-to-noise at a given beam current I ( δ ) and with voxels scaling with resolution δ , the volume rate is proportional to dV/dt = δ3*I ( δ ) . One should choose the trade-off between resolution and volume that is optimal for any given sample and line of enquiry . Figure 1 illustrates the relationship between imaging resolution and achievable volume with contours of required acquisition time . The vertical axis assumes isotropic x , y , and z resolution . Besides primary beam blur , the resolution must include the vertical and lateral extent of the exploration range of the back-scattered electron . The Monte Carlo trajectory simulation discussed previously shows that the landing energy of the primary electron must also be adjusted to lower energies to reduce the spatial spread to values consistent with the primary electron beam blur . This means landing energies of 600–1000 eV . As mentioned in the Monte Carlo discussion , contrast to the heavy metal stain is lost rapidly below 600 eV , setting a lower limit to the practical operating point . To maintain SNR at the lower contrast will require further reduction of imaging speed . How does the high-resolution/large volume approach to FIB-SEM imaging presented here compare with other forms of electron microscopy ? It turns out to have a complementary application space . Electron tomography based on high-voltage transmission EM is a standard method used to obtain high-resolution 3D images , typically superior to that of lower energy FIB-SEM imaging . This methodology typically uses ~200–500 nm thick sections . Imaging of thicker volumes requires difficult and time-consuming manual registration of multiple sections . To image thicknesses requiring more than a few sections , FIB-SEM is usually a more practical alternative . If one can sacrifice resolution in one dimension and anisotropic voxels are useful , then either traditional serial section TEM or diamond knife-trimmed block-face SEM might be a more practical alternative . The z resolution of mechanical slicing with a diamond knife can be improved with de-convolution in which the sample is imaged using various landing energies . However , this virtual slicing approach adds a substantial burden in operation and reduces throughput and has not yet been demonstrated for large volume applications . In summary , the technological developments presented here enable FIB-SEM to probe a new domain of considerable biological significance .	Precise three-dimensional imaging can help make sense of microscopic details in biology . These images are usually built up from many two-dimensional images stacked on top of each other . One approach for examining particularly fine details , such as the connections between nerve cells in the brain , is called focused ion beam scanning electron microscopy ( or FIB-SEM for short ) . This approach works by creating an image of the surface layer of a sample , which is then stripped away using a beam of charged particles to reveal the layer beneath . The new surface can then be imaged and so on , through the whole sample . Unfortunately , FIB-SEM devices are currently slow and can only run for a short time , leading to a lack of continuity in the stack of images . FIB-SEM would allow faster , more accurate and detailed studies of connections between brain cells , and other elaborate biological systems , if the technology could be made faster and more reliable over months of continuous operation . The current technical challenge is to create a system that can , for example , successfully image and analyse all the connections between the more than 100 thousand cells that make up the brain of a fruit fly – a common model organism in neurobiology . Xu et al . aimed to create a technique to image a complete fly brain , with gaps of just 8 nanometres between each image in a stack , within a reasonable timeframe . By improving how FIB-SEM signals are detected , making use of advances in ion beam controls , and by engineering ways to recover from system malfunctions , Xu et al . developed an enhanced FIB-SEM device . To demonstrate its value , the new technology was used to create images of a third of a fruit fly’s brain , parts of a mouse’s brain , and cells of a single-celled alga called Chlamydomonas reinhardtii . The results show that large and complex samples can be successfully imaged in their entirety to adequate detail , enabling high-quality reconstruction of the connections between nerve cells . The level of detail , which can be further increased for smaller samples , offers advantages in precision and image quality over other comparable techniques . As well as helping to study the brain , this approach could also be used to examine details inside cells . Future work to advance this technology will enable larger and more complete imaging of elaborate biological structures .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Technology", "and", "methods" ]	[ "cell", "biology", "tools", "and", "resources", "neuroscience" ]	2017	Enhanced FIB-SEM systems for large-volume 3D imaging
Mineral malnutrition stemming from undiversified plant-based diets is a top global challenge . In C3 plants ( e . g . , rice , wheat ) , elevated concentrations of atmospheric carbon dioxide ( eCO2 ) reduce protein and nitrogen concentrations , and can increase the total non-structural carbohydrates ( TNC; mainly starch , sugars ) . However , contradictory findings have obscured the effect of eCO2 on the ionome—the mineral and trace-element composition—of plants . Consequently , CO2-induced shifts in plant quality have been ignored in the estimation of the impact of global change on humans . This study shows that eCO2 reduces the overall mineral concentrations ( −8% , 95% confidence interval: −9 . 1 to −6 . 9 , p<0 . 00001 ) and increases TNC:minerals > carbon:minerals in C3 plants . The meta-analysis of 7761 observations , including 2264 observations at state of the art FACE centers , covers 130 species/cultivars . The attained statistical power reveals that the shift is systemic and global . Its potential to exacerbate the prevalence of ‘hidden hunger’ and obesity is discussed . The first empirical evidence of lower mineral content in plants exposed to eCO2 appeared at least over a quarter century ago ( e . g . , Porter and Grodzinski , 1984; Peet et al . , 1986; O’Neill et al . , 1987 ) . Physiological mechanisms responsible for the overall decline of plant mineral content—with expected changes being non-uniform across minerals—have been proposed: the increased carbohydrate production combined with other eCO2 effects such as reduced transpiration ( Loladze , 2002; McGrath and Lobell , 2013 ) . However , most of the experimental evidence showing CO2-induced mineral declines came from artificial facilities , mainly closed chambers and glasshouses , and many results were statistically non-significant . This led some research groups to challenge altogether the notion of lower mineral content in plants exposed to eCO2 in field conditions . Such conditions are most accurately represented in Free-Air Carbon dioxide Enrichment ( FACE ) centers , which have been established in at least 11 countries . In the grains of rice harvested at four FACE paddies in Japan , Lieffering et al . ( 2004 ) found no decline in any of the minerals but lower N content . The result disagreed with Seneweera and Conroy ( 1997 ) , who were the first to report lower iron ( Fe ) and zinc ( Zn ) in grains of rice grown at eCO2 and warned that altered rice quality can negatively affect developing countries . Lieffering et al . ( 2004 ) , however , argued that the result of Seneweera and Conroy ( 1997 ) could be an artifact of growing rice in pots , which restrict rooting volumes . They hypothesized that in FACE studies , which provide unrestricted rooting volumes , plants would increase uptake of all minerals to balance the increased carbohydrate production . This hypothesis , however , found no support in the FACE studies of Pang et al . ( 2005 ) and Yang et al . ( 2007 ) ( carried out in China and latitudinally not very far from the study in Japan ) , who found that eCO2 significantly altered the content of several minerals in rice grains . The contradictory results coming from these studies on rice seem perplexing , especially in light of the very robust effect that eCO2 has on N in non-leguminous plants . Elevated CO2 reduces N concentrations by 10–18% systemically throughout various tissues: leaves , stems , roots , tubers , reproductive and edible parts , including seeds and grains ( Cotrufo et al . , 1998; Jablonski et al . , 2002; Taub et al . , 2008 ) . If the increased carbohydrate production dilutes the nutrient content in plants , why does the dichotomy seem to exist between the responses of N and minerals to eCO2 ? In addition to the carbohydrate dilution and reduced transpiration , eCO2 can further lower N concentrations in plants by: ( 1 ) reducing concentrations of Rubisco—one of the most abundant proteins on Earth that comprises a sizable N-pool in plants ( Drake et al . , 1997 ) , and ( 2 ) inhibiting nitrate assimilation ( Bloom et al . , 2010 ) . Hence , it is reasonable to expect the effect of eCO2 on N to be larger and , thus , easier to discern than its effect on most minerals . The stronger signal for N , combined with the plentiful and less noisy data on this element , can help explain why by the end of last century the effect of eCO2 on N had been already elucidated ( Cotrufo et al . , 1998 ) , but its effect on minerals has remained elusive . The obscure nature of the effect of eCO2 on minerals becomes particularly apparent in the largest to date meta-analysis on the issue by Duval et al . ( 2011 ) , who fragmented data from 56 eCO2 studies into 67 cases . In 47 of the cases , the effect of eCO2 on minerals was statistically non-significant , that is the 95% Confidence Interval ( CI ) for the effect size overlapped with 0 . The remaining 20 cases were statistically significant but showed no pattern: for example , Fe increased in grasses but decreased in trees , Zn increased in roots but decreased in stems , while in grains only sulfur ( S ) decreased . Duval et al . ( 2011 ) concluded: “A major finding of this synthesis is the lack of effect of CO2 on crop grains nutrient concentration” . This would imply laying to rest the hypothesis that eCO2 consistently alters the plant ionome and would render mitigation efforts to combat declining crop mineral concentrations in the rising CO2 world unnecessary . However , a closer examination of the results of Duval et al . ( 2011 ) reveals that every statistically significant increase in mineral concentrations was obtained by bootstrapping a sample of size 2 , 4 or 5—a recipe for generating invalid 95% CIs . Ioannidis ( 2005 ) showed that false research findings , stemming from small sample sizes and associated low statistical power , are a persistent problem in biomedical sciences . Calling the problem as ‘power failure’ , Button et al . ( 2013 ) emphasized that the probability of a research finding to reflect a true effect drops drastically if the statistical power is reduced from 0 . 80 ( considered as appropriate ) to low levels , for example <0 . 30 . Since the power of a statistical test drops non-linearly with the effect size , a sample size that is sufficient for detecting a 15% effect , for example a decline in N content , can be inadequate for detecting a 5% effect , for example a decline in a mineral content . Considering that the standard deviation of mineral concentrations in a plant tissue can reach 25% ( Duquesnay et al . , 2000; Lahner et al . , 2003 ) , the 5% effect size standardized as Cohen's d is d = 5/25 = 0 . 2 . A t test applied for d = 0 . 2 to a sample size of 3–5—a typical size used in eCO2 studies—yields the power of 0 . 06–0 . 10 ( Faul et al . , 2007 ) . ( Unfortunately , MetaWin ( Rosenberg et al . , 2000 ) , a statistical package routinely used in meta-analytic and other CO2 studies in ecology , provides neither a priori nor post-hoc power estimates . ) Such a small power not only raises the probability of obtaining a false negative to 90–94% but also increases the likelihood that a statistically significant result does not reflect a true effect ( Button et al . , 2013 ) . As of this writing , researchers on four continents have generated data sufficient for answering with an adequate statistical power , the following questions:Does eCO2 shift the plant ionome ? If yes , what are the direction and magnitude of shifts for individual chemical elements ? How does the effect of eCO2 on N compares to its effect on minerals ? Do FACE studies differ principally from non-FACE studies in their effect on the plant ionome ? Do the plant ionomes in temperate and subtropical/tropical regions differ in their response to eCO2 ? Do the ionomes of photosynthetic tissues and edible parts differ in their response to eCO2 ? How does eCO2 affect the ionomes of various plant groups ( woody/herbaceous , wild/crops , C3/C4 ) and grains of the world's top C3 cereals—wheat , rice , and barley ? Plotting the effect sizes ( with 95% CIs ) for the 25 minerals against their respective statistical power reveals a clear pattern ( Figure 1 ) . In the very low power ( <0 . 20 ) region , the noise completely hides the CO2-induced shift of the plant ionome . In the low power region ( <0 . 40 ) , the shift still remains obscure . As the statistical power increases , so does the likelihood that a statistically significant result reflects true effect and , consequently , the direction and the magnitude of the CO2 effect on minerals become increasingly visible in the higher power regions of the plot . 10 . 7554/eLife . 02245 . 003Figure 1 . Statistical power and the effect of CO2 on the plant ionome . The effect of elevated atmospheric CO2 concentrations ( eCO2 ) on the mean concentration of minerals in plants plotted ( with the respective 95% confidence intervals [CI] ) against the power of statistical analysis . The figure reflects data on 25 minerals in edible and foliar tissues of 125 C3 plant species and cultivars . The true CO2 effect is hidden in the very low and the low power regions . As the statistical power increases , the true effect becomes progressively clearer: the systemic shift of the plant ionome . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 00310 . 7554/eLife . 02245 . 015Figure 1—source data 1 . Supportive data for Figures 1–8 . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 015 To increase the likelihood of reporting true effects , only results with the statistical power >0 . 40 are reported in this section . However , Figure 1–source data 1 lists all the results together with their p-values irrespective of the statistical power ( e . g . , results for chromium ( Cr ) or the bean ionome are not shown here due to low power , but are listed in Figure 1–source data 1 ) . Across all the data , eCO2 reduced concentrations of P , potassium ( K ) , Ca , S , magnesium ( Mg ) , Fe , Zn , and copper ( Cu ) by 6 . 5–10% ( p<0 . 0001 ) as shown on Figure 2 . Across all the 25 minerals , the mean change was ( −8% , −9 . 1 to −6 . 9 , p<0 . 00001 ) . Only manganese ( Mn ) showed no significant change . It is not clear whether the oxygen-evolving complex ( OEC ) demands for Mn separate this mineral from the pattern of declines exhibited by other minerals . Among all the measured elements , only C increased ( 6% , 2 . 6 to 10 . 4 , p<0 . 01 ) . The sharp difference between the responses of C and minerals to eCO2 is expected if a higher carbohydrate content drives the change in the plant ionome: for most plant tissues , the dilution by carbohydrates lowers the content of minerals while having little effect on C ( Loladze , 2002 ) . ( This also suggests that the increase in C concentrations found here could be a result of a higher content of lipids or lignin—the two sizable plant compounds that are very C-rich [∼60–75% C] . ) 10 . 7554/eLife . 02245 . 004Figure 2 . The effect of CO2 on individual chemical elements in plants . Change ( % ) in the mean concentration of chemical elements in plants grown in eCO2 relative to those grown at ambient levels . Unless noted otherwise , all results in this and subsequent figures are for C3 plants . Average ambient and elevated CO2 levels across all the studies are 368 ppm and 689 ppm respectively . The results reflect the plant data ( foliar and edible tissues , FACE and non-FACE studies ) from four continents . Error bars represent the standard error of the mean ( calculated using the number of mean observations for each element ) . The number of mean and total ( with all the replicates ) observations for each element is as follows: C ( 35/169 ) , N ( 140/696 ) , P ( 152/836 ) , K ( 128/605 ) , Ca ( 139/739 ) , S ( 67/373 ) , Mg ( 123/650 ) , Fe ( 125/639 ) , Zn ( 123/702 ) , Cu ( 124/612 ) , and Mn ( 101/493 ) . An element is shown individually if the statistical power for a 5% effect size for the element is >0 . 40 . The ‘ionome’ bar reflects all the data on 25 minerals ( all the elements in the dataset except of C and N ) . All the data are available at Dryad depository and at GitHub . Copies of all the original sources for the data are available upon request . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 004 The patterns of change within edible and foliar tissues are similar: N , P , Ca , Mg , Zn , and Cu declined significantly in both tissues ( Figures 3 , 4 ) . Aside from Mn , only K showed no significant decline in the edible tissues ( on Figure 1 , it is visible as one of the only two black 95% CI in the ‘High Power’ region ) . In the foliar tissues , Mg declined the most ( −12 . 3% , −16 to −8 . 7 ) , which is congruent with the hypothesis of McGrath and Lobell ( 2013 ) that Mg should exhibit a larger decline in photosynthetic tissues because ‘chlorophyll requires a large fraction of total plant Mg , and chlorophyll concentration is reduced by growth in elevated CO2’ . However , the 95% CIs for Mg and for most other minerals overlap . A richer dataset would shed more light on the issue of Mg in photosynthetic tissues . 10 . 7554/eLife . 02245 . 005Figure 3 . The effect of CO2 on foliar tissues . Change ( % ) in the mean concentration of chemical elements in foliar tissues grown in eCO2 relative to those grown at ambient levels . Average ambient and eCO2 levels across all the foliar studies are 364 ppm and 699 ppm respectively . Error bars represent 95% CI . For each element , the number of independent mean observations , m , is shown with the respective statistical power . For each plant group , m equals the sum of mean observations over all the minerals ( not including C and N ) for that group . Elements and plant groups for which the statistical power is >0 . 40 ( for a 5% effect size ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 00510 . 7554/eLife . 02245 . 006Figure 4 . The effect of CO2 on edible tissues . Change ( % ) in the mean concentration of chemical elements in edible parts of crops grown in eCO2 relative to those grown at ambient levels . Average ambient and elevated CO2 levels across all the crop edible studies are 373 ppm and 674 ppm respectively . Other details are in the legends for Figures 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 006 As expected , among all elements N declined the most ( −15% , −17 . 8 to −13 . 1 , p<0 . 00001 ) ( Figure 2 ) , matching very closely previous findings ( Figures 3–6 ) : the 17–19% decline in leaves found by Cotrufo et al . ( 1998 ) and the 14% decline in seeds found by Jablonski et al . ( 2002 ) . Since the contents of N and protein correlate strongly in plant tissues , the lower N in edible tissues ( Figure 4 ) corroborates the protein declines in crops found by Taub et al . ( 2008 ) . 10 . 7554/eLife . 02245 . 007Figure 5 . The effect of CO2 in artificial enclosures . Change ( % ) in the mean concentration of chemical elements of plants grown in chambers , greenhouses , and other artificial enclosures under eCO2 relative to those grown at ambient levels . Average ambient and eCO2 levels across all the non-FACE studies are 365 ppm and 732 ppm respectively . Other details are in the legends for Figures 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 00710 . 7554/eLife . 02245 . 008Figure 6 . The effect of CO2 at FACE centers . Change ( % ) in the mean concentration of chemical elements of plants grown in Free-Air Carbon dioxide Enrichments ( FACE ) centers relative to those grown at ambient levels . Average ambient and eCO2 levels across all the FACE studies are 376 ppm and 560 ppm respectively . Other details are in the legends for Figures 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 008 With respect to the types of experiments , the CO2 effect on the plant ionome is surprisingly robust: in both the FACE and the non-FACE studies eCO2 significantly reduced N , P , K , Ca , S , Mg , and Zn ( Figures 5 , 6 ) . The high cost of CO2 required for running free-air experiments led to a much lower average level of eCO2 in the FACE studies ( 560 ppm ) cf . 732 ppm in the non-FACE studies . It is plausible that the lower levels of CO2 in the FACE studies contributed to a smaller overall mineral decline ( −6 . 1% , −7 . 8 to −4 . 4 ) cf . ( −8 . 7% , −10 . 1 to −7 . 4 ) for the non-FACE studies . In both the FACE and the non-FACE studies , the overall mineral concentrations declined significantly in herbaceous plants and crops , foliar and edible tissues , including wheat and rice ( Figures 5 , 6 ) . The CO2 effect on the plant ionome appears to be pervasive throughout latitudes ( Figures 7 , 8 ) . With the exception of three small centers ( in Bangladesh , Japan , and the UK ) , the mean mineral concentrations declined in every FACE and open top chamber ( OTC ) center on four continents . The mineral decline in the tropics and subtropics ( −7 . 2% , −10 . 4 to −4 . 0 , p<0 . 0001 ) is comparable to the decline in the temperate region ( −6 . 4% , −7 . 9 to −5 . 0 , p<0 . 00001 ) . A finer regional fragmentation currently is not possible due to lack of data for Africa , South America , Russia , and Canada . For many existing centers the data are limited and yield a low statistical power . 10 . 7554/eLife . 02245 . 009Figure 7 . The effect of CO2 at various locations and latitudes . Locations of the FACE and Open Top Chamber ( OTC ) centers , which report concentrations of minerals in foliar or edible tissues , are shown as white dots inside colored circles . The area of a circle is proportional to the total number of observations ( counting replicates ) generated by the center . If the mean change is negative ( decline in mineral content ) , the respective circle is blue; otherwise , it is red . The figure reflects data on 21 minerals in 57 plant species and cultivars . The shaded region ( between 35 N and S latitudes ) represents tropics and subtropics . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 00910 . 7554/eLife . 02245 . 010Figure 8 . The systemic aspect of the CO2 effect . Change ( % ) in the mean concentration of minerals in plants grown in eCO2 relative to those grown at ambient levels . All the results in the figure reflect the combined data for the foliar and the edible tissues . The number of total mean observations ( m ) for all the measured minerals across all the studies for each crop/plant group , experiment type , country , or region is shown with the respective statistical power . Country specific and regional results reflect all the FACE and Open Top Chamber ( OTC ) studies carried in any given country/region . The number of total observations ( with replicates ) for all the minerals ( not counting C and N ) for each country is as follows: Australia ( 926 ) , China ( 193 ) , Finland ( 144 ) , Germany ( 908 ) , and USA ( 1156 ) . Other details are in the legends for Figures 2 and 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 010 Germany leads the world in the FACE and OTC data generation with the largest number of mean observations of mineral concentrations ( 285 ) , followed by the USA ( 218 ) ( Figure 8 ) . Though Australia generated only 30 mean observations , it stands out in the exceptional precision of some of its studies: the wheat experiments of Fernando et al . ( 2014 ) employed an unprecedented for FACE studies 48 replicates ( for this reason , the study is easily identifiable on Figure 9 ) . 10 . 7554/eLife . 02245 . 011Figure 9 . Testing for publication bias . A funnel plot of the effect size ( the natural log of the response ratio ) plotted against the number of replicates/sample sizes ( n ) for each study and each mineral in the dataset for C3 plants . The plot provides a simple visual evaluation of the distribution of effect sizes . The blue line represents the mean effect size of eCO2 on mineral concentrations: the decline of 8 . 39% ( yielding the decline of 8 . 04% when back transferred from the log-form ) . The symmetrical funnel shape of the plot around the mean effect size indicates the publication bias in the dataset is insignificant ( Egger et al . , 1997 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 011 Since eCO2 does not stimulate carbohydrate production in C4 plants to a degree that it does in C3 plants , one would expect a milder CO2 effect on minerals for C4 plants . Indeed , no statistically significant effect was found on the ionome of C4 plants ( Figure 8 ) . Note , however , that the very limited data on this plant group are insufficient for deducing the absence of the effect; rather , it is likely that the effect size <5% for C4 plants . The CO2 effect on the C3 plant ionome shows its systemic character through the analysis of various plant groups and tissues ( Figures 3 , 4 and 8 ) . Elevated CO2 reduced the overall mineral concentrations in crops ( −7 . 2% , −8 . 6 to −5 . 6 ) ; wild ( −9 . 7% , −11 . 6 to −7 . 8 ) , herbaceous ( −7 . 5% , −8 . 7 to −5 . 6 ) , and woody ( −9 . 6% , −12 . 1 to −7 . 6 ) plants; foliar ( −9 . 2% , −10 . 8 to −7 . 6 ) and edible ( −6 . 4% , −7 . 8 to −5 . 1 ) tissues , including grains ( −7 . 2% , −8 . 6 to −5 . 6 ) . The cereal specific declines in grains are as follows: wheat ( −7 . 6% , −9 . 3 to −5 . 9 ) , rice ( −7 . 2% , −11 . 3 to −3 . 1 ) , and barley ( −6 . 9% , −10 . 5 to −3 . 2 ) ( Figure 8 ) . This is notable because wheat and rice alone provide over 40% of calories to humans . Not only does eCO2 reduce the plant mineral content , but it also alters plant stoichiometry . Specifically , the effect of eCO2 on N is nearly twice as large as its mean effect on minerals . The differential effect of eCO2 on N ( 15% ) , and P ( 9% ) and S ( 9% ) translates into a ∼7% reduction in the plant N:P and N:S . In contrast to the lower N and mineral content , eCO2 increased C content by 6% ( Figures 2 , 3 and 5 ) . It follows then that eCO2 increases C:P and C:S by 16% , and C:N by 25% confirming the previous findings of 19–27% higher C:N in plants grown in eCO2 ( Poorter et al . , 1997; Stiling and Cornelissen , 2007; Robinson et al . , 2012 ) . The current dataset ( available at Dryad depository ) suffices to show the overall shift in the plant ionome . However , it would require much richer datasets to quantify differences among the shifts of various minerals and to assess shifts in the ionomes of individual species . Unfortunately , funding hurdles for analyzing fresh and archived samples harvested at FACE centers have significantly delayed progress in this area . Only two CO2 studies report selenium ( Se ) content ( Högy et al . , 2009 , 2013 ) , and none report data on tin ( Sn ) , lithium ( Li ) , and most other trace-elements . For many of the world's popular crops , pertinent data are non-existent or very limited , including ( in the descending order of calories provided to the world's population , FAO , 2013 ) : maize ( the top C4 crop ) , soybeans ( including oil ) , cassava , millet , beans , sweet potatoes , bananas , nuts , apples , yams , plantains , peas , grapes , rye , and oats . The current data scarcity , however , should not detract our attention from what is likely to be the overarching physiological driver behind the shift in the plant ionome—the CO2-induced increase in carbohydrate production and the resulting dilution by carbohydrates . Let us take a closer look at this nutritionally important issue . Carbohydrates in plants can be divided into two types: total structural carbohydrates ( TSC; e . g . , cellulose or fiber ) that human body cannot digest , and total non-structural carbohydrates ( TNC ) , most of which—including starch and several sugars ( fructose , glucose , sucrose , and maltose ) —is readily digestible and absorbed in the human gut . Hence , for humans , TNC carries the most of caloric and metabolic load of carbohydrates . Out of the two types of carbohydrates , eCO2 affects stronger the latter , boosting TNC concentration by 10–45% ( Stiling and Cornelissen , 2007; Robinson et al . , 2012 ) . Furthermore , eCO2 tends to lower protein in plant tissues ( Taub et al . , 2008 ) . Hence , we can reason that eCO2 should exacerbate the inverse relationship found between TNC and protein ( Poorter and Villar , 1997 ) . Considering that TNC and protein are two out of the three primary macronutrients ( with fats/lipids being the third ) , it becomes imperative to quantify changes in TNC:protein , when estimating the impact of altered plant quality on human nutrition in the rising CO2 world . Regrettably , TNC:protein is rarely reported by CO2 studies; instead C:N is used as a yardstick for accessing changes in the plant quality . However , C:N poorly correlates with TNC:protein because protein is more C-rich than carbohydrates ( C content in protein is 52–55% cf . 40–45% in carbohydrates ) . Thus , a higher carbohydrate:protein results in a lower C content . This means that CO2-induced changes in nutritionally and metabolically important ratios—TNC:protein and TNC:minerals—can substantially exceed the respective changes in C:N . We can calculate changes in TNC:protein using reported changes in TNC and protein ( see ‘Formula for calculating percentage changes in TNC:protein and TNC:minerals’ in ‘Materials and methods’ ) . Table 1 compares CO2-induced changes in C:N with respective changes in TNC:protein . It shows that eCO2 can elevate TNC:protein up to fivefold higher than it does C:N . 10 . 7554/eLife . 02245 . 012Table 1 . Comparing the effects of CO2 on two plant quality indicators . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 012Study/speciesC:N ( % ) TNC:protein ( % ) ReferenceArabidopsis thaliana25125Teng et al . ( 2006 ) Bromus erectus626Roumet et al . ( 1999 ) Dactylis glomerata1753Roumet et al . ( 1999 ) wheat grain ( low N ) −1047Porteaus et al . ( 2009 ) wheat grain ( high N ) −187Porteaus et al . ( 2009 ) wheat grain96Högy et al . ( 2009 ) 27 C3 species2890Poorter et al . ( 1997 ) meta-analysis2554Robinson et al . ( 2012 ) meta-analysis2739Stiling and Cornelissen ( 2007 ) CO2-induced changes ( % ) in C:N ( a quality indicator often used in CO2 studies ) and in TNC:protein ( a rarely used but nutritionally important indicator ) for wheat grains and for foliar tissues of various plants . The results shows that in the same plant tissue , eCO2 can increase TNC:protein up to several-fold > C:N . Significant CO2-induced shifts in the ratio of major macronutrients are probable . Hence , it is important for CO2 studies to start accessing and reporting changes in TNC:protein . in lieu of protein , N content is used . How shifts in TNC:protein affect human nutrition is still unknown . New evidence , however , challenges “the notion that a calorie is a calorie from a metabolic perspective” by showing that changes in dietary carbohydrate:protein:fat ratios affect metabolism and weight gain in humans ( Ebbeling et al . , 2012 ) . The new evidence supports an emerging view that while obesity is quantified as an imbalance between energy inputs and expenditures ( Hall et al . , 2011 ) , it could also be a form of malnutrition ( Wells , 2013 ) , where increased carbohydrate:protein ( Simpson and Raubenheimer , 2005 ) and excessive carbohydrate consumption ( Taubes , 2013 ) could be possible culprits . The baseline TNC content in plant tissues varies widely . In grains and tubers , it is very high , 50–85% of dry mass ( DM ) . Therefore , in these tissues a percentage increase in TNC is arithmetically limited ( e . g . , a 60% increase is impossible ) . However , even a modest percentage increase in TNC-rich tissues can be nutritionally meaningful in absolute terms . For example , the FACE study of Porteaus et al . ( 2009 ) reports a 7–8% increase in starch concentrations in wheat grains , which translates to ∼4 g of additional starch per 100 g DM . In contrast to grains and tubers , the baseline TNC level in photosynthetic tissues is small ( usually <25% ) , which makes large TNC increases possible . For example , Teng et al . ( 2006 ) reports that eCO2 increased TNC by 76% in leaves of Arabidopsis thaliana . What is interesting here is that in absolute terms ( per 100 g DM ) the ∼5 g TNC increase in Arabidopsis thaliana is comparable to the ∼4 g TNC increase in wheat grains . More generally , CO2 studies show that—irrespective of the baseline TNC content—eCO2 tends to boost TNC by a few grams ( 1–8 g ) per 100 g DM of plant tissue ( Poorter et al . , 1997; Keutgen and Chen , 2001; Katny et al . , 2005; Erbs et al . , 2010; Azam et al . , 2013 ) . Note that such an infusion of carbohydrates into plant tissues , all else being equal , dilutes the content of other nutrients by ∼1–7 . 4% . Let us compare the dilution with its pragmatic and easily graspable analog—adding a spoonful of sugar-and-starch mixture . Table 2 shows that the CO2 effect on TNC:protein and TNC:minerals is stoichiometrically similar to the effect of adding a spoonful of carbohydrates to every 100 g DM of plant tissue . 10 . 7554/eLife . 02245 . 013Table 2 . Comparing the effect of CO2 to the effect of adding ‘a spoonful of sugars . ’DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 013Plant quality indicatorEffect of adding 5g of TNC ( % ) Effect of elevated CO2 ( % ) Grains and tubers:TNC2 . 61 to 15TNC:protein76 to 47TNC:minerals76 to 28protein−4 . 8−14 to −9minerals−4 . 8−10 to −5Foliar tissues:TNC2715 to 75TNC:protein3326 to 125TNC:minerals3324 to 98protein−4 . 8−19 to −14minerals−4 . 8−12 to −5Changes ( % ) in various plant quality indicators caused by: ( 1 ) Adding a teaspoon of TNC ( ∼5g of starch-and-sugars mixture ) per 100g of dry mass ( DM ) of plant tissue , an:d ( 2 ) growing plants in twice-ambient CO2 atmosphere . Changes due to the addition of TNC are calculated assuming:the baseline TNC content of 65% for grains and tubers , and 15% for foliar tissues . The C content is assumed to be ∼42% for plant tissues and TNC . Clearly , adding a spoonful of sugar sporadically to one's diet is not a cause for concern . However , the inescapable pervasiveness of globally rising atmospheric CO2 concentrations raises new questions: What are health consequences , if any , of diluting every 100 g DM of raw plant products with a spoonful of starch-and-sugar mixture ? What are the consequences if the dilution is not sporadic but unavoidable and lifelong ? These questions are better left for nutritionists , but it is worth noting that WHO ( 2014 ) conditionally recommends that intake of free sugars not exceed 5% of total energy , which is equivalent to 5–8 teaspoons of sugar for a typical 2000–3000 kcal/day diet . Below , I shift focus on a direct consequence of the CO2-induced increase in carbohydrate production—the mineral decline in plant tissues , and explore its potential effect on human nutrition . ‘Hidden hunger’—stems from poorly diversified plant-based diets meeting caloric but not nutritional needs . It is currently the world's most widespread nutritional disorder ( Kennedy et al . , 2003; Welch and Graham , 2005 ) . It lowers the GDP of the most afflicted countries by 2–5% and is partly responsible for their Third World status ( WHO , 2002; Stein , 2009 ) . A paradoxical aspect of ‘hidden hunger’ is that the minuscule amount of minerals , which a human body requires , could be provided easily and inexpensively—at least in theory—to all people in need by fortifying foods with minerals . However , in practice , such required mineral levels do not reach large parts of the world's community . The case of iodine is illustrative: although iodized table salt nearly wiped out iodine deficiency in the industrialized world , a billion people still have no regular access to it , making iodine deficiency the leading cause of preventable brain damage , cretinism , and lower IQ in children ( Welch and Graham , 1999; WHO , 2002 ) . Hence , the reality of logistic , economic , and cultural hurdles for fortification leaves the natural and bioavailable mineral content in food , and in plants in particular , to be the major , and sometimes the only , consistent mineral supply for a large part of mankind ( White and Broadley , 2009; Bouis and Welch , 2010 ) . This supply , unfortunately , is suboptimal for human nutrition with some of the consequences outlined below . Every third person in the world is at risk of inadequate Zn intake with its deficiency substantially contributing to stunting , compromised immunity , and child mortality ( Brown et al . , 2001; UNICEF , 2009 ) . Iron deficiency affects at least 2 billion people and is the leading cause of anemia that increases maternal mortality ( WHO , 2002; UNICEF , 2009 ) . Millions are Ca , Mg , and Se deficient ( Stein , 2009; White and Broadley , 2009 ) , including some population segments of developed countries ( Rayman , 2007; Khokhar et al . , 2012 ) . Ironically , a person can be obese and mineral undernourished—the so called ‘hunger-obesity paradox’ ( Scheier , 2005 ) , for example the many homeless in the US who rely on “cheap and energy-dense but low-nutrient” foods ( Koh et al . , 2012 ) . With every third adult in the world being overweight or obese ( Keats and Wiggins , 2014 ) , WHO ranks both mineral undernutrition and obesity among the top 20 global health risks ( WHO , 2002; Hill et al . , 2003; Stein , 2009 ) . While the role of mineral deficiency in obesity is still unclear , intriguing links have been found between the lower blood serum concentrations of Ca , Cr , Fe , Mg , Mn , Se , Zn , and increased body mass index ( BMI ) , with most of the findings appearing in the last decade ( Singh et al . , 1998; Martin et al . , 2006; Arnaud et al . , 2007; García et al . , 2009; Payahoo et al . , 2013; Yerlikaya et al . , 2013 ) . How can the CO2-induced depletion of minerals in crops affect humans ? I emphasize that the impact of CO2-induced shifts in the quality of crops on human health is far from settled . The purpose of what follows is not to make definitive claims but to stimulate research into this important but unresolved issue . A randomized controlled trial for a human diet based exclusively ( directly or indirectly ) on plants grown in eCO2 is unlikely and ethically questionable; and even if feasible , the trial might take years to generate results . In lieu of relevant data , we can employ a thought experiment . While such ‘experiments’ are usually reserved for physical sciences , any living system , notwithstanding its complexity , adheres to simple but irrefutable elemental mass balance , which can help us to elucidate plausible scenarios . For simplicity , let us focus on one question: how can a 5% reduction in the plant mineral content affect human nutrition ? Thus , we ignore other potential or likely CO2 effects: for example higher agricultural yields; altered concentrations of lipids , vitamins , and polyphenols; substantially higher TNC:protein and TNC:minerals; differential C3 and C4 plant responses; changes in the phytate content that affects mineral bioavailability ( Manoj-Kumar , 2011 ) ; and multiplicative health effects of the concomitant declines of many minerals in the same tissue . Suppose that starting tomorrow and without our knowledge , the baseline mineral content of all plants on Earth drops by 5% . A self-evident but easily overlooked mass-balance law tells us that neither thermal nor mechanical processing of raw plants enriches them with minerals ( i . e . , transmutations are impossible ) . Thus , the mineral decline in raw crops will follow into plant-based foods ( except for a few food items that are fortified with certain minerals in some countries ) . We can safely assume that the individuals , whose dietary intake of each essential mineral has exceeded the recommended dietary intake ( RDI ) by >5% , will be unaffected by the depletion . This leaves us with the majority of the human population , whose diet is either at risk of deficiency or already deficient in atleast one mineral ( WHO , 2002; Kennedy et al . , 2003; Stein , 2009 ) . Though a human body can synthesize complex compounds ( e . g . , vitamins K and D , non-essential amino acids ) , the mass balance low implies that no organism can synthesize any amount of any mineral . Therefore , to compensate for the mineral deficit , an organism has to increase mineral intake ( or , otherwise , endure the consequences of the deficit ) . Taking supplements or intentionally shifting one's diet toward mineral-rich foods , for example animal products , can eliminate the deficit . Such dietary changes , however , presuppose behavioral adjustments on the part of the individuals who are aware of their mineral deficiency and have both the means and motivation to address it . A simpler way to compensate for the mineral deficit is to increase food intake , whether consciously or not . ( The notion of compensatory feeding is not entirely alien—herbivores do increase consumption by 14–16% , when consuming plants grown in eCO2; Stiling and Cornelissen , 2007; Robinson et al . , 2012 ) . For a calorie deficient person , eating 5% more ( to be exact 5 . 26% , because 1 . 0526 . 95 ≈ 1 ) is likely to be beneficial . However , for a calorie sufficient but mineral deficient person , eating 5% more could be detrimental . The dynamic mathematical model of human metabolism , which links weight changes to dietary and behavioral changes ( Hall et al . , 2011 ) , can help to quantify the effect of a prolonged 5% increase in food intake . When parameterized with anthropometric data for an average moderately active American female ( age 38 , height 163 cm , weight 76 kg , BMI 28 . 6 , energy intake 2431 kcal/day [10171 kJ] ) ( Fryar et al . , 2012; CIA , 2013 ) , the model outputs a weight gain of 4 . 8 kg over a 3-year period , provided all other aspects of behavior and diet remain unchanged . For a male , the respective weight gain is 5 . 8 kg . The results are congruent with Hill et al . ( 2003 ) , who argued that a 4–5% difference in total daily energy intake , a mere 100 kcal/day , could be responsible for most weight gain in the population . The above ‘experiment’ suggests that a systemic and sustained 5% mineral depletion in plants can be nutritionally significant . While the rise in the atmospheric CO2 concentration is expected to be nearly uniform around the globe , its impact on crop quality might unequally affect the human population: from no detrimental effects for the well-nourished to potential weight gain for the calorie-sufficient but mineral-undernourished . The rise in CO2 levels over the last 18–30 years has already been implicated in the two effects that can influence the plant ionome: higher C assimilation and plant growth ( Donohue et al . , 2013 ) , and lower transpiration ( Keenan et al . , 2013 ) . Considering that over the last 250 years , the atmospheric CO2 concentration has increased by 120 ppm—an increase that is not far from the mean 184 ppm enrichment in the FACE studies—it is plausible that plant quality has changed . Indeed , declines in mineral concentrations have been found in wild plants and in crop fruits , vegetables , and grains over 22–250 years ( Penuelas and Matamala , 1993; Duquesnay et al . , 2000; Davis et al . , 2004; Ekholm et al . , 2007; Fan et al . , 2008; Jonard et al . , 2009 ) . While the mineral declines in crops can be an unintended consequence of the Green Revolution that produced high-yield cultivars with altered mineral content ( Davis et al . , 2004; Fan et al . , 2008 ) , the reason for the mineral declines in wild plants cannot be attributed to it . Can eCO2 directly affect human health ? Hersoug et al . ( 2012 ) proposed that rising CO2 promotes weight gains and obesity in the human population directly ( via breathing ) by reducing the pH of blood and , consequently , increasing appetite and energy intake . Weight gain has been observed in wild mammals , lab animals , and humans over the last several decades ( Klimentidis et al . , 2011 ) . However , it is not clear what role , if any , the rising CO2 could have played either directly ( breathing ) or indirectly ( altered plant quality ) . And disentangling the rising CO2 effect from other plausible factors currently does not seem feasible due to scarce data . This brings us to the broader issue of detecting—amid high local noise—signals that are small in their magnitude but global in their scope . While some scientific areas ( e . g . , genomics , bioinformatics ) have experienced a data deluge , many areas of global change , including the issue of shifting plant quality , have been hindered by chronic data scarcity . Fortunately , researchers worldwide have been steadily generating data on the effects of eCO2 on the chemical composition of plants . It is their collective efforts that have made it possible to reveal the CO2-induced shift in the plant ionome . Human activities profoundly alter the biogeochemical cycle not only of C but also of N , P , and S , which are central to all known life forms . It is plausible that other subtle global shifts in the physiology and functioning of organisms lurk amid highly noisy data . The small magnitude of such shifts makes them hard to detect and easy to dismiss . But by virtue of being global and sustained , the shifts can be biologically potent . Revealing hidden shifts requires plentiful data to attain sufficient statistical power . ( For example , Rohde et al . ( 2013 ) analyzed 14 million mean monthly local temperature records to uncover the 1 . 5°C rise in the global average temperature since 1753—undoubtedly a potent but a very small change relative to the variations of tens of degrees in local temperature . ) New data on the effects of eCO2 on plant quality ( e . g . , minerals , TNC: protein , TNC:minerals , lipids , bioavailability of nutrients ) can be generated very cost-efficiently by analyzing fresh and archived plant samples collected at FACE centers worldwide ( the project leaders of many centers are keen to share such samples; PS Curtis , BA Kimball , R Oren , PB Reich , C Stokes; IL personal communication , July , 2006 ) . With regard to minerals , the application of the high-throughput techniques of ionomics ( Salt et al . , 2008 ) can generate rich phenotypic data that can be linked with functional genomics . Such analyses will shed more light on changes in plant quality in the rising CO2 world . Anticipating and assessing such changes will help not only in mitigating their effects but also in steering efforts to breed nutritionally richer crops for the improvement of human health worldwide . I searched Google Scholar , Google , PubMed , the ISI Web of Science , AGRICOLA , and Scopus to find relevant articles with sensible combinations of two or more of the following search-words: elevated , rising , CO2 , carbon dioxide , ppm , FACE , effects , content , concentration , % , mg , dry matter , micronutrients , plant ( s ) , crop ( s ) , tree ( s ) , C3 , C4 , foliar , leaves , grains , seeds , tubers , fruits , minerals , chemical elements , and names/symbols of various chemical elements ( e . g . , zinc/Zn ) . I found additional studies from references in the articles identified in the initial searches . Among all plant tissues for which mineral concentrations are reported in the literature , the most abundant data are on foliar tissues ( leaves , needles , shoots ) , and—for herbaceous plants—on above ground parts . Hence , focusing on the foliar tissues and above ground parts allows one to maximize the number of independent observations of the effect of eCO2 on each mineral . Although the data on edible parts of crops are scarcer , a dataset on crop edible tissues was compiled due to their direct relevance for human nutrition . The following objective and uniform criteria were applied for deciding which studies to include into the dataset: ( 1 ) a study grew plants at two or more CO2 levels , ( 2 ) a study directly measured the content of one or more minerals in foliar or edible plant tissues at low ( ambient ) and high ( elevated ) CO2 levels , and ( 3 ) a study reported either absolute concentrations at each treatment or relative change/lack thereof in the concentrations for each mineral between treatments . Studies that indirectly deduced mineral concentrations , reported data on N but not on any mineral , exposed only a part ( e . g . , a branch ) of the plant , used super-elevated or uncontrolled CO2 levels were not included . Table 3 lists all the studies together with their respective species/cultivars and CO2 enrichment levels ( the dataset with all the details is deposited at Dryad and GitHub ) . When a study reported the low CO2 level as ‘ambient’ with no specific numerical values , then I used the Keeling curve to approximate the ambient CO2 level for the year the study was carried out . 10 . 7554/eLife . 02245 . 014Table 3 . Studies covered in the meta-analysis of CO2 effects on the plant ionome . DOI: http://dx . doi . org/10 . 7554/eLife . 02245 . 014SpeciesCommon nameCrop+CO2CountryReferenceAcer pseudoplatanusmaple treeNo260Overdieck , 1993Acer rubrumred maple treeNo200USAFinzi et al . , 2001Agrostis capillarisgrassNo340UKBaxter et al . , 1994Agrostis capillarisgrassNo250Newbery et al . , 1995Alnus glutinosaalder treeNo350UKTemperton et al . , 2003Alphitonia petrieirainforest treeNo440Kanowski , 2001Ambrosia dumosashrubNo180USAHousman et al . , 2012Arabidopsis thalianathale cressNo450Niu et al . , 2013Arabidopsis thalianathale cressNo330Teng et al . , 2006Betula pendula 'Roth'birch treeNo349FinlandOksanen et al . , 2005Bouteloua curtipendulagrassNo230Polley et al . , 2011Bromus tectorumcheatgrassNo150Blank et al . , 2006Bromus tectorumcheatgrassNo150Blank et al . , 2011Calluna vulgarisheather shrubNo200Woodin et al . , 1992Cercis canadensisred bud treeNo200USAFinzi et al . , 2001Chrysanthemum morifoliumchrysanthNo325Kuehny et al . , 1991Cornus floridadogwood treeNo200USAFinzi et al . , 2001Fagus sylvaticabeech treeNo260Overdieck , 1993Fagus sylvaticabeech treeNo300Rodenkirchen et al . , 2009Festuca pratensismeadow fescueNo320Overdieck , 1993Festuca viviparagrassNo340UKBaxter et al . , 1994Flindersia brayleyanarainforest treeNo440Kanowski , 2001Galactia elliottiiElliott's milkpeaNo325USAHungate et al . , 2004Larix kaempferilarch treeNo335JapanShinano et al . , 2007Lepidium latifoliumpeppergrassNo339Blank and Derner , 2004Liquidambar styracifluasweetgum treeNo200USAFinzi et al . , 2001Liquidambar styracifluasweetgum treeNo167USAJohnson et al . , 2004Liquidambar styracifluasweetgum treeNo156–200USANatali et al . , 2009Liriodendron tulipiferatulip treeNo325O’Neill et al . , 1987Lolium perennegrassNo320Overdieck , 1993Lolium perennegrassNo290GermanySchenk et al . , 1997Lupinus albuswhite lupinNo550Campbell and Sage , 2002Lycium pallidumshrubNo180USAHousman et al . , 2012Nephrolepis exaltatafernNo650Nowak et al . , 2002Pelargonium x hortorum 'Maverick White'geraniumNo330Mishra et al . , 2011Picea abies 'Karst . 'spruce treeNo350Pfirrmann et al . , 1996Picea abies 'Karst . 'spruce treeNo300Rodenkirchen et al . , 2009Picea abies 'Karst . 'spruce treeNo300Weigt et al . , 2011Picea rubensspruce treeNo350Shipley et al . , 1992Pinus ponderosapine treeNo346USAWalker et al . , 2000Pinus ponderosa 'Laws . 'pine treeNo350USAJohnson et al . , 1997Pinus sylvestrispine treeNo331Luomala et al . , 2005Pinus sylvestrispine treeNo225FinlandUtriainen et al . , 2000Pinus taedaloblolly pine treeNo200USAFinzi et al . , 2001Pinus taedapine treeNo200USANatali et al . , 2009Poa alpinagrassNo340UKBaxter et al . , 1994Poa alpinagrassNo340UKBaxter et al . , 1997Pteridium aquilinumfernNo320Zheng et al . , 2008Pteridium revolutumfernNo320Zheng et al . , 2008Pteris vittatafernNo320Zheng et al . , 2008Quercus chapmaniioak treeNo350USANatali et al . , 2009Quercus geminataoak treeNo350USAJohnson et al . , 2003Quercus geminataoak treeNo350USANatali et al . , 2009Quercus myrtifoliaoak treeNo350USAJohnson et al . , 2003Quercus myrtifoliaoak treeNo350USANatali et al . , 2009Quercus subercork oak treeNo350Niinemets et al . , 1999Schizachyrium scopariumgrassNo230Polley et al . , 2011Sorghastrum nutansgrassNo230Polley et al . , 2011Sporobolus kentrophyllusgrassNo330Wilsey et al . , 1994Trifolium alexandrinum 'Pusa Jayant'berseem cloverNo250IndiaPal et al . , 2004Trifolium pratensered cloverNo320Overdieck , 1993Trifolium repenswhite cloverNo320Overdieck , 1993Trifolium repenswhite cloverNo290GermanySchenk et al . , 1997Trifolium repenswhite cloverNo615Tian et al . , 2014Trifolium repens 'Regal'white cloverNo330Heagle et al . , 1993Vallisneria spinulosamacrophyteNo610Yan et al . , 2006Apium graveolensceleryYes670Tremblay et al . , 1988Brassica juncea 'Czern'mustardYes500IndiaSingh et al . , 2013Brassica napus 'Qinyou 8'rapeseedYes615Tian et al . , 2014Brassica napus 'Rongyou 10'rapeseedYes615Tian et al . , 2014Brassica napus 'Zhongyouza 12'rapeseedYes615Tian et al . , 2014Brassica napus 'Campino'oilseed rapeYes106GermanyHögy et al . , 2010Brassica rapa 'Grabe'turnipYes600Azam et al . , 2013Citrus aurantiumorange treeYes300USAPenuelas et al . , 1997Citrus madurensiscitrus treeYes600Keutgen and Chen , 2001Cucumis sativuscucumberYes650Peet et al . , 1986Daucus carota 'T-1-111'carrotYes600Azam et al . , 2013Fragaria x ananassastrawberryYes600Keutgen et al . , 1997Glycine max 'Merr . 'soybeanYes360USAPrior et al . , 2008Glycine max 'Merr . 'soybeanYes200Rodriguez et al . , 2011Gossypium hirsutum 'Deltapine 77'cottonYes180USAHuluka et al . , 1994Hordeum vulgarebarleyYes175GermanyErbs et al . , 2010Hordeum vulgare 'Alexis'barleyYes334GermanyManderscheid et al . , 1995Hordeum vulgare 'Arena'barleyYes334GermanyManderscheid et al . , 1995Hordeum vulgare 'Europa'barleyYes400Haase et al . , 2008Hordeum vulgare 'Iranis'barleyYes350Pérez-López et al . , 2014Hordeum vulgare 'Theresa'barleyYes170GermanyWroblewitz et al . , 2013Lactuca sativa 'BRM'lettuceYes308Baslam et al . , 2012Lactuca sativa 'Mantilla'lettuceYes350Chagvardieff et al . , 1994Lactuca sativa 'MV'lettuceYes308Baslam et al . , 2012Lactuca sativa 'Waldmann's Green'lettuceYes600McKeehen et al . , 1996Lycopersicon esculentum 'Astra'tomatoYes600Khan et al . , 2013Lycopersicon esculentum 'Eureka'tomatoYes600Khan et al . , 2013Lycopersicon esculentum 'Mill . 'tomatoYes360Li et al . , 2007Lycopersicon esculentum 'Zheza 809'tomatoYes450Jin et al . , 2009Mangifera indica 'Kensington'mango treeYes350Schaffer and Whiley , 1997Mangifera indica 'Tommy Atkins'mango treeYes350Schaffer and Whiley , 1997Medicago sativaalfalfaYes615Tian et al . , 2014Medicago sativa 'Victor'alfalfaYes100UKAl-Rawahy et al . , 2013Oryza sativariceYes200ChinaPang et al . , 2005Oryza sativa 'Akitakomachi'riceYes205–260JapanLieffering et al . , 2004Oryza sativa 'Akitakomachi'riceYes250JapanYamakawa et al . , 2004Oryza sativa 'BRRIdhan 39'riceYes210BangladeshRazzaque et al . , 2009Oryza sativa 'Gui Nnong Zhan'riceYes500Li et al . , 2010Oryza sativa 'IR 72'riceYes296PhilippinesZiska et al . , 1997Oryza sativa 'Japonica'riceYes200ChinaJia et al . , 2007Oryza sativa 'Jarrah'riceYes350Seneweera and Conroy , 1997Oryza sativa 'Khaskani'riceYes210BangladeshRazzaque et al . , 2009Oryza sativa 'Rong You 398'riceYes500Li et al . , 2010Oryza sativa 'Shakkorkhora'riceYes210BangladeshRazzaque et al . , 2009Oryza sativa 'Shan You 428'riceYes500Li et al . , 2010Oryza sativa 'Tian You 390'riceYes500Li et al . , 2010Oryza sativa 'Wu Xiang jing'riceYes200ChinaGuo et al . , 2011Oryza sativa 'Wuxiangjing 14'riceYes200ChinaMa et al . , 2007Oryza sativa 'Wuxiangjing 14'riceYes200ChinaYang et al . , 2007Oryza sativa 'Yin Jing Ruan Zhan'riceYes500Li et al . , 2010Oryza sativa 'Yue Za 889'riceYes500Li et al . , 2010Phaseolus vulgaris 'Contender'beanYes340Mjwara et al . , 1996Phaseolus vulgaris 'Seafarer'beanYes870Porter and Grodzinski , 1984Raphanus sativus 'Mino'radishYes600Azam et al . , 2013Raphanus sativus 'Cherry Belle'radishYes380Barnes and Pfirrmann , 1992Raphanus sativus 'Giant White Globe'radishYes600McKeehen et al . , 1996Rumex patientia x R . Tianschanicus 'Rumex K-1'buckwheatYes615Tian et al . , 2014Secale cereale 'Wintergrazer-70'ryeYes615Tian et al . , 2014Solanum lycopersicum '76R MYC+'tomatoYes590Cavagnaro et al . , 2007Solanum lycopersicum 'rmc'tomatoYes590Cavagnaro et al . , 2007Solanum tuberosumpotatoYes500Cao and Tibbitts , 1997Solanum tuberosum 'Bintje'potatoYes170GermanyHögy and Fangmeier , 2009Solanum tuberosum 'Bintje'potatoYes278-281SwedenPiikki et al . , 2007Solanum tuberosum 'Bintje'potatoYes305-320EuropeFangmeier et al . , 2002Solanum tuberosum 'Dark Red Norland'potatoYes345USAHeagle et al . , 2003Solanum tuberosum 'Superior'potatoYes345USAHeagle et al . , 2003Sorghum bicolorsorghumYes360USAPrior et al . , 2008Spinacia oleraceaspinachYes250IndiaJain et al . , 2007Trigonella foenum-graecumfenugreekYes250IndiaJain et al . , 2007Triticum aestivumwheatYes175GermanyErbs et al . , 2010Triticum aestivum 'Ningmai 9'wheatYes200ChinaMa et al . , 2007Triticum aestivum 'Triso'wheatYes150GermanyHögy et al . , 2009Triticum aestivum 'Triso'wheatYes150GermanyHögy et al . , 2013Triticum aestivum 'Alcazar'wheatYes350de la Puente et al . , 2000Triticum aestivum 'Batis'wheatYes170GermanyWroblewitz et al . , 2013Triticum aestivum 'Dragon'wheatYes305-320SwedenPleijel and Danielsson , 2009Triticum aestivum 'HD-2285'wheatYes250IndiaPal et al . , 2003Triticum aestivum 'Janz'wheatYes166AustraliaFernando et al . , 2014Triticum aestivum 'Jinnong 4'wheatYes615Tian et al . , 2014Triticum aestivum 'Minaret'wheatYes278GermanyFangmeier et al . , 1997Triticum aestivum 'Minaret'wheatYes300EuropeFangmeier et al . , 1999Triticum aestivum 'Rinconada'wheatYes350de la Puente et al . , 2000Triticum aestivum 'Star'wheatYes334GermanyManderscheid et al . , 1995Triticum aestivum 'Turbo'wheatYes334GermanyManderscheid et al . , 1995Triticum aestivum 'Turbo'wheatYes350Wu et al . , 2004Triticum aestivum 'Veery 10'wheatYes410Carlisle et al . , 2012Triticum aestivum 'Yangmai'wheatYes200ChinaGuo et al . , 2011Triticum aestivum 'Yitpi'wheatYes166AustraliaFernando et al . , 2012aTriticum aestivum 'Yitpi'wheatYes166AustraliaFernando et al . , 2012bTriticum aestivum 'Yitpi'wheatYes166AustraliaFernando et al . , 2012cTriticum aestivum 'Yitpi'wheatYes166AustraliaFernando et al . , 2014The table provides species name , common name , the type of experimental set up , the level of CO2 enrichment , and indicates whether the species is a crop . Countries are listed only for FACE and OTC type experiments with ‘Europe’ accounting for combined data from Belgium , Denmark , Finland , Germany , Sweden , and the UK . The following data-inclusion rules were applied to the studies with multiple co-dependent datasets for the foliar dataset: ( 1 ) the lowest and the highest CO2 levels for studies with multiple CO2 levels , ( 2 ) the control and single-factor CO2 for studies with environmental co-factors ( e . g . , observations from combined eCO2 and ozone experiments were excluded ) , ( 3 ) the highest nutrient regime when the control could not be identified in a study with multiple nutrient co-factors , ( 4 ) the last point , that is the longest exposure to ambient/eCO2 for studies with time series , ( 5 ) the most mature needles/leaves for studies reporting foliar tissues of various ages . If , in rare instances , a publication reported three or more separate datasets for the same species or cultivar , the data were averaged prior to the inclusion into the foliar dataset . For the edible tissue dataset , the study inclusion rules were the same as for the foliar dataset with the following exception: due to relative scarcity of data for edible tissues , the data with co-factors were included in the dataset ( e . g . , observations from combined eCO2 and ozone experiments were included ) . The ‘Additional info’ column in the dataset specifies exactly what datasets were extracted from each study with multiple datasets . The above publication-inclusion and data-inclusion rules allow treating each study as independent in the dataset . At no instance , potentially co-dependent observations ( e . g . , multiple observations of the same plant throughout a growing season or observations of various parts of the same plant ) were included in either the foliar or the edible dataset as separate studies . I used GraphClick v . 3 . 0 and PixelStick v . 2 . 5 to digitize data presented in a graphical form , for example bar charts . The foliar dataset covers 4733 observations of 25 chemical elements in 110 species and cultivars . The edible tissues dataset covers 3028 observations of 23 elements in 41 species and cultivars . The FACE studies cover 2264 observations of 24 elements in 25 species and cultivars . The two datasets reflect data on 125 C3 and 5 C4 species/cultivars . While the amount of statistical details provided in each study varies considerably , the following data were extractable from each study: ( 1 ) the relative change ( or lack thereof ) in the mean concentration between the low and the high CO2 treatments: ( E-A ) /A , where A and E are the mean concentrations of an element at the low and the high CO2 treatments respectively , ( 2 ) the sample size or the number of replicates ( n ) . Since a decrease in the concentration of a mineral is limited to 100% , but an increase in its concentration is theoretically unlimited , a standard technique was applied to reduce biases towards increases . Specifically , the natural log of the response ratio , that is ln ( E/A ) , was used as the effect size metric ( e . g . , Hedges et al . , 1999; Jablonski et al . , 2002; Taub et al . , 2008 ) . The response ratio , r = E/A , was calculated from the relative change as follows: r = 1+ ( E-A ) /A . After performing statistical analyses , I converted all the results back from the log form to report them as ordinary percent changes . Published meta-analytic and biostatistical results need to be replicable and reproducible , and the process of replication needs to be made as easy as possible and clearly traceable to the original sources ( Peng , 2009 ) . In this regard , I have made the following efforts to ease the replication ( from the original sources ) of each and every result presented here:While copyright restrictions do not permit posting the original published data sources online , I will share , upon request , all the data sources in PDF form , where all the pertinent data are clearly marked for easy identification , thus removing any potential ambiguity about what data were extracted from each study . The entire dataset for the foliar and the edible tissues is available at Dryad digital depository , www . datadryad . org , under 10 . 5061/dryad . 6356f . The dataset is available as an Excel file ( formatted for easy viewing ) and as a ‘CSV’ file; the latter is made-ready ( tidy ) for analysis with open-source ( R Core Team , 2014 ) and commercial statistical packages ( e . g . , SPSS ) . An executable R code to generate individual results is available with the dataset at the above-mentioned depository and at GitHub: https://github . com/loladze/co2 . Assistance for replicating any result and figure presented in this study will be provided to any interested party . I performed all the analyses using R ( R Core Team , 2014 ) , SPSS v . 21 ( IBM , Armonk , NY , USA ) and G*Power 3 ( Faul et al . , 2007 ) . Meta-analytic studies often weight effect sizes by the reciprocal of their variance , which tends to give a greater weight to studies with greater precision . However , many eCO2 studies do not report measures of variation in the data ( standard error , standard deviation , or variance ) . In lieu of the measures of variance , studies can be weighted by the number of replicates ( n ) or , alternatively , each study can be assigned equal weight , that is , unweighted method ( Jablonski et al . , 2002 ) . I used both methods ( weighted and unweighted ) to calculate the means of effect sizes with 95% CIs and compared the results of both methods . Nearly in all instances , the difference between the weighted and the unweighted means was small and lesser than the standard error of the unweighted mean . For example , across all the FACE studies , the overall mineral change was −6 . 1% ( −7 . 8 to −4 . 4 ) when unweighted cf . the −6 . 5% ( −8 . 0 to −5 . 1 ) when weighted . For the reason of close similarity between weighted and unweighted approaches , I used the simpler out of the two methods , that is the unweighted one , when reporting the results . Since the distribution of effect sizes is not necessarily normal , I applied both parametric ( t test ) and non-parametric ( bootstrapping with 10 , 000 replacements ) tests for calculating the 95% CI for the mean effect size and the statistical power . The latter was calculated for: ( 1 ) an absolute effect size of 5% , and ( 2 ) the probability of Type I error , α = 0 . 05 . If the variance of a small sample << the true population variance , then this leads to substantial overestimations of Cohen's d and the statistical power . To be conservative when estimating power for small samples ( m <20 ) , I used the larger of the sample standard deviation or 0 . 21 , which is the standard deviation for the entire mineral dataset . The results from the parametric and non-parametric tests were very close . For example , for Zn in edible tissues ( sample size = 65 ) , t test yields ( −11 . 4% , −14 . 0 to −8 . 7 ) and 0 . 91 power cf . ( −11 . 4% , −13 . 9 to −8 . 7 ) and 0 . 92 power for the bootstrapping procedure . A close similarity between the results of t test and non-parametric test is expected when sample size ( m , the number of independent observations for each mineral ) is >30 , which often was the case in this study . For reporting purposes , I used the 95% CI and the power generated by the non-parametric method , that is , the bootstrapping procedure . To test for publication bias or ‘the file drawer effect’ in the dataset , I plotted effect sizes against corresponding sample sizes/replicates , n , to provide a simple visual evaluation of the distribution of effect sizes ( Figure 9 ) . The resulting cloud of points is funnel-shaped , narrowing toward larger sample sizes , and overall is symmetrical along the mean effect size . This indicates the absence of any significant publication bias ( Egger et al . , 1997 ) . Meta-analytic CO2 studies often partition their datasets into various categories ( e . g . , plant group , plant tissue , fertilization , or water regime ) to estimate effect sizes for each category . Such data fragmentation , however , is warranted only if the statistical power of the resulting test for each category is adequate . Otherwise , low power can lead to non-significant outcomes and Type II errors . As tempting as it can be to partition the current dataset into many categories and cases ( e . g . , Zn in fruits , Fe in tuber , Cu in annuals , multiple CO2 levels ) , only by fragmenting the data into sufficiently large categories an adequate statistical power can be retained . Such categories include: foliar tissues , edible tissues , woody plants ( trees and shrubs ) , herbaceous plants , FACE studies , non-FACE studies , crops , wild plants ( all non-crops , including ornamental plants ) , C3 plants , C4 plants , rice , wheat , barley , and potato . Furthermore , I fragmented the data for C3 plants , the foliar and the edible tissues , the non-FACE and the FACE studies into individual chemical elements and into individual common plant names ( e . g . , all rice cultivars grouped under ‘rice’ ) . For the regional analysis , I used only OTC and FACE studies because they reflect local environment much more accurately than studies using complete-enclosures ( e . g . , closed chamber , glasshouse ) . If an OTC or FACE study did not report precise geographic coordinates , then the latitude and longitude of a nearby research facility or city was used ( all coordinates in the dataset are in decimal units ) . Figures 1–7 include results with the statistical power >0 . 40 for each element , country , region , plant tissue or category . Generally , power >0 . 80 is considered acceptable ( Cohen , 1988 ) . Unfortunately , such a level was achievable only for elements for which the data are most abundant and for the ionomes of some plant groups and species . Note that the power was calculated for a 5% effect size , while the true effect size is likely to be larger ( ∼8% ) ; therefore , the true power is likely to be higher than the calculated power for most results . All the results , irrespective of the statistical power , can be found in Figure 1–source data 1 . Furthermore , Figure 1 shows the mean effect sizes ( with their 95% CI ) plotted against their respective statistical powers for all the minerals and all the plant groups/tissues . If the concentration of substance X in a plant increases by x% and concomitantly the concentration of substance Y decreases by y% in the plant , then the X-to-Y ratio of the plant ( X:Y ) increases by: ( 1 ) x+y100−y·100%	Rice and wheat provide two out every five calories that humans consume . Like other plants , crop plants convert carbon dioxide ( or CO2 ) from the air into sugars and other carbohydrates . They also take up minerals and other nutrients from the soil . The increase in CO2 in the atmosphere that has happened since the Industrial Revolution is thought to have increased the production of sugars and other carbohydrates in plants by up to 46% . CO2 levels are expected to rise even further in the coming decades; and higher levels of CO2 are known to lead to lower levels of proteins in plants . But less is known about the effects of CO2 levels on the concentrations of minerals and other nutrients in plants . Loladze has investigated the effect of rising CO2 levels on the nutrient levels in food plants by analyzing data on 130 varieties of plants: his dataset includes the results of 7761 observations made over the last 30 years , by researchers around the world . Elevated CO2 levels were found to reduce the overall concentration of 25 important minerals—including calcium , potassium , zinc , and iron—in plants by 8% on average . Furthermore , Loladze found that an increased exposure to CO2 also increased the ratio of carbohydrates to minerals in these plants . This reduction in the nutritional value of plants could have profound impacts on human health: a diet that is deficient in minerals and other nutrients can cause malnutrition , even if a person consumes enough calories . This type of malnutrition is common around the world because many people eat only a limited number of staple crops , and do not eat enough foods that are rich in minerals , such as fruits , vegetables , dairy and meats . Diets that are poor in minerals ( in particular , zinc and iron ) lead to reduced growth in childhood , to a reduced ability to fight off infections , and to higher rates of maternal and child deaths . Loladze argues that these changes might contribute to the rise in obesity , as people eat increasingly starchy plant-based foods , and eat more to compensate for the lower mineral levels found in crops . Looking to the future , these findings highlight the importance of breeding food crops to be more nutritious as the world's CO2 levels continue to rise .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "epidemiology", "and", "global", "health" ]	2014	Hidden shift of the ionome of plants exposed to elevated CO2 depletes minerals at the base of human nutrition
Animals , including humans , consistently exhibit myopia in two different contexts: foraging , in which they harvest locally beyond what is predicted by optimal foraging theory , and intertemporal choice , in which they exhibit a preference for immediate vs . delayed rewards beyond what is predicted by rational ( exponential ) discounting . Despite the similarity in behavior between these two contexts , previous efforts to reconcile these observations in terms of a consistent pattern of time preferences have failed . Here , via extensive behavioral testing and quantitative modeling , we show that rats exhibit similar time preferences in both contexts: they prefer immediate vs . delayed rewards and they are sensitive to opportunity costs of delays to future decisions . Further , a quasi-hyperbolic discounting model , a form of hyperbolic discounting with separate components for short- and long-term rewards , explains individual rats’ time preferences across both contexts , providing evidence for a common mechanism for myopic behavior in foraging and intertemporal choice . Serial stay-or-search problems are ubiquitous across many domains , including employment , internet search , mate search , and animal foraging . For instance , in patch foraging problems , animals must choose between an immediately available opportunity for reward or the pursuit of potentially better but more distal opportunities . It is typically assumed that animals seek to maximize the long-term average reward ( net of cost ) rate , as a proxy for reproductive fitness . The optimal behavior for maximizing this currency in foraging tasks , described by the Marginal Value Theorem ( MVT; Charnov , 1976 ) , is to choose the immediately available opportunity if it provides a reward rate greater than the average reward rate across all alternative options , which includes the costs of accessing those options . Animals tend to follow the basic predictions of long-term reward maximization: they are generally more likely to pursue opportunities for larger vs . smaller rewards and , if the cost of searching for alternatives is greater , they are more likely to pursue opportunities for smaller rewards ( Stephens and Krebs , 1986; Constantino and Daw , 2015; Hayden et al . , 2011; Kane et al . , 2017 ) . Although animal behavior follows the basic predictions of optimal foraging behavior described by MVT , in the majority of studies across a variety of species , including humans , non-human primates , and rodents , animals exhibit a consistent bias towards pursuing immediately available rewards relative to predictions of MVT , often referred to as 'overharvesting' ( Constantino and Daw , 2015; Hayden et al . , 2011; Kane et al . , 2017; Nonacs , 2001; Kolling et al . , 2012; Shenhav et al . , 2014; Wikenheiser et al . , 2013; Carter and Redish , 2016 ) . Prior studies have proposed two explanations for overharvesting: subjective costs , such as an aversion to rejecting an immediately available reward ( Wikenheiser et al . , 2013; Carter and Redish , 2016 ) ; and nonlinear reward utility or diminishing returns , by which larger rewards are not perceived as proportionally larger than smaller rewards ( Constantino and Daw , 2015 ) . But these hypotheses have never been systematically compared in a set of experiments designed to directly test their predictions . Furthermore , according to these rate-maximizing hypotheses , the perceived value of rewards does not differ between situations in which the delays occur before or after reward is received . In this respect , the predictions made by these hypotheses ( which are still grounded in a core assumption that animals attempt to maximize the long-term reward rate ) are not compatible with an otherwise seemingly similar bias that is widely observed in standard intertemporal choice tasks ( also referred to as delay discounting or self-control tasks ) : a preference for smaller , more immediate rewards over larger , delayed rewards ( Ainslie , 1992; Kirby , 1997 ) . The preference for more immediate rewards in intertemporal choice tasks is commonly explained in one of two ways , both assuming that animals choose as though they were optimizing a different currency than long-term reward rate: temporal discounting or short-term rate maximization . According to temporal discounting , the perceived value of a future reward is discounted by the time until its receipt . Temporal discounting can arise even when maximizing the long-term reward rate , for certain environments . In particular , discounting can be adaptive in unstable environments — if the environment is likely to change before future rewards can be acquired , it is appropriate to place greater value on more predictable rewards available in the near future . Under this hypothesis , and the further assumption that expected rewards disappear at a constant rate , a long-term reward rate maximizer would discount rewards exponentially in their delay ( Gallistel and Gibbon , 2000; Kacelnik and Todd , 1992 ) . However , animal preferences typically follow a hyperbolic-like form: the rate of discounting is steeper initially and decreases over time ( Gallistel and Gibbon , 2000; Kacelnik and Todd , 1992; Thaler , 1981 ) . This yields inconsistent time preferences or preference reversals: an animal may prefer to wait longer for a larger reward if both options are distant , but will change their mind and prefer the smaller reward as the time to both options draws near ( Ainslie , 1992; Kirby , 1997; Gallistel and Gibbon , 2000; Kacelnik and Todd , 1992 ) . Recent theoretical work has shown that hyperbolic time preferences may arise from imperfect foresight — if the variance in predicting the timing of future outcomes increases with the delay to the outcome , a long-term reward rate maximizer would exhibit hyperbolic time preferences ( Gabaix and Laibson , 2017 ) . Similarly , short-term maximization rules predict that animals seek to maximize reward over shorter time horizons; this may also be motivated as an approximation to long-term reward maximization as it may be difficult to accurately predict all future rewards ( Stephens , 2002; Stephens et al . , 2004 ) . Along similar lines , Namboodiri et al . ( 2014 ) argues that , rather than maximizing long-term reward rate into the future , animals may select options that maximize reward rate up to the current point in time or due to environmental factors ( e . g . non-stationarity ) or biological constraints ( e . g . computational constraints ) , over a finite interval of time . Just as hyperbolic discounting may arise from imperfect foresight ( Gabaix and Laibson , 2017 ) , maximizing reward rate over shorter time horizons predicts hyperbolic time preferences ( Namboodiri et al . , 2014 ) . An alternative explanation for the preference for immediate rewards in intertemporal choice tasks is that animals simply underestimate the duration of post-reward delays; that is , delays between receiving reward and making the next decision ( Pearson et al . , 2010; Blanchard et al . , 2013 ) . Typically , in intertemporal choice tasks , a variable post-reward delay is added to ensure that the overall amount of time for each trial is equal , regardless of the reward size or the duration of the pre-reward delay . It has been argued that it may be difficult for animals to learn these variable delays , and thus , animals may fail to consider the full duration of the delay in their decision process . Consequently , animals will perceive that it takes less time to acquire the smaller , more immediate reward and overestimate the reward rate for choosing this option . Consistent with this hypothesis , providing an explicit cue for the duration of the post-reward delay or increasing its salience by providing a small reward at the end of the post-reward delay reduces temporal discounting ( Pearson et al . , 2010; Blanchard et al . , 2013 ) . Despite the similarities between overharvesting and the preference for more immediate rewards in intertemporal choice tasks , prior attempts to use temporal discounting and/or short-term rate maximization functions fit to intertemporal choice data to predict foraging behavior have failed ( Carter et al . , 2015; Carter and Redish , 2016; Blanchard and Hayden , 2015 ) . In these studies , animals are typically closer to long-term rate maximization in foraging tasks than in intertemporal choice tasks . This has been taken as further evidence that , while animals have a good understanding of the structure of foraging tasks , they struggle to understand the structure of intertemporal choice tasks ( i . e . they fail to incorporate post-reward delays into their decision process; Blanchard and Hayden , 2015 ) . However , there are two additional possibilities for why intertemporal choice models have failed to predict foraging behavior . First , models of intertemporal choice tasks usually consider rewards for the current trial and not rewards on future trials since , in these tasks , reward opportunities on future trials are often independent of the current decision . This is not true of foraging tasks , in which future opportunities for rewards depend on the current decision . Thus , this difference in decision horizon may make it difficult to explain foraging data using discounting models fit to intertemporal choice data . Second , these studies have only examined standard , single-parameter exponential and hyperbolic discounting functions . More flexible forms of temporal discounting that produce different patterns of hyperbolic time preferences have never been tested in these contexts . More flexible discounting functions include constant sensitivity discounting , by which rewards in the distant future are discounted less than rewards in the near future due to a bias in time estimation — agents become less sensitive to longer time delays ( Ebert and Prelec , 2007; Zauberman et al . , 2009 ) ; additive-utility discounting , by which the utility of a reward , not its value , is discounted ( Killeen , 2009 ) ; or quasi-hyperbolic discounting , which has separate terms , or different discount rates , for short- or long-term rewards rewards that correlate with activity in limbic and fronto-parietal networks respectively ( Laibson , 1997; McClure et al . , 2004; McClure et al . , 2007 ) . In the present study , we found that rats exhibit similar time preferences in foraging and intertemporal choice tasks and that time preferences in both tasks can be explained by a quasi-hyperbolic discounting model that , in both contexts , considers future rewards . Rats were tested in a series of patch foraging tasks and an intertemporal choice task . In foraging tasks , they followed the basic predictions of long-term rate maximization: they stayed longer in patches that yielded greater rewards and when the cost of searching was greater . But under certain conditions , they violated these predictions in a manner consistent with time preferences: they stayed longer in patches when given larger rewards with proportionally longer delays , and they exhibited greater sensitivity to pre- vs . post-reward delays . Similarly , in an intertemporal choice task , rats exhibited greater sensitivity to pre- vs . post-reward delays . Using these data , we tested several models to determine if temporal discounting or biases in time perception , such as insensitivity to post-reward delays , could explain rats’ behavior across tasks . One model , a quasi-hyperbolic discounting model ( Laibson , 1997; McClure et al . , 2007 ) , provided the best fit to rat behavior across all experiments . Furthermore , the quasi-hyperbolic discounting model proved to be externally valid: discounting functions fit to foraging data provided as good a fit to intertemporal choice data as discounting functions fit directly to intertemporal choice data for some rats . These findings suggest that rats exhibit similar biases in the two tasks , and quasi-hyperbolic discounting may be a common mechanism for suboptimal decision-making across tasks . Long Evans rats ( n = 8 ) were tested in a series of patch foraging tasks in operant conditioning chambers ( Kane et al . , 2017 ) . To harvest reward ( 10% sucrose water ) from a patch , rats pressed a lever on one side of the front of the chamber ( left or right ) and reward was delivered in an adjacent port . After a post-reward delay ( inter-trial interval or ITI ) , rats again chose to harvest a smaller reward or to leave the patch by nose poking in the back of the chamber . A nose poke to leave the patch caused the harvest lever to retract and initiated a delay to control the time to travel to the next patch . After the delay , the opposite lever extended ( e . g . if the left lever was extended previously , the right lever would be extended now ) , and rats could then harvest from ( or leave ) this replenished patch ( Figure 1—figure supplement 1 ) . In four separate experiments , we manipulated different variables of the foraging environment: ( i ) in the ‘Travel Time Experiment , ' a 10 s vs . 30 s delay was imposed between patches , ( ii ) in the ‘Depletion Rate Experiment , ' reward depleted at a rate of 8 vs . 16 μL per harvest , ( iii ) in the ‘Scale Experiment , ' the overall magnitude of rewards and delays was varied , such that in one condition , the size of rewards and length of delays was twice that of the other . ( iv ) Finally , in the ‘Pre-vs-Post Experiment , ' the placement of delays was varied , such that the total time to harvest reward remained constant , but in one condition there was no pre-reward delay and ∼13 s post reward delay , and in the other there was a 3 s pre-reward delay and ∼10 s post-reward delay . Parameters for each experiment are shown in Table 1 . For each condition within each experiment , rats were trained for 5 days and tested for an additional 5 days; all behavioral data presented is from the 5 test days . The order of conditions within each experiment was counterbalanced across rats . Every patch visit was included for analysis; mixed effects models were used to examine the effect of task condition on the number of trials spent in each patch . Random intercepts and random slopes for the effect of task condition were used to group observations within each rat . To compare rat behavior to the optimal behavior in each condition , a mixed effects model was used to test the effect of task condition on the difference between the number of trials spent in each patch and the optimal number of trials for that patch , with random intercepts and slopes for each rat . For this mixed effects model , an intercept of zero indicates optimal performance , and the slope indicates the change in behavior relative to the optimal behavior between conditions ( see Materials and ethods for additional detail ) . The Travel Time Experiment was designed to test the two main predictions of MVT: ( i ) that animals should stay longer in patches that yield greater rewards and ( ii ) animals should stay longer in all patches when the cost of traveling to a new patch is greater . In this experiment , rats encountered three different patch types within sessions , which started with varying amount of reward ( 60 , 90 , or 120 μL ) and depleted at the same rate ( 8 μL/harvest ) . The delay between patches was either 10 s or 30 s; each travel time delay was tested in its own block of sessions and the order was counterbalanced across rats , with a range of 87–236 patches visited per condition per rat . As predicted by MVT , rats stayed for more trials in patch types that started with larger reward volume ( β = 118 . 091 trials/mL , SE = 1 . 862 , t ( 2490 . 265 ) = 63 . 423 , p < . 001 ) , indicating that rats considered reward across future patches . Rats also stayed longer in all patch types when time between patches was longer ( β = 1 . 893 trials , SE = 0 . 313 , t ( 118 . 839 ) = 6 . 040 , p < . 001; Figure 1A ) , indicating sensitivity to opportunity costs . However , rats uniformly overharvested relative to predictions of MVT ( βrat-MVT = 3 . 396 trials , SE = 0 . 176 , t ( 6 . 960 ) = 19 . 269 , p < . 001 ) . The degree to which rats overharvested was not significantly different between the 10 s and 30 s travel conditions ( β10 s-30 s = 0 . 304 trials , SE = 0 . 155 , t ( 7 . 3857 ) = 1 . 964 , p = 0 . 088 ) . The Depletion Rate Experiment tested another critical variable in foraging environments: the rate of reward depletion within a patch . Quicker reward depletion causes the local reward rate to deplete to the long-run average reward rate quicker , thus MVT predicts earlier patch leaving . Within sessions , rats encountered a single patch type ( starting volume of 90 μL ) that depleted at a rate of either 8 or 16 μL/trial , tested in separate sessions and counterbalanced , with a range of 152–283 patches visited per condition per rat . As predicted by MVT , rats left patches earlier when patches depleted more quickly ( β = 2 . 589 trials , SE = 0 . 155 , t ( 7 . 000 ) = 16 . 75 , p < . 001; Figure 1B ) . But , again , rats stayed in patches longer than is predicted by MVT ( βrat-MVT = 2 . 005 trials , SE = 0 . 134 , t ( 7 . 004 ) = 14 . 97 , p < . 001 ) . Rats overharvested to a greater degree in the 8 μL depletion condition than the 16 μL depletion condition ( β8μL-16μL = 1 . 589 trials , SE = 0 . 155 , t ( 7 . 000 ) = 10 . 28 , p < . 001 ) . These first two experiments confirm that rats qualitatively follow the predictions of MVT , but consistently overharvest . There are many possible explanations for this pattern of overharvesting , including an aversion to leaving the offer of reward within a patch and nonlinear reward utility ( Wikenheiser et al . , 2013; Carter and Redish , 2016; Constantino and Daw , 2015 ) . The Scale Experiment was conducted in an effort to distinguish between these hypotheses by manipulating the scale of time delays and rewards . Long-term rate maximization predicts that an increase in reward size in proportion to reward delay should have no effect on the number of harvests per patch , as the reward rate across trials would be equal . But if animals’ perception of reward or time is nonlinear , a manipulation of scale will affect their subjective point of equality and predict a change in behavior across the two environments . The scale of rewards and delays was manipulated in the following manner: patches started with ( A ) 90 or ( B ) 180 μL of reward , depleted at a rate of ( A ) 8 or ( B ) 16 μL/trial , and the duration of harvest trials and travel time between patches was ( A ) 10 or ( B ) 20 s . Rats visited a range of 60–212 patches per condition . They overharvested in both A and B conditions ( βrat-MVT = 4 . 374 trials , SE = 0 . 153 , t ( 6 . 900 ) = 28 . 597 , p < . 001 ) and , contrary to predictions of MVT , they stayed in patches significantly longer and overharvested to a greater degree in the B condition that provided larger rewards but at proportionately longer delays ( β = 1 . 937 trials , SE = 0 . 193 , t ( 6 . 972 ) = 9 . 996 , p < . 001; Figure 1C ) . This finding suggests that a nonlinearity in the perception of reward value and/or time contributes to overharvesting . To distinguish between biases in perception of reward , such as nonlinear reward utility , and time , such as temporal discounting or insensitivity to post-reward delays , the Pre-vs-Post Experiment directly tested rats sensitivity to time delays before vs . after reward . In this experiment , in one condition , rats received reward immediately after lever press followed by a post-reward delay of ∼13 s before the start of the next trial . In the other condition , there was a 3 s pre-reward delay between lever press and receiving reward followed by a shorter post-reward delay of ∼10 s . The total time of each trial was held constant between conditions ( 15 s total ) , so there was no difference in reward rates . Both MVT and nonlinear reward utility predict that the placement of delays is inconsequential and that rats will behave similarly in both conditions . Both temporal discounting and insensitivity to post-reward delays predict that rats will value the immediate reward more than the delayed reward and thus , would leave patches earlier in the condition with the pre-reward delay . Consistent with predictions of temporal discounting and insensitivity to post-reward delays , and contrary to predictions of MVT and nonlinear reward utility , rats left patches earlier in the environment with the pre-reward delay ( β = 2 . 345 trials , SE = 0 . 313 , t ( 7 . 017 ) = 7 . 503 , p < . 001; Figure 1D ) . This result suggests that a bias in rats’ perception of time or the way in which they perceive delayed reward values contributes to overharvesting . To determine whether the preference for immediate rewards can be explained by insensitivity to post-reward delays , a fifth foraging experiment , the ‘Post-Reward Delay Experiment , ' directly tested rats’ sensitivity to post-reward delays . A separate cohort of rats ( n = 8 ) was used for this experiment . Rats were tested in two conditions in this experiment: a short ( 3 s ) or long ( 12 s ) post-reward delay . The total time of harvest trials was not held constant; the longer post-reward delay increased the time to harvest from the patch . Since the longer post-reward delay increases the cost of harvesting from the patch relative to the cost of traveling to a new patch , MVT predicts that rats should leave patches earlier . Prior studies of intertemporal choice behavior have shown that animals are insensitive to post-reward delays , suggesting that they are only concerned with maximizing short-term reward rate ( Stephens , 2001; Bateson and Kacelnik , 1996 ) or that they may not have learned the duration of post-reward delays , and underestimate this duration in their decision process ( Pearson et al . , 2010; Blanchard et al . , 2013 ) . Consistent with MVT , rats were sensitive to the post-reward delay , leaving patches earlier in the 12 s delay condition ( β = 1 . 411 trials , SE = 0 . 254 , t ( 6 . 966 ) = 5 . 546 , p < . 001; Figure 2A ) . If rats were sensitive to the delay , but underestimated its duration , one would still expect rats to overharvest to a greater degree due to the longer delay . There was no difference in the degree to which rats overharvested between the 3 s and 12 s delay conditions ( βrat-MVT;3 s-12 s = 0 . 340 trials , SE = 0 . 286 , t ( 6 . 963 ) = 1 . 188 , p = 0 . 274 ) . This finding suggests that overharvesting in this experiment is not due to insensitivity to post-reward delays . However , it is possible that this finding could be explained by other forms of altered time perception that remain to be described . The data from the foraging experiments described above suggest that rats exhibit time preferences in the foraging task . In a final ‘Intertemporal Choice Experiment , ' we tested whether the same rats that participated in the Post-Reward Delay Experiment would exhibit similar time preferences in a standard intertemporal choice ( i . e . a delay-discounting ) task . This task consisted of a series of 20-trial episodes . On each trial , rats pressed either the left or right lever to receive a smaller-sooner ( SS ) reward of 40 μL after a 1 s delay or a larger-later ( LL ) reward of 40 , 80 , or 120 μL after a 1 , 2 , 4 , or 6 s delay . For the first 10 trials of each episode , rats were forced to press either the left or right lever to learn the value and delay associated with that lever ( only one lever extended on each of these trials ) . For the last 10 trials of an episode , both levers extended and rats were free to choose . The LL reward value and delay , and the LL lever ( left or right ) were randomly selected at the start of each episode . Rats were tested in two different versions of this task: one in which the post-reward delay was held constant , such that the longer pre-reward delays reduced reward rate ( constant delay ) ; and another in which the time of the trial was held constant , such that longer pre-reward delays resulted in shorter post-reward delays to keep reward rate constant ( constant rate ) . MVT , which maximizes long-term reward rate , predicts that rats would be sensitive to the pre-reward delay in the constant delay condition but not the constant trial condition ( in which the pre-reward delay does not affect reward rate ) . Rats were given three training sessions to learn the structure of the intertemporal choice task after previously being tested in the foraging task , then they were tested for an additional 13 sessions in each condition , participating in a range of 590–2810 free choice trials per condition ( constant delay vs . constant rate ) . Each free choice trial within each episode was counted as a separate observation . Choice data were analyzed using a generalized linear mixed-effects model ( i . e . a mixed-effects logistic regression ) to examine the effect of the size of the LL reward , the length of the LL delay , task condition ( constant delay vs . constant rate ) , and their interactions on decisions to choose the LL vs . SS option , with random intercepts and random slopes for the effects of LL reward , LL delay , and task condition for each rat . Three post-hoc comparisons were used to test the effects of ( i ) LL reward and ( ii ) LL delay within each condition , and ( iii ) LL delay between the constant delay and constant rate conditions ( Figure 2B ) . ( i ) In both conditions , rats were more likely to choose larger LL rewards ( constant delay: β = 0 . 477 , SE = 0 . 090 , χ2 ( 1 ) =28 . 320 , p < . 001; constant rate: β = 0 . 450 , SE = 0 . 089 , χ2 ( 1 ) = 25 . 378 , p < . 001 ) , showing that they were sensitive to reward magnitude . ( ii ) They were also sensitive to the pre-reward delay in both conditions ( constant delay: β = −0 . 240 , SE = 0 . 023 , χ2 ( 1 ) = 104 . 882 , p < . 001; constant rate: β = −0 . 152 , SE = 0 . 022 , χ2 ( 1 ) = 46 . 919 , p < . 001 ) . On average , rats were equally likely to select the LL option across conditions — the main effect of task condition was not significant ( β=0 . 010 , SE = 0 . 105 , z = 0 . 092 , p = 0 . 927 ) . ( iii ) However , rats were less sensitive to increasing pre-reward delays when pre-reward delays did not affect reward rate ( in the constant rate condition ) , indicated by a change in LL delay slope between conditions ( β = 0 . 088 , SE = 0 . 026 , χ2 ( 1 ) = 11 . 376 , p < . 001 ) . Overall , rats exhibited similar time preferences in the foraging and intertemporal choice tasks: they valued rewards less with longer delays until receipt but they were sensitive to opportunity costs ( e . g . time delays between receiving reward and future decisions ) . To test whether a common set of cognitive biases could explain time preferences in both the foraging and intertemporal choice tasks , both tasks were modeled as continuous time semi-markov processes . These models consisted of a set of states that represented the time between each event in each of the tasks ( e . g . cues turning on/off , lever press , reward delivery; for state space diagrams of both tasks , see Figure 3—figure supplement 1 and Figure 4—figure supplement 1 ) . These models assumed that animals have learned the appropriate structure of the task ( i . e . the time spent and reward obtained in each state ) unless otherwise noted . The value of a given state was the discounted value of all future rewards available from that state , and the agent chose the option that yielded the greatest discounted future reward via a stochastic process . As the discount factor approached 1 ( i . e . no temporal discounting ) , this model converged to long-term reward maximization , equivalent to MVT . Additional parameters were added to the model to test four specific hypotheses for suboptimal foraging behavior: ( i ) subjective costs associated with leaving a patch , in which the value of leaving was reduced by a ‘cost' term; ( ii ) nonlinear reward utility , in which the subjective utility of a reward increased sublinearly with respect to the reward magnitude; ( iii ) biased time perception , which assumed that animals underestimate post-reward delays , possibly due to insufficient learning of task structure ( Blanchard et al . , 2013; Pearson et al . , 2010 ) , or overestimate pre-reward delays; and ( iv ) temporal discounting . A brief description of each hypothesis and its general predictions can be found in Table 2 . For each model , group level parameters and parameters for each individual rat were fit simultaneously using an expectation-maximization algorithm ( Huys et al . , 2011 ) . Parameters were fit to each experiment separately ( one set of parameters for both conditions in each experiment ) . Model predictions were calculated separately for each rat , using the rat’s individual parameters . Full details for all models , fitting procedures , and model comparison can be found in the Materials and methods . Subjective costs to leave a patch and nonlinear reward utility have explained suboptimal foraging behavior in prior studies that have manipulated opportunity costs ( e . g . travel time or pre-reward delays ) and depletion rate ( Constantino and Daw , 2015; Wikenheiser et al . , 2013; Carter and Redish , 2016 ) . However , these factors are insensitive to the placement of time delays ( pre- vs . post-reward ) and thus , cannot explain the preference for more immediate rewards . Consistent with these prior studies , the subjective costs and nonlinear reward utility models explained overharvesting in the Travel Time , Depletion Rate , and Post-Reward Delay Experiments , but they failed to explain time preferences in the Pre-vs-Post foraging experiment ( Figure 3—figure supplement 2 ) . We next examined whether biased time perception and temporal discounting could explain suboptimal foraging behavior across all tasks . Three implementations for biased time perception were tested: linear underestimation of post-reward delays ( p⁢o⁢s⁢t⁢D⁢e⁢l⁢a⁢y=αp⁢o⁢s⁢t⁢D⁢e⁢l⁢a⁢y ) , non-linear underestimation of post-reward delays ( p⁢o⁢s⁢t⁢D⁢e⁢l⁢a⁢y=p⁢o⁢s⁢t⁢D⁢e⁢l⁢a⁢yα ) , and overestimation of pre-reward delays ( p⁢r⁢e⁢D⁢e⁢l⁢a⁢y=αp⁢r⁢e⁢D⁢e⁢l⁢a⁢y ) . For temporal discounting , we tested the two common single-parameter discounting functions , exponential ( v⁢a⁢l⁢u⁢e=e-βt⁢i⁢m⁢er⁢e⁢w⁢a⁢r⁢d ) and standard hyperbolic ( v⁢a⁢l⁢u⁢e=r⁢e⁢w⁢a⁢r⁢d/ ( 1+kt⁢i⁢m⁢e ) ) , and two more flexible discounting models: constant sensitivity discounting ( Ebert and Prelec , 2007; Zauberman et al . , 2009 ) , which predicts hyperbolic time preferences due to insensitivity to longer delays ( v⁢a⁢l⁢u⁢e=e-βt⁢i⁢m⁢eαr⁢e⁢w⁢a⁢r⁢d ) ; and quasi-hyperbolic discounting , formalized as two competing exponential discounting systems ( v⁢a⁢l⁢u⁢e=[ωe-βt⁢i⁢m⁢e+ ( 1-ω ) e-δt⁢i⁢m⁢e]r⁢e⁢w⁢a⁢r⁢d; ( Laibson , 1997; McClure et al . , 2007 ) . All of these models qualitatively predicted rat behavior across foraging experiments ( Figure 3 , Figure 3—figure supplement 3 , Figure 3—figure supplement 4 ) . To determine which model provided the best quantitative fit , we compared the group-level Bayesian Information Criterion ( integrated BIC or iBIC; Huys et al . , 2011; Huys et al . , 2012 ) of all models in each of the foraging tasks . To compare across tasks , we took the sum of iBIC for each model . The quasi-hyperbolic discounting model had the lowest sum of iBIC , the constant sensitivity discounting model the second lowest , and the hyperbolic discounting model third ( Figure 3 ) . These three models were also among the lowest iBIC values for each individual experiment Figure 3—figure supplement 5 ) . All three of these models predict that animals will exhibit hyperbolic time preferences , suggesting that suboptimal foraging behavior observed in these experiments is due to time preferences . Next , we tested whether the quasi-hyperbolic discounting model that provided the best fit to foraging behavior could also explain behavior in the intertemporal choice task . As in the model of the foraging task , the model of the intertemporal choice task took into account all future rewards , including rewards from future episodes ( see abbreviated state space diagram in Figure 4—figure supplement 1 ) . We tested the nonlinear reward utility , biased time perception and temporal discounting models in this task ( the subjective cost model does not apply to this task ) . Again , the quasi-hyperbolic discounting model had the lowest iBIC and hyperbolic discounting model the second lowest , but the constant sensitivity model had a higher iBIC than the biased time perception models ( Figure 4 ) . As the constant sensitivity model produces hyperbolic time preferences via insensitivity to longer delays , these results suggest that hyperbolic time preferences without insensitivity to delays is the best explanation for rat intertemporal choice behavior . If hyperbolic time preferences reflect a common explanation for suboptimal decision-making , then it might be expected that a model of behavior fit to one task could predict a rat’s behavior in the other task . To test the external validity of this hypothesis , data from each task were separated into three subsets . The best fitting model from both tasks , the quasi-hyperbolic discounting model , was fit to two subsets of data from one task , then the negative log likelihood ( -LL ) of the data was assessed on the left out sample from both tasks . This process was repeated such that each subset served as the left out sample . To determine which discount function provided the better fit to data from each task , we calculated the difference in -LL of the left out sample between the model fit to intertemporal choice data and the model fit to foraging data ( -LL difference = -LLitc - LLforage ) . Since smaller -LL indicates a better fit , a positive -LL difference indicates that the discount function fit to foraging data provided a better fit ( i . e . the foraging -LL was lower than the intertemporal choice -LL ) . For the foraging task , discounting functions fit to foraging data provided a better fit than discounting functions fit to intertemporal choice data for all eight rats . Interestingly , for the intertemporal choice task , discounting functions fit to foraging data provided a better fit than discounting functions fit to intertemporal choice data for 3 of 8 rats ( Figure 5 ) . The quasi-hyperbolic model fit to the foraging task generalized well to the intertemporal choice task , providing support for the idea that foraging and intertemporal choice can be described by a common discount function . With temporal discounting models that consider all future rewards , the more flexible quasi-hyperbolic discounting function provided the best fit to behavior across tasks . We next directly tested whether considering future rewards affects the fit of discounting models to intertemporal choice data and the estimates of discount factors compared to temporal discounting models that only consider the next reward ( one-trial horizon models ) . We fit one-trial horizon models for all of the previously tested discounting functions — exponential , hyperbolic , constant sensitivity , and quasi-hyperbolic discounting — and compared them to the discounting models that considered all future rewards ( all-future horizon models ) . For all discounting functions , the all-future horizon models had lower iBIC than one-trial horizon models ( Figure 4—figure supplement 2 ) . To compare the discount factors of each model ( for the quasi-hyperbolic function that has two discount factors , we used the slow discounting β ) , we performed paired t-tests between log transformed discount factors measured by the all-future horizon models vs . discount factors measured by the one-trial horizon models . Measured discount factors were lower for the all-future models for all discounting functions ( exponential: t ( 7 ) = 22 . 439 , p < . 001; hyperbolic: t ( 7 ) = 7 . 000 , p < . 001; constant sensitivity: t ( 7 ) = 15 . 497 , p < . 001; quasi-hyperbolic: t ( 7 ) = 25 . 322 , p < . 001; p-values adjusted using Bonferroni correction ) . Lastly , we tested whether the all-future horizon quasi-hyperbolic discounting model fit to the intertemporal choice data would predict foraging behavior better than the one-trial horizon quasi-hyperbolic discounting model fit to intertemporal choice data . For 6 of 8 rats , parameters fit to the one-trial horizon model produced a better fit to foraging behavior than parameters fit to the all-future horizon model . Overall , using full horizon temporal discounting models explained more of the intertemporal choice data , produced smaller estimates of discounting factors , but in the present study , it did not improve the ability of a model fit to intertemporal choice data to predict foraging behavior . In foraging studies , animals exhibit behavior that conforms qualitatively to predictions made by optimal foraging theory ( i . e . , the MVT ) , choosing to leave a patch when its value falls below that of the average expected value of other ( s ) available in the environment . However , an almost ubiquitous finding is that they overharvest , leaving a patch when its value falls to a value lower than the one predicted by MVT . Given that the rewards available within the current patch are generally available sooner than those at other patches due to travel time , one interpretation of overharvesting is that this reflects a similarly prevalent bias observed in intertemporal choice tasks , in which animals consistently show a greater preference for smaller more immediate rewards over later delayed rewards than would be predicted by optimal ( i . e , . exponential ) discounting of future values . However , in prior studies , models of intertemporal choice behavior have been poor predictors of foraging behavior ( Blanchard and Hayden , 2015; Carter and Redish , 2016 ) . Here , we show that in a carefully designed series of experiments , rats exhibit similar time preferences in foraging and intertemporal choice tasks , and that a quasi-hyperbolic discounting model can explain the rich pattern of behaviors observed in both tasks . The foraging behavior we observed was consistent with previous studies of foraging behavior in rats , monkeys , and humans , while also revealing novel aspects of overharvesting behavior . Consistent with prior studies , rats stayed longer in patches that yielded greater rewards , stayed longer in all patch types when the cost of traveling to a new patch was greater , left patches earlier when rewards depleted more quickly , and consistently overharvested ( Constantino and Daw , 2015; Hayden et al . , 2011; Kane et al . , 2017 ) . Our experiments also demonstrated that in certain environments rats violate qualitative predictions of MVT . Rats overharvested more when reward amount and delay were increased , even though reward rate was held constant , and they were differentially sensitive to whether the delay was before the receipt of the proximal reward or following its delivery . These findings supported the conjecture that overharvesting is related to time preferences . A number of studies have found that the preference for smaller , more immediate rewards can be explained by insensitivity to post-reward delays ( Bateson and Kacelnik , 1996; Blanchard et al . , 2013; Pearson et al . , 2010; Stephens , 2001; Mazur , 1991 ) . One hypothesis for why animals fail to incorporate post-reward delays into decisions is that they haven’t learned the structure of the task well , and thus cannot accurately predict future post-reward delays . Accordingly , providing explicit cues for the post-reward delays or increasing the salience of post-reward delays helps animals incorporate these delays into their decisions , reducing the bias towards selecting smaller , more immediate rewards over larger , delayed ones ( Pearson et al . , 2010; Blanchard et al . , 2013 ) . However , in the present study , rats were sensitive to post-reward delays in both the foraging and intertemporal choice task , providing further evidence that the preference for smaller , more immediate rewards in both tasks is due to time preferences and not a poor understanding of the task structure . Furthermore , quantitative modeling supported the hypothesis that suboptimal behavior was driven by time preferences rather than insensitivity to delays . The idea that animals exhibit similar decision biases in foraging and intertemporal choice paradigms , and that these biases can be explained by a common model of discounting , is in conflict with prior studies that found that animals are better at maximizing long-term reward rate in foraging than in intertemporal choice tasks , and that delay discounting models of intertemporal choice tasks are poor predictors of foraging behavior ( Stephens , 2008; Blanchard and Hayden , 2015; Carter et al . , 2015; Carter and Redish , 2016 ) . It has been argued that animals may perform better in foraging tasks because decision-making systems have evolved to solve foraging problems rather than two-alternative intertemporal choice problems ( Blanchard and Hayden , 2015; Stephens et al . , 2004; Stephens , 2008 ) . This idea has been challenged by a recent study of human decision-making in foraging and intertemporal choice tasks , finding that a long-term rate maximization model explained both foraging and intertemporal choice behavior better than a standard hyperbolic discounting model ( Seinstra et al . , 2018 ) . Results from the present study support the interpretation that foraging and intertemporal choice behavior can be explained via a common model , but suggest that this model is quasi-hyperbolic discounting . We found that a quasi-hyperbolic discounting model provided the best explanation to rat behavior across multiple foraging tasks and an intertemporal choice task , and that a quasi-hyperbolic discounting model fit to individual rat foraging behavior can predict their intertemporal choice behavior . Two potential explanations for why temporal discounting models have failed to predict foraging behavior in prior studies are that ( i ) prior studies have only tested single-parameter exponential and hyperbolic discounting functions , whereas the present study also tested the more flexible quasi-hyperbolic discounting function; and ( ii ) in most of these studies , models of intertemporal choice tasks have only considered the most proximal reward ( the reward received as a consequence of the decision at hand ) . This assumption seems appropriate as , in most intertemporal choice tasks , opportunities for future rewards do not depend on the current decision , so the value of rewards received for future decisions are equal for both the SS and LL rewards . But in foraging tasks , future opportunities for reward depend on current decisions , so it is critical for foraging models to include all future rewards into estimates of reward value . For this reason , comparing discount functions fit to intertemporal choice models that consider all future reward may provide better estimates of foraging behavior than discount functions fit to intertemporal choice models that only consider rewards from the most proximal decision . Consistent with this hypothesis , we found that adding the value of future rewards to intertemporal choice models reduces estimates of discount factors . However , with our data , the quasi-hyperbolic discounting model fit to the intertemporal choice task that included all future rewards did not predict foraging behavior better than an equivalent model that only considered the most proximal reward . One reason why including all future rewards may not have improved cross-task predictions is that , in the present study , the quasi-hyperbolic discounting model fit to the intertemporal choice task predicted less overharvesting than was exhibited by rats in the foraging task . Reducing estimates of the discount factor with a model that considers all future rewards predicts even less overharvesting ( i . e . behavior that is closer to long-term reward maximization ) . But in other studies , temporal discounting models typically predict greater overharvesting than is exhibited by animals ( Blanchard and Hayden , 2015; Carter et al . , 2015 ) . In these cases , obtaining smaller , potentially more accurate estimates of discount factors by including all future rewards into intertemporal choice models may improve cross-task predictions . Although quasi-hyperbolic discounting provided the best singular explanation for rat behavior across our tasks , many of the other models tested were capable of explaining some of the biases exhibited by rats . Thus , we cannot exclude the possibility that subjective costs , diminishing marginal utility , and/or biased estimation of time intervals may independently contribute to suboptimal decision-making . Furthermore , additional hypotheses or additional variants of the above-mentioned hypotheses that have not been tested in the present study may provide alternative explanations for suboptimal decision making in foraging and intertemporal choice tasks . Importantly , our data indicate that quasi-hyperbolic discounting may provide a link between foraging and intertemporal choice tasks , and it highlights the importance of future work considering the source of time preferences . These observations are buttressed by recent theoretical work demonstrating that the appearance of time preferences in intertemporal choice tasks can emerge rationally from a value construction process by which estimates increase in variability with the delay until reward receipt — an account that shares features with the short-term rate maximization hypotheses ( Stephens et al . , 2004 ) . Under this account , ‘as-if' discounting is hyperbolic when variability increases linearly with delay ( Gabaix and Laibson , 2017 ) . Further , a sequential sampling model of two-alternative forced choice ( Bogacz et al . , 2006 ) , parameterized such that outcome delay scales variability in this way , has recently been shown to capture key dynamical features of both patch foraging ( Davidson and El Hady , 2019 ) and hyperbolic discounting in intertemporal choice ( Hunter et al . , 2018 ) . Future work should build on these findings to explore directly whether the common biases identified here reflect a core computation underlying sampling and decision-making under uncertainty and across time . Adult Long-Evans rats were used ( Charles River , Kingston , NY ) . One group of eight rats participated in the scale , travel time , depletion rate , and handling time experiments ( in that order ) , a different set of eight rats were tested on the post-reward delay foraging experiment then the delay discounting task . Rats were housed on a reverse 12 hr/12 hr light/dark cycle . All behavioral testing was conducted during the dark period . Rats were food restricted to maintain a weight of 85–90% ad-lib feeding weight , and were given ad-lib access to water . All procedures were approved by the Princeton University and Rutgers University Institutional Animal Care and Use Committee . Animals were trained and tested as in Kane et al . ( 2017 ) . Rats were first trained to lever press for 10% sucrose water on an FR1 reinforcement schedule . Once exhibiting 100+ lever presses in a one hour session , rats were trained on a sudden patch depletion paradigm — the lever stopped yielding reward after 4–12 lever presses — and rats learned to nose poke to reset the lever . Next rats were tested on the full foraging task . A diagram of the foraging task is in Figure 1—figure supplement 1 . On a series of trials , rats had to repeatedly decide to lever press to harvest reward from the patch or to nose poke to travel to a new , full patch , incurring the cost of a time delay . At the start of each trial , a cue light above the lever and inside the nose poke turned on , indicating rats could now make a decision . The time from cues turning on until rats pressed a lever or nose poked was recorded as the decision time ( DT ) . A decision to harvest from the patch ( lever press ) yielded reward after a short pre-reward delay ( referred to as the handling time delay , simulating the time to ‘handle' prey after deciding to harvest ) . Reward ( sucrose water ) was delivered when the rat entered the reward magazine . The next trial began after an inter-trial interval ( ITI ) . To control the reward rate within the patch , the length of the ITI was adjusted based on the DT of the current trial , such that the length of all harvest trials was equivalent . With each consecutive harvest , the rat received a smaller volume of reward to simulate depletion from the patch . A nose poke to leave the patch caused the lever to retract for a delay period simulating the time to travel to a new patch . After the delay , the opposite lever extended , and rats could harvest from a new , replenished patch . Details of the foraging environment for each experiment can be found in Table 1 . For each experiment , rats were trained on a specific condition for 5 days , then tested for 5 days . Conditions within experiments were counterbalanced . Rat foraging behavior was assessed using linear mixed effects models . Models were fit using the lme4 package in R ( Bates et al . , 2015 ) . The lme4 package provides only t-statistics for fixed effects; p-values were calculated using the lmerTest package ( Kuznetsova et al . , 2017 ) , which uses Scatterwaithe’s method to approximate the degrees of freedom for the t-test . In the Travel Time Experiment , we assessed the effect of starting volume of the patch and the travel time on number of harvests per patch , with random intercepts and random slopes for both variables across subjects ( lme4 formula: HarvestsPerPatch∼PatchStartingVolumeTravelTime+ ( PatchStartingVolume+TravelTime\|\|Rat ) ) . In all other foraging experiments , we assessed the effect of experimental condition on harvests per patch , with random intercepts and random effect of experimental condition across subjects ( lme4 formula: H⁢a⁢r⁢v⁢e⁢s⁢t⁢s⁢P⁢e⁢r⁢P⁢a⁢t⁢c⁢h∼C⁢o⁢n⁢d⁢i⁢t⁢i⁢o⁢n+ ( C⁢o⁢n⁢d⁢i⁢t⁢i⁢o⁢n\|R⁢a⁢t ) ) . We also tested whether rats overharvested relative to MVT predictions in each experiment , and whether the degree of overharvesting was different between conditions within each experiment . To do so , we subtracted the MVT predicted number of harvests in each patch from the observed number of harvests ( see ‘Foraging Models' section for details on the calculation of the optimal number of harvests ) . Mixed effects models were used to fit an intercept and effect of experimental condition on the difference from optimal number of harvests ( lme4 formula: D⁢i⁢f⁢f⁢e⁢r⁢e⁢n⁢c⁢e⁢F⁢r⁢o⁢m⁢O⁢p⁢t⁢i⁢m⁢a⁢l∼C⁢o⁢n⁢d⁢i⁢t⁢i⁢o⁢n+ ( C⁢o⁢n⁢d⁢i⁢t⁢i⁢o⁢n\|R⁢a⁢t ) ) . In this model , an intercept greater than zeros would indicate that rats harvested more trials than was predicted by MVT , and a difference in the effect of task condition would indicate that the degree to which rats differed from optimal was affected by the task condition . Rats were immediately transferred from the foraging task to the intertemporal choice task with no special training; rats were given three 2 hr sessions to learn the structure of the new task . This task consisted of a series of episodes that lasted 20 trials . At the beginning of each episode one lever was randomly selected as the shorter-sooner lever , yielding 40 μL of reward following a 1 s delay . The other lever ( larger-later lever ) was initialized to yield a reward of 40 , 80 , or 120 μL after a 1 , 2 , 4 or 6 s delay . For the first 10 trials of each episode , only one lever extended , and rats were forced to press that lever to learn its associated reward value and delay . The last four forced trials ( trials 7–10 ) were counterbalanced to reduce the possibility of rats developing a perseveration bias . For the remaining 10 trials of each episode , both levers extended , and rats were free to choose the option they prefer . At the beginning of each trial , cue lights turned on above the lever indicating rats could now make a decision . Once the rat pressed the lever , the cue light turned off , and the delay period was initiated . A cue light turned on in the reward magazine at the end of the delay period , and rats received reward as soon as they entered the reward magazine . Reward magnitude was cued by light and tone . Following reward delivery , there was an ITI before the start of the next trial . At the completion of the episode , the levers retracted , and rats had to nose poke to begin the next episode , which reset the larger-later reward and delay . Intertemporal choice data was analyzed using a mixed effects logistic regression , examining the the effect of larger-later reward value , larger-later delay , and task condition on rats choices , with random intercepts and random effects for all three variables . This model was fit as a generalized linear mixed effects model using the lme4 package in R ( lme4 formula: ProbLL∼RewardLLDelayLLCondition+ ( RewardLL+DelayLL+Condition\|\|Rat ) ; Bates et al . , 2015 ) . Post-hoc comparisons of interest were tested using the phia package in R ( De Rosario-Martinez , 2015 ) , using Holm’s method to correct for multiple comparisons . All models were constructed as continuous time semi-markov processes . This provided a convenient way to capture the dynamics of timing in both tasks , such as slow delivery and consumption of reward ( up to 6 s for the largest rewards ) . To model the foraging task , each event within the task ( e . g . cues turning on/off , lever press , reward delivery , etc . ) marked a state transition ( abbreviated state space diagram in Figure 3—figure supplement 1 . All state transitions were deterministic , except for decisions to stay in vs . leave the patch , which occurred in ‘decision‘ states ( the time between cues turning on at the start of the trial and rats performing a lever press or nosepoke ) . In decision states , a decision to stay in the patch transitioned to the handling time state , then reward state , ITI state , and to the decision state on the next trial . A decision to leave transitioned to the travel time state , then to the first decision state in the patch . Using the notation of Bradtke and Duff , 1995 , the value of staying in state s , Q⁢ ( s⁢t⁢a⁢y , s ) , is the reward provided for staying in state s , R⁢ ( s⁢t⁢a⁢y , s ) , plus the discounted value of the next state:Q⁢ ( s⁢t⁢a⁢y , s ) =R⁢ ( s⁢t⁢a⁢y , s ) +γ⁢ ( s⁢t⁢a⁢y , s ) V⁢ ( sn⁢e⁢x⁢t ) where γ⁢ ( s⁢t⁢a⁢y , s ) is the discount applied to the value of the next state for staying in state s , and V⁢ ( sn⁢e⁢x⁢t ) is the value of the next state in the patch . For all non-decision states , rats did not have the option to leave the patch , so for these states , V⁢ ( s ) =Q⁢ ( s⁢t⁢a⁢y , s ) . For decision states , the value of the state was the greater of Q⁢ ( s⁢t⁢a⁢y , s ) and Q⁢ ( l⁢e⁢a⁢v⁢e ) . For simplicity , we assume the time spent in a given state is constant , calculated as the average amount of time a given rat spent in the state . Under this assumption , the reward in a given state , R⁢ ( s⁢t⁢a⁢y , s ) , is equal to the reward rate provided over the course of the state , r⁢ ( s ) , multiplied by the time spent in that state T⁢ ( s ) , discounted according to discount factor β:R⁢ ( s⁢t⁢a⁢y , s ) =1-e-βT⁢ ( s ) βr⁢ ( s ) , andγ⁢ ( s⁢t⁢a⁢y , s ) =e-βT⁢ ( s ) . The value of leaving a patch , Q⁢ ( l⁢e⁢a⁢v⁢e ) , was equal to the discounted value of the first state in the next patch , V⁢ ( sf⁢i⁢r⁢s⁢t ) :Q⁢ ( l⁢e⁢a⁢v⁢e ) =γ⁢ ( l⁢e⁢a⁢v⁢e ) V⁢ ( sf⁢i⁢r⁢s⁢t ) where γ⁢ ( l⁢e⁢a⁢v⁢e ) is the discount factor applied to the next state in the first patch . Assuming no variance in the travel time τ , γ⁢ ( l⁢e⁢a⁢v⁢e ) =e-βτ . Per MVT , we assumed rats left patches at the first state in the patch in which Q ( stay , s ) ≤Q ( leave ) . To model variability in the trial at which rats left patches , we added gaussian noise to Q⁢ ( l⁢e⁢a⁢v⁢e ) . As decisions within each patch are not independent , the patch leaving threshold did not vary trial-by-trial , but rather patch by patch , such that the cumulative probability that a rat has left the patch by state s , π⁢ ( l⁢e⁢a⁢v⁢e , s ) , was the probability that Q⁢ ( s⁢t⁢a⁢y , s ) ≤Q⁢ ( l⁢e⁢a⁢v⁢e ) +Q⁢ ( l⁢e⁢a⁢v⁢e ) ϵ , where ϵ∼𝒩⁢ ( 0 , σ2 ) , with free parameter σ . ϵ scaled with Q⁢ ( l⁢e⁢a⁢v⁢e ) to enable comparisons across conditions within experiments . The optimal policy for a given set of parameters was found using value iteration ( Sutton and Barto , 1998 ) . MVT predictions ( maximization of undiscounted long-term reward rate ) were determined by fixing the discount factor β= . 001 and assuming no decision noise ( ϵ=0 ) . MVT predictions were determined for each rat; the time spent in each state was taken from a given rat’s data . For each model , we fit both group level parameters and individual parameters for each rat using an expectation-maximization algorithm ( Huys et al . , 2011 ) . To model subjective costs , a free parameter c representing an aversion to leaving the patch was subtracted from the leaving threshold ( Wikenheiser et al . , 2013; Carter and Redish , 2016 ) :Qc⁢o⁢s⁢t⁢ ( l⁢e⁢a⁢v⁢e ) =-c+γ⁢ ( l⁢e⁢a⁢v⁢e ) Vc⁢o⁢s⁢t⁢ ( sf⁢i⁢r⁢s⁢t ) . To investigate whether nonlinear reward utility could explain rats' overharvesting behavior , we tested models in which the utility of a reward received in the task increased in a sublinear fashion with respect to the magnitude of the reward . Two different utility functions were tested: a power law function and a steeper constant relative risk aversion ( CRRA ) utility function that became increasingly risk averse with larger rewards , both with free parameter η:Qu⁢t⁢i⁢l⁢i⁢t⁢y⁢ ( s⁢t⁢a⁢y , s ) =U⁢ ( s⁢t⁢a⁢y , s ) +γ⁢ ( s⁢t⁢a⁢y , s ) Vu⁢t⁢i⁢l⁢i⁢t⁢y⁢ ( sn⁢e⁢x⁢t ) Up⁢o⁢w⁢e⁢r⁢ ( s⁢t⁢a⁢y , s ) =R⁢ ( s⁢t⁢a⁢y , s ) η , orUC⁢R⁢R⁢A⁢ ( s⁢t⁢a⁢y , s ) =R⁢ ( s⁢t⁢a⁢y , s ) 1-η-11-η . To examine linear and nonlinear underestimation of post-reward delays , respectively , the time spent in post-reward delay ( ITI ) states was transformed , with free parameter α:Tp⁢o⁢s⁢t-l⁢i⁢n⁢e⁢a⁢r⁢ ( sI⁢T⁢I ) =α⁢T⁢ ( sI⁢T⁢I ) , where ⁢0<α<1 , orTp⁢o⁢s⁢t-p⁢o⁢w⁢e⁢r⁢ ( sI⁢T⁢I ) =T⁢ ( sI⁢T⁢I ) α Similarly , for overestimation of pre-reward delays , the handling time and travel time were transformed:Tp⁢r⁢e-d⁢e⁢l⁢a⁢y⁢ ( sH⁢T ) =α⁢T⁢ ( sH⁢T ) , andτp⁢r⁢e-d⁢e⁢l⁢a⁢y=α⁢τ , where ⁢α>1 . For the exponential discounting model , β was fit as a free parameter . As standard hyperbolic discounting cannot conveniently be expressed recursively , this model was implemented using the μAgents model described by Kurth-Nelson and Redish ( 2009 ) . The value functions of the overall model , Qμ⁢A⁢g⁢e⁢n⁢t⁢ ( s⁢t⁢a⁢y , s ) and Qμ⁢A⁢g⁢e⁢n⁢t⁢ ( l⁢e⁢a⁢v⁢e ) , were the average of the μAgents , each with their own exponential discount factor βi , and thus individual reward functions Ri⁢ ( s⁢t⁢a⁢y , s ) , discount functions γi⁢ ( s⁢t⁢a⁢y , s ) and γi⁢ ( l⁢e⁢a⁢v⁢e ) , and value functions Qi⁢ ( s⁢t⁢a⁢y , s ) , Qi⁢ ( l⁢e⁢a⁢v⁢e ) , and Vi⁢ ( s ) :Qi⁢ ( s⁢t⁢a⁢y , s ) =Ri⁢ ( s⁢t⁢a⁢y , s ) +γi⁢ ( s⁢t⁢a⁢y , s ) Vi⁢ ( sn⁢e⁢x⁢t ) Qμ⁢A⁢g⁢e⁢n⁢t⁢ ( s⁢t⁢a⁢y , s ) =110⁢∑iRi⁢ ( s⁢t⁢a⁢y , s ) +γi⁢ ( s⁢t⁢a⁢y , s ) Vi⁢ ( sn⁢e⁢x⁢t ) Qi⁢ ( l⁢e⁢a⁢v⁢e ) =γi⁢ ( l⁢e⁢a⁢v⁢e ) Vi⁢ ( sf⁢i⁢r⁢s⁢t ) Qμ⁢A⁢g⁢e⁢n⁢t⁢ ( l⁢e⁢a⁢v⁢e ) =110⁢∑iγi⁢ ( l⁢e⁢a⁢v⁢e ) Vi⁢ ( sf⁢i⁢r⁢s⁢t ) If the μAgent discount factors , βi , are drawn from an exponential distribution with rate parameter λ>0 , the discounting function of the overall model approximated the standard hyperbolic discount function , r⁢e⁢w⁢a⁢r⁢d/ ( 1+kd⁢e⁢l⁢a⁢y ) , with discount rate k=1/λ . This model was implemented using 10 μAgents with βi equal to the 5% , 15% , … , 95% quantile of the exponential distribution . The relationship of this implementation of the μAgent model to the standard hyperbolic discount function is presented in Figure 6 . k was fit as a free parameter . The constant sensitivity discounting model was based on Ebert and Prelec ( 2007 ) . In this model , hyperbolic time preferences are produced via exponential discounting with insensitivity to longer delays . To implement this model , insensitivity to all time delays — the decision time , pre-reward delay , reward time , and post-reward delay , and travel time — was achieved using a power function , just as in the nonlinear post-reward delay model . This model was then equivalent to the exponential discounting model , replacing the time in each state T⁢ ( s ) with a power function of the time in each state T⁢ ( s ) α . Quasi-hyperbolic discounting was originally formulated for discrete time applications ( Laibson , 1997 ) . We used the continuous time formulation from McClure et al . ( 2007 ) , in which the value functions of the overall model were the weighted sum of two exponential discount systems , a steep discounting β system that prefers immediate rewards and a slower discounting δ system , each with their own reward functions , Rβ⁢ ( s⁢t⁢a⁢y , s ) and Rδ⁢ ( s⁢t⁢a⁢y , s ) , and discount functions γβ⁢ ( s⁢t⁢a⁢y , s ) , γβ⁢ ( l⁢e⁢a⁢v⁢e ) , γδ⁢ ( s⁢t⁢a⁢y , s ) , and γδ⁢ ( l⁢e⁢a⁢v⁢e ) :Qβ⁢ ( s⁢t⁢a⁢y , s ) =Rβ⁢ ( s⁢t⁢a⁢y , s ) +γβ⁢ ( s⁢t⁢a⁢y , s ) Vβ⁢ ( sn⁢e⁢x⁢t ) Qβ⁢ ( l⁢e⁢a⁢v⁢e ) =γβ⁢ ( l⁢e⁢a⁢v⁢e ) Vβ⁢ ( sf⁢i⁢r⁢s⁢t ) Qδ⁢ ( s⁢t⁢a⁢y , s ) =Rδ⁢ ( s⁢t⁢a⁢y , s ) +γδ⁢ ( s⁢t⁢a⁢y , s ) Vδ⁢ ( sn⁢e⁢x⁢t ) Qδ⁢ ( l⁢e⁢a⁢v⁢e ) =γδ⁢ ( l⁢e⁢a⁢v⁢e ) Vδ⁢ ( sf⁢i⁢r⁢s⁢t ) The value functions of the overall quasi-hyperbolic discounting model were:Qq⁢u⁢a⁢s⁢i⁢ ( s⁢t⁢a⁢y , s ) =ωQβ⁢ ( s⁢t⁢a⁢y , s ) + ( 1-ω ) Qδ⁢ ( s⁢t⁢a⁢y , s ) Qq⁢u⁢a⁢s⁢i⁢ ( l⁢e⁢a⁢v⁢e ) =ωQβ⁢ ( l⁢e⁢a⁢v⁢e ) + ( 1-ω ) Qδ⁢ ( l⁢e⁢a⁢v⁢e ) where 0<ω<1 was the weight of the β system relative to the δ system . β , δ , and ω were all free parameters . Similar to the foraging task , events within the intertemporal choice task marked state transitions , and all state transitions were deterministic except for decisions to choose the smaller-sooner option ( SS ) or larger-later option ( LL ) , which occurred only in decision states ( abbreviated state space diagram in Figure 4—figure supplement 1 ) . From decision states , animals transitioned to delay , reward , and post-reward delay ( ITI ) states for the chosen option — the delay , reward and ITI for the SS and LL options were represented by separate states . The value of choosing SS or LL in decision state s is the discounted value of the next state , the following delay state:Q⁢ ( S⁢S , s ) =γ⁢ ( s ) Q⁢ ( S⁢S⁢D⁢e⁢l⁢a⁢y ) Q⁢ ( L⁢L , s ) =γ⁢ ( s ) Q⁢ ( L⁢L⁢D⁢e⁢l⁢a⁢y ) The value of delay states were the discounted value of the reward state for that action , the value of reward states were the reward for that action plus the discounted value of the ITI state for that action , and the value of ITI states were the discounted value of the next decision state:Q⁢ ( S⁢S⁢D⁢e⁢l⁢a⁢y ) =γ⁢ ( S⁢S⁢D⁢e⁢l⁢a⁢y ) Q⁢ ( S⁢S⁢R⁢e⁢w⁢a⁢r⁢d ) Q⁢ ( S⁢S⁢R⁢e⁢w⁢a⁢r⁢d ) =R⁢ ( S⁢S⁢R⁢e⁢w⁢a⁢r⁢d ) γ⁢ ( S⁢S⁢R⁢e⁢w⁢a⁢r⁢d ) Q⁢ ( S⁢S⁢I⁢T⁢I ) Q⁢ ( S⁢S⁢I⁢T⁢I ) =γ⁢ ( S⁢S⁢I⁢T⁢I ) V⁢ ( sn⁢e⁢x⁢t⁢d⁢e⁢c ) where the value of the next decision state , V⁢ ( sn⁢e⁢x⁢t⁢d⁢e⁢c ) is the greater of Q⁢ ( S⁢S , sn⁢e⁢x⁢t⁢d⁢e⁢c ) and Q⁢ ( L⁢L , sn⁢e⁢x⁢t⁢d⁢e⁢c ) . Decisions were made assuming the value of Q⁢ ( S⁢S , s ) and Q⁢ ( L⁢L , s ) were represented as Gaussian distributions with noise that scaled with their magnitude . The probability of choosing the LL option was the probability that a random sample from the LL distribution was greater than a random sample from the SS distribution for that state:p⁢ ( c⁢h⁢o⁢o⁢s⁢e⁢L⁢L , s ) =1-ϕ⁢ ( Q ( SS , s ) -Q ( LL , s ) ) σ2[Q ( SS , s ) 2+Q ( LL , s ) 2 ) ] ) where ϕ is the normal cumulative distribution function . The nonlinear reward utility , biased time perception , and temporal discounting models were implemented as they were in the foraging task . For the one-trial horizon discounting models , the value of choosing a given option was the discounted value of the reward on the current trial only , with delay d and reward r:Qe⁢x⁢p=e-βdrQh⁢y⁢p=r1+kdQc⁢s=e-βdαrQquasi=[ω∗e−β∗d+ ( 1−ω ) ∗e−δ∗d]∗r To calculate the probability of choosing the LL option , the same decision rule was used as in the all-future horizon model . All models had two parameters except for the constant sensitivity discounting model with three and the quasi-hyperbolic discounting model with four . To determine the model that provided the best fit to the data , while accounting for the increased flexibility of these models , we calculated the Bayesian Information Criterion over the group level parameters ( iBIC ) ( MacKay , 2003; Huys et al . , 2011 ) . iBIC penalizes the log marginal likelihood , l⁢o⁢g⁢p⁢ ( D∣θ ) , which is the integral of the log likelihood of the data D over the distribution of group level parameters θ , for model complexity . Complexity is determined by the number of parameters k , and the size of the penalty depends on the total number of observations , n:i⁢B⁢I⁢C=l⁢o⁢g⁢p⁢ ( D∣θ ) +k2⁢l⁢o⁢g⁢ ( n ) . As in Huys et al . ( 2011 ) , we use a Laplace approximation to the log marginal likelihood:l⁢o⁢g⁢p⁢ ( D∣θ ) =-n2⁢l⁢o⁢g⁢ ( 2⁢π ) s+∑i=1sp⁢ ( Di∣θi ) ⁢p⁢ ( θi∣θ ) -∑i=1sl⁢o⁢g⁢d⁢e⁢t⁢ ( H⁢f⁢ ( θi ) ) 2where s is the number of subjects , and H⁢f⁢ ( θi ) is the hessian matrix of the likelihood for subject i at the individual parameters θi . To compare the fit of the quasi-hyperbolic discounting model across the foraging and intertemporal choice tasks , a cross-validation method was used . Data from each task was separated into thirds . The quasi-hyperbolic discounting model was fit to 2 of the samples from each task using maximum likelihood estimation ( fitting only individual parameters for each rat ) . The log likelihood of the data from the left out sample was evaluated . This process was repeated three times , leaving out each of the samples once , and we took the sum of the likelihood of the three left out samples . As the structure of variability was different between the foraging model ( variability in the patch leaving threshold ) and intertemporal choice models ( noise in the estimates of SS and LL values ) , to compare the discount function fit to the foraging task on intertemporal choice data , a new noise parameter was fit to the intertemporal choice data ( and vice-versa ) . We report the difference in the log likelihood of the data using parameters fit to the intertemporal choice task and of the log likelihood using parameters fit to the foraging task ( Figure 5 ) .	Often decisions have to be made on whether to stick with a resource or leave it behind to search for a better alternative . Should you book that hotel room or continue looking at others ? Is it time to start searching for a new job , or even for a new partner ? Animals face similar 'stick or twist' decisions when foraging for food . Knowing how to maximize the amount of food you obtain is key to survival . Studies have shown that most animals tend to stick with a food source for a little too long , a phenomenon known as 'overharvesting' . To find out why , Kane et al . designed carefully controlled experiments to compare foraging behavior in rats to another form of decision-making , known as intertemporal choice . The latter involves choosing between a small reward now versus a larger reward later . Given this choice , most rats opt to receive a smaller reward now rather than wait for the larger reward . This suggests that rats value rewards available in the future less than rewards they can get immediately . Kane et al . showed that this preference for short-term rewards can also explain why rats overharvest in foraging scenarios . By leaving one food source to go in search of another , rats must put up with a delay before they can access the new food supply . This delay , due to the time required to travel and search , reduces the value of the future reward . As a result , rats are more likely to stick with their current food source , even though leaving it would yield a greater reward in the long run . These findings in rats raise important questions about the mechanisms that lead to biases in thinking , and how factors like changes in the environment or specific disease states can influence these biases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2019	Rats exhibit similar biases in foraging and intertemporal choice tasks
Ecological processes underlying bacterial coexistence in the gut are not well understood . Here , we disentangled the effect of the host and the diet on the coexistence of four closely related Lactobacillus species colonizing the honey bee gut . We serially passaged the four species through gnotobiotic bees and in liquid cultures in the presence of either pollen ( bee diet ) or simple sugars . Although the four species engaged in negative interactions , they were able to stably coexist , both in vivo and in vitro . However , coexistence was only possible in the presence of pollen , and not in simple sugars , independent of the environment . Using metatranscriptomics and metabolomics , we found that the four species utilize different pollen-derived carbohydrate substrates indicating resource partitioning as the basis of coexistence . Our results show that despite longstanding host association , gut bacterial interactions can be recapitulated in vitro providing insights about bacterial coexistence when combined with in vivo experiments . Gut microbial communities are usually dominated by few bacterial phyla and families , but contain a high extent of species- and strain-level diversity ( Ley et al . , 2008; Dethlefsen et al . , 2007 ) . According to the competition-relatedness hypothesis , the more closely two organisms are related the more likely it is that they will compete and exclude each other due to overlapping niches ( Elton , 1946 ) . Therefore , it has remained unclear how closely related microbes can be maintained in the gut , or in any other natural microbial ecosystem . The high concentration of nutrients and the structured environment of the gut may allow functionally redundant species or strains to coexist ( Ley et al . , 2006 ) . The host may even select for such redundancy , as it can increase the stability and resilience of the microbiota against environmental disturbance ( Ley et al . , 2006; Foster et al . , 2017 ) . Phage predation can also contribute to the maintenance of diversity by imposing kill-the-winner dynamics and hindering the outgrowth of a single dominant strain ( Koskella and Brockhurst , 2014 ) . Another possibility is that closely related species , and even strains of the same species , have functionally diverged from each other and occupy distinct ecological niches ( Chesson , 2000; Bittleston et al . , 2019 ) . The genomic flexibility of bacteria facilitates adaptation to different nutrients , provided in the diet or by the host ( Berasategui et al . , 2017; Martens et al . , 2008 ) , or result from interactions with other bacteria ( Madi et al . , 2020 ) , such as cross-feeding ( Goldford et al . , 2018 ) or cooperative glycan breakdown ( Rakoff-Nahoum et al . , 2016 ) . Few experimental studies have investigated the coexistence of bacteria in host-associated microbial communities . The high diversity in these ecosystems and the resistance of many host-associated bacteria to experimental manipulations introduce considerable challenges for such approaches ( Ortiz et al . , 2021; Venturelli et al . , 2018 ) . Moreover , community dynamics observed in vivo can be difficult to reproduce under laboratory conditions , as the host presents a highly specialized nutritional and spatial environment . Thus , there is a need for in vitro models that can reproduce ecological interactions observed in vivo , from simple co-culturing setups ( Li et al . , 2019 ) to sophisticated ‘organoids-on-a-chip’ systems ( Jalili-Firoozinezhad et al . , 2019; Nikolaev et al . , 2020 ) . The gut microbiota of the Western honey bee ( Apis mellifera ) is composed of a few deep-branching phylogenetic lineages ( phylotypes ) belonging to the Firmicutes , Actinobacteria , and Proteobacteria phyla ( Martinson et al . , 2011; Kwong and Moran , 2016 ) . Most of these lineages are composed of several closely related sequence-discrete populations , hereafter referred to as species , each of which contains further diversity at the strain-level ( Ellegaard and Engel , 2016; Ellegaard and Engel , 2019; Engel et al . , 2012; Ellegaard et al . , 2015 ) . Microbiota-depleted bees can be generated and experimentally colonized with synthetic communities of different strains . Moreover , most community members can be cultured in pollen , which is the major dietary source of honey bees ( Kešnerová et al . , 2017 ) . This experimental tractability offers an excellent opportunity to probe the coexistence of bacteria in the gut of their native host and in controlled laboratory cultures using similar nutritional conditions . One of the most abundant and diverse phylotype of the honey bee gut microbiota is Lactobacillus Firm5 ( Ellegaard and Engel , 2019 ) . This phylotype consists of facultative anaerobes that ferment sugars into organic acids and utilize various pollen-derived glycosylated plant compounds , such as flavonoids ( Kešnerová et al . , 2017 ) . Lactobacillus Firm5 is specific to social bees but has diverged into many different species of which four are specifically associated with the Western honey bee , Apis mellifera: Lactobacillus apis ( Lapi ) , Lactobacillus helsingborgensis ( Lhel ) , Lactobacillus melliventris ( Lmel ) , and Lactobacillus kullabergensis ( Lkul ) . The four species are consistently present in the gut of individual honey bees suggesting that they can share the available niches and stably coexist despite their phylogenetic relatedness . Genomic analysis has revealed that these species share <85% pairwise average nucleotide identities ( gANI ) and exhibit high levels of genomic variation in terms of carbohydrate metabolism ( Ellegaard and Engel , 2019; Ellegaard et al . , 2015 ) . However , whether the coexistence is facilitated by adaptation to different nutritional niches , and to what extent the host environment , the diet , or the interactions with other community members matter is currently unknown . Here , we tested under which conditions the four Lactobacillus Firm5 species can coexist and investigated the underlying molecular mechanism . We serially passaged the four species in vivo through gnotobiotic bees and in vitro in liquid cultures , and applied RNA sequencing and metabolomics analysis . Our results show that the coexistence of the four species is mediated by the partitioning of nutrients derived from the pollen diet of bees and is largely independent from the presence of the host or other community members . All experiments in this study were conducted with four bacterial isolates representing the four Lactobacillus Firm5 species ( Lapi , Lhel , Lmel , and Lkul ) associated with the Western honey bee . We first tested if the four species can establish in the gut of gnotobiotic bees under two different dietary conditions . To this end , we colonized microbiota-depleted bees with each of the four species , alone or together , and fed bees either sterilized sugar water ( SW ) or sterilized sugar water and pollen ( SW+PG ) . Five days post-colonization , we assessed the bacterial loads in the gut by counting CFUs ( Figure 1A , Supplementary file 3 ) . When mono-colonized , the four species established in the gut of microbiota-depleted bees independent of the dietary treatment ( Figure 1A ) . In the SW treatment , the colonization levels were generally lower than in the SW+PG treatment ( Figure 1A , ANOVA q-value < 0 . 01 ) confirming previous results that pollen increases the total bacterial load in the gut ( Kešnerová et al . , 2020 ) . There was no statistically significant difference between the total bacterial loads of the mono-colonizations and the co-colonizations in either dietary treatment , with the exception of the mono-colonization with Lkul , which showed higher loads than the co-colonizations in SW ( Figure 1A , ANOVA q-value < 0 . 01 ) . Consequently , the sum of the bacterial loads of the mono-colonizations exceeded the total bacterial load of the co-colonizations in both dietary treatments , suggesting that the species engage in negative interactions when colonizing the honey bee gut together . To test if the four species can stably coexist in the bee gut , we serially passaged the community seven times in microbiota-depleted bees under both dietary conditions ( SW and SW+PG ) . After each passage ( i . e . after 5 days of colonization ) , we used amplicon sequencing of a discriminatory housekeeping gene fragment ( see Materials and methods ) in combination with CFU counting to determine the absolute abundance of each species in the community . We observed clear differences between the two dietary conditions in the ability of the four species to coexist across the passages ( Figure 1B–C , Supplementary file 4 ) . In the SW treatment , all species were initially detectable in most samples ( P1 , Figure 1B ) . However , three species ( Lapi , Lmel and Lkul ) steadily decreased in abundance in the subsequent passages resulting in a rapid dominance of Lhel ( Figure 1B ) . Lmel and Lkul reached the detection limit and Lapi decreased to around 104 bacteria/gut by passage five ( P5 , Figure 1B ) . Only Lhel was stably maintained across all seven passages and was present at around 1000x higher abundance than Lapi at the end of the experiment ( ~107 bacteria/gut , Figure 1B ) . In the contrary , in the SW+PG treatment , all four species were detectable in all passages at around 106 to 108 bacteria/gut , and displayed a highly stable abundance profile over time ( Figure 1C ) . In summary , these findings show that the four species can stably coexist in vivo when bees are fed pollen , but not when they are only fed sugar water . This is consistent with the idea that pollen facilitates niche partitioning in the honey bee gut by offering competing species different ecological niches facilitating their coexistence . We next tested if the four species can also coexist in vitro , outside of the host environment , under different nutrient conditions . To this end , we cultured the species alone or together in minimal medium supplemented with either glucose ( G ) , pollen extract ( PE ) , or entire pollen grains ( PG ) . All four species were able to grow when cultured alone under the three nutrient conditions ( Figure 2—figure supplement 1 , Supplementary file 3 ) . Growth yields of Lhel , Lkul , and the co-culture were slightly lower in PE and PG than in G , and Lmel showed lower growth yields than some of the other species in PE and G ( Figure 2—figure supplement 1 , ANOVA q-value < 0 . 01 ) . As in vivo , the total bacterial loads of the co-cultures were not consistently different from those of the mono-cultures ( Figure 2—figure supplement 1 ) , suggesting that the four species have overlapping metabolic niches and engage in negative interactions with each other . We then serially passaged the co-cultures 21 times under the three different nutrient conditions by transferring an aliquot after 24 hr of growth into fresh culture medium ( 1:20 ) . The absolute abundance of each strain was determined after every other passage by combining amplicon sequencing with qPCR ( see Materials and methods ) . As for the in vivo experiment , we observed clear differences in the growth dynamics of the four species , both over time and between the glucose and the pollen culture conditions ( Figure 2 , Supplementary file 4 ) . In the presence of glucose , three of the four species ( Lhel , Lmel , and Lkul ) steadily decreased in abundance over time ( Figure 2A ) , with two of them reaching the limit of detection ( <105 bacteria/ml ) after about 11 passages ( P11 ) . In contrast , Lapi was stably maintained at high abundance ( 109 bacteria/ml ) until the last passage ( Figure 2A ) and hence dominated the co-culture for most of the transfer experiment . In the presence of PE or PG , the four species revealed very different growth behaviors ( Figure 2B and C ) . None of the species decreased over time , and after 21 transfers all species still yielded between 106 and 109 bacteria/ml . To look at changes in community composition over time , we measured the community stability ( temporal mean divided by temporal standard deviation of the species abundances ) in sliding windows of five passages . Little to no change in community stability was observed for the two pollen conditions throughout the experiment , whereas in glucose the community reached a stable state after ~11 transfers ( Figure 2D ) . To compare the growth yields of each species across the three nutrient conditions , we only considered the passages after which community stability was reached ( P13-21 ) . With the exception of Lapi all species reached higher yields in the presence of pollen as compared to glucose ( Figure 2E , ANOVA q-value < 0 . 01 ) . Notably , Lmel was the only species that showed improved growth in PG as compared to PE ( Figure 2A–C ) . In summary , these findings show that the nutrient-dependent coexistence of the four Lactobacillus species observed in vivo can be recapitulated in vitro in a simple co-culture experiment , suggesting that the partitioning of pollen-derived nutrients is sufficient for enabling coexistence . Similar results were obtained for a second in vitro experiment which included the same nutrient conditions , but was only conducted for ten transfers ( Figure 2—figure supplement 2 ) . Given the impact of pollen on the coexistence of the four Lactobacillus species , we tested if genes involved in nutrient acquisition and metabolism were differentially expressed between the dietary treatments . To this end , we carried out RNA sequencing of the four-species community in honey bees that were fed either sugar water ( SW ) or sugar water and pollen grains ( SW+PG ) ( Figure 3A ) . Multidimensional scaling ( MDS ) of the normalized read counts mapped to each species revealed that most samples clustered by treatment ( SW+PG versus SW ) ( Figure 3—figure supplement 1 ) , indicating that all four species exhibited dietary-specific transcriptional responses . We found a total 687 genes ( 181 to 217 genes per species ) to be differentially expressed ( log2FC ≥ \|2\| and p-value ≤ 0 . 01 ) between the two dietary treatments ( Figure 3B ) . ‘Carbohydrate transport and metabolism’ ( Cluster of orthologous group category G , COG G ) was by far the most abundant functional category among the genes upregulated in the SW+PG treatment relative to the SW treatment ( Figure 3C , 17 . 1–37 . 6% of all upregulated genes ) . In three of the four species ( Lmel , Lhel , and Lkul ) , this category was significantly enriched among the upregulated genes ( Fisher’s exact test , p<0 . 01 , Supplementary file 6 ) . The largest fraction of the upregulated COG G genes encoded PTS transporters ( Figure 3D , Supplementary file 5 ) , followed by other sugar transporters ( e . g . ABC transporters ) , and enzymes involved in sugar cleavage and conversion ( Figure 3D ) . Among the downregulated genes , COG G genes were not abundant ( 5 . 1–7 . 8% ) ( Figure 3C ) . Instead , the category ‘Amino acid metabolism and transport’ ( COG E ) was enriched in Lapi ( Fisher’s exact test , p < 0 . 01 , Supplementary file 8 ) , and genes encoding ABC-type amino acid transporters were present among the downregulated genes in all species ( Supplementary file 5 ) . We next clustered all genes by homology into gene families . While most of the differentially expressed genes ( 89% ) belonged to gene families with homologs in multiple species , differential expression was typically observed for just one of the species ( Figure 3E–F ) . This suggests that the presence of pollen triggers distinct transcriptional changes in the four species during gut colonization . Indeed , gene annotation analysis allowed us to identify several species-specific metabolic functions among the differentially regulated genes ( Figure 4 , Supplementary file 5 ) . For example , Lhel specifically upregulated three PTS gene clusters for the uptake and metabolism of sugar alcohols and one gene cluster for ribose utilization . In contrast , Lmel upregulated several gene clusters involved in the cleavage of xylose , mannose , rhamnose , and arabinose from polysaccharides or other glycosylated compounds . Lmel also upregulated a gene cluster for the synthesis and the transport of bacteriocins in the presence of pollen . Lkul upregulated a starch utilization gene cluster , which in part was also differentially regulated in Lmel . In addition , this species upregulated an oligopeptide transporter gene cluster that was present in some of the other strains but not differentially regulated . The fourth species , Lapi , also differentially expressed genes belonging to COG ‘G’ ( mainly PTS transporters ) , but fewer ones , and with similar functional annotations as found in the other three species . However , Lapi was the only species that upregulated two conserved deoxycytidine kinase genes encoding enzymes involved in nucleoside salvage pathways . Besides these species-specific transcriptional changes , a number of interesting functions were differentially regulated in more than one species . For example , we found evidence for citrate fermentation in Lhel and Lkul . Both species upregulated genes encoding a citrate lyase for the conversion of citrate into oxaloacetate and acetate in the presence of pollen ( Supplementary file 5 ) . Lhel , Lmel , and Lkul upregulated genes for the uptake and metabolism of glycerol . Moreover , all four species upregulated gene clusters encoding surface proteins with leucine-rich repeat ( LRR ) regions , LPXT cell-wall anchoring motifs , and SLAP ( S-layer associated protein ) domains ( Figure 3G ) . Altogether , these results suggest that the four species utilize different carbohydrate-related resources from pollen , which supports the niche partitioning hypothesis as the basis for coexistence . In vivo gene expression differences between the two dietary conditions could be influenced by the host or by bacteria-bacteria interactions . Therefore , we carried out an additional transcriptomics analysis to disentangle the contribution of each of these factors to transcriptional changes in the four Lactobacillus species . We grew the four species in vitro in either co-culture or mono-culture , and with either pollen extract ( PE ) or glucose as growth substrate ( G ) ( Figure 5A ) . As for the in vivo RNA-Seq analysis , MDS plots of the normalized read counts indicated that the four species exhibit treatment-specific transcriptional responses ( Figure 5—figure supplement 1 ) . For each species , whether grown alone or in co-culture , we found between 159 and 393 genes to be differentially regulated between the PE and the G treatment ( Figure 5B , log2FC ≥ \|2\| and p-value≤0 . 01 ) . As in vivo , Carbohydrate transport and metabolism ( COG ‘G’ ) was the predominant functional category among the upregulated genes in the presence of pollen ( Figure 5C ) and enriched in all eight comparisons ( four species , each alone or in co-culture , Fisher’s exact test p-value < 0 . 01 , Supplementary file 8 ) . Moreover , 25 . 3–36 . 9% of the genes upregulated in vivo were also upregulated in vitro in the presence of pollen . In particular , the species-specific carbohydrate metabolism functions described above ( Figure 4 ) showed a similar transcriptional response to pollen in vivo and in vitro ( Figure 5D ) . In contrast , most of the putative adhesin genes upregulated in vivo were not upregulated in vitro during growth in pollen or had relatively low transcripts per million ( TPM ) . This suggests that these genes are either expressed in response to the host environment , or the presence of entire pollen grains or sugar water , both of which were only included in the in vivo but not in the in vitro experiment ( Supplementary file 9 ) . It is also noteworthy that fewer genes were downregulated than upregulated in pollen relative to glucose , and that the COG category ‘G’ was not enriched among the downregulated genes , which is concordant with our in vivo transcriptome analysis . ( Supplementary file 8 ) . Based on these results , we conclude that each species upregulates specific operons for the transport and utilization of different carbohydrates ( e . g . sugar alcohols and glycans ) in response to the presence of pollen , independent of the host environment . We found that a large fraction of the genes upregulated in PE relative to G in the mono-cultures were also upregulated in the co-cultures ( 58 . 2–87 . 8% , Figure 5E ) . In particular , the gene clusters identified to be regulated in a species-specific manner ( see above ) showed highly concordant gene expression profiles in vitro independent of the presence/absence of the other Lactobacillus species . This was confirmed by the direct comparison of mono-culture and co-culture conditions . In comparison to the nutritional treatments , fewer genes ( 9–149 genes ) were differentially expressed between co-culture and mono-culture treatments ( log2FC ≥ \|2\| and p-value≤0 . 01 ) , ( Figure 5F ) . We could not find any consistent pattern across the four species in terms of COG category enrichment ( Supplementary file 8 ) . Moreover , only a few genes were differentially expressed in more than one species ( 6 . 25–30% ) , or across both nutrient conditions ( 1 . 86–5 . 33% ) . Citrate fermentation genes were upregulated in Lkul in co-culture relative to mono-culture when grown in pollen , whereas in Lhel the opposite was observed ( Figure 5G ) . Also of note , the oligopeptide transporter system which was upregulated in vivo in Lkul in the presence of pollen , was also upregulated in vitro in the presence of pollen , but only when other species were present . These two specific examples show that a few metabolic functions are differentially regulated in response to other bacteria , but not always in the same direction across species , or only in a specific nutrient condition . We thus conclude that the main factor driving changes in gene expression in the four strains is the presence of pollen , rather than the presence of other Lactobacillus species . Our transcriptome analyses suggest that differences in sugar metabolism may enable the four species to coexist in the presence of pollen in vitro and in vivo . To assess species-specific metabolic changes when grown in pollen , we profiled the metabolome of the pollen extract medium before ( t = 0 hr ) and after bacterial growth ( t = 16 hr ) using Q-TOF-based untargeted metabolomics ( Fuhrer et al . , 2011 ) . We annotated a total of 657 ions of which 406 could be reliably categorized as pollen-derived ions , as opposed to ions originating from the base medium ( see Materials and methods , Supplementary file 10 , Figure 6—figure supplement 2 ) . The metabolomics data clearly separated the four species indicating distinctive metabolic changes and thus corroborating the transcriptome results ( Figure 6—figure supplement 1 ) . A total of 76 pollen-derived ions showed a significant decrease in abundance over-time ( log2FC ≤ −1 and p-value≤0 . 01 , Student’s t-Test , BH correction ) ( Figure 6A , Supplementary file 10 ) . Of those , 24 ions decreased in abundance in all four species , another 24 ions decreased in abundance in only a subset of the species , and the remaining 28 ions decreased in abundance in only a single species ( Figure 6A ) . Ions annotated as glycosylated flavonoids were among the top ions responsible for the separation of the four species in the PCA ( Figure 6—figure supplement 1 ) . Lmel depleted six different ions annotated as flavonoids ( isoorientin 2’’-O-rhamnoside , quercetin-3-O-glucoside , vitexin , rutin , luteolin-7-O- ( 6’’-malonylglucoside ) , and quercetin-3-O-beta-D-glucosyl- ( 1->2 ) -beta-D-glucoside ) , while Lapi depleted three ions annotated as flavonoids ( isoorientin 2’’-O-rhamnoside , quercetin-3-O-glucoside , vitexin ) ( Figure 6A , Figure 6—figure supplement 3 ) . In contrast , Lkul only depleted the flavonoid ion annotated as isoorientin 2’’-O-rhamnoside , and no flavonoid ion changes were identified for Lhel ( Figure 6A , Figure 6—figure supplement 3 ) . To corroborate the species-specific utilization of flavonoids , we incubated each of the four species in base culture medium supplemented with rutin . We observed the formation of a yellow insoluble precipitate only in the wells incubated with Lmel ( Figure 6B ) . Metabolomics analysis confirmed that rutin was depleted in these wells and that the yellow precipitate corresponded to an accumulation of quercetin , the water-insoluble , deglycosylated aglycone of rutin ( Figure 6C ) . These findings are consistent with our transcriptome results which show that Lmel is the only species that upregulated a rhamnosidase gene known to cleave rhamnose residue from rutin ( Beekwilder et al . , 2009; Figure 4 ) . Other ions with species-specific abundance changes included a plant-derived glycosylated compound belonging to the iridioids class ( i . e . antirrhinoside , depleted in the presence of Lapi ) , a component of the outer pollen wall ( i . e . 9 , 10 , 18-trihydroxystearate , accumulated in the presence of Lmel ) and cyclic nucleotides ( depleted in the presence of Lhel , Lmel , and Lkul ) ( Figure 6A and Figure 6—figure supplement 3 ) . Lhel was the only species depleting an ion corresponding to sugar alcohols ( mannitol , D-sorbitol , or L-glucitol ) ( Figure 6A and Figure 6—figure supplement 3 ) consistent with the specific upregulation of sugar alcohol PTS transporters in this species ( Figure 4 ) . Based on the untargeted metabolomics analysis , we conclude that the four species target different metabolites , in particular secondary plant metabolites present in pollen . In order to assess differences in the utilization of simple sugars and acids in more detail , we analyzed the supernatants of cultures of the four strains after 0 , 8 , 16 , and 24 hr of growth using GC-MS . We used a semi-targeted approach , where we identified a subset of metabolites by preparing analytical standards and the others by using a reference library ( see Materials and methods ) . We identified 113 metabolites of which 46 showed a significant change in abundance in at least one strain between timepoint 0 hr and 24 hr ( log2FC ≥ \|2\| and p-value ≤ 0 . 01 , Student’s t-Test , BH correction ) ( Supplementary file 10 ) . All four species showed mixed substrate utilization , that is they utilized several substrates simultaneously . Moreover , most substrates were utilized by all four species , but often at different rates . Many metabolites that we identified with the GC-MS had annotations comparable to the ones found in the Q-TOF-based experiment . For example , we detected a compound annotated as sugar alcohol , that is glucitol , that was consumed most efficiently by Lhel as observed in the previous analysis ( Figure 6C ) . Moreover , all four species consumed the carboxylic acids citrate and malate ( Figure 6C and Figure 6—figure supplement 4 ) , which corresponded with the results of the Q-TOF-based experiment . Interestingly , Lkul and Lhel consumed citrate at the fastest rate and they were also the two species that upregulated gene clusters for citrate fermentation in the presence of pollen in vivo and in vitro ( Figure 4 ) . Lmel consumed several simple monosaccharides ( such as glucose , fructose , allose , and mannose ) at a slower rate than the other species , although having a similar growth profile ( Figure 6C , Figure 6—figure supplements 4–5 ) . This could indicate that Lmel has specialized in the metabolism of pollen-derived glycosylated compounds ( such as rutin , Figure 6B–C ) at the expense of fast consumption of generic substrates , which accords with the upregulation of several gene clusters for the cleavage of such sugars from polysaccharides or other glycosides ( e . g . flavonoids ) in presence of pollen ( Figure 4 ) . In summary , our metabolomics results show that the four species specialize in the utilization of different pollen-derived compounds , and that the observed metabolite changes are to some extent consistent with the transcriptional changes observed in the presence of pollen relative to the presence of simple sugars . Ecological processes governing the coexistence of microbes have been probed in the laboratory using microbial communities of different complexity ( Goldford et al . , 2018; Ortiz et al . , 2021; Wright and Vetsigian , 2016; Friedman et al . , 2017; Piccardi et al . , 2019; Deines et al . , 2020; Logan , 2017 ) . However , few studies have examined the impact of the host on the coexistence of bacterial symbionts of animals ( Ortiz et al . , 2021; Deines et al . , 2020 ) . In particular , little is known about the extent to which closely related species and strains can be stably maintained ( Bittleston et al . , 2019 ) . We capitalized on the experimental tractability of honey bees and their gut microbiota and used a bottom-up approach to study the coexistence of four closely related , naturally co-occurring Lactobacillus species . We disentangled the effect of the diet and the host on the interactions between the four species by serially passaging them through gnotobiotic bees or in culture tubes , under two nutrient conditions ( pollen versus simple sugars ) . Our results show that the dynamics in the four-species community is governed by negative interactions , because the growth of each member was lower in co-culture than in mono-culture , independent of the environment ( host or culture tube ) and the nutrient condition ( pollen or simple sugars ) . This is consistent with previous observations that negative interactions predominate in nutrient-rich environments ( Piccardi et al . , 2019; Foster and Bell , 2012; Berry and Widder , 2014; Coyte et al . , 2015; Ghoul and Mitri , 2016 ) . Moreover , the four Lactobacillus species harbor relatively small genomes ( 1 . 5–2 Mb ) with a conserved and streamlined core metabolism and similar auxotrophies , suggesting overlapping nutritional requirements ( Ellegaard et al . , 2019; Ellegaard et al . , 2015; Kwong and Moran , 2016 ) . The coexistence of bacterial symbionts can be facilitated by the host , for example by providing a spatially structured environment that results in the physical separation of competing strains ( Gude et al . , 2020; Kim et al . , 2008; Mitri et al . , 2016; Hallatschek et al . , 2007 ) , or by secreting metabolites that support niche specialization ( Schluter and Foster , 2012; McLoughlin et al . , 2016 ) . However , in the case of the four Lactobacillus species , such host-related features seem not to be sufficient to support coexistence , because the four-species community was rapidly dominated by a single species , when passaged through gnotobiotic bees that were fed a simple sugar diet . In contrast , when providing a more diverse nutrition in the form of pollen , we found that the four species were stably maintained both in vivo and in vitro . We thus conclude that the coexistence of the four Lactobacillus species in the honey bee gut primarily depends on the pollen diet of the host and not the host environment itself . The challenges in replicating the native environment such that it is possible to study relevant interactions of host-associated microbes in vitro are formidable . These were highlighted in a recent study on the microbial community associated with the freshwater polyp hydra that could not recapitulate the coexistence of the dominant microbiota members in vitro ( Deines et al . , 2020 ) . Here , we aimed to approximate the nutritional conditions in the honey bee gut by culturing the bacteria in pollen infused media , that is the natural diet of bees . In both the in vivo and in vitro transfer experiment , we assessed the effect of pollen on the dynamics of the community by comparing it to a simple sugar treatment . Although not identical , the nutritional conditions in vitro were sufficiently similar as to recapitulate the overall community dynamics observed in vivo: pollen nutrients supported the stable coexistence of the four species , while the simple sugars led to the dominance of a single species . As the bee and members of the bee gut microbiota pre-digest pollen and sugars upstream of the rectum , it is difficult to exactly replicate the metabolic environment of the rectum . For example , sucrose is largely absorbed via the midgut epithelium and cleaved into glucose and fructose by host enzymes , while fermentative bacteria such as Gilliamella apicola degrade and modify a diverse range of carbohydrates in the ileum ( Kešnerová et al . , 2017; Crailsheim , 1988 ) . These metabolic alterations may explain some of the differences observed between the in vivo and in vitro experiments , such as the dominance of different species in the simple sugar conditions ( sucrose and glucose , respectively ) . We therefore suspect that different species would dominate in vitro or in vivo with an alternative simple sugar composition . Our findings are consistent with the consumer-resource model , which predicts that the number of species that can coexist depends on the number of available resources ( Tilman , 1986 ) . Correspondingly , in the presence of a single substrate , such as in the case of glucose in vitro , competition for the same nutrient results in the competitive exclusion of all but one species . However , depending on the nutrient availability , the dietary transit time , the crosstalk with the host , or the spatial structure of the gut , the ecological processes governing bacterial coexistence may differ across host-associated microbiomes . For example , the Lactobacillus species of the honey bee gut microbiota primarily colonize the luminal space of the rectum , where partially digested pollen accumulates . In contrast , some of the Proteobacteria of the bee gut microbiota adhere to the epithelial surface of the ileum ( Zheng et al . , 2018 ) . We expect that in the latter case interactions with the host play a more important role for microbial coexistence than in the case of the Lactobacilli in the rectum . Although ecological interactions in bacterial communities have been investigated across a wide range of experimental systems , few studies have tackled the molecular mechanisms underlying coexistence . In some cases , cross-feeding of metabolic by-products facilitates the maintenance of diversity in bacterial communities , such as after passaging leaf and soil samples in single carbon sources ( Goldford et al . , 2018 ) . However , cross-feeding does not seem to play an important role in maintaining coexistence of the four Lactobacillus species in this study . Unlike the above example , feeding a single carbon source led to the extinction of all but one species . Our metabolomics analysis also did not reveal any major metabolites that could potentially be cross-fed , that is were produced by one species and utilized by another . Finally , we identified no transcriptional changes that would suggest cross-feeding activities when comparing mono-cultures and co-cultures of the four Lactobacillus species . Instead , our combined transcriptomics and metabolomics analyses suggest that coexistence is facilitated by specialization toward distinct pollen-derived nutrients . We found that all four species upregulated carbohydrate transport and metabolism functions dedicated to the utilization of different carbon sources in the presence of pollen when colonizing the bee gut , and these changes were reproducible in vitro . Our metabolomics analysis identified a number of pollen-derived glycosides that were utilized in a species-specific manner . In particular , Lmel specialized in the utilization of flavonoids at the expense of simple sugars , which may explain why this species rapidly went extinct in presence of only simple sugars during the transfer experiments . While the importance of pollen-derived flavonoids in niche partitioning needs to be validated , the species-specific deglycosylation of these secondary plant compounds illustrates that the four species have different hydrolytic capabilities that may also be involved in the cleavage of other carbohydrates . The metabolic specialization on plant glycans may be a common phenomenon in animal gut communities , as similar transcriptional changes have been described in other gut symbionts when the host diet was supplemented with specific plant glycans ( Sonnenburg et al . , 2005; Zheng et al . , 2019 ) . In contrast to the species specific metabolism of glycoside , we observed few differences in the utilization of simple saccharides among the four species in our time-resolved GC-MS analysis . While this may seem surprising , theoretical work has established that resource preference for at least one substrate is sufficient to explain coexistence ( Meszéna et al . , 2006 ) . Moreover , it is plausible that the four species utilize the same sugars , but extract them from different pollen-derived glycans , such as starch , hemicellulose , flavonoids , or other glycosylated secondary plant metabolites . While this work focused on niche partitioning based on degradation of complex carbohydrates , it is noteworthy that all four Lactobacillus species engaged to a variable extent in co-fermentation of the carboxylic acids citrate and malate present in pollen . The two species , Lkul and Lhel , that upregulated citrate fermentation pathways in the presence of pollen also consumed citrate at the fastest rate . Citrate co-fermentation has been linked to competitive advantages in lactic acid bacteria , though whether the varying levels of co-fermentation contribute to colonization stability in this system remains an outstanding question ( Laëtitia et al . , 2014; Magni et al . , 1999; Jimeno et al . , 1995 ) . Previous work suggested that the large diversity of carbohydrate transport and metabolism functions in the accessory gene pool of Lactobacillus Firm5 is an adaptation to the pollen-based diet of the host and a consequence of the nutrient competition with closely related species ( Ellegaard and Engel , 2019; Ellegaard et al . , 2019 ) . Our findings support this hypothesis and provide the first experimental evidence for a link between the coexistence of the four Lactobacillus species , the large diversity of carbohydrate metabolism functions in their genomes , and the pollen diet of the host . Moreover , these results suggest that dietary differences between host species or natural variation in pollen diversity influence the diversity of Lactobacillus Firm5 and could , for example explain why the Asian honey bee , Apis cerana , harbors only one species of this phylotype in its gut ( Ellegaard et al . , 2020 ) . However , we have only tested a single strain of each of the four species . Therefore , given the extensive genomic diversity within these species ( Ellegaard and Engel , 2019 ) , more work is needed to determine if the identified patterns of coexistence reflect stable ecological niches occupied by the four species or are rather the result of the specific strains selected for our experiments . In a recent study on pitcher plant microbiomes , it was shown that even strains that differ by only a few base pairs can have different ecological trajectories in communities and coexist over extended period of time ( Bittleston et al . , 2019 ) . Expanding our approach to strains within species presents an exciting next step to understand at which level discrete ecological niches are defined in the bee gut and how diversity can be maintained in such ecosystems . We used the following four bacterial strains of Lhel , Lmel , Lapi , and Lkul for our experiments: ESL0183 , ESL0184 , ESL0185 , and ESL0186 ( Kešnerová et al . , 2017 ) . All strains were precultured on solid De Man – Rogosa – Sharpe agar ( MRSA ) ( supplemented with 2% w/v fructose and 0 . 2% w/v L-cysteine-HCl ) from glycerol stocks stored at −80°C . MRSA plates were incubated for three days in anaerobic conditions at 34°C to obtain single colonies . Single colonies were inoculated into a liquid carbohydrate-free MRS medium ( cfMRS; O’ Donnell et al . , 2011 ) supplemented with 4% glucose ( w/v ) , 4% fructose ( w/v ) , and 1% L-cysteine-HCl ( w/v ) and incubated at 34°C in anaerobic conditions without shaking . Bacterial colonization stocks were prepared from overnight cultures by washing the bacteria in 1xPBS , diluting them to an OD600 = 1 , and storing them in 25% glycerol at −80°C until further use . For colonization stocks containing all four species , cultures adjusted to an OD600 = one were mixed at equal proportions . Microbiota-depleted bees were obtained from colonies of Apis mellifera carnica located at the University of Lausanne following the procedure described in Kešnerová et al . , 2017 . Colonization stocks were diluted ten times in a 1:1 mixture of 1xPBS and sugar water ( 50% sucrose solution , w/v ) and 5 μL were fed to each bee using a pipette . Five days post-colonization , 10 rectums were dissected and homogenized in 1xPBS . An aliquot of each homogenized gut was used for CFU plating to enumerate the total bacterial load and for amplicon sequencing to obtain the relative abundance of each community member ( see below ) . To serial passage the community through microbiota-depleted bees , the ten homogenized gut samples from the same treatment were pooled together and stored in 25% glycerol at −80°C until a new batch of microbiota-depleted bees was available . At the day of colonization , a frozen aliquot of the pooled gut homogenate was thawed , diluted ten times in a 1:1 mixture of 1xPBS and sugar water ( 50% sucrose solution , w/v ) , and fed to newly emerged microbiota-depleted bee as described above . This was repeated for a total of six serial passages . Throughout the experiments all bees were kept on either a sugar water or a sugar water/pollen diet according to the two dietary treatment . Food was provided ad libitum . Each of the four strains was cultured in liquid medium overnight for about 16 hr as described above . The cultures were re-inoculated at an OD600 = 0 . 3 in fresh medium and let grow for another 4 hr at 34°C with shaking ( 700 rpm ) . Bacterial cells were then washed with 1xPBS , mixed in equal proportions , and inoculated at an OD600 = 0 . 05 in triplicates in 96-deep well plates ( SIGMA ) containing cfMRS medium supplemented with either 2% glucose ( w/v ) , 10% pollen extract ( v/v ) , or 10% pollen grains ( v/v ) in a final volume of 500 μL per well . Detailed information about pollen extract preparation can be found in the Supporting methods section of Kešnerová et al . , 2017 . Pollen grain solutions were prepared by adding 1 . 250 ml of ddH2O to 80 mg of pollen grains crushed with the bottom of a 15 mL falcon tube . The plates were incubated for 24 hr at 34°C under anaerobic conditions without shaking ( 300 rpm ) . After 24 hr of incubation , an aliquot of each sample was subjected to CFU plating to enumerate the total bacterial load . Then , 1% of each culture ( i . e . 5 μL ) was transferred to a plate with fresh medium supplemented with the appropriate carbon sources and incubated again . These transfers were repeated 10 , respectively , 20 times for the two independent experiments . After each transfer , cultures were washed once with 1xPBS and stored at −20°C for amplicon sequencing analysis . CFUs were counted after 24 hr and at the final transfer . The relative abundance of the four strains across all transfer experiments was obtained using amplicon sequencing of a 199 bp long fragment of a housekeeping gene encoding a DNA formamidopyrimidine-glycosylase which allows to discriminate the four strains from each other ( Ellegaard et al . , 2019 ) . For the in vitro transfer experiments , the PCR fragment was amplified from crude cell lysates . They were generated by mixing 5 μL of culture with 50 μL of lysis solution , containing 45 μL of lysis buffer ( 10 mM Tris- HCl , 1 mM EDTA , 0 . 1% Triton , pH 8 ) , 2 . 5 μL of lysozyme ( 20 mg/ml , Fluka ) , and 2 . 5 μL of Proteinase K solution ( 10 mg/ml , Roth ) . The samples were incubated for 10 min at 37°C , for 20 min at 55 °C , and for 10 min at 95 °C , followed by a short spin before preparing the PCR ( 1 min , 1500 rpm ) . For the in vivo transfer experiment , DNA was isolated from the homogenized gut samples using the hot phenol protocol used in Kešnerová et al . , 2017 . To amplify the gene fragment and to add the Illumina barcodes and adapters , the two-step PCR strategy published in Ellegaard et al . , 2019 was used . For the first PCR , 5 μL of DNA or 5 μL of cell lysate were mixed with 12 . 5 μL of GoTaq Colorless Master Mix ( Promega ) , 1 μL of forward and reverse primer ( 5 μM , see Supplementary file 1 ) and 5 . 5 μL of Nuclease-free Water ( Promega ) . The PCR I was performed as follows: initial denaturation ( 95°C – 3 min ) , 30 times denaturation-annealing-extension ( 95°C – 30 s , 64°C – 30 s , 72°C – 30 s ) , final extension ( 72 °C – 5 min ) . To purify the amplicons , 15 μL of PCR product were mixed with 5 μL of a 5X Exo-SAP solution ( 15% Shrimp Alkaline Phosphatase – 1000 U/ ml – NEB , 10% Exonuclease I – 20 , 000 U/ ml – NEB , 45% glycerol 80% and 30% dH2O ) . and incubated for 30 min at 37°C and for 15 min at 80°C . For the second PCR reaction , 5 μL of purified PCR products were mixed with the same reagents as before . The PCR program was the same as above with the exception that the annealing temperature was set to 60°C and the denaturation-annealing-extension steps were repeated for only eight times . The barcoded primers are listed in Supplementary file 1 . The amplicons of the second PCR were purified using the Exo-SAP program as described above . To prepare the sequencing of the amplicons , DNA concentrations were measured using Quant-iT PicoGreen for dsDNA ( Invitrogen ) . Each sample was adjusted to a DNA concentration of 0 . 5 ng/μL and 5 μL of each sample were pooled together . The pooled sample was loaded on a 0 . 9% agarose gel and gel-purified using the QIAquick Gel Extraction Kit ( Qiagen ) following the manufacturer’s instructions . The purified DNA was prepared for sequencing following the Illumina MiniSeq System Guide for ‘denaturate and dilute libraries’ and then loaded on a Illumina MiniSeq Mid Output Reagent Cartridge using the correspondent MiniSeq flow cell . Illumina reads were demultiplexed by retrieving the unique barcodes of the different samples and quality-filtered using Trimmomatic ( Trimmomatic-0 . 35 ) ( LEADING:28 TRAILING: 29 SLIDING WINDOW:4:15 MINLEN:90 ) . Each forward and reverse read pair was assembled using PEAR ( -m 290 n 284 j 4 -q 26 v 10 -b 33 ) ( Zhang et al . , 2014 ) , and the assembled reads were assigned to the different strains based on base pair positions with discriminatory SNP . See details in Supplementary material . To obtain absolute abundance data for each strain , we combined the relative abundance data from the amplicon sequencing with CFU counts obtained from plating homogenized bee guts in the case of the in vivo experiments ( see above ) or by carrying out qPCR with Lactobacillus-specific primers as described in Kešnerová et al . , 2017 in the case of the in vitro co-culture experiments ( Supplementary file 1 , Supplementary file 4 ) . For the in vitro transfer , the stability of the four-species community over time was calculated using the codyn R package ( Hallett et al . , 2020 ) . For the in vivo RNA sequencing , microbiota-depleted bees were colonized with the four species community as described above and fed with either sugar water and pollen grains or with sugar water only . After 5 days of colonization , the rectums of five bees per treatment ( all kept in the same cage ) were dissected and snap-frozen in liquid nitrogen in separate tubes containing glass beads ( 0 . 1 mm dia . Zirconia/Silica beads; Carl Roth ) . For RNA extraction , the tissue samples were suspended in 1 mL of TE buffer and homogenized using a bead beater ( 45 m/s , 6 s ) . Then , 200 μL of an ice-cold STOP solution ( 95% v/v ethanol , 5% v/v Aqua phenol [Roth] ) was added to 1 ml of homogenate and snap-frozen again in liquid nitrogen . Tubes were then thawed on ice and a previously developed hot phenol RNA extraction protocol was followed ( Sharma et al . , 2010 ) . For the in vitro RNA sequencing , bacterial strains were cultured in triplicates in cfMRS supplemented with either 1% w/v glucose or 1% w/v pollen extract . After 16 hr of growth , 200 μL of STOP solution was added to 1 mL of culture followed by the same steps as described above . After the precipitation step , samples were treated with DNaseI ( NEB ) to degrade DNA . RNA samples were purified using Nucleospin RNA clean-up kit ( Macherey-Nagel ) following the manufacturer’s instructions . RNA was eluted in RNase free-water and stored at −80°C until further use . RNA concentration and quality were assessed using Nanodrop ( ThermoFisher Scientific ) , Qubit ( ThermoFisher Scientific , RNA – High Sensitivity reagents and settings ) and Bioanalyzer ( Agilent ) . High-quality RNA samples were selected to prepare RNA libraries . For the in vivo RNA sequencing , libraries were prepared using the Zymo-Seq RiboFree Total RNA Library kit ( Zymo Research ) . The libraries were sequenced by the GTF facility of the University of Lausanne using HiSeq 4000 SR150 sequencing ( 150 bp reads ) ( Illumina ) . For the in vitro RNA sequencing , libraries were prepared following the protocol developed by Avraham et al . , 2016 . Libraries were then prepared for sequencing following the Illumina MiniSeq System guide for denaturate and dilute libraries . Libraries were sequenced using the Illumina MiniSeq technology using High Output Reagent Cartridges ( 150 bp reads ) and MiniSeq flow cells . For the in vitro samples , raw reads were demultiplexed using a script provided by Dr . Jelle Slager ( Personal communication ) For the in vivo samples , the reads were already demultiplexed by the sequencing facility . For both experiments , the reads were trimmed with Trimmomatic ( Trimmomatic-0 . 35 ) ( LEADING:30 TRAILING: 3 SLIDING WINDOW:4:15 MINLEN:20 ) . The quality of the reads was checked using FASTQC ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . For the in vivo samples , trimmed reads were sorted with sortmerna-4 . 2 . 0 to select only the non-rRNA reads for the downstream analysis . Reads were mapped onto the genomes of the selected strains ( Ellegaard and Engel , 2018 ) ( Lapi , Lhel , Lmel , and Lkul ) using Bowtie ( bowtie2-2 . 3 . 2 ) . Gene annotations for the four genomes were retrieved from IMG/mer ( Chen et al . , 2021 ) . Mapped reads were quality filtered for the alignment length ( CIGAR > 100 bp ) and for the allowed mismatches in the sequence ( NM = 0–1 ) . Quality filtered reads were then quantified using HTseq ( Version 0 . 7 . 2 ) . Differential gene expression between samples cultured in pollen extract and samples cultured in glucose , and between mono-cultures and co-cultures , was calculated using the R package EdgeR ( Robinson et al . , 2010 ) . Counts per million were calculated and only genes with at least one count per million were used for the analysis . EdgeR fits negative binomial models to the data . The counts were normalized for RNA composition by adjusting the log2FC according to the library size , and the quantile-adjusted conditional maximum likelihood ( qCML ) method was used to estimate the common dispersion and the tag-wise dispersion . Finally , the differential gene expression was determined using the exact test with a false discovery rate ( FDR ) <5% . COG annotations were obtained from IMG/mer , and the enrichment analysis for COG categories tested using the Fisher’s exact test . Transcripts per million ( TPM ) were visualized using the Integrated Genome Browser software ( Freese et al . , 2016 ) . Metabolites were extracted from liquid cultures supplemented with 10% ( w/v ) pollen extract at the inoculation time and after 16 hr of incubation at 34°C . For each liquid culture sample , 300 μL was collected and centrifuged ( 20 , 000 g , 4°C , 30 min ) , then 200 μL supernatant was transferred to a new tube and stored at −80°C . After collection of all samples , they were prepared for metabolomics analysis . The samples were thawed on ice and centrifuged again ( 20 , 000 g , 4°C , 5 min ) , then diluted 10 times with ddH2O . For metabolomics analysis , 25 μL of each diluted sample was sent in a 96-well plate on dry ice to the laboratory of Prof . Uwe Sauer for analysis ( ETH Zürich , Switzerland ) . Three replicates of a pollen-extract dilution series ( 10 serial 2x dilutions ) as well as undiluted pollen-extracts and water used for performing the dilution series were included in the metabolomics analysis . Because of the insolubility of flavonoid aglycones in a water matrix , metabolites from liquid cultures supplemented with rutin were extracted using a methanol-extraction protocol at the time of inoculation and after 16 hr of growth by adding 200 μL of methanol pre-cooled to −20°C to 100 μL of culture . Tubes were vortexed thoroughly and incubated for 5 min ( 4°C , shaking 14 , 000 rpm ) . Samples where then incubated at −20 °C for 1 hr and centrifuged ( 20’000 g , 5 min ) . A total of 200 μL of the supernatant was transferred to a new tube and diluted 10 times in 70% methanol and 25 μL of each diluted sample was sent to Zürich in Eppendorf tubes sealed with parafilm on dry ice . For untargeted metabolomics analysis , each sample was injected twice ( technical replicate ) into an Agilent 6550 time-of-flight mass spectrometer ( ESI-iFunnel Q-TOF , Agilent Technologies ) as detailed in Kešnerová et al . , 2017 . In brief , m/z features ( ions ) were annotated by matching their accurate mass-to-sum formulas of compounds in the KEGG database accounting for deprotonation ( -H+ ) . Alternative annotation can be found in Supplementary file 10 . When available , metabolites categories were assigned to ions based on KEGG ontology . Metabolomics data analysis was carried out using R version 3 . 6 . 3 . Variation of raw ion intensities obtained from untargeted metabolomics analysis for the two technical replicates was determined by assessing the correlation between ion intensities of the respective technical replicates . Then , mean ion intensities of technical replicates were calculated . Time point comparisons ( T = 0 hr vs T = 16 hr ) were performed using t-tests with Benjamini-Hochberg ( BH ) correction for multiple testing . log2FC values between the two time-points were calculated with respect to the mean intensity in the T0 time point . To identify pollen-derived ions , and distinguish them from background originating from culture medium and experimental noise , the ion intensities of the pollen dilution series were plotted for each ion and the R ( 2 ) of the obtained linear fit was extracted . In addition , we calculated the log2FC difference between undiluted pollen and water . The R ( 2 ) values were then plotted against the log2FC values , and stringent thresholds ( R2 > 0 . 75 and log2FC > 2 ) were chosen to discriminate ions that are likely pollen-derived ( Figure 6—figure supplement 3 ) . All ions were included for downstream analysis ( e . g . PCA ) and then they were discriminated between pollen-derived and non-pollen-derived . Soluble metabolites were extracted from liquid cultures supplemented with 10% pollen extract ( w/v ) at the inoculation time and after 8 , 16 , and 24 hr of incubation . For each liquid culture sample , 300 μL was collected and centrifuged ( 15 , 000 g , 4°C , 15 min ) . Then , 200 μL was transferred to a new tube , snap-frozen in liquid nitrogen , and stored at −80°C . Once that all the samples were collected , soluble metabolites were extracted . To extract soluble metabolites , tubes were thawed on ice , and 75 μL of sample was combined with 5 μL of 20 mM internal standard ( norleucine and norvaline , ( Sigma-Aldrich ) and U-13C6 glucose [Cambridge Isotope laboratories] ) . A volume of 825 μL of cold methanol:water:chloroform ( 5:2:2 ) solution was added to the sample and vortexed for 30 s . The tubes were incubated at −20°C for 90 min and vortexed 2x for 30 s during the incubation . Tubes were centrifuged for 5 min at 10 , 000 g at 4°C . The supernatant was removed and extraction was repeated using 400 μL of ice cold chloroform:methanol ( 1:1 ) , tubes were vortexed and left on ice for 30 min . Tubes were centrifuged 5 min at 8000 rpm at 4°C and the liquid phase was transferred to the previous extracted aqueous phase . A total of 200 μL of water was added and tubes were centrifuged 5 min at 8000 rpm . The aqueous phase was transferred to a 2 mL microcentrifuge tube . The aqueous extract was dried using a vacuum concentrator at ambient temperature overnight ( Univapo 150 ECH vacuum concentrator centrifuge ) . Once dried , the samples were dissolved in 50 μL of 20 mg/ml methoxyamine hydrochloride in pyridine for 1 . 5 hr at 33°C followed by derivatization with N-Methyl-N- ( trymethylsolyl ) trifluoroacetamide ( MSTFA , Sigma Aldrich ) for 2 hr at 35°C . Aliquots ( 1 μL ) were injected on an Agilent 8890/5977B GC-MSD . The samples were injected in split mode ( 20:1 ) with an inlet temperature of 250°C . The VF-5ms ( 30 m x 250 μm x 0 . 25 μm ) column was held initially at 125°C for 2 min , ramped at 5°C / min to 250°C , ramped at 15°C to 300°C , and held for 5 min . The MS was run in full scan mode ( 50–500 m/z ) at a speed of 5 Hz . Peaks from the total ion chromatogram ( TIC ) were identified by matching retention times and spectra to an in-house library that was built by comparing selected T=0h and T=24h samples against the NIST library , as well as our library of analytical standards . Compounds are noted as either confirmed with our own standards , or the best match and associated matching factor against the NIST library are reported ( Supplementary file 10 ) . Peaks were picked and integrated using the Agilent MassHunter Quantitative Analysis software . Peak areas were normalized to the internal standards . The data were processed using R version 3 . 6 . 3 and mean intensities and log2FC between time-points were calculated as described above for the untargeted metabolomics analysis . The complete custom code for all the analyses is available on GitHub: ( https://github . com/silviabrochet/Brochet_2021_eLife , copy archived at swh:1:rev:237a27f757296372f0333d298dfb7c765686fe03; Brochet , 2021 ) . The amplicon sequencing data and the RNA sequencing data are available under the NCBI Bioproject PRJNA700984 and the GEO record GSE166724 . All differential expression analysis results of this study are included in Supplementary file 10 .	Microbes colonize nearly every environment on Earth , from the ocean and soil to the inner and outer surfaces of animals , such as the gut or skin . They form communities that are usually made up of a diverse range of bacteria , often containing closely related species – a key factor for a successful community . But closely related bacteria can battle for the same resources , so it is unclear how they manage to live alongside each other without competing against one another . While diet is thought to play a key role in enabling closely related bacterial species to co-exist in the gut of an animal , experimental evidence is lacking , due to the difficulty in replicating these systems in the laboratory . One strategy for investigating microbial communities is using honeybees . A major dietary source for honeybees is pollen , which can also be applied in the laboratory to grow diverse types of bacteria found in the honeybee gut . In addition , scientists can generate bees that lack microbial communities in the gut , allowing them to add specific types of bacteria to study their impact . Brochet et al . used this approach with Western honeybees to assess whether diet enables closely related bacteria to live alongside one another in the gut . First , they colonized bees that lacked gut microbes with four closely related bacteria of the genus Lactobacillus , alone or together , and fed the bees either sugar water or sugar water and pollen . After five days , the gut bacteria were analysed . This revealed that bees fed on sugar water only had one dominant Lactobacillus species present in their gut , while bees fed with additional pollen harboured all four Lactobacillus species . Further analysis of these four bacterial species revealed that each of them activates distinct genes when grown on pollen , allowing the different species to consume specific nutrients from broken down pollen . These findings show that closely related bacteria can coexist in the gut by sharing the different nutrients provided in the diet of the host . Consequently , differences in dietary intake in honeybees and other animals may affect the diversity of gut bacteria , and potentially the health of an animal .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "microbiology", "and", "infectious", "disease" ]	2021	Niche partitioning facilitates coexistence of closely related honey bee gut bacteria
A polymorphism in the autophagy gene Atg16l1 is associated with susceptibility to inflammatory bowel disease ( IBD ) ; however , it remains unclear how autophagy contributes to intestinal immune homeostasis . Here , we demonstrate that autophagy is essential for maintenance of balanced CD4+ T cell responses in the intestine . Selective deletion of Atg16l1 in T cells in mice resulted in spontaneous intestinal inflammation that was characterized by aberrant type 2 responses to dietary and microbiota antigens , and by a loss of Foxp3+ Treg cells . Specific ablation of Atg16l1 in Foxp3+ Treg cells in mice demonstrated that autophagy directly promotes their survival and metabolic adaptation in the intestine . Moreover , we also identify an unexpected role for autophagy in directly limiting mucosal TH2 cell expansion . These findings provide new insights into the reciprocal control of distinct intestinal TH cell responses by autophagy , with important implications for understanding and treatment of chronic inflammatory disorders . Crohn’s disease ( CD ) and ulcerative colitis ( UC ) are the two most common forms of inflammatory bowel disease ( IBD ) , characterized by chronic inflammation of the gastrointestinal tract . IBD is a complex multifactorial disease that emerges on a background of many genetic and environmental factors ( Maloy and Powrie , 2011 ) . In recent years , tremendous efforts have been undertaken to identify the genetic factors that influence susceptibility to IBD . In particular , genome-wide association studies ( GWAS ) and subsequent meta-analyses have identified over 150 distinct loci that influence IBD susceptibility , many of which have revealed novel pathways in disease pathogenesis ( Van Limbergen et al . , 2014 ) . Among these , a single-nucleotide polymorphism ( SNP ) in the essential macroautophagy ( hereafter called 'autophagy' ) gene ATG16L1 was associated with an increased risk of CD ( Hampe et al . , 2007; Rioux et al . , 2007 ) . A recent study showed that the IBD predisposing T300A mutation in the coding region of ATG16L1 led to increased degradation of ATG16L1 protein and reduced autophagy ( Murthy et al . , 2014 ) , indicating that decreased autophagy may contribute to IBD development . Polymorphisms in several other autophagy-related genes , including IRGM , LRRK2 and SMURF1 , are also linked to IBD susceptibility ( Van Limbergen et al . , 2014 ) , suggesting that changes in the autophagy pathway alter intestinal homeostasis and predispose to chronic intestinal inflammation . Autophagy is a highly conserved cellular process that targets cytoplasmic components for lysosomal degradation and maintains homeostasis by recycling damaged organelles and large cytoplasmic protein aggregates . Autophagy becomes particularly important during metabolic or infectious stress ( Mizushima , 2007 ) . Atg16l1 forms an essential autophagy complex with Atg5 and Atg12 that facilitates elongation of the initial isolation membrane that results in engulfment of the cargo and formation of the autophagosome . Subsequent fusion with the lysosome facilitates degradation and allows nutrient recycling ( Mizushima et al . , 2003 ) . To identify the mechanisms through which autophagy may regulate intestinal tissue homeostasis , it is essential to understand the functional consequences of alterations in autophagy on both immune and tissue cells present in the gut . To date , several studies have examined the role of autophagy and Atg16l1 in intestinal epithelial cells and myeloid cells for intestinal homeostasis . In these studies , Atg16l1 was shown to play a role in Paneth cell physiology , as well as in bacterial handling and regulation of inflammatory IL-1β secretion by myeloid cells ( Cadwell et al . , 2008; Kuballa et al . , 2008; Saitoh et al . , 2008; Plantinga et al . , 2011 ) . However , the role of Atg16l1 in intestinal adaptive immune responses has not yet been addressed . CD4+ T cells constitute the largest population of intestinal lymphocytes and are central mediators of host protective and tolerogenic responses in the gut ( Shale et al . , 2013 ) . In particular , thymus-derived and peripherally induced Foxp3+ CD4+ regulatory T cells ( tTreg and pTreg cells , respectively ) are indispensable in promoting tolerance toward commensal and dietary antigens and for the prevention of aberrant effector T cell responses , including TH1 , TH2 and TH17 cell responses ( Izcue et al . , 2009 ) . An imbalance between effector and regulatory CD4+ T cells can promote chronic intestinal inflammation and accumulation of effector CD4+ T cells in the inflamed mucosa is a cardinal feature of IBD ( Abraham and Cho , 2009; Maloy and Powrie , 2011; Shale et al . , 2013 ) . Therefore , it is important to define factors that regulate aberrant CD4+ T cell responses in the gastrointestinal tract . Previous studies utilizing mice with T-cell-specific deletion of essential autophagy genes ( Atg3 , Atg5 , Atg7 , Beclin1 ) pointed to a key role of autophagy in T cell homeostasis , as these mice exhibited decreased frequencies and numbers of CD4+ and CD8+ T cells and defects in T cell proliferation in vitro ( Pua et al . , 2009; Stephenson et al . , 2009; Jia and He , 2011; Kovacs et al . , 2012 ) . In addition , recent studies highlighted the importance of autophagy in the development of memory CD8+ T cells ( Puleston et al . , 2014; Xu et al . , 2014; Schlie et al . , 2015 ) . However , the exact requirements for autophagy during different stages of T cell activation and differentiation remain poorly understood ( Xu et al . , 2014 ) . Given that the gastrointestinal tract is a site of continuous immune activation by external antigens and is therefore a challenging environment for the adaptive immune system , we hypothesized that a selective defect in autophagy may affect intestinal T cell homeostasis . We investigated the role of Atg16l1 in intestinal CD4+ T cells by generating mice that selectively lack Atg16l1 in T cells . Here , we show that T-cell-specific deletion of Atg16l1 results in chronic intestinal inflammation accompanied by increased humoral responses toward commensal and dietary antigens . We further demonstrate that Atg16l1-deficiency has opposing effects on intestinal CD4+ T cells subsets; markedly enhancing TH2 responses whilst decreasing Treg cell numbers . Through selective ablation of Atg16l1 in Treg cells , we established the importance of cell-intrinsic autophagy for intestinal Treg cell homeostasis . Furthermore , through complementary in vivo approaches we show that autophagy controls TH2 responses through two distinct mechanisms; through a cell-intrinsic pathway and by promoting extrinsic regulation by Treg cells . To investigate the role of autophagy in intestinal T cell homoeostasis , mice carrying loxP-flanked alleles of the essential autophagy gene Atg16l1 ( Atg16l1fl/fl ) ( Hwang et al . , 2012 ) were crossed with CD4-Cre mice , generating Atg16l1fl/fl::CD4-Cre mice ( hereafter denoted as Atg16l1ΔCD4 ) in which Atg16l1 is selectively ablated in T cells from the double-positive stage of thymic development . To verify functional deletion of Atg16l1 autophagy levels were analyzed by autophagosome formation and LC3 lipidation . CD4+ T cells isolated from control Atg16l1fl/fl mice exhibited increased LC3+ autophagosome formation after activation , as measured by intracellular LC3 accumulation in the presence of a lysosomal inhibitor ( Figure 1A ) . In contrast , there was no increase in intracellular LC3 accumulation in CD4+ T cells from Atg16l1ΔCD4 mice ( Figure 1A ) . To verify this finding using another method , we assessed LC3 lipidation by Western blot analysis ( Klionsky et al . , 2012 ) . Activated control Atg16l1fl/fl CD4+ T cells exhibited increased lipidated LC3 II levels in the presence of chloroquine , indicative of autophagy-mediated turnover of LC3 II after T cell activation ( Figure 1B ) . However , LC3 II levels in CD4+ T cells from Atg16l1ΔCD4 mice were barely affected by activation ( Figure 1B ) , confirming a block in autophagy . 10 . 7554/eLife . 12444 . 003Figure 1 . Aged Atg16l1ΔCD4 mice develop intestinal inflammation . ( A ) FACS analysis of LC3+ autophagosome formation in CD4+ T cells from cLP of Atg16l1ΔCD4 and Atg16l1fl/fl mice after overnight activation with or without α-CD3 ( 5 μg/ml ) and α-CD28 ( 1 μg/ml ) . ( B ) Western blot analysis of LC3 lipidation in naïve splenic CD4+ T cells isolated from Atg16l1ΔCD4 mice and Atg16l1fl/fl mice after 3hr activation with α-CD3 ( 5 μg/ml ) and α-CD28 ( 1 μg/ml ) with or without chloroquine ( CQ , inhibitor of lysosomal degradation , 50 μM ) . ( C ) Weight curves of Atg16l1ΔCD4 and Atg16l1fl/fl littermates . ( D ) Representative images of spleens and mesenteric lymph nodes ( mLN ) from aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates and ( E ) spleen weights of young and aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates . ( F , H ) Representative photomicrographs of haemotoxilin and eosin ( H&E ) stained sections of ( F ) jejunum and ( H ) mid-colon from young and aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates , scale bar 150 μm . ( G , I ) Quantification of ( G ) SI lengths and ( I ) mid-colon crypt lengths in aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates . Data are representative of at least three independent experiments ( A-E , F , H ) or combined from two ( G ) or three ( I ) independent experiments , with at least 3 mice per group . Data shown as mean ± s . e . m ( A , C ) . Each dot represents an individual mouse and horizontal bars denote means ( E , G ) . In ( I ) each dot represents an individual crypt measurement and horizontal bars denote means . Statistical significance was determined using two-way analysis of variance ( ANOVA ) with Bonferroni’s correction for multiple comparisons ( C ) or the Mann–Whitney test ( E , G , I ) , p<0 . 01; p<0 . 001 . SI LP– small intestine lamina propria , cLP – colonic lamina propria . Young mice: 8–12 weeks old , aged mice >5 months old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 003 Young Atg16l1ΔCD4 mice appeared normal , initially gained weight in a manner comparable to Atg16l1fl/fl littermates and exhibited normal intestinal morphology ( Figure 1C , F-H ) . However , from around 5 months of age , Atg16l1ΔCD4 mice stopped gaining weight ( Figure 1C ) , developed splenomegaly and lymphadenopathy ( Figure 1D , E ) and chronic intestinal pathology that progressed with age ( Figure 1F–I ) . Atg16l1ΔCD4 mice exhibited significant inflammation of both the small intestine ( SI ) and colon , characterized by increased SI length , marked lengthening of crypts , shortening of villi and epithelial hyperplasia ( Figure 1F–I ) . Thus , T-cell-specific Atg16l1 deletion resulted in spontaneous intestinal inflammation and systemic immune activation . To characterize the effects of Atg16l1 on intestinal and systemic T cell homeostasis independently from any confounding effects of ongoing tissue inflammation , we analyzed young ( 8–12 weeks old ) Atg16l1ΔCD4 mice before the onset of inflammatory pathology or systemic symptoms . Whilst thymic T cell production was unperturbed in Atg16l1ΔCD4 mice ( Figure 2—figure supplement 1A , B ) , frequencies of CD4+ and CD8+ T cells in peripheral lymphoid organs were significantly decreased compared to Atg16l1fl/fl littermates ( Figure 2A and Figure 2—figure supplement 1C ) . Furthermore , we observed significant decreases in intestinal T cell frequencies and numbers in the cLP and SI LP of Atg16l1ΔCD4 mice ( Figure 2A and Figure 2—figure supplement 1D ) . As CD4+ T cells are the main drivers and regulators of chronic intestinal inflammation ( Shale et al . , 2013 ) , we focused subsequent analyses on CD4+ T cells . 10 . 7554/eLife . 12444 . 004Figure 2 . Atg16l1ΔCD4 mice exhibit reciprocal dysregulation of intestinal TH2 and Treg cells before the onset of intestinal inflammation . ( A ) Frequencies of CD4+ T cells as a proportion of live cells in young Atg16l1ΔCD4 and Atg16l1fl/fl littermates . ( B ) Frequencies and ( C ) total numbers of IFN-γ+ TH1 , IL-17A+ TH17 and IL-13+ TH2 cells isolated from cLP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on CD4+ T cells ) . ( D ) Representative FACS plots of Gata3 and IL-13 ( top ) or IFN-γ and IL-17A ( bottom ) expression by cLP CD4+ T cells isolated from young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3- live cells ) . ( E ) Frequencies of Gata3+ CD4+ T cells in young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3- cells ) . ( F ) Representative FACS plots and ( G ) frequencies of Foxp3+ Treg cells in young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ cells ) . Data are combined from three or more independent experiments with at least two mice per group ( A , B , D , E , G ) or are representative of four independent experiments with at least four mice per group ( D , F ) . Each dot represents an individual mouse and horizontal bars denote means . Numbers indicate percentage of cells in gates or quadrants . Statistical significance was determined using the Mann–Whitney test , p<0 . 05; p<0 . 01; p<0 . 001 . SI LP– small intestine lamina propria , cLP – colonic lamina propria . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 00410 . 7554/eLife . 12444 . 005Figure 2—figure supplement 1 . Characterization of immune cell compartments in young Atg16l1ΔCD4mice . ( A ) Frequencies and ( B ) representative FACS plots of single positive CD4+ , single positive CD8+ , double positive ( DP ) CD4+ CD8+ and double negative ( DN ) CD4- CD8- thymocytes in young Atg16l1ΔCD4 and Atg16l1fl/fl littermates . ( C ) Frequencies of CD8+ T cells in young Atg16l1ΔCD4 and Atg16l1fl/fl littermates . ( D ) Total numbers of CD4+ T cells in spleen , mLN , and cLP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates . Data are combined from ( A , C , D ) or representative of ( B ) two or three independent experiments with at least 4 mice per group . Each dot represents an individual mouse and horizontal bars denote means . Numbers indicate percentage of cells in quadrants . Statistical significance was determined using the Mann–Whitney test , p<0 . 05; p<0 . 01; p<0 . 001 . cLP – colonic lamina propria , mLN - mesenteric lymph nodes . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 00510 . 7554/eLife . 12444 . 006Figure 2—figure supplement 2 . Atg16l1ΔCD4 mice have increased susceptibility to T-cell-mediated experimental IBD . Cohorts of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates were infected with Helicobacter hepaticus by oral gavage ( three feeds of 1x108 CFU ) and treated with anti-IL-10R mAb ( 1mg/mouse i . p . given weekly; H . h + αIL10R ) or left untreated ( Ctr ) . Two weeks post-infection mice were sacrificed for analyses . ( A ) Caecum and colon histopathology scores in H . h + αIL10R-treated Atg16l1ΔCD4 and Atg16l1fl/fl littermates . ( B ) Representative photomicrographs of H&E stained caecum of untreated Ctr ( left panels ) or H . h + αIL10R-treated ( middle and right panels ) Atg16l1ΔCD4 and Atg16l1fl/fl littermates , scale bar 150 μm . ( C ) Total lamina propria leukocyte numbers and ( D , E ) frequencies of ( D ) neutrophils ( Gr1hi CD11b+ ) and ( E ) CD4+ T cells in cLP isolated from untreated Ctr or H . h + αIL10R-treated Atg16l1ΔCD4 and Atg16l1fl/fl littermates . Data are combined from two or three independent experiments with at least two mice per group . Each dot represents an individual mouse and horizontal bars denote means . Statistical significance was determined using the Mann–Whitney test , p<0 . 05; *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 00610 . 7554/eLife . 12444 . 007Figure 2—figure supplement 3 . Elevated type 2 innate responses in Atg16l1ΔCD4 mice . ( A ) Frequencies of eosinophils ( Ly6Clow Ly6Glow CD11b+ F4/80- ) in spleen and mLN of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates . ( B ) Serum MCPT-1 levels in young Atg16l1ΔCD4 and Atg16l1fl/fl littermates were measured by ELISA . Data are combined from ( A ) or representative of ( B ) two or three independent experiments with at least four mice per group . Each dot represents an individual mouse and horizontal bars denote means . Statistical significance was determined using the Mann–Whitney test , p<0 . 05; *p<0 . 01 . mLN - mesenteric lymph nodes . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 00710 . 7554/eLife . 12444 . 008Figure 2—figure supplement 4 . Characterization of Atg16l1-deficient Treg cells . ( A ) Foxp3+ Treg cell numbers in spleen , mLN , cLP and SI LP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates . ( B ) Frequencies of Foxp3+ Treg in the thymus of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on single positive CD4+ TCRβ+ cells ) . ( C ) Frequencies of Neuropilin-1+ ( Nrp1+ ) Foxp3+ Treg cells in the cLP and SI LP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3+ T cells ) . ( D ) Frequencies of Helios+ Foxp3+ Treg cells in the cLP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3+ T cells ) . ( E ) Frequencies of IL-17A+ or IFN-γ+ of Foxp3+ Treg cells in the cLP and SI LP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on Foxp3+ CD4+ TCRβ+ T cells ) . ( F ) Expression of CD103 , CTLA4 , CD25 , CD69 and KLRG1 by cLP Foxp3+ Treg cells from young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on Foxp3+ CD4+ TCRβ+ T cells ) . ( G ) Representative FACS plots and ( H ) frequencies of Ki67+ Foxp3+ Treg cells in cLP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on Foxp3+ CD4+ TCRβ+ T cells ) . ( I ) Mean fluorescence intensity ( MFI ) of phospo-S6 ( P-S6 ) in Foxp3+ Treg cells in cLP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates . Data are combined from two independent experiments with at least four mice per group ( A , B , E ) , are representative from two independent experiments with at least four mice per group ( C , F-H ) , or are from one experiment ( D , I ) . Each dot represents an individual mouse and horizontal bars denote means . Numbers indicate percentage of cells in gates . Statistical significance was determined using the Mann–Whitney test , p<0 . 05; *p<0 . 01 . mLN - mesenteric lymph nodes , SI LP– small intestine lamina propria , cLP – colonic lamina propria . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 008 Despite reduced numbers of T cells , Atg16l1ΔCD4 mice developed exacerbated disease in a CD4+ T cell-mediated model of IBD , indicating that Atg16l1-deficient CD4+ T cells are capable of driving intestinal inflammation ( Figure 2—figure supplement 2 ) . Analysis of the effector CD4+ T cell compartment in Atg16l1ΔCD4 mice revealed that frequencies of colonic TH1 ( IFN-γ+ ) and TH17 ( IL-17A+ ) populations were comparable in young Atg16l1ΔCD4 mice and Atg16l1fl/fl littermates ( Figure 2B , D ) , although , due to decreased colonic CD4+ T cell numbers , total TH1 and TH17 numbers were significantly decreased ( Figure 2C ) . Conversely , both frequencies and total numbers of TH2 ( IL-13+ ) cells were significantly increased in cLP of young Atg16l1ΔCD4 mice ( Figure 2B–D ) . These IL-13-producing cells were bona fide TH2 cells , as they co-expressed the lineage-specifying transcription factor Gata3 ( Figure 2D , E ) . Interestingly , TH2 cell accumulation was primarily observed in the intestinal mucosa of Atg16l1ΔCD4 mice , as TH2 cell frequencies were only marginally increased in the mLN and remained unchanged in the spleen ( Figure 2E ) . However , the functional consequences of TH2 expansion extended beyond the intestine , as Atg16l1ΔCD4 mice had increased frequencies of eosinophils in both the spleen and mLN and elevated serum levels of mast cell protease 1 ( MCPT-1 ) , a marker of intestinal mast cell activation ( Figure 2—figure supplement 3A , B ) . As Foxp3+ Treg cells play a non-redundant role in control of effector T cells and the development of intestinal inflammation ( Izcue et al . , 2009 ) , we hypothesized that alterations in Tregs might underlie the spontaneous intestinal pathology that developed in aged Atg16l1ΔCD4 mice . Indeed , we found that the frequencies of intestinal Foxp3+ Treg cells in young Atg16l1ΔCD4 mice were severely reduced , both in the SI and cLP ( Figure 2F , G ) . Taking into account the decreased frequencies of CD4+ T cells in Atg16l1ΔCD4 mice ( Figure 2A ) , this equated to a reduction in Treg cell numbers by around 10-fold in the colonic LP and 4-fold in SI LP ( Figure 2—figure supplement 4A ) . In contrast , thymic development of Foxp3+ Treg cells was not diminished in young Atg16l1ΔCD4 mice ( Figure 2—figure supplement 4B ) , and we observed only minor , though significant , reductions in the frequencies and absolute numbers of Foxp3+ Treg cells in the spleen and mLN of Atg16l1ΔCD4 mice compared with Atg16l1fl/fl littermates ( Figure 2G and Figure 2—figure supplement 4A ) . Thus , Atg16l1-deficiency profoundly affected the maintenance of Foxp3+ Treg cells in the periphery , particularly within the intestinal mucosa . Expression of neuropilin-1 ( Nrp1 ) and Helios , putative markers proposed to distinguish pTreg and tTreg cells , were found at comparable levels on intestinal Foxp3+ Treg cells from Atg16l1fl/fl and Atg16l1ΔCD4 mice , suggesting that the local environment , rather than site of Treg induction , primarily dictates the requirement for autophagy in Treg cells ( Figure 2—figure supplement 4C , D ) . Assessment of how Atg16l1-deficiency affected intestinal Foxp3+ Treg cell phenotype showed that impaired autophagy significantly increased expression of effector TH cytokines in Treg cells from cLP and SI LP ( Figure 2—figure supplement 4E ) . We also found that cLP Treg cells from young Atg16l1ΔCD4 mice showed higher expression of CD103 and CTLA-4 , but showed decreased expression of the activation markers CD25 , CD69 , and the terminal differentiation marker KLRG-1 ( Cheng et al . , 2012 ) ( Figure 2—figure supplement 4F ) . In addition , intestinal Treg cells from young Atg16l1ΔCD4 mice had significantly increased expression of Ki67 and higher levels of phosphorylated S6 , suggesting that the majority were in cell cycle ( Figure 2—figure supplement 4G–I ) . Taken together , these results identify a crucial role for autophagy in the maintenance and functional regulation of intestinal Treg cells . Overall , these results demonstrate that selective ablation of Atg16l1 in T cells led to a decrease in Foxp3+ Treg cells and selective expansion of TH2 cells that preceded the onset of overt pathology . In addition , these perturbations in TH cell subsets were largely limited to the mucosal environment . We next assessed whether dysregulation in the intestinal Treg and TH2 compartment in Atg16l1ΔCD4 mice affected humoral responses . While at the limit of detection in Atg16l1fl/fl controls , serum IgE concentrations were significantly elevated in young Atg16l1ΔCD4 mice and increased further as the mice aged ( Figure 3A ) . Furthermore , levels of serum IgA and IgG1 in young Atg16l1ΔCD4 mice were also significantly elevated relative to Atg16l1fl/fl littermates ( Figure 3—figure supplement 1A ) and again increased as the Atg16l1ΔCD4 mice aged ( Figure 3B ) . In contrast , levels of isotypes not associated with TH2 help were identical in aged Atg16l1ΔCD4 mice and Atg16l1fl/fl littermates ( Figure 3B ) . Thus , there was a progressive dysregulation of TH2-associated antibody responses in Atg16l1ΔCD4 mice . Consistent with these elevated humoral responses , young Atg16l1ΔCD4 mice had higher frequencies of germinal center ( GC ) , memory B cells and plasma cells in the spleen and mLN compared to Atg16l1fl/fl littermates ( Figure 3—figure supplement 1B ) and markedly enlarged Peyer’s patches were observed in aged Atg16l1ΔCD4 mice ( Figure 3C ) . 10 . 7554/eLife . 12444 . 009Figure 3 . Atg16l1ΔCD4 mice develop elevated TH2-associated antibodies against intestinal luminal antigens . ( A ) Serum IgE concentrations in cohorts of young and aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates were measured by ELISA . ( B ) Serum antibody IgG1 , IgG2b , IgG2c , IgA and IgM isotype levels in aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates were measured by ELISA . ( C ) Representative photomicrographs of H&E stained sections of Peyer’s patch ( PP ) in the SI ( jejunum ) of aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates , scale bar 150 μm . ( D ) Serum levels of Soy-specific IgA , IgG1 , IgG2b , IgG2c antibodies in aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates were measured by ELISA . ( E ) Young Atg16l1ΔCD4 and Atg16l1fl/fl littermates were fed with ovalbumin ( OVA ) alone or with cholera toxin ( CT ) as described in methods and levels of OVA-specific serum IgE were measured 8 weeks after first challenge by ELISA . ( F ) Levels of CBir1-specific IgA , IgG1 , IgG2b and IgG2c antibodies in serum of aged Atg16l1ΔCD4 and Atg16l1fl/fl littermates were measured by ELISA , serum was diluted 50x . Data are representative from at least two independent experiments with at least three mice per group ( A-D ) or combined from two ( E ) or three ( F ) independent experiments with at least three mice per group . Each dot represents an individual mouse and horizontal bars denote means ( A , D , E , F ) . Serum isotype levels are shown as mean ± s . e . m ( B ) . Statistical significance was determined using the Mann–Whitney test ( A , D-F ) or two-way analysis of variance ( ANOVA ) with Bonferroni’s correction for multiple comparisons ( B ) , p<0 . 05; p<0 . 01; p<0 . 001 . SI – small intestine . Young mice: 8–12 weeks old , aged mice > 5 months old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 00910 . 7554/eLife . 12444 . 010Figure 3—figure supplement 1 . Dysregulated humoral responses in young Atg16l1ΔCD4 mice . ( A ) Serum antibody IgA and IgG1 isotype levels in young Atg16l1ΔCD4 and Atg16l1fl/fl littermates were measured by ELISA . ( B ) Frequencies of B cells ( B220+ ) , germinal center B cells ( GC: B220+ GL7+ CD95+ ) , memory B cells ( B220+ Gl7- IgM- IgG+ ) and plasma cells ( CD138+ ) in the spleen and mLN of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates ( gated on live CD45+ cells ) . ( C ) Serum levels of Soy-specific IgA , IgG1 , IgG2b , IgG2c antibodies in young Atg16l1ΔCD4 and Atg16l1fl/fl littermates were measured by ELISA . ( D ) Cohorts of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates were infected with Helicobacter hepaticus by oral gavage ( three feeds of 1x108 CFU ) and levels of serum Helicobacter-specific IgG1 , IgG2c and IgA antibodies were determined three weeks later by ELISA . ( E ) Cohorts of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates were orally infected with Trichuris muris ( 200 eggs ) and levels of T . muris-specific IgG1 in serum were determined 34 days later by ELISA . Data are from one experiment with at least three mice per group ( A-D ) or representative from two independent experiments with five mice per group ( E ) . Each dot represents an individual mouse and horizontal bars denote means ( B , C ) or data represent mean ± s . e . m ( A , D , E ) . Statistical significance was determined using the Mann–Whitney test ( A , C ) or two-way analysis of variance ( ANOVA ) with Bonferroni’s correction for multiple comparisons ( B , D , E ) : statistical significance was determined between infected groups ( b , c ) , p<0 . 05; *p<0 . 01 . mLN - mesenteric lymph nodes . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 010 Multiple studies have demonstrated the critical role played by Foxp3+ Treg cells in immune tolerance to dietary and microbial antigens within the intestine . Furthermore , changes in intestinal Treg and TH2 responses are associated with food hypersensitivities ( Berin and Sampson , 2013 ) . We hypothesized that the aberrant humoral responses in Atg16l1ΔCD4 mice might be directed against luminal antigens . Soy is the main protein source in chow , and we detected high levels of anti-soy IgG1 and IgA in sera from aged Atg16l1ΔCD4 mice , whereas these responses were undetectable in control Atg16l1fl/fl littermates ( Figure 3D ) . By contrast , we only detected marginal levels of soy-specific IgG2b or IgG2c in aged Atg16l1ΔCD4 sera ( Figure 3D ) . Importantly , elevated anti-soy IgG1 and IgA antibodies were already present in sera from young Atg16l1ΔCD4 mice , before the onset of intestinal inflammation ( Figure 3—figure supplement 1C ) . Despite the very high levels of total serum IgE in aged Atg16l1ΔCD4 mice , we did not detect elevated levels of anti-soy IgE ( data not shown ) . The absence of soy-specific IgE could be due to the inhibiting effects of persistent exposure to high-dose antigens on IgE responses ( Sudowe et al . , 1997; Riedl et al . , 2005 ) . Therefore , to test whether an IgE response was mounted during transient exposure to low-dose dietary antigens , we fed young Atg16l1ΔCD4 and Atg16l1fl/fl mice with ovalbumin ( OVA ) , either alone or in combination with the mucosal adjuvant cholera toxin ( CT ) . As expected , anti-OVA IgE responses were undetectable in control Atg16l1fl/fl mice fed OVA alone and were only marginally increased by co-administration of CT ( Figure 3E ) . In contrast , Atg16l1ΔCD4 mice exhibited significantly elevated levels of anti-OVA IgE after being fed OVA alone and developed >10-fold higher levels of OVA-specific IgE after feeding of OVA with CT ( Figure 3E ) . Together , these results indicate that Atg16l1ΔCD4 mice displayed aberrant TH2-associated antibody responses towards otherwise innocuous dietary protein antigens . Besides food antigens , the intestinal lumen harbors vast quantities of commensal-derived antigens . Thus , we measured antibodies directed against the flagellin antigen CBir1 , produced by commensal bacteria belonging to Clostridia cluster XIVa , as antibodies against flagellin are readily detected in sera of IBD patients ( Lodes et al . , 2004 ) . We detected significantly higher levels of CBir1-specific IgG1 and IgA in the serum of aged Atg16l1ΔCD4 mice compared to control Atg16l1fl/fl littermates , whereas anti-CBir1 IgG2b and IgG2c levels were comparable ( Figure 3F ) . Furthermore , CBir1-specific IgG1 and IgA were already detectable in young Atg16l1ΔCD4 mice ( data not shown ) . In contrast , increased TH2 cell-associated antibody responses were not mounted in young Atg16l1ΔCD4 mice following oral infection either with the Gram-negative bacterium Helicobacter hepaticus or with the nematode parasite Trichuris muris ( Figure 3—figure supplement 1D , E ) . Taken together , these results indicate that the abnormal TH2-associated antibody responses observed in Atg16l1ΔCD4 mice preceded the development of overt inflammation and were selectively induced towards commensal microbiota and dietary antigens . Given apparent opposing effects of Atg16l1 deficiency on TH2 and Treg cells , we questioned whether the disruption of autophagy pathway affects the differentiation of these T cell subsets . We found that , under TH2 or Treg polarizing conditions , differentiation of naïve CD4+ T cells isolated from Atg16l1ΔCD4 or Atg16l1fl/fl littermates toward the Gata3+ TH2 or Foxp3+ Treg cell phenotype was comparable ( Figure 4A–C ) . As TH2 cytokines can negatively affect Treg differentiation and stability ( Dardalhon et al . , 2008; Feng et al . , 2014 ) , it was possible that outgrowth of TH2 cells may also have contributed to the loss of intestinal Treg in Atg16l1ΔCD4 mice . We therefore isolated Foxp3+ Treg cells from Atg16l1ΔCD4 and Atg16l1fl/fl littermates and activated them in vitro in the presence of IL-4 and IL-13 . However , we did not find any evidence of Treg instability , as expression of Foxp3 and CD25 remained equally high in Atg16l1-deficient and WT Treg cells ( Figure 4D ) . 10 . 7554/eLife . 12444 . 011Figure 4 . Atg16l1 promotes survival of Treg cells and limits TH2 cell survival . ( A , B ) Atg16l1ΔCD4 or Atg16l1fl/fl naïve CD4+ T cells were cultured in TH0 , Treg , or TH2 polarizing conditions for 48 hr and analyzed by FACS . Representative FACS plots show ( A ) Foxp3 and ( B ) Gata3 expression ( gated on CD4+ TCRβ+ T cells ) . ( C ) Frequencies of Treg cells ( Foxp3+ ) and TH2 cells ( Gata3+ ) arising from Atg16l1ΔCD4 or Atg16l1fl/fl naïve CD4+ T cells cultured in Treg or TH2 polarizing conditions for 5 days . ( D ) Atg16l1ΔCD4 or Atg16l1fl/fl Treg cells were cultured with anti-CD3 ( 3 μg/ml ) and anti-CD28 ( 1 μg/ml ) for 48 hr , then maintained in the presence of IL-4 and IL-13 for a further 5 days before FACS analysis of Foxp3 and CD25 expression of live CD4+ T cells . ( E , F ) Naïve Atg16l1ΔCD4 or Atg16l1fl/fl CD4+ T cells were cultured with ( E ) 1 μg/ml or ( F ) 5 μg/ml anti-CD3 plus anti-CD28 ( 1 μg/ml ) for 48 hr in Treg or TH2 polarizing conditions , then maintained in polarizing conditions for a further 5 days before FACS analysis of cell survival . Histograms show gates and frequencies of live CD4+ T cells . ( G ) Representative FACS plots of viability dye and Annexin V staining of Treg cells and TH2 cells from the cLP of young Atg16l1ΔCD4 and Atg16l1fl/fl littermates , gated on CD4+ TCRβ+ Foxp3+ ( left panel ) , or CD4+ TCRβ+ Gata3+ ( right panel ) . Data are representative from two ( D , G ) or three independent experiments ( A , B , E , F ) , or are combined from three independent experiments ( C ) . Each dot represents an individual cell culture ( C ) or data are shown as mean ± s . e . m ( A , B , D-F ) . Numbers indicate percentage of cells in quadrants ( G ) . cLP – colonic lamina propria . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 011 We therefore examined whether autophagy deficiency influenced the survival of TH2 or Foxp3+ Treg cells . Thus , naïve CD4+ T cells isolated from Atg16l1ΔCD4 or Atg16l1fl/fl littermates were activated for 48 hr with anti-CD3 and anti-CD28 antibodies and then rested for 5 days . Cells were kept in TH2 or Treg polarizing conditions throughout the experiment . Following activation with different concentrations of anti-CD3 antibody , Atg16l1-deficient TH2 cells exhibited comparable or improved survival relative to WT TH2 cells ( Figure 4E , F ) . In contrast , there was a 50–75% decrease in survival of Atg16l1-deficient Treg cells when compared to Atg16l1-sufficient Treg cells activated under the same conditions ( Figure 4E , F ) . To establish whether autophagy-deficient Treg and TH2 cells exhibited similarly distinct survival profiles in vivo CD4+ T cells isolated from cLP of Atg16l1ΔCD4 or Atg16l1fl/fl littermates were stained with a viability dye and Annexin V . We observed that an increased proportion of Atg16l1-deficient intestinal Treg cells were dead or dying compared to WT Treg cells ( Figure 4G ) . In contrast , Atg16l1-deficiency had no negative effect on the viability of intestinal TH2 cells , which was comparable to WT controls ( Figure 4G ) . Together , these results indicate that Atg16l1-deficiency does not impair the differentiation or stability of Treg cells and does not promote differentiation towards the TH2 lineage . However , autophagy differentially impacts on the survival of mucosal Treg and TH2 cells . As pTreg cells are required to control TH2 responses at mucosal sites ( Mucida et al . , 2005; Curotto de Lafaille et al . , 2008; Josefowicz et al . , 2012 ) , we examined whether the enhanced TH2 phenotype in Atg16l1ΔCD4 mice could be corrected by reconstitution of the intestinal Foxp3+ Treg compartment . We restored the pTreg population in young Atg16l1ΔCD4 mice at the age of 10–12 weeks , before the onset of intestinal pathology , through adoptive transfer of congenic WT naïve CD45 . 1+ CD4+ T cells . Recipients were sacrificed 3 months later , when control Atg16l1ΔCD4 littermates had developed intestinal inflammation . We detected CD45 . 1+ donor CD4+ T cells in all adoptively transferred Atg16l1ΔCD4 mice , but the level of reconstitution varied by the organ examined . In reconstituted Atg16l1ΔCD4 mice , donor WT CD4+ T cells accounted for 37 ± 5% of total CD4+ T cells in spleen and 18 ± 2% in mLN , whereas in the cLP they represented 56 ± 4% ( Figure 5A ) . Thus , autophagy-deficient CD4+ T cells had a survival disadvantage when compared to WT CD4+ T cells within the intestinal mucosa . Overall , adoptive transfer of WT naïve CD4+ T cells restored the total frequencies of CD4+ T cells in the cLP to levels comparable to control Atg16l1fl/fl mice ( Figure 5—figure supplement 1A ) . 10 . 7554/eLife . 12444 . 012Figure 5 . Autophagy contributes to the elevated TH2 responses in Atg16l1ΔCD4 mice in a cell-intrinsic manner . Young Atg16l1ΔCD4 mice ( CD45 . 2+ ) were adoptively transferred with 4-5x106 naïve WT CD4+ T cells ( CD45 . 1+ ) and analyzed 3 months later . ( A ) Frequencies of WT ( CD45 . 1+ ) and Atg16l1-deficient ( CD45 . 2+ ) CD4+ T cells in the spleen , mLN and cLP . ( B ) Frequencies of WT ( CD45 . 1+ ) and Atg16l1-deficient ( CD45 . 2+ ) Foxp3+ Treg cells in the spleen , mLN and cLP ( gated on CD4+ TCRβ+ T cells ) . ( C ) Representative FACS plots showing gating of WT ( CD45 . 1+ ) and Atg16l1-deficient ( CD45 . 1- ) CD4+ T cells and expression of IL-13 ( TH2 ) , IFN-γ ( TH1 ) and IL-17A ( TH17 ) in the cLP ( gated on CD4+ TCRβ+ Foxp3- T cells ) . ( D ) Frequencies of WT ( CD45 . 1+ ) and Atg16l1-deficient ( CD45 . 2+ ) TH2 ( IL-13+ ) , TH1 ( IFN- γ+ ) and TH17 ( IL-17A+ ) cells among CD4+ TCRβ+ Foxp3- T cells in the cLP . ( E ) Frequencies of WT ( CD45 . 1+ ) and Atg16l1-deficient ( CD45 . 2+ ) Gata3+ CD4+ T cells in the spleen , mLN and cLP ( gated on CD4+ TCRβ+ Foxp3- T cells ) . ( F ) SI lengths and ( G ) representative photomicrographs of jejunum of control untreated Atg16l1fl/fl or Atg16l1ΔCD4 littermates and reconstituted Atg16l1ΔCD4 mice , scale bar 150 μm . ( H ) Serum IgE concentrations in control untreated Atg16l1fl/fl or Atg16l1ΔCD4 littermates and adoptively transferred Atg16l1ΔCD4 mice were measured by ELISA . Data are representative of two independent experiments with at least four mice per group ( A-E , G ) or combined from two independent experiments ( F , H ) . Each dot represents cells coming from the donor or the hosts within the individual transferred mouse ( A , B , D , E ) or each dot represents an individual mouse ( F , H ) , horizontal bars denote mean . Numbers indicate percentage of cells in gates . Statistical significance was determined using the Mann–Whitney test , p<0 . 05; *p<0 . 01 . mLN - mesenteric lymph nodes , cLP – colonic lamina propria . Young mice: 10–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 01210 . 7554/eLife . 12444 . 013Figure 5—figure supplement 1 . Reconstitution of intestinal CD4+ T cell compartments in adoptively transferred Atg16l1ΔCD4 mice . Young Atg16l1ΔCD4 mice ( CD45 . 2+ ) were adoptively transferred with 4-5x106 naïve WT CD4+ T cells ( CD45 . 1+ ) i . v . and cLP populations analyzed 3 months later . ( A ) Frequencies of total CD4+ T cells . ( B ) Frequencies and total numbers of Foxp3+ Treg cells ( gated on CD4+ TCRβ+ T cells ) . ( C ) Frequencies and total numbers of TH2 ( IL-13+ ) cells ( gated on CD4+ TCRβ+ Foxp3- T cells ) . Data are combined from two independent experiments with at least four mice per group . Each dot represents an individual mouse , horizontal bars denote mean . Statistical significance was determined using the Mann–Whitney test , p<0 . 05; *p<0 . 01 . cLP – colonic lamina propria . Young mice: 10–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 013 When we examined Foxp3+ Treg cells , the survival advantage conferred by autophagy was even more apparent , with around 50% of the donor WT naïve CD45 . 1+ T cells developing into Foxp3+ pTreg cells in spleen , mLN and cLP of Atg16l1ΔCD4 recipients . As a result , the majority of Foxp3+ Treg cells were of WT donor origin ( 67 ± 5% in spleen , 59 ± 4% in mLN and 80 ± 5% in cLP ) ( Figure 5B ) . Thus , adoptive transfer of WT-naïve CD4+ T cells resulted in efficient reconstitution of Foxp3+ pTreg cells in Atg16l1ΔCD4 mice; the total frequencies and numbers of Treg cells within the cLP of transferred mice were comparable with control Atg16l1fl/fl mice ( Figure 5—figure supplement 1B ) . As such , we could utilize this system to determine whether excessive TH2 cell accumulation in Atg16l1ΔCD4 mice was due to impaired mucosal pTreg cells or to a cell-intrinsic effect of Atg16l1-deficiency in TH2 cells . When we analyzed the frequencies of TH2 cells in the cLP of reconstituted Atg16l1ΔCD4 mice , we observed significantly higher frequencies of Gata3+ IL-13+ TH2 cells among Atg16l1-deficient CD45 . 2+ CD4+ T cells compared with the WT donor CD45 . 1+ CD4+ T cells ( Figure 5C–E ) . Indeed , frequencies of IL-13+ Atg16l1-deficient CD4+ T cells in the cLP of pTreg-reconstituted mice were comparable to those found in untreated Atg16l1ΔCD4 littermates ( Figure 5C , D ) . In contrast , there was no difference in TH17 cell frequencies between Atg16l1-deficient CD45 . 2+ and WT CD4+ T cells , and there was a significant decrease in TH1 frequencies among Atg16l1-deficient CD4+ T cells ( Figure 5C , D ) . In line with these observations , adoptively transferred Atg16l1ΔCD4 mice had comparable total frequencies and numbers of TH2 cells as observed in untreated Atg16l1ΔCD4 mice ( Figure 5—figure supplement 1C ) . Thus , provision of WT pTreg cells did not rescue the increased TH2 phenotype of Atg16l1-deficient CD4+ T cells , indicating that autophagy directly regulates TH2 cells through a cell-intrinsic mechanism . Consistent with this finding , Atg16l1ΔCD4 mice reconstituted with WT pTreg cells still developed intestinal pathology and elevated serum IgE levels comparable to those present in untreated Atg16l1ΔCD4 littermates ( Figure 5F–H ) . Given that Atg16l1-deficiency significantly reduced the number of intestinal Treg cells in Atg16l1ΔCD4 mice , we hypothesized that Treg cells may be particularly reliant on autophagy compared to other subsets of CD4+ T cells . Indeed , in WT mice we found that levels of autophagy were significantly higher in Foxp3+ Treg cells compared to Foxp3- CD4+ T cells , both constitutively and after TCR activation ( Figure 6A , B ) . Together with our observations of impaired survival of Atg16l1-deficient Foxp3+ Treg cells ( Figure 4E–G ) , this suggested an important cell-intrinsic role for autophagy in the maintenance of Treg cells . This hypothesis was further strengthened by analyses of mixed bone marrow ( BM ) chimeras where irradiated Rag1-/- mice were reconstituted with a 1:1 mixture of BM cells from Atg16l1ΔCD4 mice and congenic WT C57BL/6mice ( Figure 6—figure supplement 1A ) . In this setting , the reconstitution of CD4+ T cells was severely hampered in the absence of functional autophagy and this deficiency was most pronounced in the Treg compartment of the spleen and cLP ( Figure 6—figure supplement 1B–E ) , confirming that Atg16l1-deficiency decreases the ability of Foxp3+ Treg cells to compete with WT Treg cells in a cell-intrinsic manner . 10 . 7554/eLife . 12444 . 014Figure 6 . Aged Atg16l1ΔFoxp3 mice develop spontaneous multi-organ inflammation . ( A ) LC3+ autophagosome formation by Foxp3- CD4+ T cells and Foxp3+ Treg cells from cLP and mLN of WT mice in unstimulated cells or after overnight activation with α-CD3 ( 5 μg/ml ) and α-CD28 ( 1 μg/ml ) . ( B ) Representative LC3 staining of unstimulated cells ( gated on Foxp3+ CD4+ TCRβ+ Treg cells or Foxp3- CD4+ TCRβ+ T cells ) . ( C ) Weight curves and ( D ) spleen weights and representative images of spleen and mLN of aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates . ( E ) Representative photomicrographs of H&E stained sections of liver , spleen , stomach , SI ( jejunum ) , proximal colon and mid-colon of aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates , scale bar 150 μm . ( F ) Quantification of SI length . Data are combined from two to four independent experiments with two to five mice per group ( A , D , F ) or are representative of two to three independent experiments with two to five mice per group ( B , C , E ) . Each dot represents an individual mouse and horizontal bars denote means ( A , D , F ) . Data shown as mean ± s . e . m ( C ) . Statistical significance was determined using two-way analysis of variance ( ANOVA ) with Bonferroni’s correction for multiple comparisons ( C ) or using the Mann–Whitney test ( A , D , F ) , p<0 . 05; p<0 . 01; p<0 . 001 . mLN - mesenteric lymph nodes , SI – small intestine lamina propria , cLP – colonic lamina propria . Aged mice >5 months old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 01410 . 7554/eLife . 12444 . 015Figure 6—figure supplement 1 . Impaired reconstitution of mixed bone marrow chimeras by Atg16l1-deficient T cells . ( A ) Experimental design for the generation of mixed bone marrow ( BM ) chimeras . BM cells isolated from WT ( CD45 . 1+ ) and Atg16l1fl/fl or Atg16l1ΔCD4 ( CD45 . 2+ ) mice were injected at a 1:1 ratio into lethally irradiated ( 1100 Rad ) Rag1-/- recipients ( total of 1x107 cells per mouse ) . ( B ) Representative FACS plots showing frequencies of Atg16l1ΔCD4 or Atg16l1fl/fl ( CD45 . 2+ ) and WT ( CD45 . 2- ) CD4+ T cells in the thymus , spleen and cLP of mixed BM chimeras ( gated on CD4+ TCRβ+ T cells ) . ( C ) Frequencies of Atg16l1ΔCD4 or Atg16l1fl/fl ( CD45 . 2+ ) CD4+ T cells in mixed BM chimeras ( shown as percentage of total CD4+ TCRβ+ T cells ) . ( D ) Representative FACS plots and ( E ) frequencies of Foxp3+ Treg cells derived from Atg16l1ΔCD4 or Atg16l1fl/fl ( CD45 . 2+ ) cells in mixed BM chimeras ( gated on CD4+ TCRβ+ T cells ) . Highlighted top right quadrants indicate Treg cells derived from Atg16l1fl/fl orAtg16l1ΔCD4 BM . Data are representative from two independent experiments with at least seven mice per group . Each dot represents an individual mouse and horizontal bars denote means . Numbers indicate percentage of cells in gates . Statistical significance was determined using the Mann–Whitney test , p<0 . 01; *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 01510 . 7554/eLife . 12444 . 016Figure 6—figure supplement 2 . Analysis of Atg16l1 expression in Atg16l1ΔFoxp3 mice . ( A ) qPCR analysis of Atg16l1 exon 3 levels in sorted CD4+ Foxp3- T cells and Foxp3+ Treg cells from spleen and cLP of Atg16l1ΔFoxp3 and Foxp3Cre mice . Data are representative from two independent experiments with 5 mice per group . Atg16l1 exon 3 levels are shown as mean ± s . e . m of three technical replicates , normalised to expression of hprt . cLP – colonic lamina propria . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 016 To definitively assess the cell-intrinsic requirement for autophagy in Foxp3+ Treg cells we crossed Atg16l1fl/fl mice with mice expressing a YFP-Cre from the Foxp3 locus ( Foxp3Cre mice ) ( Rubtsov et al . , 2008 ) , generating Atg16l1fl/fl::Foxp3Cre mice ( hereafter denoted as Atg16l1ΔFoxp3 ) in which Atg16l1 is selectively ablated in Foxp3+ Treg cells . These mice allowed us to analyze the consequences of a lack of autophagy in Treg cells in the context of autophagy-competent CD4+ T effector cells . As expected , Atg16l1ΔFoxp3 mice showed a significant reduction of Atg16l1 expression in Foxp3+ Treg cells , but not in CD4+ Foxp3- T cells ( Figure 6—figure supplement 2A ) . Although Atg16l1ΔFoxp3 mice appeared normal in early life , at around 5 months of age they developed a severe spontaneous inflammatory disease characterized by progressive weight loss , splenomegaly , lymphadenopathy and leukocyte infiltration in multiple tissues ( Figure 6C–E ) . The gastrointestinal tract was particularly affected in aged Atg16l1ΔFoxp3 mice , with marked inflammation in the SI and colon ( Figure 6E , F ) . Intestinal inflammation in aged Atg16l1ΔFoxp3 mice was characterized by massive accumulation of activated CD4+ T cells in the intestinal LP and mLN ( Figure 7A–D and Figure 7—figure supplement 1A ) . The cLP infiltrate in aged Atg16l1ΔFoxp3 mice contained a mixed population of TH1 , TH17 and TH2 effector cells , with a significant increase in the frequencies of IL-13+ CD4+ TH2 cells ( Figure 7E , F ) , although this TH2 bias was not present in young Atg16l1ΔFoxp3 mice ( Figure 7—figure supplement 1B ) . In addition , we observed increased frequencies of Gata3+ CD4+ T cells in the spleen , mLN and cLP of aged Atg16l1ΔFoxp3 mice ( Figure 7G ) . Analyses of humoral responses in aged Atg16l1ΔFoxp3 mice revealed significantly elevated levels of circulating IgE and IgA , however IgG1 levels were not increased ( Figure 7H and Figure 7—figure supplement 1C ) . Thus , selective ablation of Atg16l1 in Foxp3+ Treg cells led to intestinal inflammation that was characterized by accumulation of all TH effector types , with a disproportionate increase in TH2 responses in aged mice . However , the breadth and magnitude of TH2-associated responses were less pronounced in Atg16l1ΔFoxp3 mice compared to those observed in Atg16l1ΔCD4 mice . 10 . 7554/eLife . 12444 . 017Figure 7 . Atg16l1ΔFoxp3 mice cannot control pro-inflammatory TH effector responses . ( A ) Representative immunofluorescence images of small intestine and proximal and mid colon of aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates stained for CD3 ( red ) , β-catenin ( green ) and DAPI ( blue ) . ( B ) Frequencies and ( C ) total numbers of cLP CD4+ TCRβ+ T cells in aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates . ( D ) Frequencies of effector ( CD44+CD62L- ) CD4+ T cells in the mLN and cLP of aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3- T cells ) . ( E ) Frequencies and ( F ) total numbers of TH1 ( IFN-γ+ ) , TH17 ( IL-17A+ ) , TH2 ( IL-13+ ) T cells in the cLP of aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3- T cells ) . ( G ) Frequencies of Gata3+ CD4+ T cells in aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3- T cells ) . ( H ) Serum IgE concentrations in Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates were measured by ELISA . Data are combined from two to four independent experiments with two to five mice per group ( B-H ) or are representative of two independent experiments with two to five mice per group ( A ) . Each dot represents an individual mouse and horizontal bars denote means . Statistical significance was determined using the Mann–Whitney test p<0 . 05; p<0 . 01; p<0 . 001 . mLN - mesenteric lymph nodes , cLP – colonic lamina propria . Young mice: 8–12 weeks old , aged mice >5 months old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 01710 . 7554/eLife . 12444 . 018Figure 7—figure supplement 1 . Additional characterization of Atg16l1ΔFoxp3 mice . ( A ) Representative FACS plots of CD44 and CD62L expression by CD4+ T cells from cLP of aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3- T cells ) . ( B ) Frequencies of TH1 ( IFN-γ+ ) , TH17 ( IL-17A+ ) , TH2 ( IL-13+ ) T cells in the cLP of young Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ Foxp3- T cells ) . ( C ) Serum antibody IgA , IgG1 , IgG2b , IgG2c isotype levels in aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates . Data are combined from three independent experiments with two to five mice per group ( B ) or are representative from two to three independent experiments with two to five mice per group ( A , C ) . Each dot represents an individual mouse and horizontal bars denote means . Numbers indicate percentage of cells in gates . Data shown as mean ± s . e . m ( C ) . Statistical significance was determined using the Mann–Whitney test , p<0 . 05; p<0 . 01; p<0 . 001 . mLN - mesenteric lymph nodes , cLP – colonic lamina propria . Young mice: 8–12 weeks old , aged mice >5 months old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 018 When we examined the Treg cell compartment in Atg16l1ΔFoxp3 mice , we found significantly decreased frequencies of Foxp3+ Treg cells in the spleen and mLN compared to Atg16l1fl/fl littermates , although thymic Treg cell frequencies were similar ( Figure 8A ) . As found in Atg16l1ΔCD4 mice , intestinal LP Foxp3+ Treg cells were severely depleted in Atg16l1ΔFoxp3 mice and those remaining exhibited significantly increased expression of effector TH cytokines ( Figure 8A , B and Figure 8—figure supplement 1A ) . Thus , Treg cell-specific deletion of Atg16l1 recapitulated the Treg cell deficits observed in Atg16l1ΔCD4 mice , showing that cell-intrinsic autophagy is essential for peripheral Treg cell homeostasis , especially in the intestine . 10 . 7554/eLife . 12444 . 019Figure 8 . Cell-intrinsic autophagy is required for metabolic adaptation and survival of intestinal Foxp3+ Treg cells . ( A ) Foxp3+ Treg cell frequencies among CD4+ TCRβ+ T cells in Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates and ( B ) representative FACS plots of Foxp3 expression in cLP CD4+ T cells from young Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates ( gated on CD4+ TCRβ+ T cells ) . ( C ) qPCR analysis of glycolytic gene levels in sorted Foxp3+ Treg cells from spleen and cLP of young Atg16l1ΔFoxp3 and Foxp3Cre mice ( sorted for CD4+ TCRβ+ YFP+ ) . ( D ) qPCR analysis of FAS and FAO gene levels in Foxp3+ Treg cells from the spleen and cLP of young Atg16l1ΔFoxp3 and Foxp3Cre mice ( sorted for CD4+ TCRβ+ YFP+ ) . FAS: fatty acid synthesis , FAO: fatty acid oxidation , Glut1: glucose transporter 1 , Slc16ac: solute carrier family 16 member 3 ( lactic acid and pyruvate transporter ) , Tpi1: triosephosphate isomerase 1 , Aldo–α: aldolase α , Ldh-α: lactate dehydrogenase α , Gpi1: Glucose phosphate isomerase 1 , Pgk1: Phosphoglycerate kinase 1 , Acc1: acetyl-CoA carboxylase 1 , Acc2: acetyl-CoA carboxylase 2 , Srebf1: sterol regulatory element binding transcription factor 1 , Srebf2: sterol regulatory element binding transcription factor 2 , Fdft1: farnesyl-diphosphate farnesyltransferase 1 , Fabp: Fatty acid-binding protein . Data are representative from two ( C , D ) or three independent experiments ( A , B ) . Each dot represents individual mouse ( A ) or data are shown as mean ± s . e . m ( C , D ) . Gene expression levels are shown as mean ± s . e . m of three technical replicates ( C , D ) . Numbers indicate percentage of cells in gates ( B ) . cLP – colonic lamina propria . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 01910 . 7554/eLife . 12444 . 020Figure 8—figure supplement 1 . Atg16l1-deficient colonic Treg cells exhibit increased cytokine secretion . ( A ) Frequencies of IFN-γ+ , IL-17A+ , IL-4+ , IL-13+ Foxp3+ Treg cells in the cLP of aged Atg16l1ΔFoxp3 and Atg16l1fl/fl littermates ( gated on Foxp3+ CD4+ TCRβ+ T cells ) . ( B ) qPCR analysis of Bcl2 , Bim and Bax levels in Foxp3+ Treg cells from young cLP of Atg16l1ΔFoxp3 and Foxp3Cre mice ( sorted for CD4+ TCRβ+ YFP+ ) . Data are combined from three independent experiments with two to five mice per group ( A ) or are representative from two independent experiments ( B ) . Each dot represents an individual mouse and horizontal bars denote means ( A ) . Gene expression levels are shown as mean ± s . e . m of three technical replicates ( B ) . cLP – colonic lamina propria . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 02010 . 7554/eLife . 12444 . 021Figure 8—figure supplement 2 . Increased lipid uptake by intestinal Treg cells . ( A , B ) Atg16l1fl/fl and Atg16l1ΔCD4 littermates were injected i . p . with 50 μg of fluorescent 16-carbon fatty acid analog BODIPY C-16 and culled 1 hr later and tissue was collected for analysis by flow cytometry . ( A ) Representative FACS plots and ( B ) quantification of C16-Bodipy uptake by Foxp3+ Treg cells in the spleen , mLN and cLP ( gated on Foxp3+ CD4+ TCRβ+ T cells ) . ( C ) Representative FACS plots and ( D ) quantification of CD36 expression by Foxp3+ Treg cells in the spleen , mLN and cLP of Atg16l1fl/fl and Atg16l1ΔCD4 littermates ( gated on Foxp3+ CD4+ TCRβ+ T cells ) . Data are combined from ( B , D ) or are representative of ( A , C ) two independent experiments with 3–5 mice per group . Each dot represents an individual mouse and horizontal bars denote means . Statistical significance was determined using the Mann–Whitney test , p<0 . 05 . mLN - mesenteric lymph nodes , cLP – colonic lamina propria . Young mice: 8–12 weeks old . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 02110 . 7554/eLife . 12444 . 022Figure 8—figure supplement 3 . TH2 cells exhibit an enhanced glycolytic metabolic profile that is independent of autophagy . ( A ) Representative FACS plot and quantification of the cell size ( FSC-H ) of naïve ( CD44- CD62L+ ) CD4+ T cells from the spleen of Atg16l1ΔCD4 or Atg16l1fl/fl littermates . ( B ) Basal level of oxygen consumption rate ( OCR ) and extracellular acidification rate ( ECAR ) in naïve ( CD44- CD62L+ ) unstimulated CD4+ T cells isolated from the spleen of Atg16l1ΔCD4 or Atg16l1fl/fl littermates measured using the Seahorse metabolic flux analyzer . ( C , D ) Expression of c-Myc ( C ) and cell size ( D ) was analyzed by FACS in Atg16l1ΔCD4 or Atg16l1fl/fl CD4+ T cells that were cultured in TH2 or Treg-polarizing conditions for 3 days and rested for one day in the presence of polarizing cytokines . ( E ) Basal level of OCR and ECAR were measured by Seahorse metabolic flux analyzer in Atg16l1ΔCD4 or Atg16l1fl/fl CD4+ T cells cultured in TH2 or Treg- polarizing conditions for 3 days and rested for 2 days in the presence of polarizing cytokines . ( F ) qPCR analysis of glycolytic gene levels in Atg16l1ΔCD4 or Atg16l1fl/fl CD4+ T cells cultured in TH2 or Treg polarizing conditions for 3 days and rested for 1 day in the presence of polarizing cytokines . Data are combined from two independent experiments ( A ) , or are representative of two independent experiments ( B-D ) , or are from one experiment ( E , F ) . Each dot represents an individual mouse ( A ) or individual cell culture ( D ) . ECAR and OCAR data represent mean ± s . e . m values of T cell populations that were assayed in triplicates or quadruplicates ( B , E ) . Gene expression data of triplicate cultures represent normalized expression values for each gene that were scaled to a mean of 0 and a standard deviation of 1 ( F ) . Statistical significance was determined using the Mann–Whitney test ( A ) or unpaired Student’s t –test ( B , D , E ) , p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12444 . 022 Finally , we investigated mechanisms that might underlie the striking survival defect of Atg16l1-deficient intestinal Treg cells . Analyses of key regulators of apoptosis revealed that Atg16l1-deficient Treg cells isolated from spleen and cLP had comparable expression of pro-apoptotic ( Bim , Bax ) and anti-apoptotic ( Bcl2 ) genes as those isolated from control mice ( Figure 8—figure supplement 1B ) . As recent evidence suggests that tissue-resident Treg cell populations may exhibit specialized metabolic adaptations ( Burzyn et al . , 2013 ) , we compared the expression of metabolic genes by WT and Atg16l1-deficient Treg cells . Analyses of genes involved in glycolysis , fatty acid synthesis ( FAS ) and fatty acid oxidation ( FAO ) , revealed that Atg16l1-deficient Treg cells had higher expression of glycolytic genes , including Glut1 , Slc16a3 , Tpi1 , Ldh-a , Aldo-a , Gpi1 and Pgk1 , than control Treg cells ( Figure 8C ) . Strikingly , this augmented glycolytic signature was much more pronounced in Atg16l1-deficient Treg cells isolated from cLP versus those from the spleen ( Figure 8C ) . Conversely , expression of many key genes involved in FAS/FAO , including Acc1 , Acc2 , Srebf1 , Srebf2 , Fdft1 , Fabp1 , Fabp4 and Fabp5 was markedly decreased in Atg16l1-deficient Treg cells ( Figure 8D ) . Again , these differences were most pronounced in the intestine; WT cLP Treg cells showed increased FAS/FAO gene expression compared to their spleen counterparts , whereas Atg16l1-deficient cLP Treg cells were not able to up-regulate the expression of FAS/FAO genes ( Figure 8D ) . Thus , Atg16l1-deficiency profoundly influenced the expression of metabolic genes in intestinal Treg cells , with an altered balance of glycolytic and FAS/FAO gene expression . Further evidence of increased reliance on lipid metabolism by colonic Treg cells was provided by our observation that Treg cells isolated from the cLP showed markedly increased lipid uptake in comparison to mLN or spleen Treg cells ( Figure 8—figure supplement 2A , B ) . A similar pattern was observed when we assayed expression of CD36 , a fatty acid translocase that enhances FA uptake: colonic Treg cells showed increased expression of CD36 compared to splenic and mLN Treg cells ( Figure 8—figure supplement 2C , D ) . Interestingly , we found that Atg16l1-deficient Treg cells showed comparable levels of lipid uptake and CD36 expression as their autophagy-sufficient counterparts ( Figure 8—figure supplement 2A–D ) , suggesting that autophagy does not affect lipid uptake per se but rather affects lipid metabolism . Together , these results demonstrate that cell-intrinsic autophagy is indispensable for Foxp3+ Treg cell maintenance and function in peripheral tissues , particularly to suppress inflammatory responses within the gastrointestinal tract . Decreased survival of Atg16l1-deficient Treg cells was associated with an altered metabolic profile , suggesting that autophagy plays an integral role in facilitating the metabolic adaptions required for long-term Treg cell survival in the intestine . We next explored whether autophagy had a general effect on T cell metabolic profile and whether this might explain the differential effects on TH2 cells and Treg cells . Evidence that this might be the case came from our observation that Atg16l1-deficient naïve CD4+ T cells exhibited increased cell size compared with naïve CD4+ T cells isolated from Atg16l1fl/fl littermates ( Figure 8—figure supplement 3A ) . We therefore measured oxygen consumption rate ( OCR ) , which is an indicator of oxidative phosphorylation ( OXPHOS ) , and extracellular acidification rate ( ECAR ) , an indirect indicator of aerobic glycolysis . We found that Atg16l1-deficient naïve CD4+ T cells exhibited significantly increased OCR and ECAR , metabolic changes that are typically observed in activated CD4+ T cells and are associated with increased aerobic glycolysis ( Figure 8—figure supplement 3B ) . As TH2 cells have previously been reported to display an increased glycolytic rate compared to other TH subsets ( Michalek et al . , 2011; Yang et al . , 2013 ) , we hypothesized that they may be more resistant to the increased glycolysis that is induced in the absence of autophagy . As it was not possible to sort TH2 cells from the cLP , we performed this analysis on in vitro cultures of TH2 and Treg cells . We found that TH2 cells were larger than Treg cells , expressed higher levels of c-Myc , a critical regulator of metabolic reprograming in activated T cells , and had markedly higher ECAR , all indicative of enhanced aerobic glycolysis ( Figure 8—figure supplement 3C–E ) . Furthermore , while Atg16l1-deficient Treg cells showed higher expression of c-Myc , significantly increased levels of ECAR and OCR , and were larger than their control Atg16l1-sufficient counterparts , we observed constitutively high and comparable levels of glycolysis in Atg16l1-deficient and Atg16l1-sufficient TH2 cells ( Figure 8—figure supplement 3C–E ) . These patterns were recapitulated when expression of key metabolic genes were analyzed; TH2 cells showed high expression of a panel of glycolytic genes irrespective of their autophagy Atg16l1 genotype , whereas Treg cell expression of glycolytic genes was generally lower , unless the Treg cells were autophagy-deficient ( Figure 8—figure supplement 3F ) . Taken together , these results suggest that the enhanced glycolytic metabolism constitutively employed by TH2 cells makes them more resistant to the metabolic changes that occur in the absence of autophagy . The unique challenges of the intestine necessitate complex mechanisms of tolerance and immune regulation to maintain homeostasis ( Izcue et al . , 2009 ) . As altered mucosal CD4+ T cell responses are implicated in intestinal diseases of increasing prevalence , including food allergies and IBD ( Maloy and Powrie , 2011; Berin and Sampson , 2013 ) , it is important to understand the factors that control effector and regulatory T cell homeostasis in the gut . Here , we identify Atg16l1 and autophagy as a new critical pathway regulating intestinal Treg and TH2 responses . Recent studies addressing the role of autophagy in distinct leukocyte populations have highlighted T cells as being very sensitive to perturbations in the autophagy pathway ( Ma et al . , 2013 ) . Our data extend these findings by showing that autophagy is particularly important for the survival of CD4+ T cells within the gut environment , as Atg16l1 deletion in T cells led to a severe reduction of CD4+ T cell numbers in the intestinal LP . This deficit was confirmed in mixed bone marrow chimeras , where Atg16l1-deficient CD4+ T cells failed to reconstitute the intestinal LP compartment , and by the rapid outgrowth of adoptively transferred WT CD4+ T cells in the colonic LP of Atg16l1ΔCD4 recipients . However , despite the reduction in intestinal CD4+ T cells , Atg16l1ΔCD4 mice spontaneously developed progressive , chronic intestinal inflammation . To confirm their increased predisposition to develop intestinal pathology , we used an experimental model of IBD triggered by infection with Helicobacter hepaticus and concomitant treatment with anti-IL-10R mAbs ( Song-Zhao and Maloy , 2014 ) . This model induces severe typholocolitis that is T cell dependent and displays several features of human IBD pathology and does not require any specific genetic manipulation or chemical barrier disruption . We found increased intestinal pathology in Atg16l1ΔCD4 mice , confirming that Atg16l1-deficient T cells could mediate potent inflammatory responses in the gut . Thus , selective autophagy deficiency within T cells decreases the competitiveness of these cells and simultaneously predisposes to intestinal inflammation . We found that Atg16l1ΔCD4 mice exhibited a drastic reduction in Foxp3+ Treg populations in the cLP and SI LP , together with marked changes in intestinal Treg phenotype , including increased cell cycling and aberrant production of TH effector cytokines . The role of autophagy in Foxp3+ Treg cells is not well defined . T cell-specific ablation of Vps34 , which encodes a class III phosphatidylinositol 3-kinase that promotes autophagy , resulted in decreased frequencies of Treg cells in the thymus , spleen and lymph nodes ( Parekh et al . , 2013 ) . However , as Vps34 also has autophagy-independent functions ( Backer , 2008 ) , it was unclear as to what extent these changes were due to impaired autophagy . Furthermore , we did not find any deficit in thymic Treg cell development in Atg16l1-deficient T cells . However , we observed that Treg cells isolated from the mLN and colonic LP had increased levels of autophagy compared to effector T cells , suggesting that autophagy is particularly important for the maintenance of Treg cells in the periphery . Indeed , we demonstrated that cell-intrinsic autophagy is indispensible for the maintenance and function of Foxp3+ Tregcells in the gastrointestinal tract , as selective deletion of Atg16l1 in the Foxp3+ Treg compartment in Atg16l1ΔFoxp3 mice led to a loss of intestinal Foxp3+ Treg cells and to severe inflammation of the small intestine and colon . In this context , it is pertinent that rapamycin , which induces autophagy through its inhibitory activity on mTOR , has been shown to promote expansion of Treg cells in vitro and in vivo ( Pollizzi and Powell , 2015 ) . Similarly , several small-molecule inducers of autophagy were shown to selectively promote the development of Treg cells in vitro ( Shaw et al . , 2013 ) . Taken together with our findings , these observations suggest that boosting autophagy may represent a rational therapeutic approach to enhance Treg responses in the intestine . How does autophagy intrinsically regulate Treg cell homeostasis ? Our data indicate that autophagy is not required for the differentiation of Foxp3+ Treg cells in vitro or in vivo for thymic generation of Treg cells in vivo . However , we found that Atg16l1-deficient Treg cells showed significantly decreased survival compared to WT Treg cells both in vitro and in vivo . As recent evidence indicates that Treg cells utilize a distinct metabolic program that favors lipid oxidation for energy provision ( MacIver et al . , 2013 ) , one potential explanation is that autophagy regulates Treg cell metabolism and thereby their survival . Indeed , we found that Atg16l1-deficient Treg cells expressed a distinct metabolic profile to their WT counterparts , exhibiting increased expression of genes involved in glycolysis and reduced expression of genes involved in FAS/FAO . Fatty acid metabolism is emerging as a potent regulator of T cell responses and preferential utilization of FAO has been linked to Treg cell induction ( Lochner et al . , 2015 ) . Although a recent report indicated that de novo FAS was not required for Foxp3+ Treg cell differentiation ( Berod et al . , 2014 ) , optimal in vivo Treg cell function was associated with intrinsic lipid synthesis ( Zeng et al . , 2013 ) . Furthermore , autophagy has been implicated in the regulation of fatty acid metabolism ( Singh et al . , 2009; Lizaso et al . , 2013; Kaur and Debnath , 2015 ) and recent studies found that autophagy plays a key role in the generation of CD8+ memory T cells ( Puleston et al . , 2014; Xu et al . , 2014 ) , which are heavily dependent on FAO for survival ( Pearce et al . , 2009; O'Sullivan et al . , 2014 ) . Thus , autophagy could play a similar survival role in Treg cells , by facilitating the degradation of intracellular lipid stores to release FAs that fuel FAO . Additionally , as degradation of intracellular lipids by autophagy is important to avoid lipotoxicity ( Galluzzi et al . , 2014 ) , defective autophagy could lead to a toxic build up of intracellular lipids in intestinal Treg cells . The imbalance between glycolysis and FAS/FAO observed in autophagy-deficient Treg cells could indicate that these cells have stalled in the activated/effector state and are unable to make the metabolic adaptations necessary for long-term survival . This is supported by our data showing that a higher proportion of autophagy-deficient Treg cells appear to be in cell cycle , but they have reduced expression of terminal differentiation markers . Consistent with our findings , a recent study reported that autophagy deficiency in Treg cells resulted in increased mTORC1 activation and glycolysis , leading to phenotypic instability , including expression of pro-inflammatory cytokines ( Wei et al . , 2016 ) . However , the molecular mechanism behind decreased survival of autophagy-deficient Treg cells was not elucidated ( Wei et al . , 2016 ) . It is striking that autophagy deficiency had a more detrimental effect on intestinal Treg cells than on those found in secondary lymphoid organs . Recent evidence suggests that tissue-resident Treg cells undergo tissue-specific adaptations , and metabolic changes are emerging as an important facet of such reprogramming ( Burzyn et al . , 2013; Liston and Gray , 2014 ) . Taken together , our results suggest that autophagy endows intestinal Treg cells with the metabolic flexibility required to survive in the gut tissue , where essential growth factors may be in short supply ( Pearce et al . , 2013 ) . Paralleling decreased Treg responses in Atg16l1ΔCD4 mice , we observed a selective expansion of TH2 cells in the intestinal LP that was already present in young mice and preceded the onset of overt pathology . Our subsequent analyses indicated that autophagy limits mucosal TH2 cells through both cell-intrinsic and cell-extrinsic ( Treg-mediated ) regulation . One possibility is that Atg16l1-deficient TH2 cells may be somewhat resistant to Treg suppression . However , when we reconstituted pTreg cells in Atg16l1ΔCD4 mice we observed a negative correlation between the numbers of intestinal Treg cells and TH2 cells ( data not shown ) , suggesting that autophagy-deficient TH2 cells are partially controlled by Treg cells . Our data strongly suggest that the intrinsic survival advantage of Atg16l-deficient TH2 cells is primarily responsible for their outgrowth in the intestine . Indeed , we observed increased survival of Atg16l1-deficient TH2 cells in vitro , suggesting that autophagy might directly inhibit TH2 cell expansion . This concept is consistent with a previous study that reported enhanced survival of TH2 cells in vitro when autophagy was inhibited and that autophagy mediated death of TH2 cells during growth-factor withdrawal ( Li et al . , 2006 ) . However , we provide evidence for an additional mechanism that could explain the preferential expansion of Atg16l1-deficient TH2 cells in the intestine , related to the unique ability of TH2 cells to cope with prolonged high levels of glycolysis . Our data contribute to accumulating evidence that a shift toward glycolysis is a general phenomenon observed when the autophagy pathway is perturbed in T cells . We observed characteristic signs of increased glycolysis in Atg16l1-deficient naïve CD4+ T cells and Treg cells , such as increases in cell size , c-Myc levels and expression of glycolytic genes , as well as elevated ECAR . Others have reported a similar glycolytic shift in autophagy-deficient CD8+ memory T cells ( Puleston et al . , 2014 ) and Treg cells ( Wei et al . , 2016 ) . Interestingly , TH2 cells have previously been shown to display an increased glycolytic rate compared to other TH subsets ( Michalek et al . , 2011; Yang et al . , 2013 ) . We confirmed the high levels of constitutive glycolysis in TH2 cells and showed that these were comparable in Atg16l1-deficient and control TH2 cells . Moreover , Gata3 activation was previously linked to induction of glycolysis after TCR activation in T cells , through induction of c-Myc , a critical regulator of metabolic reprograming ( Wang et al . , 2011; Wang et al . , 2013; Wan , 2014 ) . We therefore propose that in TH2 cells Gata3 orchestrates metabolic adaptations that enable these cells to cope with prolonged high levels of glycolysis , thus making them resistant to metabolic changes enforced by autophagy deficiency . Overall , our results indicate that autophagy is a key pathway through which TH2 responses are restrained in vivo . A lack of this restraint leads to a gradual loss of tolerance to intestinal antigens , as the excessive TH2 responses in Atg16l1ΔCD4 mice led to production of IgG1 and IgA antibodies toward commensal microbiota and dietary antigens that increased with age . Furthermore , Atg16l1ΔCD4 mice developed very high levels of circulating IgE , and mounted de novo IgE antibody responses toward introduced dietary antigen . As polymorphisms in autophagy genes are linked to IBD susceptibility , our results point towards a novel mechanism that links impaired autophagy to intestinal inflammation through dysregulation of mucosal T cell responses . Previous studies focused on the role of ATG16L1 and autophagy in myeloid cells and the intestinal epithelium . They suggested that impaired autophagy could result in reduced intestinal barrier integrity due to impaired Paneth cell function within the intestinal epithelial layer and elevated cytokine responses by macrophages and dendritic cells ( Cadwell et al . , 2008; Saitoh et al . , 2008; Lassen et al . , 2014 ) . Our data add a further layer to the control of intestinal homeostasis by autophagy , by showing that autophagy impairment alters the local T cell compartment and promotes T cell driven intestinal pathology . We present compelling evidence that autophagy deficiency in Treg cells leads to a deficit in intestinal Treg cells and the development of severe intestinal pathology . Although the contribution of the TH2 axis to IBD remains unclear ( Strober et al . , 2002; Shale et al . , 2013 ) , polymorphisms in IL-4 , IL-5 and IL-13 have been implicated by GWAS in both CD and UC ( Van Limbergen et al . , 2014 ) and elevated levels of antibodies recognizing food and commensal antigens have been detected in IBD patients ( Lodes et al . , 2004; Cai et al . , 2014 ) . Moreover , as defective Treg and increased TH2 responses at the mucosa are observed in food allergies and asthma , our findings might also have implications for these conditions . Indeed , epidemiological studies show an overlap between IBD and TH2 driven diseases , such as atopic dermatitis and asthma ( Lees et al . , 2011 ) . Furthermore , polymorphisms in the essential autophagy gene Atg5 have recently been implicated in asthma susceptibility ( Martin et al . , 2012; Poon et al . , 2012 ) . Autophagy is an attractive therapeutic target and several autophagy modulating compounds are already in clinical trials for the treatment of various disorders ( Jiang and Mizushima , 2014 ) . Furthermore , natural dietary-derived compounds , including retinoid acid ( Isakson et al . , 2010 ) and vitamin D ( Yuk et al . , 2009 ) , have been shown to enhance autophagy . Taken together with our results , these findings raise the possibility that activation of autophagy through dietary or pharmacological modulation might have beneficial effects in disorders with a signature of decreased Treg and elevated TH2 responses , including intestinal inflammation and various hypersensitivities . Atg16l1fl/fl mice were generated and provided by the H . Virgin laboratory ( Washington University , Saint Louis , MO ) , as described ( Hwang et al . , 2012 ) . Atg16l1fl/fl mice were crossed to B6 . Cg-Tg ( Cd4-cre ) 1Cwi/BfluJ ( CD4-Cre mice ) and B6 . 129 ( Cg ) -Foxp3tm4 ( YFP/cre ) Ayr/J ( Foxp3Cre mice , Jackson Laboratory , Bar Harbor , ME ) to generate Atg16l1ΔCD4 and Atg16l1ΔFoxp3 mice , respectively . All above strains , together with B6 . SJL-CD45 . 1 ( CD45 . 1+ ) , B6 Rag1-/- ( Jackson Laboratory ) , and B6 Foxp3hCD2 mice ( Komatsu et al . , 2009 ) were bred and maintained under specific pathogen-free conditions . Unless stated otherwise , mice were analyzed at 8–12 weeks ( young mice ) or > 5 months of age ( aged mice ) . In the gene expression analysis Atg16l1ΔFoxp3 mice and Foxp3Cre mice were co-housed and age- and sex- matched . In all other experiments mice used were age- and sex-matched littermates that were kept co-housed throughout the experiments . Experimental T cell-mediated colitis was induced by infection with Helicobacter hepaticus and concomitant IL-10R blockade as described ( Song-Zhao and Maloy , 2014 ) . Briefly , mice were infected with H . hepaticus ( 108 CFU per mouse ) by oral gavage on three consecutive days and anti-IL-10R mAb ( 1B1 . 2 ) was administrated via i . p . injection ( 1 mg per mouse ) on the first and seventh day of the infection . Mice were sacrificed 2 weeks after colitis induction . Mice were euthanized at indicated time points whereupon tissue sections were fixed in buffered 10% formalin and paraffin-embedded . Sections were then cut and stained with hematoxylin and eosin . Histological analysis of intestinal inflammation was performed as described ( Song-Zhao and Maloy , 2014 ) . Briefly , inflammation was graded semi-quantitatively on a scale from 0 to 3 , for four criteria; ( a ) epithelial hyperplasia and goblet cell depletion , ( b ) lamina propria leukocyte infiltration , ( c ) area of tissue affected , and ( d ) markers of severe inflammation , including crypt abscesses , sub- mucosal inflammation , and ulceration . Scores for individual criteria were totaled for an overall inflammation score between 0 and 12 . Cell suspensions were prepared from the thymus , spleen , mLN , bone marrow and intestinal lamina propria as previously described ( Uhlig et al . , 2006 ) . The following antibodies from eBioscience ( Hatfield , UK ) were used: anti-CD16/32 ( 93 ) , anti-CD4 ( GK1 . 5 ) , anti-CD8α ( 53 . 6 . 7 ) , anti-TCRβ ( H57-597 ) , anti-CD45 ( 30-F11 ) , anti-CD44 ( 1M7 ) , anti-CD62L ( MEL-14 ) , anti-CD45 . 1 ( A20 ) , anti-CD45 . 2 ( 104 ) , anti-CD103 ( 2E7 ) ) , anti-CD69 ( H1 . 2F3 ) , anti-KLRG1 ( 2F1 ) , anti-CD25 ( 7D4 ) , anti-CD36 ( No . 72–1 ) , anti-hCD2 ( RPA-2 . 10 ) , anti-CTLA4 ( UC10-4B9 ) , anti-GR . 1 ( RB6-8C5 ) , anti-CD11b ( M1/70 ) , anti-Siglec F ( E50-2440 ) , anti-Gata3 ( TWAJ ) , anti-Foxp3 ( FJK-16s ) , anti-Ki67 ( SolA15 ) , anti-Helios ( 22F6 ) , anti- Bcl2 ( 10C4 ) , anti-PS6 ( cupk43k ) , anti-IFN-γ ( XMG1 . 2 ) , anti-IL-17A ( eBio17B7 ) , anti-IL-13 ( eBio13A ) . The following antibodies were from BioLegend ( San Diego , CA ) : anti-CD138 ( 281–2 ) , anti-CD161 ( PK136 ) , anti-F4/80 ( BMB ) , anti-CD11b ( M1/70 ) . The following antibodies were from BD Biosciences ( San Jose , CA ) : anti-B220 ( RA3 6B2 ) , anti-GL7 ( GL7 ) , anti-CD95 ( Jo2 ) , anti-CD3 ( 145-2C11 ) , anti-CD19 ( 1D3 ) , anti-Ly6C ( AL-21 ) , anti-Ly6G ( 1A8 ) , anti-IgM ( R6-60 . 2 ) , anti-IgG1 ( A85-1 ) . Anti-c-Myc antibody was from Cell Signaling Technology ( D84C12 , Danvers , MA ) . Anti-Neuropilin1 polyclonal antibody was from R&D Systems ( FAB566A , Minneapolis , MN ) . Fixable Viability Dye from eBioscience was used to stain dead cells . Annexin V staining was performed using eBioscience kit ( 88–08006 ) according to manufacture instructions . For intracellular cytokine staining cells were stimulated for 3h with PMA ( 100ng/ml ) and Ionomycin ( 1 µg/ml ) in the presence of Brefeldin A ( 10 µg/ml ) . Autophagosome formation detection by flow cytometry was performed using FlowCellect Autophagy LC3 Antibody-based Assay Kit ( FCCH100171 , Merk-Millipore , Billerica , MA ) according to the manufacturer's instructions and following cell surface markers staining . The Autophagy LC3 Antibody-based Assay Kit involves a permeabilization step to wash out cytosolic LC3-I , allowing for antibody-based detection of membrane bound LC3-II . For autophagy detection in WT Treg cells B6 Foxp3hCD2 were used , as this allowed the detection of Foxp3+ Treg cells on the basis of surface expression of hCD2 marker . All data were acquired using a Cyan ADP ( Beckman Coulter , High Wycombe , UK ) and analyzed using FlowJo software ( Tree Star , Ashland , OR ) . Bulk CD4+ T cells were purified from the spleen and mLN by negative selection as previously described ( Coccia et al . , 2012 ) . Naïve CD4+ T cells were then sorted as CD4+ CD25- CD44- CD62L+ . Treg cells were sorted as CD4+ CD25+ when sorted from Atg16l1ΔCD4 and Atg16l1fl/fl mice and as CD4+ YFP+ when sorted from Atg16l1ΔFoxp3 and Foxp3Cre mice . Cells were sorted using an Astrios , Beckman Coulter MoFlo XDP or AriaIII BD Bioscience . Post-sort flow cytometry analyses confirmed that the purity of sorted populations was >97% . Naïve CD4+ T cells from WT ( CD45 . 1+ ) mice were sorted as described above and transferred to Atg16l1ΔCD4 recipient ( CD45 . 2+ ) mice via intravenous injection ( 4-5x106 cells per mouse ) . Analysis of spleen , mLN and cLP CD4+ T and Treg cells was performed 3 months after transfer . BM cells were isolated from the tibia and femur of WT ( CD45 . 1+ ) mice and Atg16l1fl/fl or Atg16l1ΔCD4 ( CD45 . 2+ ) mice and injected i . v . at 1:1 ratio ( a total of 1x107 cells per mouse ) into lethally irradiated ( 1100 Rad , split dose ) Rag1-/- recipients . Mice were allowed to reconstitute for at least 8 weeks before analysis . For induction of OVA-specific IgE antibodies two treatment regimes were utilized . For OVA only immunization mice were fed three times by oral gavage with ovalbumin grade VII ( 5 mg per mouse , Sigma-Aldrich , St Louis , MO ) with 21-day intervals between feeds . For adjuvanted immunization , mice were initially fed with OVA ( 5 mg per mouse ) plus cholera toxin ( 10 μg per mouse , Biologial Compbell ) , after which they were fed twice with OVA only ( 5mg per mouse ) , with 21-day intervals between feeds . Mice were orally infected with ~200 Trichuris muris eggs . Serum was collected on day 34-post infection and assayed by ELISA for parasite-specific IgG1 . Ninety-six-well plates were coated with 5 μg/ml T . muris excretory/secretory antigen and incubated with serial two-fold diluted serum . Bound IgG1 was detected using biotinylated anti-murine IgG1 ( AbD Serotec , Kidlington , UK ) . Atg16l1fl/fl and Atg16l1ΔCD4 mice were injected i . p . with 50 μg of fluorescent 16-carbon fatty acid analog BODIPY C-16 ( Molecular Probes ) reconstituted in DMSO . Mice were culled 1 hr later and tissue collected for analysis by flow cytometry . The real-time extracellular acidification rate ( ECAR ) and oxygen consumption rate ( OCR ) were measured using a XF 96 extracellular flux analyzer ( Seahorse Bioscience , Billerica , MA ) . Briefly , naïve ( CD62L+CD44- ) CD4+ T cells , or in vitro polarized TH2 and Treg cells , were washed twice in assay medium ( RPMI 1640 without sodium bicarbonate , 20 mM glucose , 1% FCS , 2mM pyruvate ) and seeded at 3–4 x 105 cells per well in assay medium in a 96-well XF plate coated with poly-L-lysine ( Sigma ) . T cells were rested for 1 hr at 37°C without CO2 before analysis . Naïve CD4+ T cells were cultured ( 3x105 cells/well ) in 96-well plates coated with anti-CD3 mAb ( 5 μg/ml ) and soluble anti-CD28 mAb ( 1 μg/ml ) and kept in presence of IL-2 ( 100 U/ml ) . For TH0 conditions anti-IL-4 ( 10 μg/ml ) and anti-IFN-γ ( 10 μg/ml ) mAb were added . Cultures were supplemented with IL-12 ( 10 ng/ml ) and anti-IL-4 mAb ( 10 μg/ml ) for TH1 polarization; with IL-4 ( 20 ng/ml ) , anti-IFN-γ ( 20 μg/ml ) and anti-IL-12 ( 10 μg/ml ) for TH2 polarization; and with TGF-β1 ( 5 ng/ml ) , anti-IFN-γ , anti-IL-4 mAb and anti-IL-12 ( all 10 μg/ml ) for induced Treg polarization . Sorted Treg cells were activated for 48h with anti-CD3 mAb ( 5 μg/ml ) and soluble anti-CD28 mAb ( 1 μg/ml ) plus IL-2 ( 100 U/ml ) and then cultured with IL-4 ( 10 ng/ml ) , IL-13 ( 10 ng/ml ) and IL-2 ( 100 U/ml ) for 5 days . All cytokines were from R&D Systems . Anti-CD3 ( 145-2C11 ) , anti-CD28 ( 37 . 51 ) , anti-IFN-γ ( XMG1 . 2 ) , anti-IL-12 ( C17 . 8 ) and anti-IL-4 ( 11B11 ) mAb were from eBioscience . Cells were cultured in RPMI-1640 Medium , 10% fetal calf serum , 2 mM L-glutamine , 100 U/ml of Penicillin/Streptomycin , and 0 . 05 mM 2-mercaptoethanol . All immunoglobulin isotypes except for IgE were measured by enzyme-linked immunosorbent assay ( ELISA ) using the SBA Clonotyping System ( Southern Biotech , Birmingham , AL ) . IgE concentration was determined using an anti-mouse IgE ELISA ( BioLegend ) , according to manufacturer's instructions . For the detection of soy-specific , CBir-specific and Helicobacter-specific antibodies ELISA was performed with plates coated with purified soy antigen ( 5 μg/ml ) , CBir peptide ( 10 μg/ml ) and soluble Helicobacter antigen ( sHel antigen , 10 μg/ml ) respectively . sHel antigen was prepared as previously described ( Kullberg et al . , 1998 ) . For the detection of OVA-specific IgE , a sandwich ELISA was performed with biotinylated-OVA used for detection . MCPT-1 concentrations were measured by ELISA ( eBioscience ) . Colonic and small intestine tissue samples were formalin-fixed , paraffin-embedded and sectioned as per histological analysis . Sections were deparaffinized , rehydrated , and subjected to sodium citrate-based antigen retrieval , then stained with mouse pAb anti-β-catenin ( 610153 , BD Bioscience ) , rabbit pAb anti-CD3 ( ab5690 , Abcam , Cambridge , UK ) and secondary goat antibodies conjugated to AlexaFluor488 or 555 ( Life Technologies , Carlsbad , CA ) . Slides were mounted with DAPI-containing Vectashield ( Vector Laboratories , Burlingame , CA ) . Images were acquired with an Olympus Fluoview FV1000 confocal microscope and Olympus Fluoview Software ( Olympus , Tokyo , Japan ) . CD4+ T cells purified by negative selection were lysed in RIPA buffer containing protease inhibitor cocktail ( Roche , Basel , Switzerland ) . Protein levels were normalized by Biorad DC protein assay ( Bio-Rad Laboratories , Hercules , CA ) , resolved by SDS-PAGE and , following transfer onto nitrocellulose membranes , were blotted with anti-LC3 antibody ( L7543; Sigma-Aldrich ) and anti-tubulin antibody ( sc5286 , Santa Cruz Biotechnology , Dallas , TX ) , and secondary HRP conjugated anti–rabbit antibody ( 7074S , Cell Signaling Technology ) . CD4+ T cells and Treg cells were sorted for each population based on surface marker and YFP expression from spleen and cLP of Atg16l1ΔFoxp3 and Foxp3Cre mice . Two hundred cells/population were sorted in triplicates from a total of four ( spleen ) or six ( cLP ) mice per group . Alternatively , 250 cells from in vitro polarized populations of TH2 and Treg cells were sorted from triplicate culture wells . RNA was reverse transcribed and cDNA was pre-amplified using the CellsDirect OneStep q-RT kit ( Invitrogen ) . The selected autophagy , apoptotic and metabolic genes were amplified and analyzed for expression using a dynamic 48x48 array ( Biomark Fluidigm ) as previously described ( Tehranchi et al . , 2010 ) . Data were analyzed using the 2-∆Ct method , and the results were normalized to actin or HPRTprt expression . For weight curves and antibody titers , p-values were determined by two-way ANOVA with Bonferroni post-tests . For the metabolic analysis using XF 96 extracellular flux analyzer , p-values were determined using unpaired Student’s t-test . For all other experiments , p-values were determined by nonparametric Mann–Whitney test . Differences were considered statistically significant when p<0 . 05 ( <p0 . 05 , p<0 . 01 , **p<0 . 001 ) . Data are shown as mean ± s . e . m . Statistics were calculated using GraphPad Prism 6 software . For in vivo experiments , sample size was determined by power analysis using power of trial software , which calculates a power value based on X2 test statistics . Calculated required sample sizes were applied whenever possible . No mouse was excluded from the analysis . With the exception of histological assessment of intestinal inflammation , experimenters were not 'blinded' to allocation of animals to experimental groups .	The gut presents a puzzle to our immune system . Immune cells must rapidly respond to antigens produced by harmful bacteria , but food and the beneficial bacteria that inhabit the gut also produce antigens that our immune system must tolerate . Inappropriate immune responses in the gut can lead to inflammatory bowel disease , a debilitating disease with no current cure . We do not fully understand why these harmful inflammatory responses arise , but we know that genetic factors are important . Mutations in genes that affect a process known as autophagy – a pathway that breaks down and recycles unwanted material inside cells – make inflammatory bowel disease more likely to develop , but exactly how they do so remains unclear . T helper cells are crucial controllers of intestinal immune responses and changes in their numbers and behaviour occur during inflammatory bowel disease . Kabat et al . explored how the autophagy pathway affects these key immune cells in mice . Blocking autophagy in T cells altered the balance of different types of T helper cells in the gut . A crucial population of regulatory T cells , which keep inflammatory responses in check , was lost . At the same time , another population of T cells expanded: the T helper 2 ( TH2 ) cells that are responsible for driving allergies . As a result , the mice developed intestinal inflammation and produced antibodies against gut bacteria and food . Overall , Kabat et al . ’s results show that autophagy defects can alter the balance of different types of T cells in the gut , leading to inflammation in the intestine . These observations contribute to our understanding of how genetic changes may influence susceptibility to inflammatory bowel disease . They also suggest that drugs that activate autophagy could help to treat diseases associated with changes in regulatory T cells or TH2 cells , including inflammatory bowel disease and allergies . It will now be important to test this and to confirm whether similar changes in T cells are present in humans that have mutations in autophagy genes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2016	The autophagy gene Atg16l1 differentially regulates Treg and TH2 cells to control intestinal inflammation
The unfolded protein response ( UPR ) monitors and adjusts the protein folding capacity of the endoplasmic reticulum ( ER ) . In S . pombe , the ER membrane-resident kinase/endoribonuclease Ire1 utilizes a mechanism of selective degradation of ER-bound mRNAs ( RIDD ) to maintain homeostasis . We used a genetic screen to identify factors critical to the Ire1-mediated UPR and found several proteins , Dom34 , Hbs1 and Ski complex subunits , previously implicated in ribosome rescue and mRNA no-go-decay ( NGD ) . Ribosome profiling in ER-stressed cells lacking these factors revealed that Ire1-mediated cleavage of ER-associated mRNAs results in ribosome stalling and mRNA degradation . Stalled ribosomes iteratively served as a ruler to template precise , regularly spaced upstream mRNA cleavage events . This clear signature uncovered hundreds of novel target mRNAs . Our results reveal that the UPR in S . pombe executes RIDD in an intricate interplay between Ire1 , translation , and the NGD pathway , and establish a critical role for NGD in maintaining ER homeostasis . Membrane and secreted proteins fold and mature within the endoplasmic reticulum ( ER ) before they are delivered to other compartments in the secretory pathway or to the plasma membrane . An imbalance between the protein folding load and the protein folding capacity in the ER leads to an accumulation of unfolded or misfolded proteins , a condition referred to as ‘ER stress . ’ ER stress triggers the unfolded protein response ( UPR ) , a network of signal transduction pathways that drive transcriptional programs to expand the ER’s protein folding capacity and reduce protein influx into the organelle through translational and mRNA degradative mechanisms , thereby ensuring that the organelle remains in or returns to homeostasis ( Walter and Ron , 2011 ) . In metazoans , the UPR is orchestrated by three ER-resident sensors/signal transducers: the membrane tethered transcription factor ATF6 and the transmembrane kinases PERK and IRE1 ( ‘Ire1’ in accordance with the yeast nomenclature ) ( Gardner et al . , 2013 ) . Each sensor activates a downstream transcriptional gene expression program of UPR target genes . In addition to the transcriptional response , the PERK branch induces cell-wide attenuation of translation by phosphorylating the general eukaryotic translation initiation factor 2 ( eIF2 ) , reducing the protein folding load of the ER . IRE1 catalyzes signaling through the most phylogenetically conserved branch of the UPR . It is a bifunctional transmembrane kinase/endonuclease activated by ER stress . In S . cerevisiae ( and metazoans ) , Ire1 catalyzes the unconventional splicing of the mRNA encoding the transcription factor Hac1 ( XBP-1 in metazoans ) that activates a comprehensive transcription program ( Acosta-Alvear et al . , 2007; Calfon et al . , 2002; Cox and Walter , 1996; Travers et al . , 2000 ) . Additionally , as first demonstrated in D . melanogaster , Ire1 initiates a reaction termed regulated Ire1-dependent mRNA decay , or RIDD , the selective decay of ER-bound mRNAs , thereby reducing the load of protein entering the ER ( Hollien and Weissman , 2006 ) . Conceptually , the effects of this reaction resemble that of PERK’s translational attenuation that reduces the ER’s protein folding load . RIDD occurs in mammalian cells and plants , but not in S . cerevisiae , where the HAC1 mRNA splicing reaction is the sole output of UPR signaling ( Niwa et al . , 2005 ) . In striking contrast , in S . pombe RIDD is the sole output of Ire1 . Transcriptional regulation and the otherwise conserved mRNA splicing reaction are entirely absent . S . pombe Ire1 has been shown to induce the decay of a few dozen mRNAs , cleaving them between the G and C residues of a short consensus sequence UGC ( UG/C ) ( Kimmig et al . , 2012 ) . This motif is too low in information content to specify engagement of select mRNAs with Ire1 , and thus other , still unidentified features need to contribute to bring appropriate mRNAs into juxtaposition and facilitate their engagement with Ire1 . During RIDD , mRNAs that are endonucleolytically severed by Ire1 rapidly decay through the combined actions of cellular exoribonucleases ( XRNs ) in the 5’→3’ direction and the exosome with the associated Ski complex in the 3’→5’ direction ( Hollien and Weissman , 2006; Kimmig et al . , 2012 ) . Interestingly , virtually all Ire1 cleavage sites in S . pombe mRNAs are located within the coding sequences ( CDSs ) . This observation is surprising , as mRNAs nicked within their CDSs are bound to induce translational stalls as ribosomes encounter the 3’ ends of the truncated mRNAs . This notion predicts that ribosomes stalled at the end of truncated mRNAs must be rescued by an active clearance mechanism , such as the ribosome rescue/mRNA decay pathway known as ‘no-go decay’ ( NGD ) ( Doma and Parker , 2006; Shoemaker and Green , 2012; Tsuboi et al . , 2012 ) . Recent biochemical and genome-wide ribosome foot-printing studies of NGD in S . cerevisiae showed that the Dom34/Hbs1 complex in cooperation with Rli1 promotes dissociation of stalled ribosomes on truncated mRNAs ( a form of ribosome recycling termed ‘rescue’ ) ( Guydosh and Green , 2014; Pisareva et al . , 2011; Shoemaker et al . , 2010 ) . Additionally , the NGD pathway triggers endonucleolytic cleavage of the mRNA upstream of the stalled ribosomes carried out by a still unidentified endonuclease ( which we here refer to as ‘NGDase’ ) , liberating ribosome-free mRNA fragments accessible to exonucleases ( Doma and Parker , 2006; Tsuboi et al . , 2012 ) . NGD is critical for rescuing stalled ribosomes and therefore maintaining ribosome homeostasis and is connected to the degradation of incomplete protein products through ubiquitylation and proteasome digestion ( Bengtson and Joazeiro , 2010; Brandman et al . , 2012; Shen et al . , 2015 ) . In S . cerevisiae , NGD serves as an important quality control mechanism , responding to premature polyadenylation events in ORFs , as translation of the poly ( A ) tail stalls ribosomes and triggers subsequent decay ( Guydosh and Green , 2017 ) . Here , we discovered that in S . pombe the NGD machinery Dom34/Hbs1 and the exosome-associated Ski-complex are critical players in the UPR , acting downstream of Ire1-catalyzed mRNA cleavage . Further , using short-read ribosome profiling methodology , we identified hundreds of novel mRNA targets of Ire1 . The precise and widespread nature of these target sites allowed us to show that stalled ribosomes serve as a ruler to template regularly spaced upstream mRNA cleavage events . Our results reveal that the UPR in S . pombe executes RIDD in an intimate interplay between Ire1 , translation , and the NGD surveillance pathway . To identify additional genes involved in the UPR in S . pombe , we performed a quantitative , genome-wide screen for mutants resulting in altered fitness compared to wild type ( WT ) cells when grown on limiting concentrations of tunicamycin ( Tm ) , a widely used UPR inducer that acts by inhibiting N-glycosylation in the ER lumen . To this end , we analyzed 2346 yeast strains deleted for non-essential genes by quantifying colony size differences in the absence or presence of Tm ( Figure 1A ) . By using a z-score cut-off of ±2 , we identified 180 gene deletions that showed a significant change in cell fitness by Tm treatment , including 76 gene deletions that suppressed Tm-induced growth defects ( Figure 1A , above blue dotted line ) and 104 gene deletions that sensitized cells to Tm ( Figure 1A , below red dotted line; Supplementary file 1 ) . Gene ontology ( GO ) analysis of gene deletions suppressing Tm growth defects showed enrichment for genes encoding proteins involved in vesicle transport and located on the cell surface ( Figure 1B ) . Consistent with this result , prior studies have shown that changes in cargo transport within the secretory pathway can cause this effect ( Liu and Chang , 2008 ) . GO analysis of Tm-sensitizing deletions identified an enrichment in genes encoding glycosylation enzymes and integral membrane proteins , as well as factors mediating mRNA catabolic processes ( Figure 1B ) . The latter class included the NGD components Ski2 , Ski7 , Dom34 and Hbs1 . Since the unfolded protein response in S . pombe relies exclusively on RIDD and could be affected by defects in ribosome rescue , we henceforth focused our investigation on the NGD factors to determine how their actions might synergize with the UPR . To validate that the growth defects related to RIDD and to exclude the possibility of potential false positive hits resulting from suppressor mutations in the deletion library , we reconstructed deletions of each of the four genes and plated the mutant and control strains on Tm plates . As shown in Figure 1C , the growth defects in S . pombe cells harboring hbs1 , dom34 , ski2 , and ski7 deletions impaired growth on Tm plates as compared to WT cells . Importantly , plating assays with the corresponding deletions in the NGD pathway in S . cerevisiae failed to exhibit growth defects under ER stress conditions ( Figure 1D ) . These results point to a central role for the NGD pathway in the UPR in S . pombe . In light of a potential role for NGD in the S . pombe UPR , we asked whether stalled ribosomes found on Ire1-cleaved mRNAs would accumulate in NGD defective strains . To this end , we performed ribosome profiling in S . pombe , sequencing ribosome-protected mRNA fragments ( ‘footprints’ ) ranging from 15 to 34 nucleotides in length . In our experiment , we included WT , dom34Δ , ski2Δ , and dom34Δ/ski2Δ strains in the presence and absence of DTT , which induces the UPR by disrupting disulfide bond formation in the ER lumen . Since ski2Δ S . pombe cells show a severe growth defect upon Tm treatment ( e . g . , see Figure 1C ) , we performed ribosome foot-printing after a short exposure of DTT ( 60 min ) to catch early events and minimize pleiotropic effects . We confirmed that at this early time point the Ire1 endonuclease was active by monitoring the accumulation of the cleavage product of gas2 mRNA as a reporter RIDD target ( Figure 1—figure supplement 1 ) . As is customary in ribosome profiling experiments , we controlled for changes in mRNA abundance by simultaneously performing mRNA-Seq on the same samples . Consistent with previous observations of ribosome footprint size distribution in S . cerevisiae ( Ingolia et al . , 2009 ) , we found in the dom34Δ/ski2Δ strain that most of the ribosome footprints ( ~75% ) were in the canonical range of 28–31 nucleotides ( Figure 2A ) . We also observed smaller populations of footprints either 20–22 nts in length ( ~5% of the population; corresponding to ribosomes predicted to be in an alternative state of the translation cycle [Lareau et al . , 2014] ) or 15–18 nts in length ( ~10% of the population; predicted to be stalled on truncated mRNA ends [Guydosh and Green , 2014] ) . As we further elaborate below , the population of footprints also reports on the formation of ‘disomes’ ( stacked ribosomes in direct contact with each other ) because RNase 1 , which is used to generate protected ribosome footprints , can only partially digest the mRNA between two closely-stacked ribosomes , thus yielding a larger footprint size . For reference , we also computed the size distribution for a WT strain ( Figure 2—figure supplement 1 ) . We next compared the change in mRNA levels ± UPR induction in WT cells to the change in short footprint ( 15–18 nt ) density ±UPR induction in dom34Δ/ski2Δ cells ( Figure 2B ) . The rationale of this experiment was based on the prediction that in WT cells , UPR induction leads to degradation of the Ire1-generated fragments , whereas in dom34Δ/ski2Δ cells the 5’ cleavage products are stabilized ( due to the ski2 knockout ) and ribosomes are stabilized at their 3’ ends to yield short footprints ( due to the dom34 knockout ) . Indeed , we found that a large fraction of the mRNAs that were degraded upon UPR induction in WT cells ( Figure 2B , points left of the center cloud ) were enriched in short footprints in dom34Δ/ski2Δ cells ( Figure 2B , upper left quadrant ) . Importantly , this group of mRNAs includes the majority of RIDD targets previously identified by mRNA-Seq ( Figure 2B , red dots ) ( Kimmig et al . , 2012 ) , thus validating the assumptions of our experimental strategy . In a similar analysis , we further asked whether mRNAs that are enriched in short footprints after UPR induction correspond to those that are stabilized when Ire1 is deleted ( Figure 2C , upper left quadrant ) . Again , many of these mRNAs corresponded to the previously identified RIDD transcripts ( Figure 2B right , red dots ) ( Kimmig et al . , 2012 ) . Taken together , these data establish that the accumulation of short ribosome footprints upon UPR induction is a signature of RIDD , likely because these footprints are derived from ribosomes stalled at cleavage sites generated by Ire1 . To identify the specific sites of ribosome stalling , we next examined the read distribution across genes that were enriched in short footprints upon UPR induction ( Figure 3 ) . In these examples , and in all profiling data shown in this work , we assigned the counts of mapped ribosome footprints according to their 3’ ends . This method was used for long ( 25–34 nt ) reads because our previous studies suggest that RNase 1 trims the 3’ end of the footprint more precisely than the 5’ end ( Guydosh and Green , 2017 ) . Moreover , aligning the short ( 15–18 nt ) ribosome reads by 3’ their ends allowed us to pinpoint the site of mRNA cleavage that caused ribosome stalling ( Figure 3A ) . As we show in the example of but2 mRNA in Figure 3B ( orange arrow ) , after UPR induction in dom34Δ/ski2Δ cells we typically observed a striking enrichment in short ( 15–18 nt ) reads at a UG/C site ( where cleavage occurs between the G and C ) , consistent with Ire1’s known RNA substrate specificity ( Gonzalez et al . , 1999 ) . In some cases , we also saw enrichment of long ( 25–34 nt ) footprints with a 3’ end positioned ~15 nucleotides ( roughly half of a long footprint ) upstream of the short reads ( Figure 3B , blue arrow ) . The relative positioning of these reads suggests the generation of a ‘disome’ ( two stacked ribosomes ) at this UG/C site . The significance of this observation is discussed further below . We also noted a strong UPR-dependent enrichment of short reads that extended typically hundreds of nucleotides on the 5’ side of the UG/C site . These upstream footprints are consistent with extensive mRNA cleavage events upstream of the initiating cleavage . These observations are reminiscent of NGD patterns ( Figure 3A ) that have previously been documented in S . cerevisiae ( Guydosh and Green , 2014; 2017 ) . Our data here demonstrate that this process is directional and extends over long distances from precise stall sites . In many examples ( Figure 3C ) , the pattern of cleavage showed periodicity , suggesting that iteratively stalled ribosomes in this background play a critical role in determining the no-go cleavage pattern ( further analyzed below ) . Intriguingly , we found that one of the targeted transcripts is that encoding Ire1 itself ( Figure 3—figure supplement 1 ) . We also examined the effects in WT strains or where only ski2 or dom34 was individually knocked out ( Figure 3—figure supplement 2 , Figure 3—figure supplement 3 ) . We saw little or no evidence for stalling or NGD patterns in WT and ski2Δ strains . However , in the dom34Δ strain , we observed evidence of the NGD pattern in both examples and for stalling in one case ( Figure 3—figure supplement 2 ) . To more precisely pinpoint the sites of ribosome stalling and to identify the full scope of RIDD targets , we next examined the effect of UPR induction on short ( 15–18 nt ) footprint density at every individual nucleotide position . Given the large size of the S . pombe genome ( >10 million single base positions ) , we initially focused on nucleotide positions in genes with the most reads ( >1 . 5 rpm in both , i . e . , ±UPR induction samples ) for data visualization ( Figure 4A ) . As expected , read density between UPR-induced and uninduced samples was correlated because many factors ( independent of UPR induction ) determine the pattern of read density within a gene . However , we noted a strong ( >10 x ) enrichment in read density for a distinct cluster of UG/C sites upon UPR induction ( Figure 4A , left panel; region above green diagonal line ) . Compared with cleavages at non-UG/C sites that also met this criterion ( Figure 4A , center panel; region above green diagonal line ) , UG/C sites were notably overrepresented ( 7 . 9% of UG/C sites vs . 0 . 0021% of non-UG/C sites; note that histogram is on a log scale ) , suggesting that they represent primary cleavage sites of Ire1 . Moreover , the Ire1 cleavage sites previously identified by 2’ , 3’-cyclic phosphate sequencing ( Kimmig et al . , 2012 ) that met the 1 . 5 rpm threshold were enriched in the UPR-induced sample , most more than 10-fold ( Figure 4A , right panel ) . Further analysis of the full set of data mapping to the transcriptome ( without the minimal read threshold and including sites without any reads in the case where the UPR was not induced ) revealed 5294 sites with >10 fold enrichment under UPR induction and >2 fold enrichment above the background level in their respective gene ( Supplementary file 2 ) . We found that ~22% of these corresponded to UG/C motifs ( Figure 4B , blue sector ) . Mapping these reads to individual transcripts revealed 1287 affected mRNAs , encompassing about a quarter of the S . pombe transcriptome . In particular , 471 mRNAs included reads with at least one UG/C motif , and , of these , 91% are predicted to be associated with the ER because of the presence of a signal sequence or transmembrane domain in the encoded protein ( Figure 4B , purple bar ) . This high prevalence of ER association is consistent with these genes representing a list of RIDD targets , expanded well beyond that reported previously . Consistent with this interpretation , we found these 471 mRNAs were clustered similarly to the previously-reported targets when we examined changes in overall levels of 15–18 nt footprints or mRNA-Seq ( Figure 4—figure supplement 1A ) . Similar clustering was also evident in mRNA-Seq data where ire1 was knocked out ( Figure 4—figure supplement 1B ) . In addition , the most downstream cleavage site in this subset of 471 mRNAs occurred at a UG/C about 85% of the time ( 398 genes; Figure 4B , orange bar ) , consistent with the cleavage patterns described above ( Figure 3A , B ) , where a UG/C cleavage event appears to trigger upstream no-go decay at non-UG/C sites . Of the remaining 816 transcripts where no cleavage sites mapped to UG/C sequences , about 24% of the mRNAs included at least one cleavage-site motif differing from UG/C by only a single base , suggesting that Ire1 may tolerate imperfect recognition signals . We next binned UPR-dependent strong stall sites ( >100 x the background level in a gene ) at UG/C motifs into three groups according to whether the terminal G was found in frame 0 , frame 1 , or frame 2 ( Figure 5A ) . Because the ribosome maintains the reading frame during translation and because the 5’ end of the footprint is trimmed flush against the ribosome by RNase 1 during preparation of sequencing libraries ( Guydosh and Green , 2017 ) , we expected that the length of a fragment should depend on the reading frame occupied by the terminal G . We found this to be true: the 16 nucleotide-long short reads were predominantly in frame 0 , the 17 nucleotide-long ones in frame 1 , and the 18 nucleotide-long ones in frame 2 . These observations suggest that the ribosome halts when the 3’ end of the mRNA is positioned randomly in the A site , consistent with the idea that successful decoding requires that the A site is filled with an intact codon . We also noticed a minority population of 15-nt reads for UG/C motifs in frame 2 ( Figure 5A , left tail of orange curve ) . The existence of this population suggests that the ribosome can decode the mRNA when the A site is filled with the terminal 3’ nucleotides of a truncated mRNA , which at a low efficiency of ~1/3 of the time can translocate yielding the 15 nucleotide-long footprint . Examination of ribosome footprint positions at strong stalls showed that the distribution of Ire1 UG/C cleavage sites across the 3 reading frames followed the underlying bias in the transcriptome ( Figure 5B , compare top and bottom ) . Analysis of the UG/C target through the MEME algorithm ( Moreno et al . , 1991 ) did not reveal any further strong features for the documented cleavage events . This analysis suggests that Ire1 is not influenced by recognizable sequence context immediately outside of the UG/C motif ( Figure 5C ) . By contrast to the UG/C sites , we found that the non-UG/C sites harbored some preference for a particular reading frame ( note that frame 0 non-UG/C cleavages are diminished relative to the background frequency , Figure 5B ) . Because many of these non-UG/C cleavage events likely result from NGD , we might expect that the ( in-frame ) stalled ribosomes that trigger this process guide the NGDase , which could account for this bias . This finding in fission yeast is consistent with our prior work in budding yeast that also showed a frame preference for no-go cleavage events that take place upstream of ribosomes stalled in poly ( A ) tails ( Guydosh and Green , 2017 ) . To analyze the pattern of NGD cleavage events upstream of primary Ire1 cleavages ( Figure 3 ) in further depth , we aligned 107 sequences that triggered the strongest ribosome stalling events ( >200 x above gene background level and >10 rpm upon UPR-induction ) at UG/C motifs in frame 0 and overlayed the short ( 15–18 nt ) ribosome profiling data from WT and different mutant strains ( Figure 6A , note that * symbols denote some peaks are scaled ) . In these averages , the strong peak centered at ‘0’ ( the UG/C site used as anchor in the alignment ) upon UPR induction ( ‘+DTT’ ) corresponds to a ribosome stalled at the site of Ire1 cleavage . This peak was strongest in the UPR-induced dom34Δ/ski2Δ strain ( Figure 6A , red traces ) but is also evident , albeit to a lesser extent , in the UPR-induced dom34Δ and ski2Δ strains ( Figure 6A , green and blue traces , respectively ) . Its dependence on UPR induction established that the Ire1 cleavage event is activated by ER stress and its dependence on the elimination of Dom34 and/or Ski2 establishes its dependence on the ribosome rescue and the 3’→5’ decay pathways . The variance of the peak heights is consistent with the growth data in Figure 1 showing that deletion of either dom34 or ski2 increased the mutant cells’ sensitivity to ER stress . The observation of a small peak at position 0 in the dom34Δ/ski2Δ strain in the absence ER stress ( Figure 6A , red traces ) suggests that Ire1 becomes partially activated under these conditions . These findings are further supported by analysis of the size of ribosome footprints that map to strong cleavage sites in frame 0 ( pause score >100 x background level upon UPR induction ) . The relative proportion of short reads ( e . g . 16 nucleotide reads ) increased upon dom34 deletion or UPR induction ( Figure 6B ) with respect to long ( ~28 nucleotide ) reads . This observation indicates that most ribosomes found at these sequences are stalled . We next focused on the short ( 15–18 nt ) footprint dataset in the dom34Δ/ski2Δ strain along with the corresponding long ( 25–34 nt ) footprints ( Figure 6C ) . When we turned our analysis to the regions upstream of the UG/C target sites , we noticed a striking periodicity in the 15–18 nt dataset: peaks repeated roughly every 14 nt , decreasing in abundance as one moves upstream of the initial cleavage site ( Figure 6C , orange lines ) . To confirm the accuracy of this distance measurement , we computed the power spectrum of the autocorrelation of the repeat region ( Figure 6D ) , which revealed a strong correlation every 14 nucleotides and , to a lesser extent , every 28 nucleotides ( discussed further below ) . The regularity in spacing of these peaks can be accounted for by a model wherein a ribosome that is stalled at a UG/C motif initiates NGD through endonucleolytic cleavage immediately upstream of it . This cleavage event , in turn , stalls the next ribosome behind it , generating another peak just 14-nucleotides upstream . In this way , the ribosome serves as a ruler that templates the repeat pattern . As suggested by the power spectrum peak at 28 nucleotides , the 14-nucleotide repeat pattern appeared to be superimposed by a 28-nucleotide repeat pattern , resulting in peaks with a higher amplitude at alternating 14-nucleotide positions ( Figure 6C , alternating height of orange lines ) . This trend is also visible in the data upstream of all UG/C cleavage sites ( 908 total sites , including both strong and weak stalls ) ( Figure 6—figure supplement 1 ) . A simple explanation for this observation is that two ribosomes occasionally stack — forming ‘disomes’ — when a tail-gating translating ribosome rear-ends a ribosome stalled at a mRNA truncation site before the NGDase cleaves the mRNA . This notion is supported by the analyses of the averaged dataset of the long , 25–34 nucleotide footprint data that reveal a dominating peak positioned precisely where we expect two ribosomes to collide ( Figure 6C , purple line ) . In this scenario , the upstream ribosome in the disome protects the mRNA from cleavage at the site of the 2nd NGD site ( Figure 6 , orange dot ) , thereby accounting for the reduced amplitude in arrested short-footprint ribosomes at that position . The evidence for a continued ( though diminishing ) pattern of alternating strong and weak peaks further upstream suggests that higher-order ribosome structures ( i . e . mass-collisions leading to trisome , quadrasome , etc . ) can form at a UG/C site . Alternatively , disome formation may be followed by cleavage , creating a new end on which the process repeats , leading to successive disome formation at upstream sites of no-go cleavage . We provide strong support for the coupling of NGD with Ire1-triggered endonucleolytic mRNA cleavage during the UPR in S . pombe . Our observations suggest a model ( Figure 7 ) wherein initial Ire1-triggered cleavage at UG/C sites in ER targeted mRNAs results in ribosome arrest at the truncated end of the mRNA , generating short footprints in ribosome profiling experiments . Ribosomes stalled at the end of the truncated mRNAs can trigger stalling of trailing ribosomes by a ‘fender-bender’ type collision , causing them to stack as disomes and perhaps larger stacks . Single ribosomes and ribosome stacks trigger the endonucleolytic cleavage events of the NGD pathway . Our model posits that ribosome stalling and mRNA cleavage is then reiterated until the upstream mRNA is degraded into short ribosome-associated fragments and longer ribosome-free 5’ mRNA fragments subject to unobstructed 3’→5’ mRNA decay by the cytosolic exosome . Whether an Ire1-targeted transcript is fully sliced into small fragments by NGD or whether the exosome manages to degrade unobstructed regions of some transcripts before ribosome stalling and NGD cleavage takes place likely depends on the relative kinetics of the NGDase , exosome , rescue activity of Dom34/Hbs1 , speed of elongation by the ribosome , and ribosome loading ( translational efficiency ) of the transcript . While the clearest evidence for enrichment of periodic short footprint ribosomes emerges in the dom34∆/ski2∆ mutant cells , weaker trends are seen in either single mutant alone ( Figure 6A ) , suggesting that both the NGD/ribosome rescue pathway ( Dom34/Hbs1/NGDase ) and the 3’→5’ mRNA decay pathway ( exosome/SkiX ) are critical for the elimination of the Ire1 cleavage products in WT cells . It is striking that the effects of either single knockout on ribosome stalling and upstream NGD were small compared to the effects of the double knockout . These observations suggest that the two pathways can partially substitute for each other . This finding is bolstered by the phenotypic assay showing loss of either pathway results in less tolerance for ER stress ( Figure 1 ) . The assay , in particular , showed that both Ski2 , a component of the Ski complex , and Ski7 , are involved . While Ski7 interacts with the Ski complex and the RNA exosome , it has been suggested to have additional effects on translation and to bolster tolerance to stress ( Jamar et al . , 2017; Kowalinski et al . , 2016 ) . Ski7 would therefore be an interesting candidate for further mechanistic studies . Unlike previous studies of NGD at less-well defined ribosome stall sites ( i . e . poly ( A ) stretches or long stem-loops ) , the endonucleolytic cleavage sites detected here were precise and thus offer the opportunity to more clearly examine the mechanism of long-distance NGD-cleavage events . Based on the dimension of the ribosome and positional registration of the mRNA footprint gleaned from previous ribosome profiling studies , we can infer mechanistic details regarding the still unknown and highly sought-after NGDase . For example , in the analysis of the repetitive pattern of protected fragments upstream of the primary UG/C cleavage site , we found that peaks were separated by only 14 nucleotides , which is different from a distance corresponding to the 15–18 nucleotide range observed in the footprint distributions created after RNase 1 treatment ( Figure 5A ) . This observation implies that the NGDase does not cleave flush against the upstream face of the ribosome , as occurs during experimental preparation of the 15–18 nt footprints . Instead , these data suggest that NGDase cleavage must occur inside the ribosome , at a position that is found one or two nucleotides within the channel from which the mRNA emerges the ribosome . It is also possible that the structure of ribosomes stalled in this way may be more flexible and therefore allow NGDase access to this site without having to reach inside the ribosome . This conclusion is consistent with our findings that NGD cleavage events show frame bias ( Figure 5B and previously in S . cerevisiae [Guydosh and Green , 2017] ) , reinforcing the notion that the no-go cleavage events take place on the ribosome . By identifying ribosomes stalled on mRNAs that are cleaved by Ire1 upon ER stress , our data reveal that the scope of RIDD upon UPR induction is far broader than appreciated to date . By comparison to the cohort of 39 RIDD target mRNAs identified previously by mRNA-Seq ( Kimmig et al . , 2012 ) , we here identified 471 mRNAs whose degradation is induced by Ire1-mediated cleavage at UG/C sites . This group of 471 mRNAs includes 34 of the 39 previously identified mRNAs ( three could not be evaluated due to little or no read depth; the other two were successfully identified by cleavage at non-UG/C sites ) , thus expanding the set of Ire1 substrates by over ten-fold . Almost all ( 91% ) of the identified mRNAs encode proteins that bear an ER-directed signal sequence or transmembrane domain and thus are predicted to be translated on the ER membrane in which Ire1 resides . The target list contains many mRNAs encoding proteins with functions in the secretory pathway and lipid metabolism , indicating that the regulation of these proteins’ biosynthesis may serve to fine-tune ER homeostasis , perhaps by adjusting the lipid composition of the membrane ( Volmer and Ron , 2015 ) . We also found that Ire1 appeared to cleave its own mRNA , suggesting a potential autoregulatory mechanism to limit production of this endonuclease once a threshold level is reached . We note that our method of detecting mRNA cleavage via the presence of a stalled ribosome limits our ability to detect cleavages outside canonical coding regions . It has previously been shown that Ire1 can target the 3’UTR ( Kimmig et al . , 2012 ) , and it is therefore reasonable to assume that additional cleavage sites may be found in the 5’UTR or 3’UTR regions . We also identified 816 mRNAs that are cleaved upon UPR induction , even though the cleavage sites did not correspond to UG/C motifs . While 193 of these mRNAs include sites matching UG/C with only a single base change and are therefore putative targets of Ire1 , the remaining 623 mRNAs exhibit ribosome stalling solely at other classes of endonucleolytic cleavage sites that are induced upon UPR induction . One possible explanation is that Ire1 or another , yet to be identified endonuclease that is activated by ER stress , can cleave at these other sites . While UG/C sites were strongly enriched in the set of Ire1 target mRNAs , the vast majority of UG/C motifs that we evaluated did not meet our threshold criteria for inclusion: reads representing only 7 . 9% of the mRNAs containing UG/C motifs shown in Figure 4A ( left panel ) were enriched >10 x ( i . e . , fell above the green line ) when the UPR was induced . It is possible that this number could be increased with improved methodology for cleavage site detection; yet it seems more plausible that other features of these sites are critical in determining the efficiency of Ire1 target selection . From the data shown in Figure 5C , it seems unlikely that such features lie in the immediate sequence context of the site . Other possibilities that could offer an explanation include higher-order mRNA structural features , mRNA localization to the ER , a specific complement of RNA binding proteins , or features in the nascent polypeptide chains . Since it is clear that ER-associated factors are enriched in the set of targeted mRNAs , we asked whether the ER-associated mRNAs that manage to escape cleavage showed any particular functional enrichment . We were unable to reveal any trend , further implying that additional properties of these transcripts are involved in specifying targeting . The biological importance of this broadened spectrum of Ire1 RIDD targets is underscored by the genetic screen that identified components in ribosome rescue and nonstop decay in an unbiased fashion . In particular , mutants in which the Ski-complex was defective showed strong sensitivity to ER stress . One potential explanation for why the failure to clear truncated mRNAs at the ER membrane may be so severe is that the stalled ribosomes may clog translocons and limit protein flux into the ER ( Arakawa et al . , 2016 ) , including that of newly synthesized chaperones and other factors required to restore ER homeostasis . If this were the case , we expect these findings will apply to higher eukaryotes where the RIDD pathway is also active and , as such , have broad implications for human disease . The UPR triggers the integrated stress response ( ISR ) , and many recent reports have suggested that chronic ISR activation by unfolded proteins or other stresses can lead to a number of diseases , including atherosclerosis ( Tufanli et al . , 2017 ) and many forms of neurological dysfunction ( Scheper and Hoozemans , 2015 ) . The ribosome recycling and mRNA decay pathways that we have shown here to be intricately intertwined are likely to be important for maintaining fitness of the proteome and human health . All high throughput data have been deposited with NCBI GEO with accession number GSE98934 . Standard cloning and yeast techniques were used for construction , transformation and gene deletions as described previously ( Moreno et al . , 1991 ) . Strains used in this study are listed in Supplementary file 3 . All non-ribosome profiling experiments were carried out in yeast extract complex media ( YE5S ) supplemented with 0 . 225 mg/ml of l-histidine , l-leucine , l-lysine , adenine and uracil at 30°C , unless otherwise described . The Schizosaccharomyces pombe Haploid Deletion Mutant Set version 2 . 0 ( M-2030H; Bioneer Corporation ) was accessible through the Azzalin lab ( ETHZ ) . The original library contained 3006 non-essential gene deletions , but only 2346 non-essential gene deletions were viable in this study . The library was spotted in duplicates on a 384-array format with YE5S media supplemented with or without 0 . 15 µg/ml tunicamycin . Plates were incubated at 30°C and after 3 days pictures were taken by the Fusion solo S system . Colony sizes were quantified and analyzed with Balony software ( https://code . google . com/p/balony/ ) . Resistant and sensitive gene hits were identified by the described z-score threshold . Growth rates for each deletion strain are listed in Supplementary file 1 . All cells were grown in YES 225 media ( Sunrise bioscience ) . All media was sterile filtered and cultures were grown at 30°C . Cultures were harvested at an OD of ~0 . 6 after~5 doubling times . DTT was added to 1 mM at 1 hr prior to harvest . Ribosome profiling libraries were prepared as described ( Guydosh and Green , 2014 ) by using a protocol very similar that used by Ingolia and coworkers ( Ingolia et al . , 2012 ) . All RNA size separation gels were cut as a single slice from 15 to 34 nt for footprints and 40–60 nt for mRNA-Seq . All footprint samples were lysed and separated over sucrose gradients in the presence of 0 . 1 mg/ml CHX . Total mRNA for mRNA-Seq was isolated from cells using hot SDS/acid phenol and chloroform , as previously described . Footprint samples here were subject to rRNA subtraction by using a yeast Ribo-Zero kit ( Epicentre ) . Subtraction of rRNA for all footprint samples was performed prior to linker ligation with the exception of the dom34Δ/ski2Δ strains where it was performed after linker ligation . This change was implemented because the RiboZero kit introduces a variety of short sequences that map at random across the genome , leading to occasional spikes in the data ( Guydosh and Green , 2017 ) . Samples for mRNA-Seq were subject to subtraction after purification of total RNA ( mRNA-Seq ) . The 50°C incubation step for standard footprint preparation was skipped in the Ribo-Zero-modified protocol , as recommended by the manufacturer . Sequencing and demultiplexing were performed on an Illumina HiSeq2500 at the Johns Hopkins Institute of Genetic Medicine . Analysis of footprints was essentially as described ( Guydosh and Green , 2014 ) with modifications as noted below . The PomBase ASM294 v2 . 22 genome assembly was the reference genome used for analysis ( Wood et al . , 2012 ) . De-multiplexed sequences were processed to remove reads with any position with Phred score <20 or assigned N as a quality filter step . Following a search for the linker and sorting of reads into short ( 15–18 nt ) or long ( 25–34 nt ) populations , contaminating ladder oligonucleotides were removed and alignment to a database of rRNA and tRNA spliced genes was performed . Following this step , a second round of subtraction for short , 15–18 nt , reads was performed by aligning to all the tRNA gene sequences plus extension with CCA on their 3’ ends . This enhanced the removal of cytoplasmic tRNA fragments . The remaining reads were mapped to the genome and those that failed to match were aligned to a custom transcriptome , created by splicing together annotated exons . Read lengths were assessed with the FastQC software ( Babraham Bioinformatics ) . All reads that aligned to multiple coding sequences were discarded . Read occupancy was determined by giving one count per read at its 3’ end and in some cases shifted to align with various active sites in the ribosome ( i . e . start of the P site ) as described below . However , reads were assigned to 5’ ends for mRNA-Seq analysis . These mRNA-Seq reads were also trimmed of 3’ consecutive As after alignment and remapped to include those near poly ( A ) tails , as was done previously ( Guydosh and Green , 2014 ) . Read counts were then normalized by dividing by the total number of million mapped reads in a sample . Alignments were performed with Bowtie ( Langmead et al . , 2009 ) and included the parameters: -y -a -m 1 --best --strata . All footprint alignments to coding sequences allowed for no mismatches; mRNA-Seq alignments allowed two mismatches . All other analysis software was custom coded in the Python 2 . 7 programming language and Biopython ( Supplementary file 4 ) . Plot construction and correlation analysis was done with Igor Pro ( Wavemetrics ) . In general , regions of transcript analysis that overlapped with other transcripts on the same strand or marked dubious were ignored in the analysis . Gene quantitation ( shown in Figure 2B and others , and used in calculations elsewhere ) relied on a shift of −2 for 3’ assignment ( short reads ) or −14 ( long reads ) and therefore aligns read density roughly with the P site . Total gene ribosome occupancy was quantitated into density units of reads per kilobase per million mapped reads ( rpkm ) by taking reads mapping to an annotated sequence and dividing by the gene length in kilobases . The reads from the first and last five amino acids were not included to prevent known artifacts around start and stop codons from skewing results . Ratio analysis was performed by taking the ratio at every point in the transcriptome between datasets from yeast with and without DTT treatment ( Figure 4A and Supplementary file 2 ) . To be included in Figure 4A , >1 . 5 rpm of density had to be present in both datasets . The threshold for detection under DTT exposure ( Supplementary file 2 ) was that the ratio between datasets must be >10 and the size of the peak in the +DTT dataset must be at least 2x higher than the average of reads that map to the gene . There were no read density thresholds in the –DTT dataset but positions without any reads were assigned one read ( 0 . 244 rpm ) so that a lower limit to the ratio could be computed . For some analyses , these thresholds were raised higher as noted . These ratios were corrected for changes in mRNA levels ( using mRNA-Seq ) between the two datasets . Position-average plots were created by averaging together ( with equal weight ) reads in a defined window for every occurrence of a particular motif ( i . e . UG/C ) in a list of targets ( Figure 6 ) . MEME 4 . 11 . 2 was run with these parameters: -dna -oc . -nostatus -time 18000 -maxsize 60000 -mod zoops -nmotifs 5 -minw 34 -maxw 34 .	Most proteins need to fold into a specific shape in order to work properly . As such , cells have developed a number of ways to sense and respond to stressful conditions that cause their proteins to fold incorrectly . One place this happens is within a network of tubes inside the cell called the endoplasmic reticulum; this is where proteins that are destined for the cell surface or other compartments in the cell become folded . The endoplasmic reticulum has a limited capacity to fold proteins . When it is overwhelmed , the cell temporarily stops making the proteins that use up this capacity . This action makes up part of a larger set of responses collectively referred to as the “unfolded protein response” . During the unfolded protein response , the production of some proteins is turned off when an enzyme called Ire1 cuts the transcript molecules that contain the instructions to build these proteins . Cutting these transcripts , however , creates a problem: it interrupts the translation of the transcript by the ribosome , the molecular machine that reads the genetic code to build proteins . Usually , a ribosome only comes off of a transcript when it arrives at a specific stop signal . Yet , ribosomes that run to the ends of broken transcripts never reach this signal and instead have to be rescued . If left without rescue , these stalled ribosomes could never be used again for translation of other transcripts , and the cell would lose the ability to make more proteins . Guydosh , Kimmig et al . searched for new genes in the yeast Schizosaccharomyces pombe that are involved in the part of the unfolded protein response that occurs after the actions of the Ire1 enzyme . This search revealed that cells missing so-called ribosome rescue proteins ( namely Dom34 and Hbs1 ) grow slowly under conditions that cause proteins to fold incorrectly . Guydosh , Kimmig et al . then looked to see where on the transcripts the ribosomes stall and remain un-rescued in the absence of these ribosome rescue proteins . These sites corresponded to places that were cut by Ire1 , the majority of which were previously unknown . Together these findings indicate that ribosome rescue is a key part of the unfolded protein response in S . pombe because it removes ribosomes that stall at the broken ends of transcript molecules cut by the Ire1 enzyme . An efficient and well-controlled response to conditions that cause proteins to fold incorrectly is important for human health . Loss of this control can lead to disorders as diverse as atherosclerosis , cancer and neurological diseases . By revealing that the unfolded protein response uses the ribosome rescue pathway , the findings improve our understanding of these health conditions and may open the door to new research and treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	Regulated Ire1-dependent mRNA decay requires no-go mRNA degradation to maintain endoplasmic reticulum homeostasis in S. pombe
Skeletal muscle comprises a family of diverse tissues with highly specialized functions . Many acquired diseases , including HIV and COPD , affect specific muscles while sparing others . Even monogenic muscular dystrophies selectively affect certain muscle groups . These observations suggest that factors intrinsic to muscle tissues influence their resistance to disease . Nevertheless , most studies have not addressed transcriptional diversity among skeletal muscles . Here we use RNAseq to profile mRNA expression in skeletal , smooth , and cardiac muscle tissues from mice and rats . Our data set , MuscleDB , reveals extensive transcriptional diversity , with greater than 50% of transcripts differentially expressed among skeletal muscle tissues . We detect mRNA expression of hundreds of putative myokines that may underlie the endocrine functions of skeletal muscle . We identify candidate genes that may drive tissue specialization , including Smarca4 , Vegfa , and Myostatin . By demonstrating the intrinsic diversity of skeletal muscles , these data provide a resource for studying the mechanisms of tissue specialization . Gene expression atlases have made enormous contributions to our understanding of genetic regulatory mechanisms . The field of functional genomics was set in motion by the completion of the Human Genome Project and the coincident development of high throughput gene expression profiling technologies . The overriding goals of this field are to understand how genes and proteins interact at a whole-genome scale and to define how these interactions change across time , space , and different disease states . The development of SymAtlas was an early , influential effort to address these questions ( Su et al . , 2002 ) . Custom microarrays were used to systematically profile mRNA expression in dozens of tissues and cell lines from humans and mice . Besides describing tissue-specific expression patterns , these data provided essential insights into the relationship between chromosomal structure and transcriptional regulation ( Su et al . , 2004 ) . Related approaches have had a similarly high impact on biomedical research . For example , microRNA expression throughout mammalian tissues was described in an expression atlas that has revolutionized the study of regulatory RNAs ( Landgraf et al . , 2007 ) . Our lab has contributed to this growing literature with the creation of CircaDB , a database of tissue-specific mRNA rhythms in mice ( Hughes et al . , 2009; Zhang et al . , 2014 ) . Taken together , these projects demonstrate that publicly available functional genomics data have enduring value as a resource for the research community . Nevertheless , previous gene expression atlases have largely ignored skeletal muscle . CircaDB includes gene expression data from the heart and whole calf muscle , but it does not distinguish between their constituent tissues . Similarly , SymAtlas profiles nearly 100 different tissues , but there is only a single representative sample for either cardiac or skeletal muscle . The microRNA Atlas includes over 250 human , mouse , and rat tissues; however , skeletal muscle is entirely absent from these data . More recent human gene expression atlases have similar biases ( Evangelista et al . , 2015; Lindholm et al . , 2014; Vissing and Schjerling , 2014 ) . Many studies have compared muscle-specific gene expression in different tissues ( Porter et al . , 2001 ) , fiber-types ( Chemello et al . , 2011 ) , or disease states ( Chen et al . , 2000; Colantuoni et al . , 2001 ) , but on the whole , there is no systematic analysis of transcriptional diversity in skeletal muscle . This gap in the literature is problematic since skeletal muscle comprises a remarkably diverse group of tissues ( Tirrell et al . , 2012 ) . Skeletal muscle groups originate from different regions of the developing embryo , and they have characteristic morphological specializations ( Merrell and Kardon , 2013; Murphy and Kardon , 2011; Noden and Francis-West , 2006 ) . Their physiological functions are similarly diversified . For example , extraocular muscles govern precise eye movements , the diaphragm drives rhythmic breathing , and limb skeletal muscles are involved in either fast bursts of motion or sustained contractions underlying posture . These intrinsic differences contribute to differential susceptibility of muscle groups to injury and disease . For example , there are six major classes of muscular dystrophy , and each one afflicts a characteristic pattern of skeletal and cardiac muscle tissues ( Ciciliot et al . , 2013; Emery , 2002 ) . Since the causative mutations underlying congenital muscular dystrophies are germline and present in all cells , this observation indicates that there are properties intrinsic to different muscle tissues that regulate their sensitivity or resistance to different pathological mechanisms . Acquired diseases show similar specificity in which muscle tissues they affect and which they spare . Patients with chronic obstructive pulmonary disorder ( COPD ) typically have pronounced myopathy in their quadriceps and dorsiflexors ( Barreiro and Gea , 2016; Clark et al . , 2000; Gagnon et al . , 2013 ) . Critical illness myopathy , a debilitating condition caused by mechanical ventilation and steroid treatment , causes muscle weakness in limb and respiratory muscles while sparing facial muscles ( Aare et al . , 2011; Hermans and Van den Berghe , 2015; Latronico et al . , 2012 ) . Cancer cachexia ( Acharyya et al . , 2005 ) , HIV ( Serrano et al . , 2008 ) , and sepsis ( Tiao et al . , 1997 ) cause muscle wasting , typically affecting fast twitch fibers more severely than slow twitch fibers . In fact , histologically identical muscle fibers show widely divergent responses to injury and disease depending on the muscle group in which they reside ( Ciciliot et al . , 2013; Aravamudan et al . , 2006 ) . Taken together , these observations strongly suggest that the intrinsic diversity of skeletal muscle has important consequences for human health and disease . However , the mechanisms through which disease susceptibility and functional specialization are established are unknown . Here we present the first systematic examination of transcriptional programming in different skeletal muscle tissues . We find that more than 50% of transcripts are differentially expressed among skeletal muscle tissues , an observation that cannot be explained by fiber type composition or developmental history alone . We show conservation of gene expression profiles across species and sexes , suggesting that these data may reveal conserved functional elements relevant to human health . Finally , we discuss how this unique data set may be applied to the study of disease , particularly regarding muscular dystrophy and regenerative medicine . To determine which skeletal muscle tissues are of the broadest general interest , we distributed a Google poll ( Figure 1—figure supplement 1 ) to leading investigators in the skeletal muscle field . Having recorded over 100 individual responses , we selected 11 mouse skeletal muscle tissues ( Table 1 ) for study in order to span the functional , developmental , and anatomical diversity of skeletal muscles . To hedge against selection bias from the investigators asked to vote in the poll , we cross-correlated these results with papers indexed in NCBI’s PubMed ( Figure 1—figure supplement 1 ) . In addition to these 11 mouse skeletal muscle tissues , we also identified representative smooth and cardiac muscle tissues for collection from mouse , and two skeletal muscle tissues ( EDL and soleus ) from male and female rats to permit inter-species and inter-sex comparisons ( Table 1 ) . Adult mice and rats were sacrificed and whole muscle tissues were dissected to include the entire muscle body from tendon to tendon . Each sample included six biological replicates from three animals apiece . Therefore , tissues were collected from 18 individual animals for every sample . RNA was purified from these tissues , and RNAseq was used to measure global gene expression ( Materials and methods and Supplementary file 1 ) . On average , every muscle sample was covered by greater than 200 million aligned short nucleotide reads , for a grand total of over 4 . 4 billion aligned reads in the entire data set ( Table 1 ) . Empirical simulations indicate that this experimental design reaches saturation with respect to identification of expressed exons ( Figure 1—figure supplement 2 ) . Pairwise comparisons of each replicate sample further indicate a high degree of reproducibility in expression level measurements among biological replicates ( median R2 value >0 . 93 , Figure 1—figure supplement 3 ) . Over 80% of transcripts encoded in the genome are expressed in at least one skeletal muscle ( Figure 1A ) . Comparing the transcripts expressed in all tissues identifies a core group of ~21 , 000 transcripts found in every skeletal , cardiac , and smooth muscle ( Figure 1B ) . Presumably , these mRNAs include the minimal set of genes required for a cell to generate contractile force . Differential expression analysis shows that even at extremely stringent ( q < 10−6 ) statistical thresholds , 55 . 2% of mouse transcripts are differentially expressed among skeletal muscle tissues ( Figure 1C ) . Phrased differently , over half of all transcripts are statistically different when comparing mRNA expression among the 11 mouse skeletal muscles in this study . To validate a subset of these data , we selected 10 genes differentially expressed between EDL and soleus across a range of different fold changes and performed quantitative PCR ( qPCR ) on independent biological samples . All 10 replicated , and manual examination of internal controls for cardiac and smooth muscle agreed with a priori expectations ( Figure 1—figure supplement 4 ) . To explore similarity between tissues , we calculated a pairwise Euclidean distance between every tissue ( Figure 1—figure supplement 5 ) . From these data , we generated a dendrogram that clusters tissues based on overall transcriptome similarity ( Figure 1D ) . Smooth and cardiac tissues clustered together as expected , and skeletal muscles were found in two primary clusters , presumably based on the proportion of fast twitch fibers they include ( i . e . the proportion of Myosin heavy chain 4 ( Myh4 ) expression ) . We then calculated the proportion of differentially expressed transcripts in every pairwise tissue comparison ( Figure 1E ) . Some surprising observations emerged from this analysis . For example , masseter , a head/neck muscle , is more similar to limb muscles like EDL than muscles that share a similar developmental history , such as the tongue . In contrast , the flexor digitorm brevis ( FDB ) , a muscle necessary for flexing the toes , is more similar to extraocular muscles than limb muscles . On average , 13% of transcripts are differentially expressed between any two skeletal muscles , with a maximum of 36 . 5% and a minimum less than 1% . A list of all transcripts differentially expressed among skeletal muscles is provided in Supplementary file 2 , and the analyzed expression data can be found in our online database . Skeletal muscle contains numerous non-muscle cells types . Therefore , a key question is whether the transcriptome diversity described above reflects genuine expression differences in muscle cells or alternatively differential cellular composition of the bulk tissues . To answer this we performed principal component analysis ( Figure 2 ) . The first principal component ( PC1 ) accounts for nearly 80% of the variance in our data and separates skeletal muscles into two groups with striking similarity to skeletal muscle clusters 1 and 2 in Figure 1E . The second principal component ( PC2 ) accounts for roughly 8% of variance and separates cardiac and smooth muscles from skeletal muscle ( Figure 2A ) . We then calculated the correlation ( expressed as R2 values ) between each transcript’s expression and PC1 ( Figure 2B ) . A small number of transcripts are highly ( R2 >0 . 90 ) correlated with PC1 and are thereby predictive of skeletal muscle identity . From the literature , we identified a number of ‘marker’ genes commonly used to separate non-muscle cells from skeletal muscle in flow cytometry ( Liu et al . , 2015 ) . We then compared the expression these ‘marker’ genes to PC1 to identify non-muscle cell types that may be partially responsible for the overall transcriptome diversity in our data . Pecam1 ( CD31 ) , a marker of endothelial cells , has an R2 value of 0 . 56 , reflecting modest correlation with PC1 . These data suggest that different proportions of endothelial cells in skeletal muscle contribute in part to the overall mRNA diversity observed . All other known marker genes had lower R2 values than Pecam1/CD31 , reflecting weaker correlation with PC1 . These include markers of muscle stem cells ( Vcam1 , R2 = 0 . 37 ) , fibroblasts ( Ly6a/Sca1 , R2 = 0 . 19 ) , and blood cells ( Ptprc/CD45 , R2 <0 . 01 ) . This observation supports the interpretation that some gene expression differences observed in this study are due to differential cellular composition . Nevertheless , every one of the top ten genes most strongly correlated with PC1 ( Figure 2D ) encodes a characteristic skeletal muscle protein that is unlikely to have been transcribed by non-muscle cell types . To extend on this observation , Supplementary file 3 provides a list of the top 100 genes most strongly correlated with PC1 . GO pathway analysis ( Huang et al . , 2009a ) reveals that the most significantly enriched term among this list of genes is ‘muscle protein’ ( enrichment 6 . 58 , FDR < 10−46 ) . Moreover , some of these genes are positively correlated with PC1 while others are negatively correlated . This observation implies that PC1 is more than simply a gross measure of the relative contribution of muscle cell RNA in any given sample . Taken as a whole , these observations indicate that many of the gene expression differences in this study reflect bona fide differences among muscle cell transcriptional programs rather than contamination from non-muscle mRNAs . To extend on this analysis , we performed Gene Ontology ( GO ) analysis ( Huang et al . , 2009a ) on every pairwise comparison of skeletal muscle tissues . We then consolidated these results by identifying GO terms that were repeatedly enriched in pairwise comparisons . Figure 2—source data 1 summarizes these results for all 11 skeletal muscle tissues . Consistent with the principal component analysis described above , many of these GO terms were related to structural components of the sarcomere , including Z-disc , A-band , M-band , and others . We also observed enrichment in terms related to extracellular proteins such as Integrin , Collagen , and Fibronectin . More general terms were also enriched , including various pathways involved in mitochondrial function , fatty acid metabolism , and neuromuscular junction assembly . To extend on these observations , we performed co-expression analysis using WGCNA ( Langfelder and Horvath , 2008 ) . As an internal control , we first identified clusters of genes that differentiate smooth , cardiac , skeletal muscle cluster one and skeletal muscle cluster 2 ( Figure 2—figure supplement 1A ) . As expected , GO analysis of the most statistically significant clusters of co-expressed genes ( Supplementaryl file 4 ) included genes conventionally associated with smooth , cardiac , and skeletal muscle physiology respectively . Co-expression analysis of all 17 mouse tissues revealed a smaller number of statistically significant gene clusters ( Figure 2—figure supplement 1B ) , and analysis of the resulting gene lists revealed enrichment in genes typically believed to be specific for skeletal muscle tissue ( Supplementary file 5 ) . These observations are consistent with the notion that many of the gene expression differences observed herein are genuine products of differential muscle cell gene expression . Moreover , close examination of the lists of genes identified by principal component , gene ontology , and co-expression analyses reveals that the most informative genes regarding skeletal muscle identity come from families of genes known for their function in skeletal muscle , such as Troponin , Tropomyosin , Calsequestrin , Myosin heavy chain ( Myh ) , and Myosin light chain . Skeletal muscle is comprised of myofibers which are typically classified into one of four types in mice based on their Myh expression ( Haizlip et al . , 2015 ) . We used our data to quantify the relative abundance of different fiber types across skeletal muscle tissues ( Figure 3A ) . These results were cross-correlated with legacy data from histological studies , showing close agreement ( Figure 3—figure supplement 1 ) . One hypothesis is that fiber type composition ( i . e . the relative amounts of fast versus slow twitch fibers ) establishes the observed diversity of gene expression profiles . To test this , we clustered skeletal muscle tissues based on similarity of Myh expression . The resulting dendrograms are grossly similar to those based on global gene expression ( Figure 3B; compare with Figure 1D ) , as predominantly fast twitch muscles expressing high levels of Myh4 ( Type IIB fibers ) tend to cluster together . Nevertheless , clustering within the two major skeletal muscle groups reveals important differences , such as the high similarity of diaphragm and FDB based on Myh expression , but their dissimilarity based on global gene expression ( Figure 1E ) . Similarly , the clustering of tongue and extraocular eye muscles varies considerably depending on whether Myh or global gene expression establishes the pairwise Euclidean distance . To explore these observations further , we made use of a single-fiber microarray study that identified genes enriched in slow ( Type I ) versus fast twitch ( Type IIB ) skeletal muscle fibers ( Chemello et al . , 2011 ) . These legacy data reveal that Myostatin ( Mstn ) is almost exclusively expressed in fast twitch fibers . Therefore , if fiber type composition determines gene expression patterns , we would expect close correlation between Mstn and Myh4 . In actuality , there is only moderate correlation ( R2 = 0 . 47 , Figure 3C ) . Expanding on this result , we find weak correlation between genes enriched in fast twitch fibers and Myh4 ( median R2 = 0 . 132 , Figure 3D ) . The low correlation between fiber type composition and gene expression signatures is also seen in slow twitch fibers ( Figure 3E and F ) . Taken together , we conclude that fiber type based on Myh expression contributes to tissue-specific gene expression but is insufficient to establish the diversity of transcriptional patterns we observe . Alternatively , developmental history may play an essential role in defining gene expression patterns in adult skeletal muscle . Hox genes are a family of transcription factors that establish anterior/posterior patterning and skeletal muscle specification during development ( Krumlauf , 1994 ) . We therefore examined the expression of Hox genes in muscle tissues ( Figure 4A ) . Clustering of tissues based on Hox gene expression reveals that the head and neck muscles cluster more closely with cardiac tissues than other skeletal muscles . Similarly , the diaphragm is more similar to the aorta than limb skeletal muscle . This is in marked contrast to clustering by the entire transcriptome , where each skeletal muscle is more similar to other skeletal muscles than any cardiac or smooth tissue . Hox gene expression in muscle recapitulates the developmental history of these tissues with respect to anterior/posterior axis formation . But , as these clusters are significantly different from those seen when comparing whole transcriptome expression patterns ( Figure 1D ) , we conclude that Hox genes are insufficient to explain the mRNA diversity among adult skeletal muscle tissues . Similar to Hox family members , many developmentally significant genes maintain expression in adult muscle tissues . These include the myogenic regulatory factor Myf6 that is expressed in , and exclusive to , skeletal muscle ( Figure 4B ) . Related differentiation factors , Myod1 , Myf5 , and Myog , are also expressed in adult muscle , albeit at roughly 10-fold lower levels than Myf6 . Hotair is a noncoding RNA that regulates Hox gene expression in development ( Tsumagari et al . , 2013 ) ; as expected , it is expressed in all limb muscles ( Figure 4C ) . In contrast , Pitx2 is a transcription factor with a key role in driving extraocular muscle development ( Zhou et al . , 2012 ) , and Lhx2 is a transcription factor important for masseter muscle development ( Buckingham and Rigby , 2014 ) . The expression of both genes is consistent with regulatory activity that continues into adulthood and may contribute to maintaining proper mRNA expression patterns ( Figure 4D and E ) . Notably , Lhx2 is also involved in limb muscle development ( Hobert and Westphal , 2000 ) ; it is of interest to determine how and why its expression is maintained in adult masseter but not adult limb tissues . We speculate that computational modeling of transcription factor ( TF ) networks may resolve the mechanisms underlying tissue-specific mRNA profiles , particularly since neither developmental history nor fiber type composition are sufficient to explain the diversity of the skeletal muscle transcriptome . As a first step to this end , we used Ingenuity Pathway Analysis ( IPA ) ( Krämer et al . , 2014 ) to predict TFs upstream of differentially expressed genes among skeletal muscles . We then consolidated these predictions by restricting our analysis to TFs that are predicted to drive differential gene expression in at least five of the 11 total skeletal muscle tissues . A summary of these results is provided in Table 2 , which includes 20 TFs predicted to contribute to skeletal muscle specialization . A simplified visual display of these data is provided in Figure 5 , which highlights a complex web of 1 ) predicted upstream transcription factors , 2 ) numerous differentially expressed genes that we predict act as effectors of tissue specialization ( including Myostatin and Vegfa ) , and 3 ) downstream processes involved in skeletal muscle disease and physiology . We note that Smarca4 is predicted to be upstream of tissue-specific gene expression in nine of 11 tissues , and that it ranked as the most statistically confident prediction by IPA in four of these tissues . As Smarca4 ( Brg1 ) is involved in early transcriptional patterning of skeletal muscle ( Albini et al . , 2015 ) , we believe it is a promising candidate for establishing transcriptional specialization in skeletal muscles . The generation of tissue-specific promoter lines in transgenic mice will facilitate testing these candidates . Therefore , we identified transcripts expressed in a tissue-specific fashion in skeletal muscle . Plotting tissue specificity versus average expression reveals a bimodal distribution among all transcripts ( Figure 1—figure supplement 6A and B ) . The majority of transcripts in skeletal muscle are expressed evenly across most skeletal muscle tissues , consistent with the notion that there is a minimal set of genes required to form sarcomeres and to generate contractile force . Nevertheless , a smaller set of transcripts ( ~5% ) show nearly exclusive expression in a single skeletal muscle tissue . Manual examination of these tissue-specific genes reveals that many are found in head and neck muscles that have the most divergent developmental history of tissues in this study . Nonetheless , we found examples of tissue-specific genes in the diaphragm ( Figure 1—figure supplement 6C ) and the FDB ( Figure 1—figure supplement 6D ) . A list of the most specific genes for any given muscle tissue is presented in Supplementary file 6 . We note that one of the most specific genes for soleus , Myh7 , has been validated previously using measures of protein expression ( Burkholder et al . , 1994 ) . Nevertheless , we caution readers that independent validation using alternative approaches is recommended for the observations herein . The presence of tissue-specific gene expression is consistent with a skeletal muscle transcriptome that is considerably more complex than previously appreciated . Moreover , the unprecedented depth of sequencing coverage in muscle permits the discovery of heretofore-unannotated transcripts . Manual examination of ‘gapped reads’ spanning two or more loci in the genome indicates that hundreds to thousands of splicing events occur in muscle that have not been previously described ( Figure 1—figure supplement 7A ) . The majority of these novel splicing events result in either inclusion of novel exons or the exclusion of exons previously thought to be constitutive ( Figure 1—figure supplement 7B ) . To validate this observation , we designed oligonucleotide primers ( Figure 1—figure supplement 7C ) specific for two novel exons of Myosin light chain kinase 4 ( Mylk4 ) , a gene with high expression ( >200 FPKM ) in many skeletal muscle tissues . PCR amplification and molecular cloning confirmed that these putative exons are included in full-length Mylk4 transcripts in two different skeletal muscle tissues , EDL and soleus ( Figure 1—figure supplement 7D ) . As the previously canonical splicing event linking exons 2 and 3 is detected by RNAseq in every muscle sample in this study , albeit at levels below the threshold for detection in RT-PCR , we conclude that transcripts including these putative exons are in actuality the predominant species of Mylk4 mRNA . A list of all novel splice junctions is provided in Supplementary file 7 and 8 . There are significant differences in fiber type composition between analogous muscle tissues of different mammals . For example , mouse soleus is a mixture of fast- and slow fibers ( Figure 3A ) , while rat soleus is almost entirely slow twitch ( Figure 6A ) . Because of this , we asked whether gene expression differences in mice are conserved across species . Expression of orthologous genes in mouse versus rat tissues has moderate levels of correlation ( R2 >0 . 6 , Figure 6B ) , despite the difference in fiber type composition noted above . This reinforces the observation that tissue identity , rather than fiber-type composition , drives transcriptome diversity in muscle . Moreover , the vast majority of genes differentially expressed between EDL and soleus in both mice and rats changed in the same direction ( Figure 6C ) . Taking this observation one step further , the fold change of all genes differentially expressed in mouse EDL compared to soleus ( Figure 6D ) is largely consistent with the fold change of same genes in rat EDL versus soleus ( Figure 6E ) . Sex differences did not dramatically influence differential gene expression between EDL and soleus ( Figure 6—figure supplement 1 ) . Fewer than 3% of transcripts were differentially expressed between male and female rat EDL ( 2 . 7% ) and male and female rat soleus ( 1 . 9% ) . Of these differentially expressed genes , most are up-regulated in males . This observation agrees with previous studies ( Roth et al . , 2002 ) and is consistent with the possibility that androgen response elements influence sex-specific gene expression differences in skeletal muscle . These results indicate that differentially expressed genes are largely conserved between mice and rats and suggest that these data may predict gene expression in related species . Taken as a whole , there is considerable variance in gene expression profiles among skeletal muscle tissues , in stark disagreement with the assumptions of previous gene expression atlases . In the remainder of this paper , we will explore several analyses illustrating the utility of these data as resource for generating testable hypotheses related to tissue specialization . Based on the likely conservation of gene expression patterns in humans , we speculate that genes associated with disease and those encoding drug targets will be of particular importance to follow-up studies . Of the roughly 23 , 000 genes encoded by the mouse genome , over 50% are differentially expressed among mouse skeletal muscle tissues . Of these , 3370 differentially expressed genes have human orthologs associated with disease , and 556 of those encode molecular targets of drugs on the market today ( Figure 7A ) . These genes may contribute to the molecular mechanisms underlying differential disease susceptibility and pharmaceutical sensitivity in skeletal muscle tissues . As a resource for investigators , we provide a list of all differentially expressed genes specifically involved in human skeletal muscle disorders ( Supplementary file 9 ) . To give one example of differential disease susceptibility , the aberrant expression of an embryonic isoform of Pyruvate kinase ( Pkm ) is involved in the mechanism of myotonic dystrophy ( Gao and Cooper , 2013 ) . Our data show that isoforms of Pkm are up-regulated by several standard deviations in EDL compared to all other muscle groups ( Figure 7B ) . Myotonic dystrophy disrupts normal splicing and pathologically elevates Pkm , which in turn disrupts normal metabolism , decreasing oxygen consumption and increasing glucose consumption . Based on these observations , elevated expression of Pkm in adult tissues is hypothesized to be a critical step in the pathology of myotonic dystrophy ( Gao and Cooper , 2013 ) . Since Pkm is expressed much higher in EDL than all other muscle types ( Figure 7B ) , we predict that EDL would be more sensitive to degeneration than other muscles . As myotonic dystrophy most dramatically affects certain subsets of muscle tissues , these observations suggest testable hypotheses regarding the underlying mechanism of disease susceptibility . Consistent with this possibility , we note with great interest that in mouse models , EDL is considerably more susceptible to muscle weakness than either diaphragm or soleus ( Moyer et al . , 2011 ) . In addition to disease susceptibility , these data may help explain differential drug sensitivity in muscle . To give one example , the drug chlorzoxazone ( brand name: Lorzone ) is used to treat muscle spasms . It is thought to act on the central nervous system ( CNS ) by regulating a potassium channel encoded by the gene Kcnma1 ( Dong et al . , 2006 ) . Although Kcnma1 is expressed throughout the central nervous system ( ENCODE Project Consortium , 2012 ) , it is also found at comparable levels in most skeletal muscle tissues ( Figure 7C ) . The two exceptions are extraocular eye muscles and the soleus that have greater than three-fold higher Kcnma1 expression than other muscle tissues . The differential abundance of chlorozoxazone’s target could influence either the efficacy of this drug or the severity of its side effects in different muscles . The expression in skeletal muscle of mRNAs encoding chlorzoxazone’s protein target calls into question the assumption in the literature that this drug acts exclusively through the CNS . Moreover , given that some commonly prescribed drugs , such as glucocorticoids , cause muscle wasting in specific subsets of muscle tissues ( Schakman et al . , 2008 ) , we speculate that this data set will be a valuable resource for exploring the mechanisms underlying differential drug sensitivity among skeletal muscles . Skeletal muscles are endocrine tissues , secreting numerous hormones that influence the physiology and metabolism throughout the body . Termed ‘myokines’ , these signaling molecules are key regulators of human health and disease ( Lightfoot and Cooper , 2016 ) . The total number of myokines is currently unknown . Our data contribute to this field by comprehensively defining mRNA expression of candidate myokines for future study ( Figure 7—figure supplement 1 and Supplementary file 10 ) . Dozens of genes encoding secreted proteins are differentially expressed among skeletal muscles at relatively high levels ( N = 42 unique genes , FPKM >10 ) . One notable example is Vegfa ( q-value ~10−30 ) , a gene involved in angiogenesis , cardiac disease , cancer progression , and many other normal and pathological processes ( Smith et al . , 2015 ) . Average expression of its predominant isoform is in the top 98th percentile of all transcripts expressed by skeletal muscle . Moreover , maximal expression of Vegfa in the diaphragm is nearly 10-fold greater than its minimal expression in the FDB , indicating that there are muscle-specific mechanisms for regulating Vegfa levels in particular and myokine levels in general . The serum concentration of VEGF-A in healthy adult humans is considerably greater than serum levels of IL6 , a canonical myokine ( Lightfoot and Cooper , 2016 ) . Given that roughly 40% of the human body is comprised of skeletal muscle expressing high levels of Vegfa , we speculate that muscle may be a heretofore under-appreciated source of VEGF-A in circulation . Whether Vegfa expression by skeletal muscle has endocrine as well as paracrine functions is unknown . Regenerative medicine has made considerable progress in generating skeletal muscle from stem cells ( Qazi et al . , 2015 ) . Nevertheless , engineered tissues have important deficiencies in generating sufficient force and forming appropriate neuromuscular synapses ( Juhas and Bursac , 2013 ) . Given the extensive diversity observed among skeletal muscle tissues , we speculate that the inability to form proper synaptic connections in engineered tissue may be due to the expression of inappropriate or incomplete transcriptional programs . In other words , differentiating stem cells into generic skeletal muscle may not recapitulate all the cues necessary for muscle-specific synapse formation . Therefore , we examined the differential expression of genes implicated in synapse assembly ( Figure 7—figure supplement 2 ) . Dozens of transcripts involved in synapse formation are expressed at high levels ( FPKM >10 ) , suggesting that they are skeletal muscle mRNAs , rather than contamination from nearby neurons . One illustrative example , Fbxo45 , is an E3 ligase involved in synapse formation . Mouse knock-outs of Fbxo45 have disrupted neuromuscular junction ( NMJ ) formation in the diaphragm , resulting in early lethality ( Saiga et al . , 2009 ) . Moreover , the C . elegans ortholog , FSN-1 , also disrupts NMJ formation , with some synapses being over-developed while others are under-developed ( Liao et al . , 2004 ) . We speculate that this protein and its orthologs may play a conserved role in tissue-specific NMJ assembly . General textbook discussions of skeletal muscle typically focus on developmental patterning , neuromuscular synapse physiology , or the biophysics of contractile functions ( Alberts , 2014 ) . This reflects the generally held belief that adult skeletal muscle is interesting only as a mechanical output of the nervous system . Functional genomics studies have acted on this assumption to the extent that every gene expression atlas generated to-date has selected at most one skeletal muscle as representative of the entire family of tissues . As such , the null hypothesis of this study was that gene expression profiles would be largely similar among skeletal muscle tissues . However , as more than 50% of transcripts are differentially expressed among skeletal muscles ( Figure 1C ) , and 13% of transcripts are differentially expressed between any two skeletal muscle tissues on average ( Figure 1E ) , the data are entirely inconsistent with the null hypothesis . These results indicate that there is no such thing as a representative skeletal muscle tissue . Instead , skeletal muscle should be viewed as a family of related tissues with a common contractile function but widely divergent physiology , metabolism , morphology , and developmental history . Based on these antecedents , it comes as no surprise that the transcriptional programs maintaining skeletal muscle specialization in adults are highly divergent as well . This study is the first systematic examination of transcriptome diversity in skeletal muscle . At greater than 200 million aligned short nucleotide reads per tissue and six biological replicates apiece , this data set is unprecedented in its scope , accuracy , and reproducibility ( Figure 1—figure supplement 4 , Figure 1—figure supplement 4 , and Figure 3—figure supplement 1 ) . Moreover , the depth of sequencing allows the detection of previously unannotated transcripts that may play a role in muscle physiology ( Figure 1—figure supplement 2 and Figure 1—figure supplement 7 , Supplementary file 7 and 8 ) . Besides establishing that skeletal muscles have considerable differences in their transcriptomes , the key significance of this paper will be as a resource for future studies . Therefore , we have made our analyzed data freely available ( http://muscledb . org ) , and all raw data may be downloaded from NCBI’s GEO . This resource will allow investigators to perform analyses beyond the scope of this paper , such as generating muscle-specific Cre-recombinase mouse strains for genetically manipulating specific muscle groups . Most importantly , these data will provide the foundation for computational modeling of transcription factor networks , a method we believe will uncover the genetic mechanisms that establish and maintain muscle specialization . To this end , we have used principal component analysis ( Figure 2 ) and related approaches ( Figure 1—figure supplement 6 and Figure 1—figure supplement 7 ) to show that expression of key skeletal muscle genes including different versions of Troponin , Tropomyosin , and Calsequestrin are highly predictive of skeletal muscle identity . Moreover , we used pathway analysis ( Figure 5 and Table 2 ) to identify 20 candidate transcription factors that may drive transcriptional specialization in muscle cells . One potential criticism is that gene expression does not always predict protein level and therefore function . We acknowledge that many processes besides steady-state mRNA levels regulate protein expression . Nevertheless , regulation of mRNA expression is unquestionably of biological importance in muscle cells , and transcriptional profiling predicts the majority of protein expression levels even in highly dynamic settings ( Robles et al . , 2014 ) . Moreover , the larger dynamic range of RNAseq measurements permits a more comprehensive description of expression profiles than would be possible with existing proteomic technology . We look forward to proteomic studies making use of these mRNA data , especially the identification of novel spliceforms , to generate improved catalogs of protein expression in skeletal muscle . We further acknowledge that the samples collected and analyzed in this study are bulk tissues rather than single-fiber preparations , and that contaminating tissues such as vasculature and immune cells may influence some gene expression measurements . However , our data agrees with the few single-fiber profiling papers in the literature ( Chemello et al . , 2011 ) , indicating that this potential bias is unlikely to confound the major observations herein . Consistent with this , the genes whose expression is most predictive of skeletal muscle identity are almost unanimously canonical skeletal muscle genes ( Supplementary file 3 ) . Furthermore , we emphasize that the majority of studies in the field use bulk tissues rather than single cell preparations . As such , our experimental design yields the greatest possible consistency with previous and future studies . In short , we believe whole tissue expression profiling provides a critical reference point and the rationale for follow-up work examining single-fiber gene expression . Acquired and genetic diseases show remarkable selectivity in which muscles they affect and which muscles they spare . For example , Duchenne muscular dystrophy severely affects the diaphragm and proximal limb extensors , while oculopharyngeal dystrophy causes weakness in the neck , facial , and extraocular muscles ( Emery , 2002 ) . At present , there is no satisfying explanation for how this occurs . A reasonable hypothesis is that intrinsic properties of muscle cells , such as gene expression , determine their sensitivity to different pathological mechanisms . These data are a starting point for future studies on how specialized transcriptional programs in muscles are maintained and how they ultimately influence disease . As an illustrative example , we note that the naturally elevated expression of Pkm in EDL may in part explain how this muscle is most dramatically affected in mouse models of myotonic dystrophy ( Figure 7 ) . Finally , skeletal muscle is an endocrine organ that regulates many normal and pathological processes , including sleep , bone health , diabetes , cancer , and cardiovascular disease ( Giudice and Taylor , 2017; Iizuka et al . , 2014; Karsenty and Olson , 2016 ) . At roughly 40% of an adult human’s body weight , skeletal muscle has an enormous capacity to influence other tissues through the expression of local or systemic signaling molecules . This study reveals extensive differential expression of putative myokines with largely unexplored functional significance ( Figure 7—figure supplement 1 and Supplementary file 10 ) . As such , we predict these data will be instrumental in future studies of the endocrine mechanisms through which skeletal muscle regulates health and disease . Adult male C57Bl6J mice were acquired from Jackson Laboratories at ten weeks of age . They were housed in light tight cages with a 12L:12D light schedule for four weeks with water and normal chow ad libitum . At 14 weeks of age , mice were sacrificed at between two and five hours after lights-on ( i . e . ZT 2–5 ) , and muscle tissues were rapidly dissected and flash frozen in liquid nitrogen for subsequent purification of total RNA . Adult male and female Sprague Dawley rats were obtained from Charles River ( Wilmington , MA ) at 12 weeks of age . Rats were housed in pairs with a 12 hr:12 hr light/dark cycle , and standard rat chow and water were provided ad libitum . At 14 weeks of age , rats were sacrificed at between two and five hours after lights-on ( i . e . ZT 2–5 ) , and muscle tissues were rapidly dissected and flash frozen in liquid nitrogen for subsequent purification of total RNA . Three animals were sacrificed per biological replicate . All animal procedures were conducted in compliance with the guidelines of the Association for Assessment and Accreditation of Laboratory Animal Care ( AAALAC ) and were approved by the Institutional Animal Care and Use Committee at University of Kentucky . Between 5–20 mg of frozen tissue were manually homogenized in Trizol reagent ( Invitrogen ) , and total RNA was purified using a standard chloroform extraction . RNA samples for gene expression analysis were mixed with an equal volume of 70% ethanol and further purified with RNEasy columns ( Qiagen , Germany ) using the manufacturer’s protocol . RNA was purified from tissue collected from individual mice; samples from the three individual mice of each biological replicate were then pooled together in equimolar amounts for further analysis . To assess RNA integrity , aliquots of each sample were denatured for 2 min at 70°C and analyzed on the Agilent 2100 Bioanalyzer using Eukaryote Total RNA Nano chips according to manufacturer’s protocol . RNA integrity numbers ( RINs ) for all samples were above 8 . 0 with a median RIN of 9 . 2 . Libraries were prepared using the Illumina Truseq Stranded mRNA LT kit using single end indexes according to manufacturer’s protocol . Approximately 500 ng of total RNA was used as starting material and amplified with 13 cycles of PCR . Libraries were validated for size and purity on the Agilent 2100 Bioanalyzer using DNA 1000 chips according to manufacturer’s protocol . Pilot runs to verify library integrity were sequenced on an Illumina MiSeq ( University of Missouri—St . Louis ) , and subsequent sequencing was performed on an Illumina HiSeq 2500 ( University of Michigan ) or HiSeq 3000 ( Washington University in St . Louis ) . All sequenced reads were 50 bp , single-end . Raw reads were aligned to the genome and transcriptome of Mus musculus ( build mm10 ) or Rattus norvegius ( build rn5 ) using RNAseq Unified Mapper ( RUM ) ( Grant et al . , 2011 ) with the following parameters: ‘--strand-specific --variable-length-reads --bowtie-nu-limit 10 --nu-limit 10’ . Mouse and rat gene models for RUM alignments were based on UCSC gene models updated as of December 2014 . 65%–92% of reads uniquely mapped to the genome/transcriptome , and the total number of aligned reads ( including unique and non-uniquely aligned reads ) was at least 94 . 7% for every replicate sample ( Supplementary file 1 ) . FPKM values for each transcript/exon/intron were calculated by RUM using strand-specific unique reads normalized to the total number of reads uniquely aligned to the nuclear genome . To account for variability in the mitochondrial content of different muscles , uniquely aligned mitochondrial reads were excluded from the denominator . R-squared values of log-transformed FPKMs between biological replicates of the same tissue were generally ~0 . 93 ( Figure 1—figure supplement 3 ) , and internal controls ( e . g . Figure 1—figure supplement 4 ) were used to verify the biological validity of these measurements . Differential expression between tissues was determined by one-way ANOVA of log-transformed FPKM values and adjusted for multiple testing using a Benjamini-Hochberg q-value ( Hochberg and Benjamini , 1990 ) . Unless otherwise noted , a q-value less than 0 . 01 and fold-change greater than 2 . 0 among transcripts expressed with an FPKM > 1 . 0 was deemed statistically significant . Transcripts were deemed to be expressed at FPKM > 1 . 0 , unless otherwise noted . Disease and drug target associations were identified using public data sets ( DrugBank v5 . 0 . 9 and DisGeNET or Ingenuity Pathway Analysis ) . Mouse genes were mapped to human orthologs using the BiomaRt package for R ( Durinck et al . , 2009 ) . Dendrograms were produced in R . A distance matrix was calculated for all ( Figure 1D ) or subsets ( Figures 3B and 4A ) of transcripts using the dist function using the standard options ( Euclidean distance; Figure 1—figure supplement 5 ) . Hierarchical clustering was performed using the hclust function using the complete agglomeration method . The code used to produce the heatmaps and dendrograms is freely available on GitHub ( https://github . com/flaneuse/muscleDB [Hughes , 2017]; copy archived at https://github . com/elifesciences-publications/muscleDB ) . For principal component analysis , log transformed FPKM values of expressed transcripts were passed to the bootPCA function from the bootSVD R library . A transcript was considered expressed if its mean log2 ( FPKM +1 ) value across replicates was >1 . 0 for at least one tissue . To estimate sampling variability , 10 , 000 bootstrap replicates were generated , and 99% confidence intervals computed for the relevant statistical functionals ( Fisher et al . , 2016 ) . Component loadings onto PC1 ( i . e . , the correlation between expression levels and PC1 ) were computed using R's cor function with default options . The PC plot was created using the autoplot function from the package ggfortify . Ingenuity Pathway Analysis ( v . 43605602 ) was used to evaluate major transcription factor networks involved in the observed transcriptional variation between skeletal muscle tissues ( Krämer et al . , 2014 ) . Lists of differentially upregulated transcripts were downloaded from pairwise comparisons within MuscleDB ( N = 110; FC > 2; q < 0 . 01 ) . Each list was analyzed using log2 fold change in IPA’s core analysis feature . If the list of upregulated genes contained more than 3000 genes , the q-value corresponding to the 3000th gene was used as an alternative threshold . Comparison analyses were run in IPA by grouping the core analyses for a single tissue ( N = 11 ) . The upstream analysis tool within the comparison analysis inferred potential transcriptional regulators based on the gene expression patterns within and between samples for a single tissue . The top 25 transcription factors consistent with the pattern of upregulation were recorded and the 20 most common TFs across all tissues were compiled into a table . The interactions shown were derived from the known interactions listed on Ingenuity Knowledge Base’s ( IKB ) summary pages for each TF . The network shown was constructed using IPA’s pathway designer . The regulated genes in the network were included if two independent pairwise comparisons ( using four unique tissues ) showed differential expression in MuscleDB ( q < 0 . 01; FC > 1 . 7 ) . The diseases and functions were added using IPA’s data overlay tool , which is based on the known interactions in the IKB . For the sake of clarity , only diseases and functions highly related to skeletal muscle physiology were included in Figure 5 . Using the same gene lists described above , we used DAVID Bioinformatics Resources 6 . 8 for functional GO analysis ( Huang et al . , 2009a; Huang et al . , 2009b ) . To generate an appropriate background list for analysis , genes expressed in at least one of eleven skeletal muscle tissues were selected and converted into ENSEMBL gene IDs using either an ENSEMBL annotation file or the DAVID ID conversion tool . Genes with ambiguous accessions during conversion were removed . Enriched GO terms were filtered using a fold enrichment threshold of 2 and a false-discovery threshold of 0 . 05 . For specific tissues , GO terms enriched in pairwise comparisons with ten other skeletal tissues were pooled to determine the most enriched GO terms . We used package WGCNA 1 . 63 in R version 3 . 3 . 3 to detect trait related modules using the ‘automatic network construction and module detection’ method with a soft-thresholding power of 18 and a minimum module size of 30 ( Langfelder and Horvath , 2008; Zhang and Horvath , 2005 ) . Transcripts expressed in at least one of all 17 muscle tissues were collected as input . External traits were specific tissues or one of the four muscle categories identified in Figure 1E . For modules of interest , transcripts were converted from either Refseq or UCSC IDs to gene symbols using the biotools . fr online converter . Independent biological replicates of mouse EDL and soleus tissues were collected as described above . Total RNA was purified and integrity was verified as described above . Reverse transcription reactions were performed with 500 ng starting material using the manufacturer’s protocol ( TaqMan Fast Universal PCR Master Mix , Applied Biosystems ) . qPCR was performed with ABI Taqman probes on a Stratagene MX3005 instrument ( Pcp4l1:mm01295270_m1 , Fam129a:mm00452065_m1 , Fhl2 , mm00515781_m1 , Tsga10:mm01228282_m1 , Stau2:mm00491782_m1 , Prkag3:mm00463997_m1 , Mstn:mm01254559_m1 , Plcd4:mm00455768_m1 , Myl1:mm00659043_m1 , Igfbp5:mm00516037_m1 , Ipo8:mm01255158_m1 ) using the manufacturer’s recommendations . To identify novel splicing events , previously unknown introns were identified from the junctions_all . rum file in RUM’s output . These results were then sorted by the number of reads aligning to that splice junction and manually curated for follow-up studies . To validate novel splicing events , PCR reactions were performed using 1 ul of the reverse transcription reaction using the manufacturer’s protocol ( Clontech Takara PCR kit ) . PCR products were visualized on a 1% agarose gel using conventional methods . The primary PCR products for EDL and soleus ( Figure 1—figure supplement 7D ) from primers Exon 2- > Exon 2 . 1 and Exon2 - > Exon3 were excised , purified , and TOPO cloned using the manufacturer’s protocol ( TOPO TA , Thermo Fisher ) . Cloned fragments were sequenced using conventional Sanger methods . All PCR primers were ordered from IDT; primer locations are described in Figure 1—figure supplement 7C . Primer sequences as follows: Exon 2 – AGGATCTCAGATTTGCTCACG , Exon 2 . 1 – GGATCCACTTTCCAGAATGC , Exon 2 . 2 – CATCTTTGCACCTGCATTC , Exon 3 – TATGGTCCAACCGTGCACTA , CtrlF – AGTGTGGGCGTCATCACAT , CtrlR – GTGGAGCTTGTGGTCTGACA . Validation experiments using immunohistochemistry were performed for Figure 1—figure supplement 6 using the following antibodies: UPK1B polyclonal antibody 1:200 ( PAB25730 , Abnova ) and BBOX1 polyclonal antibody 1:400 ( NBP1-32327 , Novus Biologicals ) . We note that neither antibody is well characterized for immunostains in any tissue , and that our experiments revealed no specific staining . All raw data and . bed files are available on NCBI’s Gene Expression Omnibus ( accession number: GSE100505 ) , and transcript-level expression values can also be downloaded from MuscleDB ( http://muscledb . org/ ) , a web application built using the ExpressionDB platform ( Hughes et al . , 2017 ) .	About 40% of our weight is formed of skeletal muscles , the hundreds of muscles in our bodies that can be voluntarily controlled by our nervous system . At the moment , the research community largely sees all these muscles as a single group whose tissues are virtually interchangeable . Yet , skeletal muscles have highly diverse origins , shapes and roles . For example , our diaphragm is a long muscle that contracts slowly and rhythmically so we can draw breaths , while tiny muscles in our eyes generate the short and precise movements of our eyeballs . Different skeletal muscles also respond in distinct ways to injuries , drugs and diseases . This suggests that these muscles may be diverse at the genetic level . While all the cells in our body have the same genetic information , exactly which genes are turned on and off ( or ‘expressed’ ) changes between types of cells . On top of this ‘on or off’ regulation , the level of expression of a gene – how active it is – can also differ . However , the studies that examine the differences in gene expression between tissues usually overlook skeletal muscles . Here , Terry et al . use genetic techniques to measure how genes are expressed in over 20 types of muscle in mice and rats . The results show that the expression levels of over 50% of all the animals’ genes vary between muscles . In fact , any two types of muscles express on average 13% of their genes differently from each other . The analyses yield further unexpected findings . For example , the expression levels in a muscle in the foot that helps to flex the rodents’ toes are more similar to those found in eye muscles than to the ones observed in limb muscles . These conclusions indicate that skeletal muscles are a widely diverse family of tissues . The research community will be able to use the data collected by Terry et al . to explore further the origins and the consequences of the differences between skeletal muscles . This could help researchers to understand why specific groups of muscles are more susceptible to disease , or react differently to a drug . This knowledge could also be exploited to refine approaches in tissue engineering , which aims to replace damaged muscles in the body .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "cell", "biology", "tools", "and", "resources" ]	2018	Transcriptional profiling reveals extraordinary diversity among skeletal muscle tissues
Bioimage analysis of fluorescent labels is widely used in the life sciences . Recent advances in deep learning ( DL ) allow automating time-consuming manual image analysis processes based on annotated training data . However , manual annotation of fluorescent features with a low signal-to-noise ratio is somewhat subjective . Training DL models on subjective annotations may be instable or yield biased models . In turn , these models may be unable to reliably detect biological effects . An analysis pipeline integrating data annotation , ground truth estimation , and model training can mitigate this risk . To evaluate this integrated process , we compared different DL-based analysis approaches . With data from two model organisms ( mice , zebrafish ) and five laboratories , we show that ground truth estimation from multiple human annotators helps to establish objectivity in fluorescent feature annotations . Furthermore , ensembles of multiple models trained on the estimated ground truth establish reliability and validity . Our research provides guidelines for reproducible DL-based bioimage analyses . Modern microscopy methods enable researchers to capture images that describe cellular and molecular features in biological samples at an unprecedented scale . One of the most frequently used imaging methods is fluorescent labeling of biological macromolecules , both in vitro and in vivo . In order to test a biological hypothesis , fluorescent features have to be interpreted and analyzed quantitatively , a process known as bioimage analysis ( Meijering et al . , 2016 ) . However , fluorescence does not provide clear signal-to-noise borders , forcing human experts to utilize individual heuristic criteria , such as morphology , size , or signal intensity to classify fluorescent signals as background , or to , often manually , annotate them as a region of interest ( ROI ) . This cognitive decision process depends on the graphical perception capabilities of the individual annotator ( Cleveland and McGill , 1985 ) . Constant technological advances in fluorescence microscopy facilitate the automatized acquisition of large image datasets , even at high resolution and with high throughput ( Li et al . , 2010; McDole et al . , 2018; Osten and Margrie , 2013 ) . The ever increasing workload associated with image feature annotation therefore calls for computer-aided automated bioimage analysis . However , attempts to replace human experts and to automate the annotation process using traditional image thresholding techniques ( e . g . histogram shape- , entropy- , or clustering-based methods [Sezgin and Sankur , 2004] ) frequently lack flexibility , as they rely on a high signal-to-noise ratio in the images or require computational expertise for user-based adaptation to individual datasets ( von Chamier et al . , 2019 ) . In recent years , deep learning ( DL ) and in particular deep convolutional neural networks have shown remarkable capacities in image recognition tasks , opening new possibilities to perform automatized image analysis . DL-based approaches have emerged as an alternative to conventional feature annotation or segmentation methods ( Caicedo et al . , 2019 ) and are even capable of performing complex tasks such as artificial labeling of plain bright-field images ( von Chamier et al . , 2019; Christiansen et al . , 2018; Ounkomol et al . , 2018 ) . The main difference between conventional and DL algorithms is that conventional algorithms follow predefined rules ( hard-coded ) , while DL algorithms are flexible to learn the respective task on base of a training dataset ( LeCun et al . , 2015 ) . Yet , deployment of DL approaches necessitates both computational expertise and suitable computing resources . These requirements frequently prevent non-AI experts from applying DL to routine image analysis tasks . Initial efforts have already been made to break down these barriers , for instance , by integration into prevalent bioimaging tools such as ImageJ ( Falk et al . , 2019 ) and CellProfiler ( McQuin et al . , 2018 ) , or using cloud-based approaches ( Haberl et al . , 2018 ) . To harness the potentials of these DL-based methods , they require integration into the bioimage analysis pipeline . We argue that such an integration into the scientific process ultimately necessitates DL-based approaches to meet the same standards as any method in an empirical study . We can derive these standards from the general quality criteria of qualitative and quantitative research: objectivity , reliability , and validity ( Frambach et al . , 2013 ) . Objectivity refers to the neutrality of evidence , with the aim to reduce personal preferences , emotions , or simply limitations introduced by the context in which manual feature annotation is performed ( Frambach et al . , 2013 ) . Manual annotation of fluorescent features has long been known to be subjective , especially in the case of weak signal-to-noise thresholds ( Schmitz et al . , 1999; Collier et al . , 2003; Niedworok et al . , 2016 ) . Notably , there is no objective ground truth reference in the particular case of fluorescent label segmentation , causing a critical problem for training and evaluation of DL algorithms . As multiple studies have pointed out that annotations of low quality can cause DL algorithms to either fail to train or to reproduce inconsistent annotations on new data ( von Chamier et al . , 2019; Falk et al . , 2019 ) , this is a crucial obstacle for applying DL to bioimage analysis processes . Reliability is concerned with the consistency of evidence ( Frambach et al . , 2013 ) . To allow an unambiguous understanding of this concept , we further distinguish between repeatability and reproducibility . Repeatability or test-retest reliability is defined as 'closeness of the agreement between the results of successive measurements of the same measure and carried out under the same conditions' ( Taylor and Kuyatt , 1994 , 14 ) , which is guaranteed for any deterministic DL model . Reproducibility , on the other hand , is specified as 'closeness of the agreement between the results of measurements of the same measure and carried out under changed conditions' ( Taylor and Kuyatt , 1994 , 14 ) , for example , different observer , or different apparatus . This is a critical point , since the output of different DL models trained on the same training dataset can vary significantly . This is caused by the stochastic training procedure ( e . g . random initialization , random sampling and data augmentation [Ronneberger et al . , 2015] ) , the choice of model parameters ( e . g . model architecture , weights , activation functions ) , and the choice of hyperparameters ( e . g . learning rate , mini-batch size , training epochs ) . Consequently , the reproducibility of DL models merits careful investigation . Finally , validity relates to the truth value of evidence , that is , whether we in fact measured what we intended to . Moreover , validity implies reliability - but not vice versa ( Frambach et al . , 2013 ) . On a basis of a given ground truth , validity is typically measured using appropriate similarity measures such as F1 score for detection and Intersection over Union ( IoU ) for segmentation purposes ( Ronneberger et al . , 2015; Falk et al . , 2019; Caicedo et al . , 2019 ) . In addition , the DL community has established widely accepted standards for training models . These comprise , among other things , techniques to avoid overfitting ( regularization techniques and cross-validation ) , tuning hyperparameters , and selecting appropriate metrics for model evaluation . However , these standards do not apply for the training and evaluation of a DL model in the absence of a ground truth , like in the case of fluorescent features . Taken together and with regard to the discussion about a reproducibility crisis in the fields of biology , medicine and artificial intelligence ( Siebert et al . , 2015; Baker , 2016; Ioannidis , 2016; Hutson , 2018; Fanelli , 2018; Chen et al . , 2019 ) , these limitations indicate that DL could aggravate this crisis by adding even more unknowns and uncertainties to bioimage analyses . However , the present study asks whether DL , if instantiated in an appropriate manner , also holds the potential to instead enhance the objectivity , reproducibility and validity of bioimage analysis . To tackle this conundrum , we investigated different DL-based strategies on five fluorescence image datasets . We show that training of DL models on the pooled input of multiple human experts utilizing ground truth estimation ( consensus models ) increases objectivity of fluorescent feature segmentation . Furthermore , we demonstrate that ensembles of consensus models are even capable of enhancing the reliability and validity of bioimage analysis of ambiguous image data , such as fluorescence features in histological tissue sections . The primary goal in bioimage analysis is to rigorously test a biological hypothesis . To leverage the potentials of DL models within this procedure , we need to trust our model – by establishing objectivity , reliability , and validity . Pertaining to the case of fluorescent labels , validity ( measuring what is intended to be measured ) requires objectivity to know what exactly we intend to measure in the absence of a ground truth . Similarly , reliability in terms of repeatability and reproducibility is a prerequisite for a valid and trustworthy model . Starting from the expert model strategy , we seek to establish objectivity ( consensus models ) and , successively , reliability and validity in the consensus ensemble strategy . In the following analysis , we first turn toward a comprehensive evaluation of the objectivity and its relation to validity before moving on to the concept of reliability . To assess the three different strategies , a training dataset of 36 images and a test set of nine microscopy images ( 1024 × 1024 px , 1 . 61 px/µm , on average ∼35 nuclei per image , see also Figure 1—figure supplement 2 ) showing cFOS immunoreactivity were manually annotated by five independent experts ( experts 1–5 ) . In absence of a rigorously objective ground truth , we used STAPLE ( Warfield et al . , 2004 ) to compute an estimated ground truth ( est . GT ) based on all expert annotations for each image . First , we trained a set of DL models on the 36 training images and corresponding annotations , either made by an individual human expert or as reflected in the est . GT ( see Materials and methods for the data set and detailed training , evaluation and model selection strategy ) . Then , we used our test set to evaluate the segmentation ( Mean IoU ) and detection ( F1 score ) performance of human experts and all trained models by means of similarity analysis on the level of individual images . For the pairwise comparison of annotations ( segmentation masks ) , we calculated the intersection over union ( IoU ) for all overlapping pairs of ROIs between two segmentation masks ( Figure 2A; see 7 . 9 . 1 Segmentation and detection ) . Following Maška et al . , 2014 , we consider two ROIs with an IoU of at least 0 . 5 as matching and calculated the F1 score MF1 score as the harmonic mean of precision and recall ( Figure 2B; see 7 . 9 . 1 Segmentation and detection ) . As bioimaging studies predominantly use measures related to counting ROIs in their analyses , we also focused on the feature detection performance ( MF1 score ) . The color coding ( gray , blue , orange ) introduced in Figure 2C refers to the different strategies depicted in Figure 1 and applies to all figures , if not indicated otherwise . To better grasp the difficulties in annotating cFOS-positive nuclei as fluorescent features in these images , we first compared manual expert annotations ( Figure 2D ) . The analysis revealed substantial differences between the annotations of the different experts and shows varying inter-rater agreement ( Schmitz et al . , 1999; Collier et al . , 2003; Niedworok et al . , 2016 ) . The level of inter-rater variability was inversely correlated with the relative signal intensities ( Figure 2—figure supplement 1; Niedworok et al . , 2016 ) . By comparing the annotations of the expert models ( gray ) to the annotations of the respective expert ( Figure 2E ) , we observed a higher MF1 score median compared to the inter-rater agreement ( Figure 2D ) in the majority of cases . Furthermore , comparing the similarity analysis results of human experts with those of their respective expert-specific models revealed that they are closely related ( Figure 2F , Figure 2—figure supplement 3 , and Figure 2—figure supplement 4 ) . As pointed out by von Chamier et al . , 2019 , this indicates that our expert models are able to learn and reproduce the annotation behavior of the individual experts . This becomes particularly evident in the annotations of the DL models trained on expert 1 ( Figure 2F , Figure 2—figure supplement 3 , and Figure 2—figure supplement 4 ) . Overall , the expert models yield a lower similarity to the est . GT compared to the consensus models ( blue ) or consensus ensembles ( orange ) . Notably , both consensus models and consensus ensembles perform on par with human experts . Hereby , the consensus ensembles outperform all other strategies , even at varying IoU thresholds ( Figure 2F and Figure 2G ) . In order to test for reliability of our analysis , we measured the repeatability and reproducibility of fluorescent feature annotation of our DL strategies . We assumed that the repeatability is assured for all our strategies due to the deterministic nature of our DL models ( unchanged conditions imply unchanged model weights ) . Hence , our evaluation was focused on the reproducibility , meaning the impact of the stochastic training process on the output . Inter-expert and inter-model comparisons within each strategy unveiled a better performance of the consensus ensembles strategy concerning both detection ( MF1 score ) and segmentation ( M¯IoU ) of the fluorescent features ( Figure 2H ) . Calculating the Fleiss’ kappa value ( Fleiss and Cohen , 1973 ) revealed that consensus ensemble annotations show a high reliability of agreement ( Figure 2H ) . Following the Fleiss’ kappa interpretation from Landis and Koch , 1977 , the results for the consensus ensembles indicate a substantial or almost perfect agreement . In contrast , the Fleiss’ kappa values for human experts refer to a fair agreement while the results for the alternative DL strategies lead to a moderate agreement ( Figure 2H ) . In summary , the similarity analysis of the three strategies shows that training of DL models solely on the input of a single human expert imposes a high risk of incorporating an intrinsic bias and therefore resembles , as hypothesized , a mere automation of manual image annotation . Both consensus models and consensus ensembles perform on par with human experts regarding the similarity to the est . GT , but the consensus ensembles yield by far the best results regarding their reproducibility . We conclude that , in terms of similarity metrics , only the consensus ensemble strategy meet the bioimaging standards for objectivity , reliability , and validity . Similarity analysis is inevitable to assess the quality of a model’s output , that is , the predicted segmentations ( Ronneberger et al . , 2015; Caicedo et al . , 2019; Falk et al . , 2019 ) . However , the primary goal of bioimage analysis is the unbiased quantification of distinct image features that correlate with experimental conditions . So far , it has remained unclear whether objectivity , reliability , and validity for bioimage analysis can be inferred directly from similarity metrics . In order to systematically address this question , we used our image dataset to quantify the abundance of cFOS in brain sections of mice after Pavlovian contextual fear conditioning . It is well established in the neuroscientific literature that rodents show changes in the distribution and abundance of cFOS in a specific brain region , namely the hippocampus , after processing information about places and contexts ( Keiser et al . , 2017; Campeau et al . , 1997; Huff et al . , 2006; Ramamoorthi et al . , 2011; Tayler et al . , 2013; Murawski et al . , 2012; Guzowski et al . , 2001 ) . Consequently , our experimental dataset offered us a second line of evidence , the objective analysis of mouse behavior , in addition to the changes of fluorescent features to validate the bioimage analyses results of our DL-based strategies . Our dataset comprised three experimental groups ( Figure 3A ) . In one group , mice were directly taken from their homecage as naive learning controls ( H ) . In the second group , mice were re-exposed to a previously explored training context as context controls ( C- ) . Mice in the third group underwent Pavlovian fear conditioning and were also re-exposed to the training context ( C+ ) ( Figure 3A ) . These three groups of mice showed different behavioral responses . For instance , fear ( threat; LeDoux , 2014 ) conditioned mice ( C+ ) showed increased freezing behavior after fear acquisition and showed strong freezing responses when re-exposed to the training context 24 hr later ( Figure 3—figure supplement 1 ) . After behavioral testing , brain sections of the different groups of mice were prepared and labeled for the neuronal activity-related protein cFOS by indirect immunofluorescence . Sections were also labeled with the neuronal marker NeuN ( Fox3 ) , allowing the anatomical identification of hippocampal subregions of interest . Images were acquired as confocal microscopy image stacks ( x , y-z ) and maximum intensity projections were used for subsequent bioimage analysis ( Figure 1—figure supplement 2 ) . Overall , we quantified the number of cFOS-positive nuclei and their mean signal intensity in five regions of the dorsal hippocampus ( DG as a whole , suprapyramidal DG , infrapyramidal DG , CA3 , and CA1 ) , and tested for significant differences between the three experimental groups ( Figure 3B–D ) . To extend this analysis beyond hypothesis testing at a certain significance level , we calculated the effect size ( η2 ) for each of these 30 pairwise comparisons . We illustrate our metrics with the detailed quantification of cFOS-positive nuclei in the stratum pyramidale of CA1 as a representative example and show two analyses for each DL strategy ( Figure 3E ) . These two examples represent those two models of each strategy that yielded the lowest and the highest effect sizes , respectively ( Figure 3E ) . Despite a general consensus of all models and ensembles on a context-dependent increase in the number of cFOS-positive nuclei , these quantifications already indicate that the variability of effect sizes decreases from expert models to consensus models and is lowest for consensus ensembles ( Figure 3E ) . The analysis in Figure 4 allows us to further explore the impact of the different DL strategies on the bioimage analysis results for each hippocampal subregion . Here , we display a high-level comparison of the effect sizes and corresponding significance levels of 20 independently trained expert models ( 4 per expert ) , 36 consensus models , and 9 consensus ensembles ( each derived from four consensus models ) . In contrast to the detailed illustration of selected models in Figure 3E , Figure 4A , for instance , summarizes the results for all analyses of the stratum pyramidale of CA1 . As indicated before , all models and ensembles show a highly significant context-dependent increase in the number of cFOS-positive nuclei , but also a notable variation in effect sizes for both expert and consensus models . Moreover , we identify a significant context-dependent increase in the mean signal intensity of cFOS-positive nuclei for all consensus models and ensembles . The expert models , by contrast , yield a high variation in effect sizes at different significance levels . Interestingly , all four expert models trained on the annotations of expert 1 ( and two other expert models only in the case of H vs . C+ ) did not yield a significant increase , indicating that expert 1’s annotation behavior was incorporated into the expert-1-specific models and that this also affects the bioimage analysis results ( Figure 4A ) . The meta analysis discloses a context-dependent increase of cFOS in almost all analyzed hippocampal regions ( Figure 4A–D ) , except for the infrapyramidal blade of the dentate gyrus ( Figure 4E ) . Notably , the majority votes of all three strategies at a significance level of p ≤ 0 . 05 ( after Bonferroni correction for multiple comparisons ) are identical for each pairwise comparison ( Figure 4A–E ) . However , the results can vary between individual models or ensembles ( Figure 4A–E ) . In order to assess the reliability of bioimage analysis results of the individual strategies , we further examined the variation per effect and variation per model in Figure 4F . For the variation per effect , we calculated the standard deviation of reported effect sizes within each strategy for every pairwise comparison ( effect ) . This confirmed the visual impression from Figure 4A–E as the consensus ensembles yield a significantly lower standard deviation compared to both alternative strategies ( Figure 4F ) . To illustrate the variation per model , we show the interaction between the number of biological effects that the corresponding model ( or ensemble ) reported differently compared to the congruent majority votes versus the standard deviation of its centered effect sizes across all 30 analyzed effects . This analysis shows that no expert model detected all biological effects in the microscopy images that were defined by the majority votes of all models . This is in stark contrast to the consistency of effect interpretation across the consensus ensembles ( Figure 4F ) . Consequently , we conclude that the consensus ensemble strategy is best suited to satisfy the bioimaging standards for objectivity , reliability , and validity . Bioimage analysis of fluorescent labels comes with a huge variability in terms of investigated model organisms , analyzed fluorescent features and applied image acquisition techniques ( Meijering et al . , 2016 ) . In order to assess our consensus ensemble strategy across these varying parameters , we tested it on four external datasets that were created in a fully independent manner and according to individual protocols ( Lab-Mue , Lab-Inns1 , Lab-Inns2 , and Lab-Wue2; see Materials and methods and Figure 5—source data 2 ) . Due to limited dataset sizes , the lab-specific training datasets consisted of just five microscopy images each and the corresponding est . GT based on the annotations from multiple experts . In the biomedical research field , the limited availability of training data is a common problem when training DL algorithms . For this reason , extensive data augmentation and regularization techniques , as well as transfer learning strategies are widely used to cope with small datasets ( Ronneberger et al . , 2015; Christiansen et al . , 2018; Falk et al . , 2019 ) . Transfer learning is a technique that enables DL models to reuse the image feature representations learned on another source , such as a task ( e . g . image segmentation ) or a domain ( e . g . the fluorescent feature , here cFOS-positive nuclei ) . This is particularly advantageous when applied to a task or domain where limited training data is available ( Yosinski et al . , 2014; Oquab et al . , 2014 ) . Moreover , transfer learning might be used to reduce observer variability and to increase feature annotation objectivity ( Bayramoglu and Heikkilä , 2016 ) . There are typically two ways to implement transfer learning for DL models , either by fine-tuning or by freezing features ( i . e . model weights ) ( Yosinski et al . , 2014 ) . The latter approach , if applied to the same task ( e . g . image segmentation ) , does not require any further model training . These out-of-the-box models reduce time and hardware requirements and may further increase objectivity of image analysis , by altogether excluding the need for any additional manual input . Consequently , we hypothesized that transfer learning from pretrained model ensembles would substantially reduce the training efforts ( Falk et al . , 2019 ) and might even increase objectivity of bioimage analysis . To test this , we followed three different initialization variants of the consensus ensemble strategy ( Figure 5A ) . In addition to starting the training of DL models with randomly initialized weights ( Figure 5A - from scratch ) , we reused the consensus ensemble weights from the previous evaluation ( Lab-Wue1 ) by means of fine-tuning ( Figure 5A - fine-tuned ) and freezing of all model layers ( Figure 5A - frozen ) . Although no training of the frozen model is required , we tested and evaluated the performance of frozen models to ensure their validity . After performing the similarity analysis , we compared the full bioimage analyses , including quantification and hypothesis testing , of the different initialization variants . Finally , to establish a notion of external validity , we also compared these results with the manually and independently performed bioimage analysis of a lab-specific expert ( Figure 5 , Figure 5—figure supplement 1 , and Figure 5—figure supplement 2 ) . In accordance with previous studies , similarity analyses revealed a substantial level of inter-rater variability in the heuristic annotations of the single experts ( Schmitz et al . , 1999; Collier et al . , 2003; Niedworok et al . , 2016 ) . Furthermore , we confirmed the concerns already put forward by others ( Falk et al . , 2019; von Chamier et al . , 2019 ) that training of DL models solely on the input of a single human expert imposes a high risk of incorporating an individual human bias into the trained models . We therefore conclude that models trained on single expert annotations resemble an automation of manual image annotation , but cannot remove subjective biases from bioimage analyses . Importantly , only consensus ensembles led to a coincident significant increase also in the reliability and validity of fluorescent feature annotations . Our analyses also show that annotations of multiple experts are imperative for two reasons: first , they mitigate or even eliminate the bias of expert-specific annotations and , second , are essential for the assessment of the model performance . Our bioimage dataset from Lab-Wue1 enabled us to look at the impact of different DL-based strategies on the results of bioimage analyses . This revealed a striking model-to-model variability as the main factor impairing the reproducibility of DL-based bioimage analyses . Convincingly , the majority votes for each effect were identical for all three strategies . However , the variance within the reported effect sizes differed significantly for each strategy . This entailed , for example , that no expert model was in full agreement with the congruent majority votes . On the contrary , consensus ensembles detected all effects with significantly higher reliability . Thus , our data indicates that bioimage analysis performed with a consensus ensemble significantly reduces the risk of obtaining irreproducible results . We then tested our consensus ensemble approach and three initialization variants on four external datasets with limited training data and varying similarities in terms of image parameters to our original dataset ( Lab-Wue1 ) . In line with previous studies on transfer learning , we demonstrate that the adaptation of models from pretrained weights to new , yet similar data requires less training iterations , compared to the training of models from scratch ( Falk et al . , 2019 ) . We extend these analyses and show that the reliability of fine-tuned ensembles was at least equivalent to from scratch ensembles , if not higher . Furthermore , we also provide initial evidence that pretrained ensembles can be used even without any adaptation , if task similarity is sufficiently high . Our data suggest that this component in the analysis pipeline could further increase the objectivity of bioimage analyses . Sharing model weights from validated models in open-source libraries , similarly to TensorFlow Hub ( https://www . tensorflow . org/hub ) or PyTorch Hub ( https://pytorch . org/hub/ ) , offers a great opportunity to provide annotation experience across labs in an open science community . In this study , for instance , we used the nuclear label of cFOS , an activity-dependent transcription factor , as fluorescent feature of interest . This label is in its signature indistinguishable from a variety of other fluorescent labels , like those of transcription factors ( CREB , phospho-CREB , Pax6 , NeuroG2 , or Brain3a ) , cell division markers ( phospho-histone H3 ) , apotposis markers ( Caspase-3 ) , and multiple others . Similarly to the pretrained and shared models of Falk et al . , 2019 , we surmise that the learned feature representations ( i . e . model weights ) of our cFOS consensus ensembles may serve as a good initialization for models that aim at performing nucleosomatic fluorescent label segmentation in brain slices . In line with the results of the Kaggle Data Science Bowl 2018 ( Caicedo et al . , 2019 ) , however , our findings indicate that a model adapted to a specific data set usually outperforms a general model trained on different datasets from different domains . To use and share frozen out-of-the-box models across the science community , we therefore need to create a well-documented library that contains validated model weights for each specific task and domain ( e . g . for each organism , marker type , image resolution , etc . ) . In conjunction with data repositories , this would also allow retrospective data analysis of prior studies . In summary , open-source model libraries may contribute to a better reproducibility of scientific experiments ( Fanelli , 2018 ) by improving the objectivity in bioimage analyses , by offering openness to analysis criteria , and by sharing pretrained models for ( re- ) evaluation . This paper describes a blueprint for the evaluation of DL models in biomedical imaging . Therefore , some of our methodological decisions were shaped by standardization considerations concerning the future deployment in bioimage analysis pipelines . The project was triggered by segmentation tasks for fluorescent labels ( cFOS ) in the cell nucleus . These are rather simple features , and we could readily annotate data from different labs , which facilitated the evaluation . However , this limits the generalizability to more complex image segmentation tasks , where training data annotation is slow and tedious . In particular , human perceptive capabilities for richer graphical features , such as area , volume , or density , is much worse than for regular , linear image features ( Cleveland and McGill , 1985; Feldman-Stewart et al . , 2000 ) . A case in point is the annotation of images showing ramified neurons or astrocytes . Such tasks would cause an enormous workload rendering complete human annotation virtually impossible . In this respect , we concur with prior research asserting that DL models based on human annotations will not be an option in these settings ( Driscoll et al . , 2019 ) . The characteristics of our examined strategies are based on best practices in the field of DL and derived from extant literature ( Meijering et al . , 2016; Falk et al . , 2019; Caicedo et al . , 2019 ) . The focus on the U-Net model architecture ( Ronneberger et al . , 2015 ) is a direct consequence of this standardization idea . Yet , it is also an important limitation of our study . Unlike more conventional studies that introduce a new method and provide a comprehensive performance comparison to the state of the art , we rely on U-net as the widely studied de facto standard for biomedical image segmentation purposes ( McQuin et al . , 2018; Falk et al . , 2019; Caicedo et al . , 2019 ) . Similarly , we chose to use STAPLE ( Warfield et al . , 2004 ) as the benchmark procedure for ground truth estimation . Thereby , we forwent considering alternatives and variants ( Lampert et al . , 2016 ) . In addition , we tried different ways to incorporate the single expert annotations into one DL model . For instance , we followed the approach of Guan et al . , 2018 by modeling individual experts in a multi-head DL model instead of pooling them in the first place . However , we decided to discard the approach as our tests did not improve the results but increased complexity . To enable other researchers to easily access , to interact with , and to reproduce our results and to share our trained models , we provide an open-source Python library that is easily accessible for both local installation or cloud-based deployment . With Jupyter Notebooks becoming the computational notebook of choice for data scientists ( Perkel , 2018 ) , we also implemented a training pipeline for non-AI experts in a Jupyter Notebook optimized for Google Colab , providing free access to the required computational resources ( e . g . , GPUs and TPUs ) . In summary , we recommend to use the annotations of multiple human experts to train and evaluate DL consensus model ensembles . In such a way , DL can be used to increase the objectivity , reliability , and validity of bioimage analyses and pave the way for higher reproducibility in science . Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact , Robert Blum ( Blum_R@ukw . de ) . Requests regarding the machine learning model and infrastructure should be directed to Christoph M . Flath ( christoph . flath@uni-wuerzburg . de ) . MS-275 ( Entinostat , Selleck Chemicals , Vienna , Austria; 10 mg kg−1 dissolved in saline +25% dimethylsulfoxide vehicle ) was administered immediately ( <1 min ) following an extinction training session and L-DOPA ( Sigma-Aldrich , Vienna , Austria; 20 mg kg−1 dissolved in saline ) was administered 1 hr before an extinction training session . All drugs were administered intraperitoneally in a volume of 10 ml kg−1 body weight . Control animals received saline . Mice were randomly selected to be administered either vehicle or pharmacological compound ( Whittle et al . , 2016 ) . In absence of an objective ground truth , we derived a probabilistic estimate of the ground truth by running the expectation-maximization algorithm for simultaneous truth and performance level estimation ( STAPLE , Warfield et al . , 2004 ) . The STAPLE algorithm iteratively estimates the ground truth segmentation ( est . GT ) based on the expert segmentation maps . During each algorithm iteration two steps are performed: Estimation step: The ground truth segmentation’s conditional probability is estimated based on the expert decisions and previous performance parameter estimates . Maximization step: The performance parameters ( sensitivity and specificity ) for each expert segmentation are estimated by maximizing the conditional expectation . Iterations are repeated until convergence is reached . We implemented the algorithm using the simplified interface to the Insight Toolkit ( SimpleITK 1 . 2 . 4 , Lowekamp et al . , 2013 ) . All evaluation metrics were calculated using Python ( version 3 . 7 . 3 ) , SciPy ( version 1 . 4 . 1 ) , and scikit-image ( version 0 . 16 . 2 ) . We provide the source code in Jupyter Notebooks ( see 7 . 13 Data and software availability ) . The deep learning pipeline was implemented in Python ( version 3 . 7 . 3 ) , Tensorflow ( version 1 . 14 . 0 ) , Keras ( version 2 . 2 . 4 ) , scikit-image ( version 0 . 16 . 2 ) , and scikit-learn ( version 0 . 21 . 2 ) . We provide the source code in Jupyter Notebooks ( see 7 . 13 Data and software availability ) . Fluorescent features were analyzed on base of the binary segmentation masks derived from the output of DL models or model ensembles , or counted manually by lab-specific experts . In order to compare the number of fluorescent features across images , we normalized in each image the number of annotated fluorescent features to the area of the analyzed region ( e . g . the number of cFOS-positive features per NeuN-positive area for Lab-Wue1 ) . For one set of experiment , we pooled this data for each condition ( e . g . H , C- and C+ for Lab-Wue1 ) and the analyzed brain region ( e . g . whole DG , infrapyramidal DG , suprapyramidal DG , CA3 , or CA1 for Lab-Wue1 ) . To compare different sets of experiments with each other , we normalized all relative fluorescent feature counts to the mean value of the respective control group ( e . g . H for Lab-Wue1 ) . The mean signal intensity for each image was calculated as the mean signal intensity of all ROIs annotated within the analyzed NeuN-positive region ( only performed for Lab-Wue1 ) . Subsequent pooling steps were identical as described above for the count of fluorescent features . All statistical analyses were performed using Python ( version 3 . 7 . 3 ) , SciPy ( version 1 . 4 . 1 ) , and Pingouin ( version 0 . 3 . 4 ) . We provide all source datasets and source codes in Jupyter Notebooks ( see 7 . 13 Data and software availability ) . In box plots , the area of the box represents the interquartile range ( IQR , 1st to 3rd quartile ) and whiskers extend to the maximal or minimal values , but no longer than 1 . 5 × IQR .	Research in biology generates many image datasets , mostly from microscopy . These images have to be analyzed , and much of this analysis relies on a human expert looking at the images and manually annotating features . Image datasets are often large , and human annotation can be subjective , so automating image analysis is highly desirable . This is where machine learning algorithms , such as deep learning , have proven to be useful . In order for deep learning algorithms to work first they have to be ‘trained’ . Deep learning algorithms are trained by being given a training dataset that has been annotated by human experts . The algorithms extract the relevant features to look out for from this training dataset and can then look for these features in other image data . However , it is also worth noting that because these models try to mimic the annotation behavior presented to them during training as well as possible , they can sometimes also mimic an expert’s subjectivity when annotating data . Segebarth , Griebel et al . asked whether this was the case , whether it had an impact on the outcome of the image data analysis , and whether it was possible to avoid this problem when using deep learning for imaging dataset analysis . For this research , Segebarth , Griebel et al . used microscopy images of mouse brain sections , where a protein called cFOS had been labeled with a fluorescent tag . This protein typically controls the rate at which DNA information is copied into RNA , leading to the production of proteins . Its activity can be influenced experimentally by testing the behaviors of mice . Thus , this experimental manipulation can be used to evaluate the results of deep learning-based image analyses . First , the fluorescent images were interpreted manually by a group of human experts . Then , their results were used to train a large variety of deep learning models . Models were trained either on the results of an individual expert or on the results pooled from all experts to come up with a consensus model , a deep learning model that learned from the personal annotation preferences of all experts . This made it possible to test whether training a model on multiple experts reduces the risk of subjectivity . As the training of deep learning models is random , Segebarth , Griebel et al . also tested whether combining the predictions from multiple models in a so-called model ensemble improves the consistency of the analyses . For evaluation , the annotations of the deep learning models were compared to those of the human experts , to ensure that the results were not influenced by the subjective behavior of one person . The results of all bioimage annotations were finally compared to the experimental results from analyzing the mice’s behaviors in order to check whether the models were able to find the behavioral effect on cFOS . Segebarth , Griebel et al . concluded that combining the expert knowledge of multiple experts reduces the subjectivity of bioimage annotation by deep learning algorithms . Combining such consensus information in a group of deep learning models improves the quality of bioimage analysis , so that the results are reliable , transparent and less subjective .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2020	On the objectivity, reliability, and validity of deep learning enabled bioimage analyses
Recent reports have suggested that self-propagating CRISPR-based gene drive systems are unlikely to efficiently invade wild populations due to drive-resistant alleles that prevent cutting . Here we develop mathematical models based on existing empirical data to explicitly test this assumption for population alteration drives . Our models show that although resistance prevents spread to fixation in large populations , even the least effective drive systems reported to date are likely to be highly invasive . Releasing a small number of organisms will often cause invasion of the local population , followed by invasion of additional populations connected by very low rates of gene flow . Hence , initiating contained field trials as tentatively endorsed by the National Academies report on gene drive could potentially result in unintended spread to additional populations . Our mathematical results suggest that self-propagating gene drive is best suited to applications such as malaria prevention that seek to affect all wild populations of the target species . CRISPR-based gene drive systems can bias inheritance of desired traits by cutting a wild-type allele and copying the drive system in its place ( Esvelt et al . , 2014 ) . Following reports of successful CRISPR gene drive systems in yeast ( DiCarlo et al . , 2015 ) and fruit flies ( Gantz and Bier , 2015 ) , scientists emphasized the need to employ strategies beyond traditional barrier containment as a laboratory safeguard ( National Academies of Sciences , Engineering , and Medicine , 2016; Akbari et al . , 2015 ) . These precautions were judged necessary to prevent unintended ecological effects , but also because any unauthorized release affecting a wild population could severely damage trust in scientists and governance , significantly delaying or even precluding applications of gene drive and other biotechnologies . Drive resistance can result from mutations that block cutting by the CRISPR nuclease . Recent examinations of the phenomenon by experiments and deterministic models have generated substantial media attention ( Champer et al . , 2017; Unckless et al . , 2017; Drury et al . , 2017; Noble et al . , 2017 ) . Resistance can arise from standing genetic variation at the drive locus or because the drive mechanism is not perfectly efficient and is predicted to prevent drive fixation in wild populations unless additional mitigating strategies are employed ( Burt , 2003; Deredec et al . , 2008; Esvelt et al . , 2014; Noble et al . , 2017; Marshall et al . , 2017 ) . Recent articles highlighting the problem of resistance for self-propagating gene drives have suggested that it might prevent drive invasion in wild populations—with some even implying that resistance could serve as an experimental safeguard . While we agree that resistance should prevent drive fixation , an allele can nonetheless spread to significant frequency without fixing . To clarify this point , we sought to quantify the likelihood and magnitude of spread in the most likely unauthorized release scenario—a small number of engineered individuals released into a wild population . CRISPR-based gene drive systems function by converting drive-heterozygotes into homozygotes in the late germline or early embryo ( Esvelt et al . , 2014 ) ( Figure 1A ) . First , a CRISPR nuclease encoded in the drive construct cuts at the corresponding wild-type allele—its target prescribed by an independently expressed guide RNA ( gRNA ) —producing a double-strand break ( Jinek et al . , 2012 ) . This break is then repaired either through homology-directed repair , producing a second copy of the gene drive construct , or through a nonhomologous repair pathway ( non-homologous end joining , NHEJ , or microhomology-mediated end joining , MMEJ ) , which typically introduces a mutation at the target site ( Mali et al . , 2013; Cong et al . , 2013 ) . Because the drive target is determined through sequence homology , such a mutation generally results in resistance to future cutting by the gene drive . Thus , the allele converts from a wild-type to resistant allele if it undergoes repair by a pathway other than homology-directed repair . Moreover , drive-resistant alleles are expected to exist in wild populations simply due to standing genetic variation ( Unckless et al . , 2017; Drury et al . , 2017 ) . Deterministic models , which assume an infinite , well-mixed population , predict whether an allele is favored to increase in frequency when initially rare in a wild population . Whether gene drives are predicted to invade by deterministic models depends on two key parameters: the homing efficiency ( P ) , or the probability of undergoing homology-directed repair instead of nonhomologous repair , and fitness ( f ) , or the relative fecundity or death rate the drive and its cargo confer on their organism compared to the wild-type . Mathematically , drives are initially favored by selection if f ( 1+P ) >1 , i . e . , if the inheritance bias of the drive exceeds its fitness penalty ( Noble et al . , 2017; Deredec et al . , 2008; Unckless et al . , 2015 ) . Given that the homing efficiencies of reported drive systems typically range from 0 . 37 to 0 . 99 ( Appendix 1—table 1 ) , current drive systems can clearly invade in deterministic models . Although the fitness parameter , f , is typically not measured in proof-of-concept studies , a substantial fitness cost is tolerable by all reported CRISPR drive constructs ( DiCarlo et al . , 2015; Gantz and Bier , 2015; Champer et al . , 2017; Gantz et al . , 2015; Hammond et al . , 2016 ) ( Figure 1B ) . However , in finite populations , the fate of initially rare alleles is determined not only by selection but also by stochastic fluctuations ( Wright , 1931; Fisher , 1930; Haldane , 1927 ) . Therefore , stochastic models are required to predict the probability that a drive will invade a population upon the introduction of a very small number of individuals , even when deterministic models predict that they are to invade . A previous , and arguably prescient , stochastic model of endonuclease drive containment found that homing-based drives , such as those subsequently developed using CRISPR , were among the likeliest to invade of the various drive alternatives ( Marshall , 2009 ) . To determine whether self-propagating homing drives are still able to invade in the presence of resistance , we formulated a finite population , stochastic , Moran-based model that allows us to study small releases in finite and structured populations ( Materials and methods ) . Our model considers three distinct allelic classes: wild-type ( W ) , gene drive ( D ) , and resistant ( R ) . Consistent with experiments , we assume that the drive invariably cuts the wild-type allele in the germline of a heterozygous WD individual , converting to a drive allele with probability P , or a resistant allele with probability 1−P . Each genotype , AB , has a relative reproductive rate , fAB , corresponding to its fitness in deterministic models , normalized such that the wild-type homozygote has fitness one ( fWW=1 ) , the drive confers a dominant cost ( fDW=fDD=fDR<1 ) , and resistance is neutral ( fWR=fRR=1 ) . This ordering of the parameters conservatively represents the worst-case scenario for drive spread ( Comparison with deterministic model ) . At the population level , our basic model considers N diploid individuals mating randomly . The process unfolds in discrete steps , during which parents are chosen for reproduction , an offspring is chosen according to the mechanism above , and another individual is replaced by the offspring ( Figure 1C and Materials and methods ) . These steps are repeated until one allele fixes . A generation is N time-steps , which corresponds to the mean lifespan of an individual . Code to perform numerical simulations of this model and all model extensions described below ( C++ , Matlab ) , as well as data files , documentation , and code to reproduce all of the figures shown here ( Matlab ) can be found at GitHub ( Noble , 2018; copy archived at https://github . com/elifesciences-publications/drive-invasiveness ) . Figure 1D shows typical simulations for drive efficiencies of 0 . 15 , 0 . 5 , and 0 . 9 , which correspond respectively to a constitutively active drive system targeting a common insertion site , and conservative and high efficiency systems ( based on previous experimental studies , Appendix 1—table 1 , Figure 1B , Empirical data supplement ) . These simulations assume a dominant drive fitness cost of 10% , a population of size 500 , and a release of 15 drive-homozygous individuals . ( Note that the dynamics are similar for larger population sizes; see Population size and Figure 3 ) . In all three cases , the drive , on average , irreversibly alters a majority of the population , either via invasion of the drive itself or via spread of drive-created resistant alleles . We call the maximum frequency of drive alleles reached during a simulation the peak drive , and we say a drive has invaded if it establishes in the population , ensuring behavior qualitatively similar to deterministic models ( Comparison with deterministic model ) . Notably , for sufficiently large populations , arbitrarily low frequencies meet this standard , as it depends on the absolute number of drive alleles rather than their frequency ( Analytic formulae for the escape probability in structured populations ) . Note also that each of these examples is chosen from the parameter regime in which invasion is predicted by deterministic models , since invasion is very unlikely outside of this regime . We next calculated the distribution of peak drive while varying the number of organisms released ( Figure 1E and F ) . We find that these distributions are bimodal , with one mode centered around the initial frequency ( corresponding to drift leading rapidly to extinction ) and one centered roughly around the maximum values observed in the large-release scenarios in Figure 1D . The former mode shrinks rapidly as more organisms are released , and for the parameters studied , a release of 10 individuals nearly guarantees invasion with substantial peak drive ( Comparison with deterministic model , Figure 10 ) . To understand the extent to which isolation might prevent invasion of other populations connected by gene flow , we introduced population structure . Our model consists of five subpopulations ( or islands ) that are equally connected by migration ( Figure 2A and Finite population model with population structure ) . Typical dynamics are illustrated in Figure 2C . Figure 2B and D show the escape probability , or the probability of the drive invading ( arbitrarily defined as attaining a frequency of 0 . 1 ) at least one subpopulation other than its originating one , and Figure 2E shows the probability of invading a varying number of subpopulations . Our results in Figure 2 suggest that if the migration rate is extremely low , then the drive is effectively contained in the initial subpopulation . If the migration rate is high , the drive is almost guaranteed to invade all subpopulations linked to the originating one . For intermediate migration rates—characterized roughly by migration rates on the order of the inverse of the drive extinction time—both outcomes occur . In the scenario studied in Figure 2 , a migration rate of 10−3 , which corresponds to a single migration event every 2 generations on average ( Materials and methods ) , virtually guarantees escape for moderate drive efficiencies ( Materials and methods ) . For further details and analytical formulae allowing rapid estimation of escape probabilities , see Analytic formulae for the escape probability in structured populations . Finally , we sought to understand the effects of additional mitigating factors that could potentially affect peak drive or invasion . We considered the most prominent factors that have arisen in previous papers , and we studied each by varying parameters in our basic model and developing model extensions . Our results are explored in detail in the Materials and methods . First , we considered preexisting drive resistance resulting from standing genetic variation ( Unckless et al . , 2017; Drury et al . , 2017 ) ( Standing genetic variation ) . We find that increasing the proportion of the population that is initially resistant linearly decreases the mean peak drive ( R2=0 . 996 ) . Using the parameters in Figure 1E and considering a release of 15 individuals , more than 50% preexisting resistance is required to contain average peak drive below 10% ( Figure 4 ) . Second , we studied the effect of varying family size , which may be relevant to species such as mosquitoes with large egg batch sizes ( Hammond et al . , 2016; Yaro et al . , 2006 ) . We extended the model so that k ( adult ) offspring are produced from a reproduction event , rather than one . We find that this effect scales the release and population sizes ( Hill , 1972 ) by a factor of 4/ ( 2k+6 ) . For illustration , we estimated k for Anopheles gambiae to be roughly 10 ( Offspring number distribution ) , so that a release of 7 individuals roughly corresponds to a release of 1 individual in our basic model . While this effect somewhat reduces the chance of drive invasion for small release sizes , it does not preclude it . Third , we varied drive fitness , resistant-individual fitness and homing efficiency across their entire parameter regimes and recorded peak drive ( Effect of varying fitness and homing efficiency , Figure 7 , Figure 8 ) . While varying drive fitness , we find that peak drive is on average greater than 30% across the majority of the regime and almost always greater than 10% ( Figure 7 , left ) —and , as a technical aside , we find that this is the case whether the fitness cost of the drive manifests itself via a reduction in birth rate or via increase in death rate ( Figure 7 , right ) . Moreover , in line with previous deterministic results , we find that peak drive can be substantially increased by associating a fitness cost with resistance ( Figure 8 ) , which could be expected for drive constructs intended for large-scale application , utilizing methods such as multiplex targeting of essential genes ( Esvelt et al . , 2014; Noble et al . , 2017; Marshall et al . , 2017 ) . Fourth , we studied the effect of inbreeding , which has been shown in several recent theoretical studies ( Bull , 2017; Drury et al . , 2017 ) to impede drive spread ( Inbreeding ) . We extended the model to include a probability s of an individual selfing rather than mating with a second individual ( Bull , 2017 ) . The model assumes no inbreeding depression and thus considers the worst-case scenario for drive ( Bull , 2017 ) . We find that even in this scenario , high selfing probabilities are required to reduce peak drive and the probability of invasion for moderate drive costs . There are a variety of other phenomena that could affect invasiveness , e . g . , density dependence ( Deredec et al . , 2011 ) , environment ( Tanaka et al . , 2017 ) , costly resistance ( Traulsen and Reed , 2012 ) , local ecology , and even mating incompatibilities between some laboratory strains and wild individuals . Such effects should be carefully studied in subsequent papers . Most importantly , the drive architecture itself should affect invasiveness; we consider here only alteration-type drive systems , while others , e . g . , sex-ratio distorters and genetic load drives , would be expected to yield different dynamics . In particular , population suppression drive systems may locally self-extinguish before invading new populations . However , for alteration drives , our key qualitative finding—that peak drive is difficult to reliably contain below a socially tolerable threshold following a very small release of organisms—appears robust to a variety of mitigating factors . Fundamentally , we exercise caution by omitting application-specific phenomena that might aid containment in particular instances but not in general . Our results suggest that current first-generation CRISPR-based gene drive systems for population alteration are capable of far-reaching—perhaps , for species distributed worldwide , global—spread , even for very small releases . A simple , constitutively expressed CRISPR nuclease and guide RNA cassette targeting the neutral site of insertion—an arrangement that could occur accidentally—may be capable of altering many populations of the target species depending on the homing efficiency of the organism in question . More generally , resistance can be problematic for intentional applications of gene drives , but we find that it is not a major impediment to invasion of unintended populations . These findings raise two important questions: ( 1 ) How likely are unauthorized releases of self-propagating gene drive systems in the first place ? ( 2 ) How likely are serious negative consequences given the apparently high likelihood of spread to most populations of the target species ? Rigorously addressing these questions is an important direction for future work , and we can offer only opinions here . The answer to the first question likely depends on a large number of factors , such as species , application , containment strategies , economic motivations , drive development stages , geography , and the caution of the investigators , so we omit speculation here . However , we consider the answer to the second question to be clearer: although most laboratory gene drive systems are unlikely to cause ecological changes—they are typically predicted to be transient and are not designed to alter traits of the host organism , least of all interactions with other species—the history of genetic engineering offers many examples suggesting that substantial social backlash could be triggered by unauthorized spread of a self-propagating gene drive ( Funk and Rainie , 2015; Couzin and Kaiser , 2005 ) . Any such event could significantly reduce public support for interventions against diseases such as malaria that could possibly save millions of lives . We believe it would be profoundly unwise to proceed with anything less than an abundance of caution . On a more technical note , our findings are specific to population alteration drive and cannot be directly generalized to self-propagating suppression drive , which could potentially self-extinguish before invading other populations . However , our results suggest a method for rough comparison between these scenarios: we find that the primary factor in determining drive spread between adjacent populations is the average number of migrants per generation ( Analytic formulae for the escape probability in structured populations ) , which can , in principle , be compared between models . For example , an earlier model of suppression drive systems ( Deredec et al . , 2011 ) predicted a total number of drive-carrying organisms over time which is remarkably similar to our example of an inefficient alteration drive system that is rapidly outcompeted by resistant alleles ( Figure 1D , middle ) . Thus , assuming comparable migration rates , it might not be surprising to see qualitatively similar levels of invasiveness . Accordingly , we urge researchers to exercise caution in developing or advocating for self-propagating suppression drives for applications other than malaria prevention—or similar projects intended to affect an entire species—until explicit models of invasiveness are available . Additionally , our findings emphasize the importance of the containment strategy known as ‘ecological confinement’ , which was proposed previously ( Esvelt et al . , 2014; Akbari et al . , 2015 ) . Given the risk that organisms may escape through accidents or outside intervention , laboratories in regions with endemic wild populations may wish to refrain from constructing self-propagating systems capable of invading those populations and undergoing unwanted spread . Laboratories in regions with endemic wild populations can reliably prevent accidental invasion by employing intrinsic molecular confinement mechanisms such as synthetic site targeting or split drive as recommended by the National Academies’ report on gene drives ( National Academies of Sciences , Engineering , and Medicine , 2016 ) . Perhaps most importantly , any development efforts looking ahead toward field trials , a component of the staged testing strategy outlined by the National Academies report , should be aware that there could be a high likelihood of unwanted spread across international borders , even from ostensibly isolated islands . The development of ‘local’ , intrinsically self-exhausting gene drive systems ( Chen et al . , 2007; Akbari et al . , 2014; Noble et al . , 2016; Magori and Gould , 2006; Gould et al . , 2008 ) , sensitive methods of monitoring population genetics , and strategies for countering self-propagating drive systems and removing all engineered genes from wild populations should be correspondingly high priorities . To model gene drives in finite populations , we introduce a Moran-type model with sexual reproduction ( illustrated in Figure 1C ) . We consider a population of N individuals , each of which is diploid . We focus on a locus with three allelic classes: wild-type ( W ) , CRISPR gene drive element ( D ) and drive-resistant ( R ) . There are six possible genotypes: WW , WD , WR , DD , DR , and RR . We assign to each genotype α a reproductive rate fα . The process proceeds in discrete time-steps , during each of which three events occur in succession ( Figure 1C ) . First , two individuals are chosen without replacement for mating with probabilities proportional to their reproductive rates , so that genotype α is selected with probability ( 1 ) fαNα∑β fβNβ . Here Nα is the number of individuals having genotype α , and the sum in the denominator is over all six genotypes . Second , after selecting the two parents , the offspring genotype is chosen randomly based on the genotypes of the two parents . To proceed , we introduce notation α=AB to mean that genotype α consists of alleles A and B , and we index these alleles via α1=A and α2=B . Note that we track only one genotype for each heterozygote , implicitly combining counts for genotypes AB and BA . Using this notation , the probability that an offspring of genotype γ is chosen given a mating between parents of genotypes α and β is given by the quantity qαβγ , which is equal to ( 2 ) qαγ1qβγ2+qαγ2qβγ11+δγ1γ2 . Here qαA is a gamete production probability—the probability that a parent with genotype α produces a gamete with haplotype A—and δAB is the Kronecker delta , defined by δAB=1 if A=B ( i . e . , if the offspring under consideration is a homozygote ) , and δAB=0 otherwise . The gamete production probabilities , qαA , are determined by accounting for the gene drive process described above . They are given by: qWWW=qDDD=qRRR=1 , qWDD= ( 1+P ) /2 , qWDR= ( 1−P ) /2 , qWRW=qWRR=qDRD=qDRR=1/2 . The remaining values not listed , e . g . , qWWR , are zero . Third , an individual is chosen uniformly at random for death . Thus , the population size remains constant . The resulting counts become the starting abundances for the next iteration of the process . The process is initialized with a small number , i , of drive homozygotes ( DD ) and the remaining population , N−i , wild-type homozygotes ( WW ) . The process continues as described above either until a specified number of time steps have elapsed or until one of the three alleles has fixed . Any of the alleles can fix , but typically either the wild-type or resistant alleles fix , due to the emergence of resistance . To study the effects of population structure on drive containment , we extended the well-mixed model from the previous section . We now consider l well-mixed subpopulations , each consisting initially of N/l individuals . The process proceeds in discrete time steps , as before . In each time step , we either migrate an individual from one population to another , or we choose a particular subpopulation and proceed through one mating and replacement iteration , as outlined above . More specifically , one step of the process proceeds as follows ( illustrated in Figure 11 ) . With probability m , we initiate a migration event . In this case , we perform three steps . First , we choose a source population with probability proportional to its size . Second , we choose an individual uniformly at random from the source population for migration . Finally , we move the chosen individual to a linked subpopulation uniformly at random . Or , with probability 1−m , we initiate a mating event as described in the well-mixed section . To carry this out , we first choose the population in which the event will occur . We choose this population with probability proportional to the square of its total fitness , since this counts the rate of reproduction for every possible mating pair in the population ( as matings occur with rates proportional to the fitness of each parent ) . We then step through one iteration of the well-mixed mating process within this subpopulation . Note that in this model the migration rate has a simple interpretation . The time between migrations is geometrically distributed with parameter m , so the mean time between migrations is 1/m time steps . Recall that a ‘generation’ is equal to the mean lifespan of an individual , that is , N reproduction events or N/ ( 1−m ) time steps . Then the typical time between migrations can be expressed with the units as generations: ( 3 ) E[T]=1−mNm . To compare our stochastic simulations with deterministic results , we use a recently published model ( Noble et al . , 2017 ) . From that work , we employ the ‘previous drive’ model , as it was designed to agree with the existing proof-of-concept CRISPR drive constructs that we consider here . Specifically , we consider the case of 1 guide RNA ( n=1 in that work’s notation ) , and zero production of costly resistant alleles ( γ=1 ) . Above , we present results from simulations which assume populations of size N=500 . We claim that N=500 is a reasonable approximation for the dynamics in the large-population limit , which is the relevant regime for widespread invasion or for species with very large population sizes , e . g . , mosquitoes . Here we briefly evaluate this claim . Figure 3 recreates Figure 1E from the main text with additional population sizes overlaid: N=1000 , 2500 , 5000 , and 10000 . The distributions narrow for larger N until plateauing at roughly N=5000 . However , the central tendencies show little change with increasing N . Several recent studies have explored the effect of pre-existing drive resistant alleles in a population brought about by standing genetic variation ( SGV ) at the target locus ( Unckless et al . , 2017; Drury et al . , 2017 ) . These studies developed deterministic models and showed that pre-existing resistant alleles—presumably neutral—should rapidly outcompete costly drives due to selection , resulting in rapid drive extinction . The study by Drury et al . ( Drury et al . , 2017 ) used sequencing to quantify this standing variation in diverse populations of flour beetles and found resistance-conferring mutations to exist at a wide range of frequencies , from 0 to 0 . 375 , with an average of roughly 0 . 1 . However , these studies were primarily concerned with long-term outcomes following drive release , in which case resistance certainly outcompetes the drive . For our purposes , however , we are concerned with the intermediate time regime in which the dynamics of resistance are less clear . Moreover , these studies employed deterministic models , whereas our model is stochastic . Here , we seek to understand the effect of SGV in our model . To incorporate SGV , we simply alter the initial conditions: rather than introducing i drive homozygotes into a population of N−i wild-type homozygotes , we introduce i drive homozygotes into a population consisting of j resistant homozygotes ( we choose resistant homozygotes for simplicity , since they rapidly go to Hardy-Weinberg equilibrium following release ) and N−i−j wild-type homozygotes . Figure 4 shows the effect of SGV on peak drive for pre-existing resistance frequencies up to 0 . 5 . We find that the effect of SGV is to linearly decrease the mean peak drive ( R2=0 . 996 ) . Our intuition for this result is as follows . Because the population is well-mixed , the effect of resistance is simply to decrease the size of the population that is susceptible to the effects of the drive . This can be roughly viewed as linearly scaling the drive-frequency axis . For example , if the population has a 0 . 1 frequency of resistant alleles immediately prior to release , then the population that is susceptible to drive is roughly 90% of the census population size , and the drive undergoes its usual dynamics within this subpopulation . There are of course complications to this simplistic explanation , e . g . , selection increasing the size of the resistant population and diploidy mixing resistant and drive alleles . Furthermore , the linear relationship only holds for sufficiently low levels of SGV . In our example here , the relationship holds to roughly 0 . 5 initial resistance frequency . However , this is still higher than would be anticipated for drives engineered to spread in the wild . Overall , our results suggest that a high level of SGV would be required to protect against drive invasion . In our conservative example ( Figure 4 ) assuming 0 . 5 homing efficiency , 0 . 9 drive fitness , and neutral resistance , pre-existing resistance of greater than 0 . 5 frequency is required to contain peak drive to below 10% of the population , compared to 35% in the absence of SGV . In the model presented above , we assume that each mating produces one offspring . However , a variety of application-relevant species are known to produce many offspring per mating . For example , female Anopheles gambiae mosquitoes can lay hundreds of eggs per lifetime ( Hammond et al . , 2016 ) . It is not clear , a priori , how varying the offspring number distribution in our model would affect the results presented above . Thus we here analyze a simple extension of the model which allows us to vary the number of offspring following a given mating event . To begin , recall our model . We consider a population of constant size N with the following process: At each time-step , two individuals are chosen for mating; an offspring is sampled according to the parental genotypes; a third individual is chosen for removal from the population , and the parents’ offspring takes its place . ( We implicitly assume that these offspring are only the offspring which successfully reach adulthood , i . e . , reproductive age ) . We now add a new parameter , k , which determines number of ( adult ) offspring produced by a mating pair . The process proceeds as before , except now k offspring are independently sampled from the parental genotypes , and k individuals are chosen uniformly ( without replacement ) for removal from the population . Clearly the model presented in the main text is the special case k=1 . Note that this parameter k is not equivalent to brood size , clutch size , egg batch size , etc . —values often considered in the ecological literature—in that k describes the number of offspring produced per mating which successfully attain reproductive age . This number can of course be much lower than these other parameters due to death during juvenile life stages . We provide an example calculation for this parameter in An . gambiae at the end of this section . We now argue that increasing the number of offspring per mating , k , corresponds to decreasing the effective size of the population , Ne . We omit rigorous proof here , but we provide a formula for the effective population size in our model and present numerical simulations as support . To begin , Hill showed in 1972 that the variance effective population size in the standard Moran model is ( Hill , 1972 ) ( 4 ) Ne=4N2+σX2 . Here N is the census population size , and σX2 is the variance in the distribution of the total number of offspring produced by an individual over the course of its lifetime ( i . e . , its lifetime reproductive success ) . It was proven that this formula holds both for the Wright-Fisher model with discrete generations and for the Moran model with overlapping generations , provided that σX2 is the same and that the total number of individuals entering the population in each generation is equal ( Hill , 1972 ) . Our model meets both of these requirements—indeed , the only difference is that two parents are chosen to sample offspring types , rather than one , and this has no bearing on the number of offspring produced—so we conjecture that Equation ( 4 ) holds for our case as well . To proceed , we calculate σX2 for our extended model and employ the variance effective population size given by Equation ( 4 ) . Consider one particular individual in the population , and let t=1 , 2 , … count time-steps . As described , in each step , k individuals are uniformly sampled ( without replacement ) for removal . Thus , an individual has probability k/N of dying in each step . Its lifespan , T , is thus geometrically distributed , T∼Geometric ( k/N ) . Next , let X be a random variable describing the number of offspring an individual produces in its lifetime , so that X\|T is the number of such events given that the individual survives T time-steps . Because each mating event is independent , ( X\|T ) ∼k⋅Bin ( T , 2/N ) . The success probability derives from the fact that two individuals are chosen for mating in each time-step and that the process is neutral . Thus , EX=EE[X\|T]=Ek ( 2/N ) T=k ( 2/N ) N/k=2andVar ( X ) =EVar ( X∣T ) +Var ( E ( X∣T ) ) =Ek2T ( 2/N ) ( 1−2/N ) +Var ( k ( 2/N ) T ) =kN ( 2/N ) ( 1−2/N ) + ( 2k/N ) 2N ( N−k ) /k2=4+2k ( N−4 ) /N . Returning to the variance effective population size expression in Equation ( 4 ) , we obtain for our model: ( 5 ) Ne=4N2k+6 . Note that in the case k=1 we recover Ne=N/2 , which is the variance effective population size for the standard Moran model . In Figure 5 , we present peak drive distributions ( as in Figures 1E and 3 ) for varying values of k with the effective population size , Ne , and effective release size , ie , both determined by Equation ( 5 ) , held constant . In this case we used Ne=250 and ie=8 , which correspond to N=500 and an initial release of i=16 in our standard model with k=1 . The peak drive distributions for all values of k studied are approximately identical . This suggests that the dynamics for larger k can indeed be inferred from the standard model with k=1 and population/release sizes appropriately scaled via Equation ( 5 ) . An immediate consequence of this result is that releases of organisms which have many offspring ( e . g . , mosquitoes ) are effectively smaller than would be expected from simply counting . For example , an organism which typically has 100 offspring that survive to adulthood would need a release size of roughly 258 to surpass the 10-individual initial release threshold we have observed . Note that the 10-individual threshold discussed throughout the text is the census release size; the effective release size is ie=5 . In Figure 6 , we recalculate the distributions in Figure 5 holding the actual population and release sizes constant , rather than their effective values . Two effects are apparent . First , the decrease in effective population size , Ne , leads to greater variation in peak drive among simulations that invade , i . e . , the distribution centered around ≈0 . 4 widens . Second , the decrease in effective release size , ie , leads to a greater probability of simulations immediately going extinct , i . e . , the relative mass of the mode centered around ≈0 increases . In sufficiently large populations the first effect would be less pronounced—see Figure 3—while the second effect should apply for any small release . Finally , as an example , we provide an estimate of our model’s k parameter for a particularly relevant species , An . gambiae . To do this , we find the typical size , n , of egg batches laid by females following a particular mating event; then we estimate the total number of these which survive to adulthood using parameters from the literature . The first number , n , varies according to a variety of environmental and ecological factors ( Hammond et al . , 2016; Yaro et al . , 2006 ) , so we assume a large but reasonable value in order to avoid underestimating our parameter k . For this , we assume that n≈186 , which is roughly the highest value observed by Hammond et al . in the CRISPR drive study ( Hammond et al . , 2016 ) and is in line with previous field work ( Yaro et al . , 2006 ) . To estimate the survival probability for each egg to adulthood , we employ the method and parameters presented by Deredec et al . ( Deredec et al . , 2011 ) Each egg goes through three juvenile stages before reaching adulthood—the egg stage , the larva stage , and the pupae stage . We denote the probabilities of surviving each of these stages by θ0 , θL , and θP , respectively . The probability of a particular egg reaching adulthood is then p=θ0θLθP . These parameters were estimated to be θ0=0 . 831 , θL=0 . 076 , and θP=0 . 831 . Thus we have p=0 . 0525 . Given this formulation , the number of eggs laid per mating event which reach adulthood is distributed according to Bin ( n , p ) . We take the mean of this distribution to obtain:k≈np=9 . 76 . Therefore , while An . gambiae females exhibit large egg batch sizes , the value of k for our model is much lower—indeed , low enough that the central tendency of the peak drive distribution remains roughly unchanged in Figure 6 . Above , we study various values of the homing efficiency , P , but we perform less exploration of the parameters governing drive fitness , f , and resistance cost , s . This is motivated primarily by the abundance of data for the former—see Appendix 1—table 1—and the lack of data for the latter parameters . In addition , we have assumed throughout that death rates are identical for the various genotypes , while reproductive events occur with probabilities proportional to fitness . On the other hand , some drive constructs might behave the opposite way: reducing fitness by increasing an organism’s death rate , while leaving its birth rate unchanged . In this section we explore these three effects: ( i ) varying drive fitness across its entire range , ( ii ) varying the fitness cost of resistance across its entire range , and ( iii ) modifying the model so that death rates are affected by fitness , rather than birth rates . To begin , we consider our standard model for fitness and study drive spread across the entire range of values for drive fitness , f , and homing efficiency , P . In particular , we consider 51 values of each parameter: P∈[0 , 1] and f∈[0 . 5 , 1] , both evenly spaced , for a total of 2601 parameter pairs . For each pair , the average peak drive is calculated over 100 simulations , and the results are shown in Figure 7 , left . We find that maximum drive frequencies of greater than 0 . 3 are common across a wide range of drive fitness values . In particular , for our lower-bound estimate of empirical drive efficiency ( P=0 . 5 ) , drives can confer fitness costs as high as 20% before the peak drive drops below 0 . 3 . For more typical empirical efficiencies ( P>0 . 8 ) , the peak drive is typically greater than 0 . 5 even for costly drives ( f≈0 . 7 ) , and low-cost drives ( f>0 . 9 ) have peak drive of greater than 0 . 9 . We next modified our standard well-mixed model in the following way . Recall that the model involves choosing two parents to mate , then choosing an individual to die and be replaced by the parents’ offspring . In our standard model , the two parents are chosen to reproduce with probabilities proportional to their fitnesses , and an individual is chosen to die uniformly . In our modified model , we choose the two parents uniformly and then choose the individual to die with probability proportional to the inverse of its fitness . Results from the modified model are shown in Figure 7 , right and are nearly identical to the results from the standard model . In both cases , it is important to note that the peak drive and likelihood of invasion deemed socially acceptable for accidental release would likely be lower than those discussed above . With this in mind , our simulations suggest that if a drive is predicted to invade by deterministic models ( i . e . , if it lies above the boundary in Figure 7 ) , then it will almost certainly reach a maximum frequency greater than 0 . 1 . While acceptable levels of peak drive are as-yet unknown and will likely vary between species , applications , jurisdictions and so on , spread to this extent will likely surpass it . Finally , we sought to understand the effect of varying the fitness cost associated with drive-resistance . Throughout the text above we have assumed that resistance is neutral , as this presumably represents the best case for containment . However , drive constructs developed for applications are likely to employ resistance-mitigating strategies , such as multiplex targeting of essential genes ( Esvelt et al . , 2014; Noble et al . , 2017 ) , which essentially increase the fitness cost associated with drive-resistance . Thus , we ran simulations varying drive-individual fitness , f , in the range f∈[0 . 5 , 1] , and resistant-individual ( RR ) fitness in the range [0 , 1] , assuming conservative drive efficiency , P=0 . 5 . In both dimensions we considered 51 parameter values , evenly spaced , for a total of 2601 parameter pairs . For each pair , the average peak drive is calculated over 100 simulations , and the results are shown in Figure 8 . We find qualitatively that there are two regimes , determined by the fitness cost of resistance , s ( i . e . , individuals with genotype RR have fitness 1−s ) , and the deterministic invasion condition , f ( 1+P ) >1 . In the figure , we assume that P=1/2 , so the deterministic invasion condition is simply f>2/3 . When the fitness cost of resistance , s , is sufficiently low ( s<1/3 ) , then the dynamics are determined by the relationship between the fitness of drive individuals and the fitness of resistant individuals: if the fitness of drive individuals is greater than the fitness of resistant individuals , then the spread of the drive is dramatically improved—typically reaching fixation—compared to the baseline neutral-resistance case . However , if the fitness cost of resistance is sufficiently high ( s>1/3 ) , then the improvement in drive spread brought about by increasing the cost of resistance saturates , since the drive can now be less costly than resistance ( f>1−s ) but also too costly to invade ( f<2/3 ) . That is , for resistance costs higher than 1/3 , the mean peak drive as a function of drive fitness , f , remains essentially unchanged with increasing s , since the deterministic invasion condition can no longer be satisfied when the drive has fitness f<2/3 , no matter the cost of resistance . Since the drive functions only in heterozygotes , inbreeding in a population—which in effect reduces the frequency of heterozygotes—would be expected to impact drive invasiveness . Indeed , this has been shown in recent theoretical studies by Bull , 2017 and Drury et al . ( 2017 ) Thus we here extend our well-mixed model to include inbreeding and study its effect . For simplicity , we consider a partial selfing model . In each update step of our process ( see Figure 1C ) , we typically choose two parents for mating with probabilities proportional to their fitnesses . To include selfing , we instead choose the first parent as usual , with probability proportional to its fitness . We then choose the first parent as the second parent as well with probability s; or , with probability 1−s , we choose a second parent from the remaining population , with probability proportional to its fitness . Note that the fitness of each offspring is determined entirely by its genotype and does not account for inbreeding depression . Implicitly , we thus consider the case of zero inbreeding depression . As this effect helps protect against drive invasion , we essentially consider the worst-case scenario for drive containment ( Bull , 2017 ) . Using our extended model , we then computed peak drive distributions for values of s between 0 and 1 and for the three values of P explored above: P=0 . 15 , 0 . 5 , 0 . 9 . The results are shown in Figure 9 . We find that a fairly high degree of selfing is required to impact the peak drive distribution in a meaningful way . For highly effective drive , P=0 . 9 , the mass of the upper mode in the frequency distribution is larger than the lower mode until roughly s≈0 . 75 . For conservative drive , P=0 . 5 , this occurs at roughly s≈0 . 6 , and for ineffective drive there is little change , as the maximum frequency begins very near zero . To compare with previous results , we can consider the inbreeding coefficient rather than the selfing probability . In our model , the inbreeding coefficient , F , is given by s/ ( 2−s ) . Thus highly effective drive can tolerate inbreeding of F≈0 . 6 and conservative drive can tolerate F≈0 . 43 . To show that the deterministic ODE solutions provide reasonable approximations to the typical behavior of our stochastic model , we overlay numerical solutions to the ODEs for the systems studied in Figure 1D of the main text . The results are shown in Figure 10 . Throughout we have assumed that resistance is neutral with respect to the wild-type . This assumption is biologically realizable as resistance is conferred by changing sequence homology to the drive’s gRNA—something that could be achieved with synonymous codon substitutions , for example . In practice , some resistance mutations could be costly and those that are neutral could be rare . However , assuming resistance is always neutral represents the worst-case scenario for drive invasiveness , as resistance can increase in frequency without being selected against with respect to the wild-type . When resistance is no longer assumed to be neutral , other interesting dynamics can occur ( Traulsen and Reed , 2012 ) . In particular , when resistance is costly with respect to the wild-type , but not so costly as the drive and its cargo , the dynamics resemble the Rock-Paper-Scissors game . This allows the drive to avoid extinction indefinitely . We consider a deme structured population , where each subpopulation has size N and there are n demes . We define a Moran-type process , where in each time step either a reproduction or migration event takes place ( illustrated in Figure 11 ) . A reproduction event occurs with probability 1−m and a migration event occurs otherwise . If a reproduction occurs , then a subpopulation is selected proportional to the square of its total fitness . Next , two individuals in the subpopulation are selected proportional to their fitnesses and they produce an offspring according to the mechanism above . Finally , another individual from the subpopulation is chosen uniformly at random for death . If a migration event occurs , then an individual is selected uniformly at random and migrates to a new subpopulation uniformly at random . We denote the proportion of genotype α at time t in the initial subpopulation by Ptα . The process begins with i drive homozygotes and N−i wild-type homozygotes in a single subpopulation . The remaining subpopulations consist only of wild-type homozygotes . Let E be the event that the frequency of drive alleles reaches 10% in a subpopulation other than where the drive was released , given that i drive homozygotes were released in the initial subpopulation . We assume that i is small with respect to N . As an aside , note that the choice of 10% is arbitrary—any other percentage ( less than the peak drive in the deterministic model , c ) would be equivalent if N is large enough . This is clear from Figure 1E , where either the drive does not invade and so peak drive is roughly equal to the initial frequency or the drive does invade and the peak drive is close to c . This claim is equivalent to stating that the probability that the drive starting at frequency c0 attains frequency c1 ( such that c0<c1<c ) before going extinct tends to 1 . This behavior is typical of Moran-type models , since the extinction probability of i drive homozygotes rapidly approaches 0 , even in an infinite population , as i increases ( Marshall , 2009 ) . Specifically , if we have i=c0N , then the extinction probability approaches 0 as N becomes large , and moreover , if the drive does not go extinct , then it behaves almost deterministically and will reach frequency c and thus also c1 . Returning to approximating the probability of E , note that for E to take place a drive allele has to migrate from the initial subpopulation and this allele has to survive stochastic fluctuations and avoid extinction in its new subpopulation . The drive alleles do not last indefinitely in the initial population . We denote the random time at which the drive alleles go extinct by T . As long as the initial drives do not go extinct due to stochastic fluctuations , the frequency of the drive increases rapidly , as it outcompetes the wild-type . Concurrently , resistant alleles are produced that eventually push the drive to extinction . This means that the drive has a finite time to migrate to other subpopulations . Although this process is stochastic it shows fairly deterministic behavior once there are a sufficient number of drive alleles ( see Figure 10 ) —that is , if the drive avoids immediate extinction . Let ei , j , be the probability that the drive survives stochastic fluctuations and avoids immediate extinction when starting with i drive homozygotes and j heterozygotes . Implicitly , here we = assume that ei , j does not depend on whether the heterozygotes are wild-type or resistant heterozygotes . Note that when i or j are 𝒪 ( N ) , ei , j is approximately 1 , so when i , j≪N , we assume that the probability that the drive migrates is approximately 0 . Moreover , since the drive will almost certainly go extinct , there is some time where the frequency of drive alleles is again much less than 𝒪 ( N ) . We also assume here that the probability that the drive migrates is approximately 0 . At each time step , there is a small probability that the drive migrates from the initial population and invades another subpopulation . To calculate , we first condition on the non-extinction of the initial i drive homozygotes . Second , we note that if the drive does not migrate and avoid extinction in another subpopulation , then it does not do so at any particular time t . Third , we assume that these events for each t are approximately independent . Finally , we numerically solve a deterministic ODE system representing the dynamics ( Noble et al . , 2017 ) to approximate the probability that the drive does not migrate at time t . Thus , P{E}=P{E\|drive avoids extinction}ei , 0+P{E\|drive does not avoid extinction} ( 1−ei , 0 ) ≈P{E\|drive avoids extinction}ei , 0≈ei , 0 ( 1−∏t=1T P{drive does not migrate and invade at time t} ) =ei , 0 ( 1−∏t=1T ( 1−P{drive invades\|drive migrates at time t}P{drive migrates at time t} ) ) =ei , 0 ( 1−∏t=1T ( 1−me1 , 0EPtDD−me0 , 1 ( EPtWD+EPtDR ) ) ) , since if the drive avoids extinction it will invade . Now we substitute the ODE solution ptαβ for EPtαβ in the above expression to find thatP{E}≈ei , 0 ( 1−exp⁡ ( N∫0T/ ( 1−λ ) dtlog⁡ ( 1−λe1 , 0p ( 1−λ ) tDD−λe0 , 1 ( p ( 1−λ ) tWD+p ( 1−λ ) tDR ) ) ) ) ≈ei , 0 ( 1−exp⁡ ( N1−λ∫0T dtlog⁡ ( 1−λe1 , 0ptDD−λe0 , 1 ( ptWD+ptDR ) ) ) ) . Here we approximated the product with an integral and used a change of variables . Note that if m=𝒪 ( 1/T ) and heuristically we replace EPtα in the above expressions with its time average , denoted ϕα , thenei , 0[1−∏t=1T ( 1−me1 , 0EPtDD−me0 , 1 ( EPtWD+EPtDR ) ) ]≈ei , 0[1− ( 1−e1 , 0ϕDD+e0 , 1 ( ϕWD+ϕDR ) T ) T]≈ei , 0[1−exp⁡ ( −e1 , 0ϕDD+e0 , 1 ( ϕWD+ϕDR ) ) ] . Thus , when the migration rate is on the order of the inverse of the drive extinction time , the invasion probability is order 1 .	Gene drive is a genetic engineering technology that can spread a particular suite of genes throughout a population . Among the types of gene drive systems , those based on the CRISPR genome editing technology are predicted to be able to spread genes particularly rapidly . This is because components of the CRISPR system can be tailored to replace alternative copies of a particular gene , ensuring that only the desired version is passed on to offspring . In this way , for example , a gene that prevents mosquitoes from carrying or transmitting the malaria parasite could be introduced to a very large wild population to reduce the incidence of the disease among humans . Gene drives can be “self-propagating” or “self-exhausting”: the former are designed so that they can always spread as long as there are wild organisms around , whereas the latter are expected to lose their ability to spread over time . Self-propagating CRISPR gene drives have been shown to work in controlled populations of fruit flies , mosquitoes and yeast . These experiments happen in a controlled environment in the laboratory , so the organisms edited to have the gene drive elements do not come in contact with susceptible wild organisms . However , if just a few were to escape , the gene drive could theoretically spread quickly outside the laboratory . Noble , Adlam et al . investigated , using mathematical models , whether or not – and how fast – a self-propagating CRISPR-based gene drive would spread if a number of organisms with the gene-drive elements were released into the wild . The models showed that the release of just a few of the edited organisms would result in the gene drive spreading to most populations that interbreed . This happened regardless of the structure of the wild populations or whether a degree of resistance to the drive emerged . As a result , even the smallest breach of a contained trial could lead to significant gene drive spread in the wild . The findings suggest that self-propagating gene drive technologies would be most useful where the invasion of most wild populations of the target species is the intended purpose , rather than a risk to be avoided . As a result , a self-propagating CRISPR-based gene drive could be well suited to spreading among mosquitoes to impede the malaria parasite , provided there were strong international agreements in place . The findings also underline the difficulty of carrying out safe field trials of self-propagating gene drives , and the need for very tight control of laboratories carrying out experiments in this field . Lastly , they highlight the importance of developing and testing the evolutionary stability of self-exhausting gene drives , which could be better contained to local populations .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2018	Current CRISPR gene drive systems are likely to be highly invasive in wild populations
The mechanistic Target of Rapamycin complex 1 ( mTORC1 ) senses intracellular amino acid levels through an intricate machinery , which includes the Rag GTPases , Ragulator and vacuolar ATPase ( V-ATPase ) . The membrane-associated E3 ubiquitin ligase ZNRF2 is released into the cytosol upon its phosphorylation by Akt . In this study , we show that ZNRF2 interacts with mTOR on membranes , promoting the amino acid-stimulated translocation of mTORC1 to lysosomes and its activation in human cells . ZNRF2 also interacts with the V-ATPase and preserves lysosomal acidity . Moreover , knockdown of ZNRF2 decreases cell size and cell proliferation . Upon growth factor and amino acid stimulation , mTORC1 phosphorylates ZNRF2 on Ser145 , and this phosphosite is dephosphorylated by protein phosphatase 6 . Ser145 phosphorylation stimulates vesicle-to-cytosol translocation of ZNRF2 and forms a novel negative feedback on mTORC1 . Our findings uncover ZNRF2 as a component of the amino acid sensing machinery that acts upstream of Rag-GTPases and the V-ATPase to activate mTORC1 . Mechanistic target of rapamycin ( mTOR ) is a conserved serine/threonine protein kinase of the phosphatidylinositol 3-kinase ( PI3K ) -related kinase family ( PIKK ) , which functions as the catalytic subunit of two distinct complexes , mTORC1 and mTORC2 ( Wullschleger et al . , 2006 ) . In mTOR complex 1 ( mTORC1 ) , mTOR associates with Raptor , mLST8 , PRAS40 and Deptor . The second mTOR-containing complex , mTORC2 , comprises mTOR , Rictor , mSIN1 , mLST8 , Deptor and Protor . The functions of the two complexes can be distinguished through the use of the mTOR inhibitor , rapamycin , which acutely and specifically inhibits mTORC1 , but has no short-term effect on mTORC2 ( Sarbassov et al . , 2006 ) . mTORC1 is a central integrator of regulatory inputs from intracellular amino acids , ATP , O2 and extracellular signals such as insulin and growth factors ( Dibble and Manning , 2013 ) . When active , mTORC1 initiates transcriptional , translational and post-translational responses to promote a multitude of anabolic activities and to suppress catabolic processes ( Duvel et al . , 2010; Howell et al . , 2013; Sarbassov et al . , 2005 ) . Through phosphorylation of its substrates , including ribosomal protein S6 kinase ( S6K ) and the eIF4E-binding proteins ( 4-EBP ) , mTORC1 stimulates mRNA translation and , ultimately , cell growth and proliferation ( Fingar and Blenis , 2004; Showkat et al . , 2014 ) . mTORC1 activation involves its recruitment onto late endosomes and lysosomes ( Ogmundsdottir et al . , 2012; Sancak et al . , 2010; Zoncu et al . , 2011 ) . Growth stimuli regulate mTORC1 via the heterotrimeric TSC1-TSC2-TBC1D7 complex , which negatively regulates the Rheb-GTPase , an activator of mTORC1 ( Dibble and Manning , 2013 ) . In parallel , amino acids signal to mTORC1 via the Rag GTPases , which consist of RagA or RagB bound to RagC or RagD ( Hirose et al . , 1998; Kim et al . , 2008; Sancak et al . , 2008 ) . Amino acids , through the Rag-GTPases , promote mTORC1 recruitment and activation at the lysosomal surface , where Rheb resides ( Dibble and Manning , 2013; Menon et al . , 2014; Sancak et al . , 2008 ) . Rag-GTPases are anchored to the lysosomes via the Ragulator complex that comprises of p18 , p14 , MP1 , HBXIP and C7orf59 proteins ( Bar-Peled et al . , 2012; Sancak et al . , 2010 ) . This pentameric Ragulator complex was shown to possess guanine nucleotide exchange factor ( GEF ) activity towards RagA and RagB ( Bar-Peled et al . , 2012 ) . More recently , work from the Sabatini group identified the GATOR1 complex as a GTPase-activating protein ( GAP ) for the RagA/B GTPases , and as a major negative regulator of the amino acid sensing pathway , loss of which causes mTORC1 signalling to be completely insensitive to amino acid deprivation ( Bar-Peled et al . , 2013 ) . Interestingly , the vacuolar proton-ATPase ( V-ATPase ) was reported to be involved in activation of mTORC1 by amino acids ( Zoncu et al . , 2011 ) , however the mechanisms linking the V-ATPase , amino acids and mTORC1 activation are still unclear . The V-ATPase pump was shown to interact with the Ragulator complex in an amino acid-sensitive manner . This pump also supports the amino acid-promoted interaction between Ragulator and Rags , and the export of amino acids from the lysosome via PAT1 ( proton-assisted transporter 1 ) ( Bar-Peled and Sabatini , 2014; Bar-Peled et al . , 2012; Nada et al . , 2014; Zoncu et al . , 2011 ) . The V-ATPase is a large , multisubunit H+ pump composed of V1 ( catalytic ) and V0 ( membrane-spanning ) subcomplexes . At the surface of membrane vesicles , V-ATPase couples the energy of ATP hydrolysis to proton translocation across plasma and intracellular membranes , which results in acidification of intracellular compartments such as secretory vesicles , early and late endosomes and lysosomes ( Forgac , 2007 ) . Inhibition of the V-ATPase by compounds such as conconamycin A or bafilomycin A results in increased lysosomal pH , as well as inhibition of the mTORC1 ( Hinton et al . , 2009 ) . Our previous study showed that the E3 ubiquitin ligase ZNRF2 is an enzyme tethered to intracellular membranes , via an N-myristoyl moiety , where it ubiquitylates the Na+/K+ATPase pump ( Hoxhaj et al . , 2012 ) . ZNRF2 is robustly phosphorylated on Ser19 , Ser82 and Ser145 in response to growth factors , phorbol ester ( PMA ) and forskolin . Akt and PKC were identified as kinases phosphorylating of Ser19 and Ser82 , respectively , and these sites are responsible for mediating the binding of ZNRF2 to 14-3-3 proteins ( Hoxhaj et al . , 2012 ) . Furthermore , the phosphorylations of Ser19 and Ser145 promote the release of ZNRF2 from intracellular membranes into the cytosol in an Akt-dependent manner ( Hoxhaj et al . , 2012 ) . Here , we show that ZNRF2 is a regulator of mTORC1 activation by amino acids . Upon growth factor and amino acid stimulation , mTORC1 phosphorylates ZNRF2 at Ser145 promoting its dissociation from membranes . We also show that the protein phosphatase 6 ( PP6 ) dephosphorylates ZNRF2 at Ser145 , re-localizing ZNRF2 to the membranes . Interestingly , we also find that on membranes ZNRF2 interacts with the V-ATPase and positively regulates its functions . Our findings present ZNRF2 as a positive regulator of nutrient-mediated mTORC1 signalling , which is also a negative feedback target of mTORC1 signalling . To better understand the molecular function of ZNRF2 , we aimed to identify ZNRF2-interacting proteins . To do this , extracts of HEK293 cells stably expressing GFP-ZNRF2 ( N-terminal tag , non-myristoylated ) and ZNRF2-GFP ( C-terminal tag , myristoylated ) were subjected to immunoprecipitation . After SDS-PAGE , strong bands at the molecular weights expected for the GFP-tagged ZNRF2 proteins were identified as such by mass spectrometric analyses ( Figure 1—figure supplement 1a , b ) . As reported previously , the E2 conjugating enzyme UBE2N/UBC13 co-purified with both forms of ZNRF2 , whereas the Na+/K+ATPase ATP1A1 subunit co-purified only with the N-myristoylated ZNRF2-GFP protein ( Hoxhaj et al . , 2012 ) . In addition , we identified mTOR as a high-score hit in the immunoprecipitates of N-myristoylated ZNRF2-GFP protein ( Figure 1—figure supplement 1b ) . The interactions of mTOR with ZNRF2 was confirmed by Western blotting , which showed that endogenous mTOR bind to ZNRF2-GFP , but not to the GFP-only control nor to an N-myristoylation-defective mutant ( G2A ) of ZNRF2 ( Hoxhaj et al . , 2012 ) , indicating that N-myristoylation of ZNRF2 is important for this interaction ( Figure 1a ) . ZNRF2 also interacted with other components of the mTORC1 complex , namely raptor and mLST8 ( Figure 1b and Figure 1—figure supplement 1c ) and showed co-localization with mTOR in HEK293 cells ( Figure 1—figure supplement 1d ) . To test whether the binding of ZNRF2 to mTOR was direct or mediated by one of the mTORC1 or mTORC2 components , we immunoprecipitated ZNRF2-GFP from cells depleted of Raptor or from mouse embryonic fibroblast ( MEF ) cells lacking rictor , Sin1 and mLST8 ( Figure 1—figure supplement 1e , f , respectively ) . ZNRF2 interacted with mTOR under all these conditions , indicating that raptor , rictor , Sin1 and mLST8 do not mediate the binding of ZNRF2 to mTOR ( Figure 1—figure supplement 1e , f ) . We next aimed to identify regions in ZNRF2 responsible for interacting with mTOR . mTOR bound to both the N-terminal ( 1 to 156 ) region of ZNRF2 and C-terminal RING ( catalytic ) domain-containing region of ZNRF2 , provided these fragments also carried an N-terminal myristoyl group ( Figure 1c ) , but was not able to interact to ZNRF2 with a defective UBZ domain ( Cys160Ala/Cys163Ala mutations ) or to the RING domain alone ( residues 156-end ) . These data indicate that membranal localization is required for the mTOR-ZNRF2 interaction , consistent with lack of mTOR binding to non-N- myristoylated ZNRF2 ( Figure 1a ) , and suggest that while the UBZ domain is involved , more than one domain of ZNRF2 contributes to mTOR binding . 10 . 7554/eLife . 12278 . 003Figure 1 . ZNRF2 binds to mTOR and is a substrate of mTORC1 . ( a ) Lysates of cells expressing GFP control , ZNRF2-GFP and a myristoylation mutant G2A-GFP ( with GFP-tag at the C-terminus ) were subject to immunoprecipitation with GFP-Trap beads . Precipitates were immunoblotted with the indicated antibodies . ( b ) HEK293 Flp-In cells that stably express GFP or GFP-tagged ( N-terminus and C-terminus ) ZNRF2 were harvested in 0 . 3% CHAPS lysis buffer and GFP tagged proteins were immunoprecipitated with GFP-Trap beads . Precipitates were analyzed by Western blotting with the indicated antibodies . ( c ) Full-length ZNRF2 and the indicated fragments of ZNRF2 ( all with C-terminal GFP tags ) were tested for binding to endogenous mTOR . C160A/C163A represents ZNRF2 with the zinc-coordinating cysteines of the UBZ domain changed to alanines . Below , schematic diagram of full-length ZNRF2 containing UBZ ( zinc finger ) and Ring ( catalytic ) domains is shown . ( d ) Kinase assays were performed using mTOR immunopurified from HEK293 cells and bacterially purified GST-ZNRF2 as substrate in the presence of Mg2+[γ-32P]ATP for 30 min . Where indicated , 1 µM of catalytic mTOR inhibitor ( Ku-0063794 ) was added to the reaction and incubated for 10 min at 4oC before the addition of the GST-ZNRF2 . Reactions were subjected to SDS-PAGE and autoradiography . CBB , Coommassie brilliant blue . ( e ) As in ( d ) , except that phosphorylated GST-ZNRF2 was digested with trypsin and tryptic peptides were analyzed by LC-MS on an ABI 4000 Q-TRAP system using precursor ion scanning in negative mode , searching for the ( PO3- ) ion ( -79 Da ) . The extracted ion chromatograph for phospho-Ser145 is presented . ( f ) HEK293 cells stimulated with IGF1 , in the presence or absence of the indicated amounts of catalytic mTOR inhibitor ( Ku-0063794 ) , were tested for phosphorylation of S145 by Western blotting after immunoprecipitating endogenous ZNRF2 . The phosphorylation status of Thr389 p70S6-kinase ( p-S6K ) was assayed . p-S6K is a marker of mTORC1 activation . S6K and GAPDH were used as controls . ( g ) HEK293 cells were starved of amino acids ( 1 . 5hr ) and stimulated with IGF1 or amino acids , in the presence or absence of rapamycin . Endogenous ZNRF2 was immunoprecipitated and its phosphorylation ( p-S19 and p-S145 ) was assayed by Western blotting . ( h ) HEK293 cells were starved of amino acids as in ( g ) and cycloheximide was added in the presence or absence of rapamycin . Phosphorylation of ZNRF2 was assayed as in ( g ) . ( i ) Cell lysates of HEK293 Flp-In cells stably expressing ZNRF2-GFP were treated with 20 nM rapamycin for the indicated times . Phosphorylation of ZNRF2 was assayed as in ( g ) . ( j ) HEK293 Flp-In cells that stably express ZNRF2-GFP were stimulated with IGF1 or serum in the presence or absence of the indicated amounts of S6K1 inhibitor ( PF-4708671 ) . Cell lysates were tested for phosphorylation of S145 of ZNRF2 . ( k–l ) Rictor and mLST8 wild-type and knock-out MEFs were transfected with ZNRF2-GFP for 36 hr . Cells were treated with 20 nM rapamycin for 30 min in fresh media containing 10% FBS . ZNRF2-GFP was immunoprecipitated and precipitates were immunoblotted with p-S145 antibody of ZNRF2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 00310 . 7554/eLife . 12278 . 004Figure 1—figure supplement 1 . ZNRF2 is a target of mTORC1 and interacts with mTOR . ( a ) Plasmids pcDNA5 . 1–FRT–TO ( Invitrogen ) , expressing ZNRF2 with a N-terminal GFP tag ( top ) or a C-terminal GFP tag ( bottom ) , were stably integrated into HEK293 Flp-In T-Rex cells . The expression of proteins was induced with tetracycline ( 1 µg/ml ) for the indicated times . ( b ) HEK293 Flp-In T-Rex cells that stably express GFP ( control ) and GFP-tagged ( N-terminus and C-terminus ) ZNRF2 were induced with tetracycline ( 1 µg/ml ) for 36 hr before lysis . These extracts were subjected to immunoprecipitation with GFP-Trap beads and precipitates were subjected to SDS-PAGE . The gel was fixed and stained with Colloidal Blue . The gel lanes were cut into slices , as indicated , and the proteins were digested with trypsin before mass spectrometric fingerprinting . ( c ) Input blots of Figure 1b . ( d ) HeLa cells were transfected with ZNRF2-FLAG for 12 hr . Images of cells co-immunostained for FLAG ( red ) and mTOR ( green ) are shown , together with the merged images . Inset shows a higher magnification of a selected field . ( e ) HEK293 Flp-In cells stably expressing ZNRF2-GFP were transfected with siRNAs targeting raptor . Cell extracts were subjected to immunoprecipitation with GFP-Trap and analyzed by Western blotting with the antibodies indicated . ( f ) Rictor , Sin1 and mLST8 wild-type and knockout MEFs were transfected with ZNRF2-GFP for 48 hr . Cells were lysed in 1% Triton lysis buffer . ZNRF2-GFP was immunoprecipitated with GFP-Trap® beads and the binding to mTOR was tested . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 00410 . 7554/eLife . 12278 . 005Figure 1—figure supplement 2 . Conservation of ZNRF2 in eukaryotes . ( a ) Phylogeny tree depicting conservation of amino acid sequence of ZNRF2 orthologues among eukaryotes . ( b ) Matrix plot showing percentage amino acid identity among ZNRF2 orthologues . ( c ) Amino acid sequence alignment showing conservation of N-myristoylation signal , Ser145 residue ( asterisk ) , UBZ and RING domains . Putative ZNRF2 orthologue in D . melanogaster is an uncharacterized protein called CG14435 , which shows 73% amino acid identity with human ZNRF2 , with an E-value of 10–49 and sequence coverage of 40% . The putative ZNRF2 orthologue in C . elegans is an uncharacterized protein called Y71F9AL . 10 , which shows 56% amino acid identity with human ZNRF2 , with an E-value of 6x10-42 and sequence coverage of 48% . The putative ZNRF2 orthologue in S . cerevisiae is a protein called Pib1p , which shows 47% amino acid identity with human ZNRF2 , with an E-value of 2x10-6 and sequence coverage of 14% , according to BLAST alignment . Importantly , putative ZNRF2 orthologues in vertebrates , D . melanogaster , C . elegans and S . cerevisiae share the highest similarity with human ZNRF2 in the UBZ and RING domains . Notably , the fruit fly and nematode orthologues of ZNRF2 also have a conserved N-myristoylation signal . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 00510 . 7554/eLife . 12278 . 006Figure 1—figure supplement 3 . Ser145 phosphorylation of ZNRF2 promotes its dissociation from membranes . ( a ) Live cells from U2OS-Flp-In expressing stably ZNRF2-GFP were imaged under serum conditions in the presence or absence of 200 nM rapamycin . The colocalization with Golgi marker ( CellLight Golgi-RFP ) and lysosome marker ( Lysotracker ) , was tested . ( b ) Live cells from ZNRF2-Ser145A U2OS-Flp-In stables were imaged under serum conditions . Colocalization with Golgi marker ( CellLight Golgi-RFP ) and lysosome marker ( lysotracker ) is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 006 We hypothesized that ZNRF2 is a substrate of mTOR . To explore this , we first performed in vitro kinase assay using bacterially purified ZNRF2-GST as a substrate of mTOR , which was immunoprecipitated from HEK293 cells . mTOR could phosphorylate ZNRF2-GST in vitro , and this was prevented by the mTOR-specific inhibitor Ku-0063794 ( Figure 1d ) . Mass spectrometric analysis revealed a single phosphorylated residue , namely Ser145 ( Figure 1e ) . Though ZNRF2 is conserved across eukaryotes ( Figure 1—figure supplement 2a–c ) , Ser145 is conserved only in the vertebrate proteins ( Figure 1—figure supplement 2c ) . To validate this phosphorylation , we raised a phospho-specific antibody against Ser145 . In cells , the mTOR catalytic inhibitor , Ku-0063794 , decreased Ser145 phosphorylation in response to IGF1 ( Figure 1f ) . Since this compound inhibits both mTORC1 and mTORC2 , to distinguish between these complexes , we selectively stimulated mTORC1 with either amino acids or cycloheximide ( CHX ) in the absence of growth factors ( Figure 1g , h , respectively ) . Ser145 was markedly phosphorylated in response to these treatments and this effect was blocked by the mTORC1 specific inhibitor , rapamycin ( Figure 1g , h ) , suggesting that mTORC1 rather than mTORC2 mediates Ser145 phosphorylation . However , while mTORC1 is solely responsible for amino acid-stimulated phosphorylation of Ser145 , rapamycin decreased , but did not abolish Ser145 phosphorylation in response to IGF1 ( Figures 1g ) , which suggests that another IGF1-activated kinase can also phosphorylate this site . The phosphorylation of Ser145 was not affected by the p70 S6 kinase 1 ( S6K1 ) -specific inhibitor PF-4708671 ( Pearce et al . , 2010 ) , indicating that ZNRF2 is directly phosphorylated by mTORC1 rather than via S6K ( Figure 1j ) . Moreover , phosphorylation of Ser145 was not decreased in rictor or mLST8 knock-out MEFs , ruling out the possibility that mTORC2 phosphorylates this site ( Figures 1k and 1l , respectively ) . Together , these data are consistent with Ser145 of ZNRF2 being phosphorylated by mTORC1 in response to growth factors and amino acids . Previously , we reported that ZNRF2 is translocated from intracellular membranes to the cytosol upon IGF1 stimulation and that Ser145Ala mutation hampered the IGF1-stimulated dissociation of ZNRF2 into the cytoplasm ( Hoxhaj et al . , 2012 ) . Consistent with mTORC1 being a Ser145 kinase , rapamycin increased the membranal localization of ZNRF2-GFP ( Figure 1—figure supplement 3a; Hoxhaj et al . , 2012 ) . In addition , Ser145Ala mutation of ZNRF2 resulted in a more distinct membranal localization in comparison to the wild-type protein ( Figure 1—figure supplement 3b ) . The expressed ZNRF2-GFP and Ser145Ala-ZNRF2-GFP co-localized predominantly with Golgi and to a lesser extent with the lysosomes ( Figure 1—figure supplement 3a , b ) . In addition to mTOR , several components of the protein serine/threonine phosphatase PP6 complex were identified as high-score hits in ZNRF2 immunoprecipitates analyzed by mass spectrometry ( Figure 1—figure supplement 1b ) . The PP6 holoenzyme is a heteromeric complex , comprising the catalytic ( PPP6C ) , three regulatory subunits PP6R1/2/3 ( also called SIT4 phosphatase–associated protein ( SAPS1/2/3 ) and three ankyrin repeat-domain containing regulatory subunits ( ANKRD28/44/52 ) ( Stefansson and Brautigan , 2006; Stefansson et al . , 2008 ) . The binding of PP6 complex components to ZNRF2-GFP was unaffected by the position of the GFP tag ( N- or C-terminus ) suggesting that the N-myristoylation of ZNRF2 is not required for the interaction of PP6 to ZNRF2 ( Figure 2a and Figure 2—figure supplement 1a ) . Interestingly , binding of the PP6 complex to ZNRF2 was compromised by the E3 ligase-dead ( Cys199Ala ) mutation ( Figure 2a and Figure 2—figure supplement 1a ) . Consistent with this , ZNRF2 lacking the catalytic RING domain ( fragments 1–180 and 1–156 ) did not bind to PP6 ( Figure 2b and Figure 2—figure supplement 1b ) . 10 . 7554/eLife . 12278 . 007Figure 2 . PP6 interacts with ZNRF2 and dephosphorylates phosphoS145 of ZNRF2 . ( a ) Lysates of cells expressing GFP , GFP-ZNRF2 and ZNRF2-GFP ( with GFP-tag at N- and C-terminus , respectively ) , and ligase-dead C199A-ZNRF2-GFP were subject to immunoprecipitation with GFP-Trap beads . Precipitates were immunoblotted with the indicated antibodies against the PP6 complex components . ( b ) As in Figure 1c , except that the immunoprecipitates were blotted for the PP6 complex component . ( c ) HEK293 Flp-In cells stably expressing ZNRF2-GFP were transfected with siRNAs targeting individual PP6 components or control ( Ctl ) . Cell extracts were subjected to immunoprecipitation with GFP-Trap and analyzed by Western blotting with the antibodies indicated . ( d ) Purified ZNRF2-GFP was incubated with wild-type or catalytically-inactive ( H55Q/R85A ) recombinant PPP6C-FLAG for 2 hr and then analyzed for phosphorylation of p-S145 and p-S19 of ZNRF2 . ( e ) HEK293 Flp-In cells expressing ZNRF2-GFP were transfected with control ( Ctl ) siRNA or siRNA targeting PPP6C and were incubated with two different concentrations of rapamycin ( 10 and 50 nM ) to inhibit mTORC1 . p-S145 and total ZNRF2 were assessed . ( f ) The ratio of p-S145 ZNRF2/ZNRF2 of Figure 2e is presented . ( g ) HEK293 cells were transfected to express GFP ( control ) and untagged ZNRF2 ( wild-type and S145A mutant ) were starved of amino acids ( 1 . 5hr ) , and stimulated with amino acids for 10 or 20 min . Western blotting was used to measure the phosphorylation status of S6K and levels of S6K and ZNRF2 . GAPDH was used as the loading control . ( h ) HEK293 transfected with GFP control or S145A mutant or a combination of S145A with myristoylation mutant G2A were starved of amino acids and stimulated with amino acids for 20 min . Cell lysates were immunoblotted as in ( g ) . ( i ) Model depicting ZNRF2 which is phosphorylated on Ser145 by mTORC1 and this site is dephosphorylated by PP6 . Phosphorylation of Ser145 contributes to the release of ZNRF2 from membranes into cytosol . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 00710 . 7554/eLife . 12278 . 008Figure 2—figure supplement 1 . PP6 interacts with and dephosphorylate ZNRF2 . ( a ) Input blots of Figure 2a . ( b ) Input blots of Figure 2b . ( c ) Input blots of Figure 2c . ( d ) HEK293 cells transfected with control siRNA ( Ctl ) or siRNA targeting PP6 were serum starved . Cell lysates were analyzed for p-S6K ( Th389 ) , p-ULK1 ( S757 ) , 4EBP1 , S6K , PP6 and GAPDH . ( e ) p-S6K/S6K ratio of immunoblots in Figure 2f , representative of 3 independent experiments . ( f ) HEK293 transfected with GFP control , WT , E3 ligase dead mutant ( C199A ) , S145A mutant or a combination of S145A with E3 ligase dead mutant ( S145A/C199A ) were starved of amino acids and stimulated with amino acids for 20 min . Cell lysates were immunoblotted with the indicated proteins from biological duplicates . The ratio of p-S6/S6 of immunoblots is presented in the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 008 To identify which subunits mediate the interaction of the PP6 holoenzyme with ZNRF2 , cells stably expressing ZNRF2-GFP were transfected with siRNAs targeting individual PP6 components ( Figure 2c and Figure 2—figure supplement 1c ) . Only knockdown of PPP6C resulted in a decrease in the levels of other subunits in cell lysates ( Figure 2—figure supplement 1c ) . Depletion of PPP6C , PP6R1 ( SAPS1 ) and PP6R3 ( SAPS3 ) did not affect the ability of other PP6 holoenzyme components to interact with ZNRF2 . In contrast , knockdown of PP6R2 ( SAPS2 ) and ANKRD28 decreased the amounts of PPP6C , SAPS1 and SAPS3 that co-immunoprecipitated with ZNRF2-GFP . These data suggest that SAPS2 and ANKRD28 mediate the interaction of the PP6 complex with ZNRF2 ( Figure 2c ) . The yeast orthologue of PP6 ( Sit4 ) , functions downstream of the rapamycin-sensitive TOR complex 1 to regulate G1 to S phase cell cycle progression , Gcn2-regulated translation and expression of certain nitrogen catabolite-repressed genes ( Beck and Hall , 1999; Morales-Johansson et al . , 2009 ) . PP6 is also proposed to dephosphorylate GCN2 when mTORC1 is inhibited in mammalian cells ( Wengrod et al . , 2015 ) . Since we found that ZNRF2 is a substrate of mTORC1 and interacts with PP6 , we tested whether PP6 could dephosphorylate ZNRF2 in vitro . Active PP6 holoenzyme , but not the catalytically inactive mutant , dephosphorylated phosphoSer145 , but not phosphoSer19 ( Akt site ) in vitro ( Figure 2d ) . Importantly , rapamycin treatment of HEK293 cells decreased phosphorylation of Ser145 and this was rescued by PPP6C knockdown ( Figure 2e , f ) . These data suggest that Ser145 is phosphorylated by mTOR and dephosphorylated by PP6 in vivo ( Figure 2i ) . We also tested the effect of PP6 knock-down on known substrates of mTORC1 such as p-S6K , p-ULK1 and 4EBP1 , and observed that while PP6 knockdown increases S6K phosphorylation , it does not affect ULK1 phosphorylation or 4EBP1 mobility shift ( Figure 2—figure supplement 1d ) . To our knowledge , this is the first report of PP6-dependent changes in S6K phosphorylation . Furthermore , we investigated the role of ZNRF2 Ser145 phosphorylation on mTORC1 signaling and found that overexpression of untagged ZNRF2 significantly increased the activation of mTORC1 by amino acids , while overexpression of ZNRF2-Ser145Ala augmented this effect , suggesting that phosphorylation of Ser145 negatively regulates ZNRF2 function towards mTORC1 ( Figure 2g , i and Figure 2—figure supplement 1e ) . This augmented activation of mTORC1 did not occur when the overexpressed ZNRF2-Ser145Ala mutant also carried a Gly2Ala mutation , rendering ZNRF2 non-myristoylatable ( Figure 2h ) . Interestingly , enhancement of mTORC1 activation by amino acids upon expression of wild-type or Ser145A ZNRF2 was decreased 2-fold by ligase dead C199A mutation ( Figure 2—figure supplement 1f ) . These findings suggest that both membranal localization of ZNRF2 and its E3 ubiquitin ligase activity are important for its effect on mTORC1 activation by amino acids . We next explored whether ZNRF2 plays a role in mTORC1 signalling . Activation of mTORC1 by amino acids was decreased in HeLa and HEK293 cells upon knockdown of ZNRF2 , as detected by a decrease in phosphorylation of S6K and 4E-BP1 ( Figure 3a and Figure 3—figure supplement 1a–d ) . Since amino acids stimulate lysosomal recruitment and activation of mTORC1 ( Sancak et al . , 2008; 2010 ) we asked whether ZNRF2 affected the lysosomal translocation of mTOR . After addition of amino acids , mTOR appeared in puncta-like structures that colocalize with lysosomes . The mTOR-LAMP2 colocalization was reduced by 30% upon knockdown of ZNRF2 ( Figure 3b , c ) . Moreover , ZNRF2 knockdown resulted in decreased cell proliferation and cell size in HeLa cells ( Figure 3d and Figure 3—figure supplement 1e , respectively ) . In contrast , ZNRF2 knockdown had no negative effect on activation of the PI 3-kinase pathway in response to IGF1 ( Figure 3—figure supplement 1f ) . In fact , we noticed an increase in the phosphorylation of Ser473 and Thr308 of Akt and a decrease in the phosphorylation of 4E-BP1 , upon ZNRF2 knockdown ( Figure 3—figure supplement 1f ) , consistent with lower mTORC1 signalling relieving negative feedback via IRS1 and Grb10 ( Harrington et al . , 2004; Hsu et al . , 2011; Manning , 2004; Shah et al . , 2004; Yu et al . , 2011 ) . 10 . 7554/eLife . 12278 . 009Figure 3 . Depletion of ZNRF2 attenuates the activation of mTORC1 by amino acids . ( a ) HeLa cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were starved of amino acids ( 1 . 5 hr ) , and stimulated with amino acids for 10 or 20 min . Cell lysates were subjected to immunoblotting with the indicated antibodies . p-S6K is a marker of mTORC1 activation . ( b ) HeLa cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were starved of amino acids ( -AA ) , and stimulated with amino acids for 20 min . Images of HeLa cells ( with or without amino acids ) co-immunostained for mTOR ( green ) and LAMP2 ( red ) are shown , together with the merged images . Inset shows a higher magnification of a selected field . ( c ) The graph displays the quantification of cells displaying lysosomal spots of mTOR fluorescence after addition of amino acids . N = ~100 cells per condition . Data are presented as mean ± S . E . M from two independent experiments . Two-tailed Student’s t tests were used for the pairwise comparison . *p < 0 . 0001 . ( d ) Cell viability of HeLa cells after knockdown of ZNRF2 was assessed daily using CellTiter-Glo Luminescent Assay . Data are mean ± S . E . M of biological triplicates from two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 00910 . 7554/eLife . 12278 . 010Figure 3—figure supplement 1 . Depletion of ZNRF2 decrease the activation of mTORC1 by amino acids . ( a ) The p-S6K/S6K ratio of immunoblots in Figure 3a . ( b ) HEK293 cells expressing control shRNA ( Ctl ) or two independent shRNAs targeting ZNRF2 were starved of amino acids for 1 . 5 hr , and stimulated with amino acids for 10 or 20 min . Western blotting was used to measure the phosphorylation status and levels of the indicated proteins . ( c ) HEK293 cells expressing control shRNA ( Ctl ) and shRNAs targeting ZNRF2 were starved of amino acids in the presence of 10% dialyzed FBS for 1 . 5 hr and stimulated with amino acids for 15 min . Western blotting was used to measure the phosphorylation status of S6K and levels of the indicated proteins . ( d ) HeLa cells expressing control shRNA ( Ctl ) and shRNAs targeting ZNRF2 were amino acids starved for 1 . 5 hr , and stimulated with the indicated concentrations of amino acids for 20 min . Western blotting was used to measure the phosphorylation status and levels of the indicated proteins . ( e ) Cell size of HeLa cells was measured after ZNRF2 knockdown using flow cytometry . The x-axis shows Forward Scatter ( FSC ) , which reflects cell size . The counts are shown on the y-axis . Data shown are representative of two independent experiments performed in duplicate . ( f ) HEK293 cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were serum starved ( 15hr ) and stimulated with the indicated amount of IGF1 for 20 min . Lysates were immunoblotted with the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 010 In the ZNRF2-GFP immunoprecipitates , we also identified components of the V-ATPase and the Ragulator complex ( Figure 4a and Supplementary file 1 ) . In particular , the V0A1/A2 components of the V-ATPase and p18 subunit of Ragulator were identified with high scores . Other V-ATPase subunits , namely V0D1 , V1A , V0C and V1E; and the Mp1 and p14 subunits of Ragulator were also identified with relatively lower scores ( Figure 4a and Supplementary file 1 ) . 10 . 7554/eLife . 12278 . 011Figure 4 . ZNRF2 interacts with V-ATPase and Ragulator . ( a ) Cartoon representation of V-ATPase and Ragulator subunits from mass spectrometry analyses of immunoprecipitates of ZNRF2-GFP overexpressed in HEK293 cells . The subunits are colour-coded according to the number of peptides identified by mass spectrometry and a scale is shown at the bottom . V0 and V1 subunits of the V-ATPase are presented in small or capital letters , respectively . ( b ) Lysates of cells expressing GFP , GFP-ZNRF2 , ZNRF2-GFP ( with GFP-tag at N- and C-terminus , respectively ) , or ligase-dead C199A-ZNRF2-GFP were subject to immunoprecipitation with GFP-Trap beads . The immunoprecipitates were immunoblotted with antibodies against the V0A2 component of V-ATPase and the p18 component of Ragulator . ( c ) HEK293 Flp-In cells that stably express ZNRF2-GFP or G2A-ZNRF2-GFP were treated with 2 µM bafilomycin for 60 min in the presence of serum or amino acid free medium , or amino acid free medium with 15 min stimulation with amino acids . The lysates were subjected to immunoprecipitation with GFP-Trap beads and the precipitates were blotted with V0A2 subunit of V-ATPase and p18 . ( d ) HEK293 Flp-In cells that stably express ZNRF2-GFP were treated with 1 µM concanamycin A ( ConA ) for 90 min . ZNRF2–GFP was immunoprecipitated and the interaction with V0A2 subunit of V-ATPase and p18 was tested . ( e ) Full-length ZNRF2 and the indicated fragments of ZNRF2 ( all with C-terminal GFP tags ) were tested for binding to endogenous V0A2 ( V-ATPase component ) and p18 subunit of Ragulator . ( f ) HEK293 Flp-In cells stably expressing ZNRF2-GFP were transfected with siRNA targeting control ( Ctl ) , p18 or V0A2 . The lysates were subjected to immunoprecipitation with GFP-Trap beads . Precipitates were immunoblotted with the indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 011 Similar to mTOR , V0A2 and p18 co-purified with C-terminally tagged ZNRF2-GFP , but not N-terminally tagged GFP-ZNRF2 ( Figure 4b ) nor with the Gly2Ala myristoylation-defective mutant of ZNRF2 ( Figure 4c ) , indicating that myristoylation of ZNRF2 and/or its localization to membranes is required for its binding to V-ATPase and p18 . The ligase-dead ZNRF2 ( Cys199Ala ) also displayed decreased binding to V0A2 and p18 ( Figure 4b ) . Similar to mTOR , V0A2 and p18 also bind to ZNRF2 mainly through the N-terminal region of ZNRF2 ( residues 1–156 ) , although residual binding was also observed with the RING domain N-terminally fused to the N-myristoylation motif ( Figure 4e ) . These experiments indicate that the binding of ZNRF2 to V-ATPase is specific and occurs at the membranes . The binding of ZNRF2 to V0A2 was enhanced when cells were treated with bafilomycin A ( Baf . ) or concanamycin A ( Con A ) , specific V-ATPase inhibitors ( Figure 4c and 4d , respectively ) . However , the interactions of ZNRF2 with V-ATPase and Ragulator were not markedly regulated by amino acids ( Figure 4c ) . Since p18 binds to V-ATPase , we tested whether the interaction of ZNRF2 with the V0A2 was direct or mediated by p18 . Knockdown of p18 did not affect interaction of ZNRF2-GFP with V0A2 , whereas knockdown of V0A2 decreased the levels of p18 in ZNRF2-GFP immunoprecipitates ( Figure 4f ) . Thus , binding of ZNRF2 to V-ATPase is likely to be direct and not mediated by p18 . We next investigated whether ZNRF2 knockdown influences V-ATPase function , particularly in the context of its role in activating mTORC1 by amino acids . Known mechanisms for regulation of the V-ATPase include ( i ) reversible dissociation of the V0 and V1 subcomplexes , and ( ii ) trafficking of the V-ATPase from its site of synthesis in the endoplasmic reticulum ( ER ) to other membranal compartments ( Forgac , 2007 ) . ZNRF2 knockdown did not alter the elution profiles of the V0 and V1 subunits of V-ATPase , as assessed by size-exclusion chromatography , suggesting that ZNRF2 does not affect the association between V0 and V1 compartments ( Figure 5—figure supplement 1a ) . Likewise , ZNRF2 knockdown caused no significant changes in the elution profiles of Ragulator subunit p18 , mTOR , raptor or rictor ( Figure 5—figure supplement 1a ) . ZNRF2 knockdown did not alter the protein levels of V-ATPase or Ragulator ( Figure 5—figure supplement 1b ) . These experiments suggest that lack of ZNRF2 does not compromise the abundance or formation of V-ATPase and mTORC1 complexes . We next assessed whether ZNRF2 is able to affect the trafficking of V-ATPase . To do this we utilized concanamycin A ( ConA ) , a specific inhibitor of V-ATPase and Brefeldin A ( BFA ) , a lactone antibiotic which arrests trafficking of membranal proteins from ER ( Klausner et al . , 1992 ) . ConA binds tightly to V-ATPase and it has been proposed that de novo synthesis of the pump ( and therefore its trafficking from ER to lysosomes ) is required for the recovery from the ConA treatment ( Hanada et al . , 1990 ) . Therefore , we knocked down ZNRF2 , treated the cells with ConA or BFA , and assessed the activation of mTORC1 in the presence of amino acids after ConA or BFA were washed off . We found that the recovery of mTORC1 activity after ConA or BFA chase was markedly attenuated by ZNRF2 knockdown ( Figure 5a , b , respectively ) , and enhanced by ZNRF2-Ser145Ala overexpression ( Figure 5c , d ) . 10 . 7554/eLife . 12278 . 012Figure 5 . Knockdown of ZNRF2 affects V-ATPase function and lysosomal pH . ( a ) HEK293 cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were starved of amino acids in the presence of 1 µM concanamycin A ( conA ) for 90 min . Then , ConA containing media was washed off and cells were stimulated with amino acids for the indicated times . Cell lysates were immunoblotted with the indicated antibodies . ( b ) As in ( a ) except that cells were incubated with brefeldin A ( BFA ) instead of ConA . ( c ) HEK293 cells transfected with untagged ZNRF2 Ser145Ala or mock ( Ctl ) and were treated as in ( a ) . ( d ) HEK293 cells transfected with untagged ZNRF2 Ser145Ala or mock ( Ctl ) and were treated as in ( b ) . ( e ) HeLa cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were subjected to lysis in hypotonic buffer with chemical crosslinker DSP as described in Materials and methods . After immunoprecipitation with IgG ( control ) or V0A2 antibody , the LAMP2 containing compartments were assessed by immunoblotting with the indicated antibodies . ( f ) HEK293 cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were incubated for overnight with dextran-Oregon Green 514 . Prior to pH measurements , the cell culture media was replaced with fresh media containing serum or amino acid-free media for 2 hr . The 490/440 fluorescence ratios were plotted as a function of pH and fitted to a Boltzmann sigmoid curve . Data are mean ± S . D . of biological triplicates and are representative of three independent experiments . ( g ) Model depicting the positive influence of ZNRF2 on V-ATPase , and hence amino acid activation of mTORC1 . Also shown is the negative feedback in which mTORC1 phosphorylates and inhibits ZNRF2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 01210 . 7554/eLife . 12278 . 013Figure 5—figure supplement 1 . Knockdown of ZNRF2 affects V-ATPase function and lysosomal pH . ( a ) HEK293 cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were harvested in 0 . 3% CHAPS lysis buffer and 1 . 2 mg of each sample was injected onto a Superose 610/300GL column and analysed by size-exclusion chromatography . Fractions of 500 µl were collected , diluted 1:1 in sample buffer , and subject to immunoblotting with the indicated antibodies . ( b ) HEK293 cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 grown under 10% FBS were immunoblotted for the indicated V-ATPase subunits and Ragulator protein levels . ( c ) A calibration curve used for Figure 5f was created by loading HEK293 cells with Dextran-Oregon Green , washing out excess dye and resuspending the cells in K+ isotonic buffers at different pH values . The 490/440 fluorescence ratios were plotted as a function of pH and fitted to a Boltzmann sigmoid curve . ( d ) HEK293 cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were transfected with FLAG only vector ( control ) or RagBGTP/RagCGDP-FLAG tagged proteins for 36 hr . The cells were starved of amino acids , and stimulated for 20 min with amino acids . The cell lysates were immunoblotted as indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 12278 . 013 Consistent with the ConA and BFA chase experiments , the knockdown of ZNRF2 resulted in decreased association of V-ATPase with lysosomes ( the LAMP2 containing compartment ) when V0A2 was immunopurified in the presence of a cross-linking agent ( Figure 5e ) . Since one of the major functions of V-ATPase is to maintain the lysosomal pH ( Nishi and Forgac , 2002 ) , we tested whether knockdown of ZNRF2 affects this function of V-ATPase . In cells depleted of ZNRF2 , we observed a marked increase in the lysosomal pH from 4 . 3 up to 5 . 5 in cells grown under standard conditions , and from 4 . 7 to 6 in amino acid-starved cells ( Figure 5f and Figure 5—figure supplement 1b ) , indicating that ZNRF2 is crucial for the V-ATPase function . To test whether ZNRF2 acts upstream of Rag-GTPases , we used a constitutively-active RagBGTP-RagCGDP complex ( Kim et al . , 2008; Sancak et al . , 2008 ) that renders mTORC1 insensitive to amino acid deprivation ( Figure 5—figure supplement 1c ) . Overexpression of these Rag mutants rescued the decrease in mTORC1 activity caused by ZNRF2 knockdown , suggesting that ZNRF2 acts upstream of the Rag GTPases . These experiments reinforce a model in which ZNRF2 contributes to mTORC1 activation by amino acids upstream of V-ATPase and Rag GTPases , via regulating V-ATPase localization and function ( Figure 5g ) . The activation of mTOR , as part of mTORC1 complex , is dependent on lysosomal localization and the vacuolar H+-ATPase ( V-ATPase ) activity; however , the precise mechanism underlying regulation of mTORC1 by the V-ATPase remains unclear . In this study , we identified a new binding partner of mTOR , ZNRF2 , which acts as a positive effector functioning upstream of V-ATPase to mediate amino acid-activation of mTORC1 . Furthermore , we show that ZNRF2 is also a new substrate of mTORC1 and PP6 that promotes negative feedback regulation of mTORC1 ( Figure 2i and Figure 5g ) . Similar to mTOR , ZNRF2 is conserved among eukaryotes ( Figure 1—figure supplement 2 ) . Putative ZNRF2 orthologues in S . cerevisiae ( Pib1p ) , D . melanogaster , C . elegans and vertebrates share most similarity in their UBZ and RING E3 ligase domains . The D . melanogaster and C . elegans orthologues of ZNRF2 also have a conserved N-myristoylation signal , which is critical for membranal localization of ZNRF2 . Notably , the S288C strain of S . cerevisiae that carries a null allele for ZNRF2 orthologue Pib1p has abnormal vacuolar morphology , increased replicative lifespan and decreased growth in minimal medium , indicative of roles analogous to the V-ATPase- and mTOR-related functions of ZNRF2 that we describe in this study ( http://www . yeastgenome . org ) . Our data suggest that ZNRF2 is targeted via N-myristoylation to Golgi and other intracellular membranes where it interacts with the V-ATPase . According to size-exclusion chromatography analysis , knockdown of ZNRF2 has no effect on V-ATPase or mTORC1 complex assembly . However , several observations indicate that ZNRF2 promotes the trafficking of newly-synthesized V-ATPase to the lysosomes . When ER to Golgi trafficking was resumed after a BFA block and also after V-ATPase was released from inhibition by ConA , we found that ZNRF2 depletion prevented the recovery of amino acid-stimulation of mTORC1 . Furthermore , ZNRF2 knockdown reduced the association of the V-ATPase V0A2 subunit with lysosomes , with a parallel increase in lysosomal pH . These data suggest that ZNRF2 is a key regulator of V-ATPase trafficking , impacting on its key function of maintaining vesicle pH . Lysosomes are acidic ( pH ~4 . 5 ) intracellular organelles and their primary function is degradation of extracellular and intracellular material ( Appelqvist et al . , 2013; Settembre et al . , 2013 ) . Concomitantly with increased lysosomal pH , ZNRF2 depletion decreased mTORC1 signalling , consistent with requirement of functional lysosomes to sustain mTORC1 activation . We also demonstrate that ZNRF2 knockdown affects the lysosomal recruitment of mTORC1 , which is mediated by the Rag GTPases , the Ragulator complex and the V-ATPase . Zoncu et al . ( 2011 ) proposed an inside-out mechanism for sensing amino acids that requires the V-ATPase , which in turns interacts with the Ragulator complex . Constitutively active Rag GTPase mutants can rescue the decrease in mTORC1 activity due to knockdown of ZNRF2 . This finding suggests that ZNRF2 regulates mTORC1 recruitment to lysosomes upstream of Rag GTPases . We therefore hypothesize that attenuation of mTORC1 recruitment to the lysosomal membranes upon ZNRF2 knockdown is due to impaired function of V-ATPase and subsequent effects on the Ragulator and Rag GTPases . As well as being a positive regulator of mTORC1 , ZNRF2 is also a new target of negative regulation by mTORC1 , indicative of a self-regulating feedback mechanism . Our findings are consistent with ZNRF2 having been identified in an unbiased quantitative phosphoproteomics screen aiming to find mTOR substrates ( Yu et al . , 2011 ) . Together with previous work ( Hoxhaj et al . , 2012 ) , this study suggests that mTORC1-mediated phosphorylation of ZNRF2 on Ser145 enhances the release of ZNRF2 from intracellular membranes into the cytosol . We therefore propose the existence of an auto-inhibitory feedback loop in which the release of ZNRF2 from membranes limits the activation of mTORC1 in response to amino acids and growth factors , perhaps preventing unrestrained cell growth . Consistent with this proposal , overexpressing wild-type ZNRF2 enhanced cellular mTORC1 activation by amino acids , and expressing a Ser145Ala mutant did so to greater extent . This finely balanced dynamic regulation of ZNRF2 is reminiscent of the negative feedbacks by mTORC1 via S6K on IRS1 ( Harrington et al . , 2004; Shah et al . , 2004 ) and Grb10 ( Hsu et al . , 2011; Yu et al . , 2011 ) . In our model , mTORC1 mediate release of ZNRF2 from membranes into the cytosol , resulting in decreased mTORC1 activity through lysosome basification via the V-ATPase pathway . Hampering activation of mTORC1 by amino acids , during strong activation of the pathway by growth factors/insulin , may ensure that hyperactivation of mTORC1 does not occur . Consistent with this proposal , ZNRF2 knockdown results in decreased mTORC1 signalling in cells lacking TSC2 , that have constitutively active mTORC1 ( data not shown ) . In other words , ZNRF2 adjusts the homeostatic set point for mTORC1 . We also show that the mTORC1-phosphorylated Ser145 on ZNRF2 is dephosphorylated by PP6 , making PP6 a positive regulator of ZNRF2 by promoting its rebinding to membranes where it can regulate the V-ATPase and mTORC1 . PhosphoSer145 of ZNRF2 should provide a useful reporter for studying how PP6 is regulated , which is critical because the relative activity of PP6 versus mTORC1 will set the level of membrane bound ZNRF2 , mTORC1 activity and cell growth . In summary , this study introduces ZNRF2 as a positive regulator of mTORC1 activation by amino acids , which functions upstream of the V-ATPase and of Rag-GTPases . ZNRF2 is also identified as a negative feedback target of mTORC1 signalling . Antibodies to mTOR , rictor , raptor , mLST8 , p18 , phosphoThr1135 rictor , phospho-Ser240/244 S6 ribosomal protein , total S6 ribosomal protein , phospho-Thr389 p70 S6 kinase 1 , phospho-Ser65 and phospho-Thr37/46 4E-BP1 , p-Ser757 ULK1 , phospho-Ser473 Akt , phospho-Thr308 Akt and GAPDH were from Cell Signaling Technology . Monoclonal anti-FLAGM2-Peroxidase ( HRP ) antibody , Flag M2 antibody , anti-FLAGM2 Affinity Gel , anti-HA and dimethyl pimelimidate were from Sigma Aldrich; PP6 subunit antibodies ( SAPS1 , SAPS2 , SAPS3 and ANKRD28 ) from Bethyl laboratories; antibodies to ATP6V1B2 , ATP6V0A2 and LAMP2 from Abcam; antibody to ATP6V1A was from GeneTex; polyethylenimine ( PEI ) from Polysciences; Protein G-Sepharose , glutathione-Sepharose and enhanced chemiluminescence Western blotting kit were from Amersham Bioscience; [32P-γ]ATP was from Perkin Elmer; protein G-Sepharose and immobilized glutathione from Pierce; Precast NuPAGE polyacrylamide Bis-Tris gels , Colloidal Coomassie , LDS sample buffer and Dextran Oregon Green 514 ( D-7176 ) were from Invitrogen; dialyzed Fetal Bovine Serum ( Cat # 26400036 ) was from ThermoFisher Scientific; sequencing-grade trypsin and CellTiter-Glo was from Promega; and microcystin-LR was purchased from Dr Linda Lawton ( Robert Gordon University , Aberdeen , UK ) . mTOR antibody was from Cell Signaling Technology ( #2983 ) and LAMP2 antibody from Abcam #ab25631 . Secondary antibodies were from Life Technologies ( Alexa Fluor 488 Donkey Anti-Rabbit IgG and Alexa Fluor 594 Donkey Anti-Mouse IgG ) . Cells were maintained in Dulbecco’s Modified Eagle’s Medium ( DMEM ) with 10% foetal bovine serum ( FBS ) . For amino acid-stimulation experiments , cells were incubated in amino acid-free EBSS for 1 . 5 hr , followed by addition of amino acids ( 50X stock of GIBCO MEM Amino Acids Solution ) to 1X final concentration for 20 min , unless otherwise noted . For IGF1 stimulation , cells were serum starved in DMEM ( 12 hr ) and stimulated with 50 ng/ml for 20 min . The following wild-type and knock-out mouse embryonic fibroblasts ( MEFs ) were obtained from other institutions: mLST8 MEFs ( Guertin et al . , 2006 ) from David Sabatini ( Whitehead Institute for Biomedical Research , USA ) , Rictor MEFs ( Shiota et al . , 2006 ) provided by Manus Magnuson ( Vanderbilt University School of Medicine , USA ) and Sin1 MEFs ( Jacinto et al . , 2006 ) from Bing Su ( Yale School of Medicine , USA ) . HEK293 and HeLa cells were purchased from ATCC and supplied by the Division of Signal Transduction Therapy ( DSTT ) , University of Dundee . Flp-In T-REx 293 stable cell lines with tetracycline-inducible wild type or mutant forms of ZNRF2 have been described previously ( Hoxhaj et al . , 2012 ) . All cells described above were regularly tested for mycoplasma contamination . After treatments , cells were rinsed with ice-cold PBS and lysed in ice-cold Triton X-100 lysis buffer . Protein concentrations were normalized prior to SDS-PAGE and immunoblotting . Lysates were pre-cleared by incubating with protein G-Sepharose beads . GFP-tagged proteins were isolated from 2 to 4 mg lysates using 15 μl of GFP-Trap-agarose , incubated at 4°C for 2 hr , and washed thrice with lysis buffer and eluted in 2x LDS sample buffer . For immunoprecipitations of endogenous mTOR , the mTOR antibody was covalently coupled to Protein G-Sepharose . mTOR kinase assays were performed as previously ( Sancak et al . , 2007 ) . Briefly , endogenous mTOR or control ( HA ) was immunoprecipitated for 2 hr at 4°C . The kinase reactions were carried at 30°C in Hepes kinase buffer ( 25 mM Hepes ( pH7 . 5 ) , 50 mM KCl ) containing 2 µg of substrate ( GST-ZNRF2 or GST-p70 S6 kinase 1 ) , 0 . 1 mM non-radioactive or [γ-32P] ATP , and 10 mM MgCl2 in a total volume of 50 µl . Cells grown on coverslips were fixed with 4% paraformaldehyde , permeabilized with 0 . 2% Triton X-100 for 5 min , blocked in 10% donkey serum in PBS for 30 min and incubated with primary antibody in 1% BSA/PBS overnight at 4°C . The coverslips were washed 3 times with 1% BSA/PBS and incubated with secondary antibodies ( 1:500 ) . After 3 washes with 1% BSA/PBS , cells were stained with DAPI and mounted using Vectashield ( Vector Laboratories , CA , USA ) . The slides were viewed under a Zeiss LSM700 microscope using an alpha Plan-Apochromat ×100 NA ( numerical aperture ) 1 . 46 objective . HEK293 Flp-In T-Rex cells ( Invitrogen ) stably expressing tetracycline inducible GFP-tagged ( N/C-terminus ) ZNRF2 or GFP were induced with tetracycline for 36 hr and lysed in Triton X-100 lysis buffer . Clarified cell lysates ( 200 mg ) from each cell line were subjected to immunoprecipitation using of 80 µl GFP-Trap beads for 2 hr at 4°C . Beads were collected by centrifugation for 5 min at 4000 rpm and washed thrice with washing buffer 1 ( 0 . 27 M sucrose , 50 mM Tris-Cl ( pH7 . 4 ) , 150 mM NaCl , 1% Triton X-100 , 0 . 1% 2-mercaptoethanol ) followed by two washes with washing buffer 2 ( 0 . 27 M sucrose , 50 mM Tris-Cl ( pH 7 . 4 ) , 0 . 1% 2-mercaptoethanol ) . Proteins were eluted with 150 µl of 2x LDS sample buffer . Proteins ( 90% of each sample ) were separated on SDS-polyacrylamide gels and stained with Coomassie brilliant blue . Mass spectrometric analyses were performed as in Hoxhaj et al . ( 2012 ) . Immunopurification of lysosomes was performed as previously ( Menon et al . , 2014 ) . HeLa cells from two 15-cm dishes ( ~90% confluent ) were washed once with cold PBS , gently scraped into 10 ml cold fractionation buffer ( 140 mM KCl , 250 mM sucrose , 2 mM EGTA , 10 mM MgCl2 , 25 mM HEPES , pH 7 . 4 , 5 mM glucose ) , pelleted by centrifugation at 400 g for 3 min at 4ºC , and resuspended in 600 μl of lysis buffer containing 140 mM KCl , 250 mM sucrose , 2 mM EGTA , 10 mM MgCl2 , 25 mM HEPES , pH 7 . 4 , 5 mM glucose , 1 mM orthovanadate , 1 μM microcystin , 2x protease inhibitors and 2 . 5 mg/ml of the cross-linking agent DSP . Cells were mechanically lysed by drawing through a 23G needle 8 times , Lysates were centrifuged at 700 g for 10 min at 4ºC , yielding a post-nuclear supernatant ( PNS ) . Normalized PNS samples of equal volume were pre-cleared for 1 hr with protein A/G-agarose beads and incubated with V0A2 antibody ( Abcam ) with rocking for 12 hr at 4ºC and then for an additional 3 hr following addition of 20 μl of a 1:1 slurry of buffer and pre-washed protein A/G-agarose beads . Bead-immunocomplexes were washed five times in fractionation buffer . siRNA ( ON-TARGETplusSMARTpool ) oligos towards PPP6C , SAPS1 , SAPS2 , SAPS3 , Raptor , TSC2 and control were from Thermo Scientific . siRNA towards p18 and ATP6V0A2 were from Sigma . Cells at 40–50% confluency were transfected using Dharmafect 1 transfection reagent following manufacturer's instructions . Briefly , for transfection of cells in a 10 cm-dish , siRNA oligos ( 100 nM final concentration ) and Dharmafect 1 ( 50 µl ) were incubated separately in 1 ml of Opti-MEM for 5 min , mixed gently and incubated at room temperature for a further 20 min . The mixture was added slowly to the cells and the culture media replaced with fresh after 12 hr of siRNA transfection . The sequences of the lentiviral shRNAs targeting ZNRF2 from Sigma were: ZNRF2#1:CCGGCCGCACATGTTTGGAGGATTTCTCGAGAAATCCTCCAAACATGTGCGGTTTTT ZNRF2#2:CCGGGCTCGGATCTACCTTCCAGTACTCGAGTACTGGAAGGTAGATCCGAGCTTTTT The MISSION pLKO . 1-puro lentivirus plasmid vector ( 7 µg ) containing the shRNA sequence was transfected together with packaging ( 7 µg ) and envelope ( 7 µg ) plasmids in HEK293T cells ( T75 flasks ) of 70% confluency ( 12 ml final ) using 60 µl of PEI ( 1 mg/ml ) . The lentiviral particles were collected 72 hr after transfection , filtered ( 0 . 45 μm pore size ) and used to infect HEK293 or HeLa cells in the presence of 10 μg/ml Polybrene . Lentivirus particles ( 5 ml/10 cm-dish ) were used to infect cells at 60–70% confluency . The cells were selected with 3 µg/ml puromycin and experiments carried out within a week of infection . Lysosomal pH was measured as in ( Zoncu et al . , 2011 ) . Briefly , 106 HEK293 cells were seeded in each well of a 6-well plate and treated after 8–12 hr with 30 μg/ml of Dextran-Oregon Green 514 ( D-7176 , Invitrogen ) for 6 to 12h . Excess dye was washed out by three rinses in PBS , and cells incubated for a further 2 hr in amino acid-free EBSS or in full medium and collected via pipetting . The media was removed by centrifugation at 1000 rpm for 1 min and cells were resuspended in 200 μl physiological buffer containing 10 mM HEPES pH7 . 4 , 2 . 5 mM KCl , 2 mM CaCl2 , 136 mM NaCl , 1 . 3 mM MgCl2 and 5 mM glucose . The fluorescence of Dextran Oregon Green was measured in black 96-multiwell plates . The samples were excited at 440 nm and 490 nm , and data collected at 530 nm . To calculate pH , a calibration curve was created by loading HEK293 cells with Dextran-Oregon Green , washing out excess dye and resuspending the cells in K+ isotonic buffers at different pH values . The calibration standards were generated using a K+ solution consisting of 145 mM KCl , 10 mM glucose , 1 mM MgCl2 , buffered with 20 mM HEPES ( for pH 7–8 ) , 20 mM of MES ( for pH 5–6 . 5 ) , or 20 mM acetate ( for pH 3 . 5–4 . 5 ) . Nigericin ( 10 μg/ml ) was also added to the calibration samples , and 490/440 ratios were fitted to a Boltzmann sigmoid using Prism and plotted as function of pH . After treatments , the medium was aspirated and cells were rinsed once with ice-cold PBS and lysed in 0 . 35 ml ( 10-cm dish ) or 0 . 5 ml ( 15-cm dish ) of ice-cold Triton X-100 lysis buffer . For mTOR kinase assays CHAPS lysis buffer was used . Lysis buffer contained: 25 mM Hepes ( pH 7 . 5 ) , 1 mM EDTA , 1 mM EGTA , 1% Triton X-100 or 0 . 3% ( w/v ) CHAPS , 50 mM NaF , 5 mM sodium pyrophosphate , 1 mM sodium orthovanadate , 1 mM benzamidine , 0 . 2 mM PMSF , 0 . 1% 2-mercaptoethanol , 1 μM microcystin-LR , 0 . 27 M sucrose and one mini Complete protease inhibitor cocktail tablet ( #1697498 , Roche ) per 10 ml of lysis buffer . Lysates were clarified , snap frozen and stored at -80°C . HEK293 cells were transfected to express wild-type and catalytically-inactive PPP6C-FLAG ( 15-cm dishes ) for 36 hr and harvested in 1% Triton X-100 lysis buffer containing 50 mM Tris-HCl , pH 7 . 4 , 150 mM NaCl , and one tablet of protease inhibitor cocktail per 50 ml . The FLAG tagged proteins were purified using FLAG beads and eluted with FLAG peptide . These eluates were incubated at 30oC with ZNRF2-GFP ( purified from cells fed with serum and used as a substrate ) in a 50 µl reaction , which contained 50 mM HEPES , 100 mM NaCl , 2 mM DTT , 0 . 01% Brij-35 pH 7 . 5 and 1 mM MnCl2 . The proteins were analysed for phosphorylation of ZNRF2 . HeLa cells expressing control shRNA ( Ctl ) or shRNAs targeting ZNRF2 were plated in 96 well plates . Cell viability was measured in medium containing 1% FBS using CellTiter-Glo every 24 hr for 3 constitutive days . All proliferation assays were done with biological triplicates seeded in technical quadruplicates for each condition . Cells were trypsinized , washed twice in ice-cold PBS , resuspended in 1 mL PBS , and fixed with 3-ml of absolute cold-ethanol for 15 hr . Cells were washed twice with cold PBS and incubated with 1 mL of Propidium iodide solution ( Sodium citrate 3 , 8 mM , Propidium iodide 40 μg/ml in PBS ) for 45 min in the dark at 4°C prior to analysis by a Flow cytometer . Amino acid sequence alignments , percent identity matrix and phylogeny tree compilations were performed using BLAST ( http://blast . ncbi . nlm . nih . gov/Blast . cgi; blastp ( protein-protein BLAST ) ) and Clustal Omega ( http://www . ebi . ac . uk/Tools/msa/clustalo/ ) . Amino acid sequences were retrieved from the NCBI HomoloGene database ( http://www . ncbi . nlm . nih . gov/homologene ) .	During digestion , proteins are broken down into their constituent parts called amino acids . Amino acids are transported in the bloodstream and are used to build up new cells and repair old ones . Optimal regulation of the cellular rates of amino acid uptake and protein synthesis is critical to the overall health of our bodies . Inside each of our cells is a molecule called mammalian target of rapamycin ( mTOR for short ) , which acts as a controller that receives information about amino acid availability . mTOR also senses how much of each amino acid the cell needs and calibrates the cell’s amino acid uptake and protein synthesis machineries accordingly . When investigating an enzyme named ZNRF2 , Hoxhaj et al . discovered that it interacts with mTOR on membranes inside cells . This raised questions about how ZNRF2 might work with mTOR to sense amino acid supplies and regulate cell growth . Hoxhaj et al . found that when cells are provided with amino acids and growth-stimulating hormones , mTOR is activated and attaches a phosphate group to ZNRF2 . This chemical modification promotes the release of ZNRF2 from membranes so that ZNRF2 separates from mTOR . In contrast , when cells are starved of amino acids , this phosphate group is removed from ZNRF2 , which then returns to the membranes . On membranes , ZNRF2 also influences the activity of a pump called V-ATPase , which controls the internal acidity of the membrane-enclosed vesicles named lysosomes that help to recycle amino acids inside cells . The action of ZNRF2 on the pump may help to prime mTOR so that it is ready to sense amino acids . These findings by Hoxhaj et al . suggest that ZNRF2 and mTOR may ‘tune’ each other , making constant to-and-fro adjustments to help ensure that levels of amino acid uptake and cell growth are set just right . However , many questions about ZNRF2 still remain to be addressed . For example , are genetic mutations in ZNRF2 involved in cancers , developmental disorders or growth syndromes ? Is ZNRF2 most important in the brain , where it is particularly abundant ? And how does ZNRF2 affect acidity within the lysosomes ?	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2016	The E3 ubiquitin ligase ZNRF2 is a substrate of mTORC1 and regulates its activation by amino acids
RAB5 is a key regulator of endosomal functions in eukaryotic cells . Plants possess two different RAB5 groups , canonical and plant-unique types , which act via unknown counteracting mechanisms . Here , we identified an effector molecule of the plant-unique RAB5 in Arabidopsis thaliana , ARA6 , which we designated PLANT-UNIQUE RAB5 EFFECTOR 2 ( PUF2 ) . Preferential colocalization with canonical RAB5 on endosomes and genetic interaction analysis indicated that PUF2 coordinates vacuolar transport with canonical RAB5 , although PUF2 was identified as an effector of ARA6 . Competitive binding of PUF2 with GTP-bound ARA6 and GDP-bound canonical RAB5 , together interacting with the shared activating factor VPS9a , showed that ARA6 negatively regulates canonical RAB5-mediated vacuolar transport by titrating PUF2 and VPS9a . These results suggest a unique and unprecedented function for a RAB effector involving the integration of two RAB groups to orchestrate endosomal trafficking in plant cells . Eukaryotic cells contain various single-membrane-bound organelles , each of which possesses distinctive constituents and functions . For an organelle to maintain a specific identity and function , the protein and lipid content must be strictly regulated; however , organelles are actively interconnected , exchanging various substances to fulfill complex and diverse cellular activities . The inter-organelle trafficking system mediated by vesicular and/or tubular trafficking intermediates plays an integral role in ensuring proper organelle function . A single round of membrane trafficking involves several sequential steps: budding of vesicles/tubules from a donor organelle , delivery of the transport vesicles , and tethering and fusion of the vesicles with the target membrane . RAB GTPase is an evolutionarily conserved key regulator of the targeting/tethering step and is also responsible for other cellular activities , such as organelle movement and inter- and intracellular signaling ( Saito and Ueda , 2009; Stenmark , 2009 ) . These functions of RAB GTPases are fulfilled through nucleotide state-dependent binding with specific sets of proteins , collectively called effector proteins . To ensure the organized and specialized functions of each RAB GTPase , interactions with effectors should be temporally , spatially , and combinatorially regulated for each trafficking event ( Grosshans et al . , 2006; Zerial and McBride , 2001 ) . The early endosome acts as a communication platform between the intracellular environment and the cell surface and/or extracellular space . RAB5 is a key regulator of a wide spectrum of early endosomal functions in animal cells , including homotypic fusion between early endosomes , endosomal motility , regulation of lipid metabolism at the endosomal membrane , and signal transduction via the early endosome ( Grosshans et al . , 2006; Miaczynska et al . , 2004; Zerial and McBride , 2001 ) . RAB5 is also conserved in a broad range of eukaryotic systems , suggesting its ancient origin in the evolution of eukaryotes ( Brighouse et al . , 2010; Dacks and Field , 2007; Elias , 2010; Pereira-Leal and Seabra , 2001 ) . Endocytic trafficking pathways demonstrate surprising diversity among lineages , representing uniquely acquired endosomal and endocytic functions in each lineage . The endocytic pathway in plant cells provides a striking example: internalized proteins are initially delivered to the trans-Golgi network ( Dettmer et al . , 2006 ) , some parts of which mature into multivesiculated late endosomes that acquire RAB5 ( Scheuring et al . , 2011 ) . Conversely , in animal cells , sequestered proteins are initially delivered to RAB5-positive early endosomes , which mature into RAB7-positive late endosomes , mediated by a RAB5 effector complex comprising SAND1/Mon1 and CCZ1 ( Kinchen and Ravichandran , 2010 ) . This complex , which is also conserved in plants , distinctly regulates endosomal transport from animal and yeast systems ( Ebine et al . , 2014; Singh et al . , 2014; Takemoto et al . , 2018 ) . Plants are equipped with multiple vacuolar trafficking pathways that involve RAB5 and RAB7 in unique ways ( Bottanelli et al . , 2012; Ebine et al . , 2014; Singh et al . , 2014 ) , further underpinning the diversification of endosomal trafficking systems between plant and animal systems . However , the mechanisms underlying diversified endosomal trafficking systems involving evolutionarily conserved RAB5 remain unresolved . The presence of the plant-unique RAB5 group , ARA6/RABF1 group , is a remarkable feature of plant endosomal trafficking . Arabidopsis thaliana has three RAB5 members: two canonical RAB5 members , ARA7 ( aka RABF2b ) and RHA1 ( aka RABF2a ) , which act in vacuolar and endocytic transport ( Dhonukshe et al . , 2006; Ebine et al . , 2011; Kotzer et al . , 2004; Sohn et al . , 2003 ) , and plant-unique ARA6 ( aka RABF1 ) ( Ueda et al . , 2001 ) . In lieu of cysteine residues that are isoprenylated at the C-terminus , which are essential for membrane binding and functions of canonical RAB GTPases ( Seabra , 1998 ) , ARA6 harbors an extra stretch in the N-terminus , where this protein is N-myristoylated and palmitoylated to target a distinct subpopulation of endosomes from canonical RAB5 with substantial overlap ( Haas et al . , 2007; Ueda et al . , 2001 ) . ARA6 promotes formation of the SNARE complex , which contains plant-specific R-SNARE VAMP727 at the plasma membrane ( Ebine et al . , 2011 ) , and an endosomal function has also been described; overexpression of the nucleotide-free mutant form of ARA6 results in impaired vacuolar trafficking ( Bolte et al . , 2004; Bottanelli et al . , 2012 ) . Despite their distinct functions , ARA6 and canonical RAB5 share the common activating factor , VPS9a ( Goh et al . , 2007 ) . Intriguingly , loss-of-function mutations of ARA6 and canonical RAB5 confer counteracting effects in a vps9a mutant , further highlighting the distinct functions of these two plant RAB5 groups ( Ebine et al . , 2011 ) , although the molecular mechanism integrating these two groups in endosomal trafficking remains unexplored . In the present study , we identified and characterized the first effector molecule of ARA6 , PLANT-UNIQUE RAB5 EFFECTOR 2 ( PUF2 ) . Based on these results , PUF2 is a key integrator of the two RAB5 groups in the unique endosomal trafficking system of plants . Consistent with the notion that endosomal trafficking pathways in animals and plants have substantially diverged , close homologs of well-characterized RAB5 effectors , such as EEA1 and Rabaptin-5 , do not exist in plants . To identify effector molecules of ARA6 , we performed yeast two-hybrid screening using a GTP-fixed mutant of ARA6 ( ARA6Q93L ) as bait . After screening 4 . 57 × 105 independent clones , we obtained a candidate clone encoding the C-terminal region of At1g24560 ( Figure 1A ) . At1g24560 encodes a 678-amino-acid protein of unknown function , without clear homologs in animals or yeasts . We designated this protein PUF2 . Bacterially expressed and purified full-length PUF2 also bound to GST-tagged ARA6Q93L but not to GST-ARA6S47N or -ARA7Q69L in vitro ( Figure 1B ) . Surprisingly , PUF2 also interacted with ARA7S24N ( Figure 1B ) . Based on a co-immunoprecipitation analysis using a lysate prepared from a transgenic plant expressing PUF2-GFP and an anti-GFP antibody , PUF2-GFP and ARA6 form a complex in planta , whereas canonical RAB5 members ARA7 and RHA1 were not co-precipitated with PUF2-GFP under these experimental conditions ( Figure 1C , left panels ) . However , when a crosslinker ( dithiobis succinimidyl propionate; DSP ) was added to the reaction , canonical RAB5 was also co-precipitated ( Figure 1C , right panels ) . Thus , PUF2 interacts with active GTP-bound ARA6 , and PUF2 also weakly and/or transiently forms a complex with GDP-bound canonical RAB5 . Four coiled-coil domains were predicted in the PUF2 protein using a simple modular architecture research tool ( SMART , http://smart . embl . de/ ) ( Letunic et al . , 2015; Schultz et al . , 1998 ) , although PUF2 contained no known functional domain . We subsequently examined whether these coiled-coil regions are responsible for the interaction with ARA6 . Truncated PUF2 containing only the fourth coiled-coil region was isolated in the yeast two-hybrid screening . Consistently , truncated PUF2 containing the other coiled-coil region did not interact with ARA6 ( Figure 1D ) . Furthermore , full-length PUF2 did not interact with ARA6 , which may indicate that these regions negatively regulate the interaction between ARA6 and PUF2 . We did not detect interactions between any PUF2 constructs and ARA7 ( Figure 1—figure supplement 1 ) . To investigate the subcellular localization of PUF2 , we observed transgenic plants expressing GFP-tagged PUF2 under the regulation of its own regulatory elements ( promoter , introns , and terminator ) , thus retaining the authentic function of PUF2 as described below . As shown in Figure 2A , PUF2 localized to punctate organelles in the cytoplasm , which dilated with the application of wortmannin ( Wm; a phosphatidylinositol-3 and -4 kinase inhibitor ) and aggregated into so-called BFA bodies with brefeldin A ( BFA; an ARF GEF inhibitor ) treatment . These drug responses were similar to those of multivesicular endosomes bearing ARA6 and/or ARA7 ( Ebine et al . , 2011; Grebe et al . , 2003; Ito et al . , 2012; Jaillais et al . , 2008 ) . The endosomal nature of the PUF2-positive compartments was also supported by their accessibility to an endocytic tracer FM4-64 ( Figure 2A , arrowheads ) . We next compared the subcellular localization of PUF2 and ARA6 or ARA7 in transgenic plants coexpressing fluorescently tagged proteins . PUF2 exhibited good colocalization with both RAB5 members ( Figure 2B ) . PUF2 also colocalized with GTP-fixed ARA6 but not with GDP-fixed ARA6 . Interestingly , we occasionally observed PUF2-positive puncta without ARA6 ( Figure 2B , arrows in upper panels ) , whereas PUF2 demonstrated almost complete colocalization with ARA7 ( Figure 2B , lower left panel ) . Quantification of the colocalization confirmed our observation; specifically , 83 . 4 ± 4 . 2% ( mean ±SD; n = 4 independent images , including 70 to 377 endosomes ) of PUF2-GFP puncta overlapped with ARA6-mRFP , whereas 96 . 7 ± 1 . 7% ( mean ±SD; n = 4 independent images , including 151 to 304 endosomes ) of PUF2-GFP puncta colocalized with mRFP-ARA7 ( Figure 2—figure supplement 1 ) . Thus , PUF2 colocalized more fully with ARA7 than with ARA6 ( p<0 . 05 , Tukey’s test ) . The colocalization ratio between ARA7 and PUF2 was comparable to that between GFP- and mRFP-tagged ARA6 , indicating nearly perfect colocalization . Next , we generated transgenic plants coexpressing PUF2-GFP , ARA6-Venus , and mRFP-ARA7 and again observed endosomes bearing only ARA6 ( Figure 2C , arrowheads ) , whereas PUF2 exhibited nearly perfect colocalization with ARA7 ( Figure 2—figure supplement 1 ) . PUF2 did not colocalize with the trans-Golgi network marker mRFP-SYP43 and the trans-Golgi cisternae marker ST-mRFP ( Boevink et al . , 1998; Uemura et al . , 2012; Uemura et al . , 2004; Wee et al . , 1998 ) ; Figure 2B , lower middle and right panels ) . To determine whether the endosomal localization of PUF2 requires RAB5 activity , we observed the localization of PUF2-GFP expressed in mutants of ARA6 and the common activator of all RAB5 members in Arabidopsis , VPS9a ( Goh et al . , 2007 ) . In ara6-1 mutant plants , PUF2-GFP localized to punctate compartments , which were swollen and aggregated after treatment with Wm and BFA , respectively , and were labeled with FM4-64 as observed in wild-type plants ( Figure 2D , arrowheads ) . PUF2-GFP also exhibited the same responses to drugs and FM4-64 accessibility in the vps9a-2 mutant ( Figure 2E ) . Thus , ARA6 and canonical RAB5 activities are not required for the recruitment of PUF2 onto endosomes . We next examined the effects of the loss of PUF2 function using a puf2 mutant , in which PUF2 was not detected at either the mRNA or protein level ( Figure 3—figure supplement 1A–C ) . The puf2 mutant exhibited no macroscopic abnormalities under normal laboratory conditions ( Figure 3—figure supplement 1D ) . Thus , we explored the potential genetic interactions between PUF2 and the two RAB5 groups . The puf2 ara6-1 double mutant was indistinguishable from wild-type , puf2 , and ara6-1 plants . By contrast , the puf2 rha1 double mutation resulted in dwarfism ( Figure 3A ) , suggesting that the functions of PUF2 and canonical RAB5 lie in the same trafficking pathway . We previously demonstrated that ara6-1 and ara7/rha1 mutations exerted opposing effects on syp22-1 , a mutation in a SNARE protein mediating membrane fusion at the vacuole , and vps9a-2; ara7 and rha1 exaggerated the deleterious phenotypes of syp22-1 and vps9a-2 , whereas ara6-1 remedied these phenotypes ( Ebine et al . , 2011 ) . If PUF2 is a downstream effector mediating ARA6 function , then loss of PUF2 function should exert effects similar to that of ara6-1 on syp22-1 and vps9a-2 mutations . However , contrary to this hypothesis , the puf2 syp22-1 double mutant exhibited more severe wavy-rosette and late-flowering phenotypes than the syp22-1 mutant ( Figure 3B ) , which were restored by introducing the genomic sequence of PUF2 ( Figure 3—figure supplement 1E and F ) . A similar synergistic genetic interaction was also observed between puf2 and vps9a-2; the hemizygous puf2 mutation aggravated the growth phenotypes of vps9a-2 , resulting in seedling lethality ( Figure 3C ) . Conversely , the puf2 mutant harboring the hemizygous vps9a-2 mutation ( puf2 vps9a-2+/− ) exhibited no discernable abnormality , enabling segregation analysis . The observation of developing seeds in seedpods of the puf2 vps9a-2+/− mutant revealed that approximately one-fourth of the seeds exhibited a brownish and shriveled appearance ( normal:abnormal = 76:25 , n = 101 seeds , p<0 . 001 , chi-square test , χ2 = 67 . 7 ) ( Figure 3D ) , indicating that the double mutation resulted in halted embryogenesis . We subsequently observed embryos collected from puf2 vps9a+/− seedpods . Although we were unable to distinguish double mutants from their normal siblings at the globular stage , double-mutant embryos exhibited severe developmental retardation at later developmental stages ( Figure 3E ) . The embryonic lethality of the puf2 vps9a-2 mutant was restored after introducing a genomic fragment containing PUF2 ( Figure 3—figure supplement 1G and H ) . Similar complementation was also observed for the genomic PUF2 fragment , in which cDNA for GFP was inserted after the start codon or in front of the stop codon , indicating the functionality of GFP-tagged PUF2 . An intimate genetic interaction between PUF2 and VPS9a was also observed in this overexpression experiment: overexpressed PUF2 partially rescued the defective growth of vps9a-2 ( Figure 3F , Figure 3—figure supplement 1I and J ) . As the deleterious phenotype of vps9a-2 reflects the defective activation of canonical RAB5s rather than ARA6 ( Goh et al . , 2007 ) , this genetic evidence indicates that PUF2 acts in the same trafficking event as canonical RAB5 , although PUF2 preferentially interacts with GTP-bound ARA6 . Canonical RAB5 acts during the transportation of soluble cargos such as 12S globulin and GFP-CT24 to protein storage vacuoles in Arabidopsis seeds , a process that also involves SYP22 ( Ebine et al . , 2011; Ebine et al . , 2014; Ebine et al . , 2008 ) . If PUF2 indeed acts in the same trafficking pathway as canonical RAB5 and SYP22 , then mutations in PUF2 would synergistically affect the transport of vacuolar cargos in syp22-1 . We verified this hypothesis by examining the accumulation of storage protein precursors in a puf2 syp22-1 double mutant . Although processing of storage proteins was not markedly affected in the puf2 mutant ( Figure 4A ) , precursor proteins with larger molecular masses were observed in the puf2 syp22-1 double mutant ( Figure 4A , arrowhead ) . Immunoblot analysis using an anti-12S globulin antibody also revealed the accumulation of 12S globulin precursors in puf2 syp22-1 and puf2+/− vps9a-2 , as observed in rha1 syp22-1 and vps9a-2 mutants ( Figure 4B and Figure 4—figure supplement 1A ) . The synergistic interaction between puf2 and syp22 or vps9a-2 mutations was also evident in the mis-secreted GFP-CT24 observed in the seeds of puf2 syp22-1 and puf2+/− vps9a-2 plants ( Figure 4C , Figure 4—figure supplement 1B ) . Synergistic enhancement of the trafficking defect in vps9a-2 by puf2 was also observed in vegetative tissue . It is possible to monitor the constitutive transport of PIN2-GFP from the plasma membrane to the vacuole by the accumulation of GFP fluorescence in the vacuole of root epidermal cells of dark-grown wild-type and puf2 mutant plants ( Kleine-Vehn et al . , 2008; Tamura et al . , 2003; Figure 4D ) . Weaker fluorescence was observed in vps9a-2 plants than in wild-type and puf2 plants , indicating partial impairment of the endocytic pathway as previously reported ( Inoue et al . , 2013 ) . This trafficking defect was substantially enhanced by hemizygous puf2 mutation; fluorescence in the vacuole was barely detected , and PIN2-GFP accumulated at punctate or ring-shaped compartments , which likely represent endosomal compartments in the puf2+/− vps9a-2 mutant ( Figure 4D , Figure 4—figure supplement 1C ) . These results confirmed that PUF2 acts in the vacuolar/endocytic trafficking pathway with canonical RAB5 and SYP22 . The observation that PUF2 acts in the same trafficking pathway as canonical RAB5 , even though this protein was isolated as an ARA6 effector , prompted us to determine whether PUF2 integrates the functions of plant-unique and canonical RAB5 groups at endosomes . Therefore , we examined the interactions between PUF2 and other RAB5-related molecules and observed that PUF2 also interacted with VPS9a in a yeast two-hybrid assay ( Figure 5A ) . Deletion analysis revealed that this interaction was mediated by the N-terminal coiled-coil region of PUF2 and was distinct from the interaction with GTP-bound ARA6 mediated by the C-terminal coiled-coil region ( Figure 1D , Figure 5A , Figure 5—figure supplement 1 ) . The in planta interaction between PUF2 and VPS9a was also confirmed by performing co-immunoprecipitation analysis ( Figure 5B ) . This interaction was direct: bacterially expressed and purified GST-tagged PUF237-127 , which contained the N-terminal coiled-coil region , pulled down purified HA-tagged VPS9a , while PUF2461-639 , which contained the C-terminal coiled-coil region , did not ( Figure 5C ) . Furthermore , VPS9a and PUF2 colocalized at a subpopulation of endosomes bearing ARA6 ( Figure 5D , arrowheads ) . To verify the significance of the interaction between PUF2 and VPS9a in vacuolar trafficking , we tested the effect of overexpression of the N-terminal coiled-coil region of PUF2 on vacuolar transport of sporamin . When expressed in Arabidopsis suspension cultured cells , sporamin tagged with Venus predominantly accumulated in vacuoles , which was inhibited by co-expression of dominant-negative ARA7 ( ARA7S24N ) ( Figure 5—figure supplement 1B and D ) . The N-terminal coiled-coil of PUF2 exerted a similar effect to ARA7S24N ( Figure 5—figure supplement 1B and D ) . This effect should be attributed to titration of VPS9a by the PUF2 N-terminus , because the deleterious effect was suppressed by co-expressing VPS9a-tagRFP , but not by tagRFP alone ( Figure 5—figure supplement 1C and D ) . Co-expression of GFP , GFP-tagged full-length PUF2 , and the C-terminal coiled-coil region of PUF2 where PUF2 interacts with ARA6 did not exhibit an inhibitory effect on vacuolar trafficking of sporamin . These results indicated functional significance of the interaction of PUF2 with VPS9a at the N-terminal region in vacuolar transport in Arabidopsis . To investigate the effects of the puf2 mutation on VPS9a localization , we expressed VPS9a-GFP in the null alleles of the vps9a mutant vps9a-1 ( Goh et al . , 2007 ) and the puf2 vps9a-1 double mutant under regulation of the VPS9a promoter , which complemented VPS9a function by comparable expression levels of VPS9a-GFP ( Figure 6—figure supplement 1A ) . In root epidermal cells of puf2 vps9a-1 , the endosomal population of VPS9a-GFP was significantly reduced compared to that in vps9a-1 cells [14 . 3 ± 1 . 9 puncta per cell slice in vps9a-1 ( mean ±SD , n = 6 independent images , each of which contained five cells ) versus 5 . 4 ± 2 . 1 puncta in puf2 vps9a-1 ( mean ±SD , n = 6 independent images , each of which contained five cells ) , p<0 . 01 , Student’s t-test] ( Figure 6A ) . The remaining VPS9a-GFP-positive dots in puf2 vps9a-1 plants were sensitive to BFA treatment; however , endosome dilation was rarely observed after Wm treatment ( Figure 6A , middle and right panels ) , although VPS9a-positive puncta retained their endosomal identity as demonstrated by ARA6 localization ( Figure 6B ) . We did not observe a noticeable difference in VPS9a-GFP fluorescence in lateral root cap cells between the vps9a-1 and puf2 vps9a-1 mutants ( Figure 6—figure supplement 1B ) , likely reflecting distinct PUF2 requirements in different tissues . According to the results described above , PUF2 promotes the recruitment of VPS9a onto the endosome . This was further confirmed using a transient expression system in protoplasts , in which fluorescently tagged VPS9a ( VPS9a-tagRFP ) failed to localize on endosomes with dispersed distribution in the cytosol ( Figure 6C , upper panels ) . When PUF2-GFP was coexpressed with VPS9a-tagRFP , VPS9a was recruited to PUF2-positive compartments ( Figure 6C , lower panels ) , whereas coexpression of another endosomal protein , GFP-VAMP727 , did not affect the cytosolic localization of VPS9a ( Figure 6—figure supplement 1C ) . Thus , PUF2 facilitates the localization of VPS9a onto endosomes . Reduced endosomal localization of VPS9a in puf2 may lead to inefficient activation of RAB5 on the endosomal membrane . This hypothesis was verified by examining Wm-induced endosomal fusion ( Wang et al . , 2009 ) in the puf2 mutant . Although the distribution of ARA6-GFP and mRFP-ARA7 was not markedly affected in DMSO-treated puf2 mutant cells ( Figure 6D , left panels ) , the diameters of the dilated endosomes induced by Wm treatment were significantly reduced by approximately 30% [1 . 98 ± 0 . 06 µm in wild-type ( n = 83 dilated endosomes ) versus 1 . 60 ± 0 . 03 µm in mutant cells ( n = 206 dilated endosomes ) , mean ±SD , p<0 . 01 , Student’s t-test] ( Figure 6D , right panels ) . A histogram of the diameters of the dilated endosomes also showed a shift in the peak to the smaller population in the puf2 mutant ( Figure 6—figure supplement 1D ) . Thus , recruitment of VPS9a to the endosomal membrane by PUF2 is required for efficient RAB5 activation , which is likely necessary for Wm-mediated endosomal fusion . According to the results described above , PUF2 integrates the functions of ARA6 and canonical RAB5 via the common activator VPS9a; therefore , we next undertook further genetic and biochemical tests . We previously reported that loss of function of ARA6 suppressed the vps9a-2 mutation , whose mechanism remains unknown ( Ebine et al . , 2011; Figure 7 ) . Here , we examined whether the suppression exerted by the ara6-1 mutation requires PUF2 . We analyzed genotypes of progenies generated by the self-pollination of ara6-1 vps9a-2 puf2+/− plants and observed that the ara6-1 puf2 vps9a-2 triple mutant was embryonically lethal ( ara6-1 vps9a-2 puf2+/+:ara6-1 vps9a-2 puf2+/−:ara6-1 vps9a-2 puf2−/− = 59:110:0 , n = 169 , p<0 . 001 , chi-square test , χ2 = 15 . 4 ) ( Figure 7A ) . Thus , PUF2 is required for the suppression of vps9a-2 by ara6-1 , and the suppression activity of ara6-1 is exerted through PUF2 . Consistently , the puf2+/− vps9a-2 growth defect was partially suppressed by the ara6-1 mutation ( Figure 7B ) . Together , these results suggest that PUF2 promotes canonical RAB5-mediated endosomal transport by assembling VPS9a and GDP-bound canonical RAB5 on the endosomal membrane , thereby facilitating the activation of canonical RAB5 on the endosome . GTP-bound ARA6 negatively regulates this process by titrating PUF2 , as shown by the effects of different concentrations of ARA6Q93L on binding between GST-ARA7S24N and PUF2 . As the amount of ARA6Q93L increased , the amount of PUF2 pulled down by GST-ARA7S24N decreased ( 57 . 5 ± 34 . 1% band intensity , n = 4 independent experiments , p<0 . 05; Figure 7C and D , Figure 7—figure supplement 1A ) . ARA6S47N and ARA7Q69L , non-interactive RAB5 partners of PUF2 , did not hamper the interaction between PUF2 and ARA7S24N ( Figure 7—figure supplement 1B ) . Thus , ARA6 in its active state interferes with the assembly of GDP-bound ARA7 , PUF2 , and likely VPS9a by competitively binding to PUF2 to diminish endosomal transport mediated by canonical RAB5 . The existence of two RAB5 groups with distinct functions , specifically the canonical and plant-unique RAB5 groups , is a unique characteristic of the plant membrane trafficking system . These RAB5 groups share the upstream regulator VPS9a , containing the VPS9 domain , which is also conserved in activating factors for RAB5 in animal systems ( Carney et al . , 2006; Goh et al . , 2007; Ishida et al . , 2016 ) . In animal cells , several VPS9 domain-containing proteins with distinct domain structures regulate Rab5 activity in different endocytic trafficking processes . In plant cells , by contrast , the single activating factor VPS9a coordinates two plant RAB5 groups during vacuolar and endosomal transport , whose molecular mechanisms remain unknown . In the present study , we identified a novel ARA6 effector , PUF2 , which acts as a keystone to integrate the functions of two plant RAB5 groups together with VPS9a on the endosomal membrane . PUF2 captures inactive canonical RAB5 and VPS9a on the endosomal membrane , leading to efficient activation of canonical RAB5 and the enhancement of vacuolar transport from endosomes to the vacuole . ARA6 , the plant-unique RAB5 in Arabidopsis , diminishes this process by competitively binding to PUF2 in its GTP-bound state against GDP-bound canonical RAB5 , thus balancing the distinct canonical RAB5- and ARA6-mediated endosomal transport pathways in plant cells ( Figure 7E ) . No striking homologs of PUF2 have been observed in sequence databases of non-plant systems , suggesting that PUF2 was uniquely acquired by the plant lineage . However , PUF2 shares several characteristics with the animal Rab5 effector protein Rabaptin-5/RABEP1 , and both proteins contain four coiled-coil regions . Furthermore , both PUF2 and Rabaptin-5/RABEP1 interact with GTP-bound Rab5 members at one of the coiled-coil regions and also interact with their activating factors at another coiled-coil region ( Horiuchi et al . , 1997 ) . Rabaptin-5 directly binds to the active form of another Rab GTPase , Rab4 , at a distinct coiled-coil region from Rab5 , suggesting coordination of the two endosomal Rab GTPase functions by this molecule . Although these data suggest functional similarities between PUF2 and Rabaptin-5 , PUF2 appears to preferentially interact with inactive canonical RAB5 , and GTP-bound active ARA6 represses the action of canonical RAB5 via competitive binding to PUF2 . Furthermore , PUF2 interacts with ARA6 and VPS9a at its C-terminal and N-terminal coiled-coil regions , respectively , whereas Rabaptin-5 interacts with Rab5 and Rabex-5 at its fourth and third coiled-coil regions , respectively ( Mattera et al . , 2006; Vitale et al . , 1998 ) . Rabaptin-5 promotes Rab5 activation by alleviating the autoinhibition of Rabex-5 ( Delprato and Lambright , 2007; Lippé et al . , 2001 ) , resulting in additional Rab5 activation to form an active RAB5 domain on the endosomal membrane . Thus , the mechanisms utilized by PUF2 and Rabaptin-5 are completely different . The downregulation of a Rab GTPase by another Rab GTPase via competitive binding with an effector and a common GEF has not previously been reported . These findings demonstrate that plants have developed specific mechanisms to regulate RAB5 GTPases by acquiring two plant-unique machinery components , ARA6 and PUF2 , to participate in the underlying mechanism of endocytic pathway regulation in plants . Although efficient recruitment of VPS9a to the endosome requires PUF2 , it is not yet clear how PUF2 is targeted to the endosomal membrane . In the case of Rabaptin-5 , Rabaptin-5 interacts with several proteins in addition to Rab5 , Rab4 , and Rabex-5 , including Rabphillin-3 ( a Rab3 effector: Ohya et al . , 1998 ) , γ-adaptin , and GGA ( Golgi-localizing , γ-adaptin ear homology domain , ARF-binding proteins ) ( Mattera et al . , 2003 ) . The similarity between PUF2 and Rabaptin-5 at the amino acid sequence level is low , and their modes of actions are divergent , but PUF2 may also have other interacting partners because proteins with multiple coiled-coil regions frequently interact with multiple proteins . In the future , it would be interesting to isolate molecules responsible for the upstream regulation of plant RAB5 groups by recruiting PUF2 onto the endosomal membrane . Further screening and characterization of other binding partners of PUF2 is needed . We recently observed an effector of canonical RAB5 in Arabidopsis , EREX , which appears to act only during embryogenesis and early developmental stages ( Sakurai et al . , 2016 ) , suggesting canonical RAB5 utilizes several distinct effector proteins at different developmental stages . ARA6 may also exert its function by interacting with multiple effector proteins , and the identification and analysis of other ARA6 effectors will allow us to understand the precise molecular mechanisms underlying ARA6 functioning . Such approaches will reveal how plants have pioneered plant-specific endosomal trafficking pathways and provide valuable information regarding the general mechanisms involved in membrane trafficking system diversification during the evolution of eukaryotic organisms . Individual interaction assays were performed using GAL4 Two-Hybrid Phagemid Vector Kits ( Agilent Technologies , Santa Clara , California , USA ) and the AH109 strain ( Takara Clontech , Kusatsu , Siga , Japan ) . The colonies were cultured in selective medium without leucine and tryptophan ( designated as ‘+H’ ) , diluted to OD600 = 0 . 5 , and spotted onto plates containing media without leucine , tryptophan , and histidine ( designated as ‘–H’ ) as well as +H plates . More than three independent colonies were tested for each interaction . The Arabidopsis puf2 ( SAIL_24_C10 ) mutant was obtained from ABRC ( Alonso et al . , 2003 ) and backcrossed more than three times with wild-type Arabidopsis ( Col-0 ) prior to use in subsequent experiments . ara6-1 , rha1 , syp22-1 , and vps9a-2 were obtained from our lab stock . A 5 . 9-kbp PUF2 genomic fragment , including 2 . 0-kbp 5’ and 0 . 9-kbp 3’ flanking sequences , was subcloned into the pHGW vector ( Karimi et al . , 2002 ) and used for the complementation assay and overexpression analysis . The translational fusions of PUF2 with GFP and ARA6 with Venus were prepared by adding fluorescence tags to the full-length proteins ( Tian et al . , 2004 ) . The cDNA for GFP was fused in front of the initiation or stop codons in the PUF2 genomic sequence as described above to produce GFP-PUF2 and PUF2-GFP , respectively . The cDNA for Venus was inserted in front of the stop codon of the 5 . 7 kb genomic fragment of ARA6 ( At3g54840 ) , which included a 2 . 7 kb promoter , exons , introns , and 3’ flanking region , to generate ARA6-Venus . The chimeric fragments were subcloned into the pGWB1 vector , a kind gift from Dr T . Nakagawa ( Shimane University , Japan ) . Transformation of Arabidopsis plants was performed by floral dipping ( Clough and Bent , 1998 ) , using the Agrobacterium tumefaciens strain GV3101::pMP90 . Transgenic lines expressing ARA6-mRFP , ARA6Q93L-mRFP , ARA6S47N-mRFP , mRFP-ARA7 , mRFP-SYP43 , ST-mRFP , VPS9a-GFP and/or VPS9a-Venus were generated as described previously ( Ebine et al . , 2011; Ebine et al . , 2008; Inada et al . , 2016; Sunada et al . , 2016; Uemura et al . , 2012 ) . Transgenic plants expressing SP-GFP-CT24 and PIN2-GFP were kind gifts from Dr S . Utsumi ( Kyoto University , Japan ) and Dr J . Friml ( IST , Austria ) , respectively , and were crossed with puf2 vps9a-2+/− and puf2 syp22-1 . The plants were grown on Murashige and Skoog medium [MS medium: 1 × MS salt ( Wako ) , 2% sucrose , 1 × Gamborg’s vitamin solution ( Sigma ) , adjusted to pH 6 . 3] at 23°C under constant light . Transgenic plants expressing GFP- , Venus- , and/or mRFP-tagged proteins were mounted in 1/2 × MS liquid medium and observed under an LSM710 , LSM780 ( Carl Zeiss , Oberkochen , Germany ) , or a microscope ( model BX51; Olympus , Shinjuku , Tokyo , Japan ) equipped with a confocal scanner unit ( model ORCA-AG; Yokogawa Electric , Musashino , Tokyo , Japan ) . At least three different seedlings from three independent transgenic lines were observed for microscopy . For drug treatments , 5 day-old seedlings were soaked in 1/2 × MS liquid medium containing 3 . 3 µM wortmannin ( Sigma-Aldrich , St . Louis , Missouri , USA ) or 50 µM brefeldin A ( Sigma-Aldrich ) for two hours or one hour , respectively . For FM4-64 labeling , 5-day-old seedlings were treated with 4 µM FM4-64 ( Thermo Fisher Scientific , Waltham , MA ) at 23°C for 30 min . The Feret’s diameters of Wm rings were measured using ImageJ software ( National Institutes of Health , Maryland , Washington , DC ) . Transient expression of GFP- and/or tagRFP-tagged proteins in Arabidopsis suspension cultured cells was performed as described previously ( Ueda et al . , 2001; Ueda et al . , 2004 ) . The Arabidopsis Col-0 suspension cultured line ( Deep ) was described in Mathur et al . ( 1998 ) . The amounts of plasmids used for transformation were as follows: 10 µg for GFP , PUF2-GFP , PUF21-211-GFP , PUF2388-672-GFP , and Sporamin-Venus subcloned in pHTS13 , 30 µg for tagRFP , VPS9a-tagRFP , and tagRFP-ARA7S24N subcloned in pHTS13 , and 1 µg for GFP-VAMP727 subcloned in pUC18 ( Ueda et al . , 2004; Uemura et al . , 2004 ) . Ten different cells that were successfully transformed were observed . Whole-mount visualization of embryos was performed as previously described ( Aida et al . , 1997 ) . To monitor the vacuolar targeting of PIN2-GFP , transgenic plants were initially grown vertically for 4 days under constant light and subsequently incubated in the dark for 48 hr . Vacuolar accumulation of PIN2-GFP or sporamin-Venus was measured using Fiji ( Schindelin et al . , 2012 ) . At least five different seedlings form each line were observed . ARA6Q93L , ARA6S47N , ARA7Q69L , ARA7S24N , PUF2 ( 37–127 aa ) , and PUF2 ( 461–639 aa ) were expressed as glutathione S-transferase ( GST ) -fusion proteins in Escherichia coli strain DH5α and purified according to the manufacturer’s instructions ( GE Healthcare , Little Chalfont , Buckinghamshire , England ) . To obtain full-length PUF2 protein , the codon usage of PUF2 cDNA was optimized to that of E . coli , and subsequently codon-optimized PUF2 was subcloned into the pGEX6P-1 vector ( GE Healthcare ) . GST-fusion protein was expressed in E . coli strain Rosetta-gami ( DE3 ) ( Merck , Darmstadt , Germany ) , and GST-PUF2 was digested using the PreScission protease ( GE Healthcare ) on resin . Eluted PUF2 was concentrated using Amicon Ultra-15 Centrifuge Filter Units ( Merck ) . GST-RAB5s ( 0 . 2 nmol ) , pre-bound to glutathione sepharose 4B resin ( GE Healthcare ) , were incubated in buffer A [20 mM Tris-HCl pH , 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , and 0 . 05% Tween-20] containing 50 mM EDTA and 100 µM GTPγS or GDP and then incubated with PUF2 ( 0 . 266 nmol ) in buffer A for 30 min at room temperature . The beads were washed three times with buffer A containing 10 µM GTPγS or GDP , and bound proteins were subjected to immunoblotting analysis . For an in vitro competition assay , the ARA6Q93L , ARA6S47N , or ARA7Q69L were mixed with PUF2 and incubated in buffer A containing 10 µM GTPγS ( for ARA6Q93L and ARA7Q69L ) or GDP ( for ARA6S47N ) prior to mixing with GDP-preincubated GST-ARA7S24N . HA-tagged VPS9a was expressed in the yeast strain YPH414 ( MATa Δpep4:TRP1 ura3 lys2 ade2 trp1 his3 leu2 ) under the control of the GAL1 promoter . Yeast cells were collapsed by vortexing with glass beads in PBS with protease inhibitor cocktail ( GE Healthcare ) . The collected lysates ( 475 µg ) were mixed with GST or GST-tagged truncate PUF2 ( 0 . 2 nmol ) , which were pre-bound to glutathione-Sepharose 4B resin ( GE Healthcare ) , and incubated for 60 min at 4°C in PBS buffer containing 0 . 05% Tween-20 and protease inhibitor cocktail ( GE Healthcare ) . The beads were washed three times with the same PBS buffer , and the bound proteins were subjected to immunoblotting analysis . At least three independent experiments were performed . T3 plants expressing PUF2-GFP were grown vertically on MS medium plates for 16 days , and 0 . 6 g of each sample was collected and ground in 1 ml of extraction buffer [50 mM HEPES-KOH , pH 7 . 5 , 0 . 4 M sucrose , 5 mM MgCl2 , protease inhibitor cocktail ( Roche , Basel , Switzerland ) ] using sea sand . The lysates were centrifuged at ×1000 g to remove debris . For analysis using a chemical cross-linker , 1 mM DSP ( Thermo Fisher Scientific ) was added to the supernatants and incubated for 30 min at 4°C , followed by quenching using 50 mM Tris-HCl , pH 7 . 5 , and 0 . 5% CHAPS was added to the lysates to solubilize the membranes . Next , 750 µl of each sample was incubated with 50 µl of anti-GFP micro beads ( Miltenyi Biotec , Bergisch Gladbach , Germany ) for 30 min at 4°C . The samples were loaded onto microcolumns attached to the magnetic field of a micro-MACS separator ( Miltenyi Biotec ) and washed four times with extraction buffer supplemented with 0 . 5% CHAPS . Immunoprecipitates were eluted according to the manufacturer’s instructions , followed by immunoblotting . At least three independent experiments were performed . MBP-tagged truncated PUF2 , which included amino acid residues 133 to 361 , was expressed in E . coli Rosetta-gami ( DE3 ) ( Merck ) , purified according to the manufacturer’s instructions ( New England Biolabs , Ipswich , Massachsetts , USA ) , and subsequently used as an antigen to generate an anti-PUF2 polyclonal antibody . The obtained anti-PUF2 antibody was purified via protein G affinity column chromatography ( GE Healthcare ) . The anti-GFP antibody was raised against GST-tagged GFP , which was expressed in E . coli Rosetta ( DE3 ) ( Merck ) according to the manufacture’s instructions ( GE Healthcare ) . The obtained anti-GFP antibody was purified with HiTrap NHS-activated HP columns ( GE Healthcare ) conjugated with the purified GFP protein . The anti-12S globulin antibody was a kind gift from Dr I . Hara-Nishimura ( Kyoto University , Japan ) . Anti-HA , anti-GST , and anti-H3B antibodies were purchased from Thermo Fisher Scientific , Santa Cruz Biotechnology ( Dallas , TX ) , and Merck , respectively . The following dilution ratios were used for each antibody in the immunoblotting experiments: anti-ARA6 ( Haas et al . , 2007 ) , 1:200; anti-RAB5 ( mixture of anti-RHA1 ( Ebine et al . , 2011 ) , 1:1000 and anti-ARA7 ( Haas et al . , 2007 ) , 1:500 ) ; anti-GFP , 1:1 , 000; anti-12S globulin ( Shimada et al . , 2003 ) , 1:10 , 000; anti-VPS9a ( Goh et al . , 2007 ) , 1:1 , 000; anti-HA , 1:500; anti-GST , 1:1 , 000; anti-H3B , 1:1 , 000; and anti-PUF2 , 1:200 . The Arabidopsis Genome Initiative locus identifiers for the genes utilized in this study are At1g24560 ( PUF2 ) , At3g54840 ( ARA6/RABF1 ) , At5g45130 ( RHA1/RABF2a ) , At4g19640 ( ARA7/RABF2b ) , At3g05710 ( SYP43 ) , At3g19770 ( VPS9a ) and At5g46860 ( VAM3/SYP22 ) .	Living cells often contain compartments that pass proteins , fats and other biological molecules to one another via a process called membrane trafficking . Endosomes are one of the key platforms of membrane trafficking . These structures accumulate molecules from the outside of the cell , sort them , and then redirect them back to the cell surface or send them to other compartments within the cell where they can be broken down . Proteins known as RAB5s regulate many of the activities of endosomes . Some are found in a wide range of organisms , including animals , fungi , and plants , and are referred to as the “canonical” RAB5 group . Another group of RAB5 proteins are unique to land plants and some green algae . The existence of two RAB5 groups ( i . e . canonical and plant-unique ) is a distinctive feature of plant cells . In 2011 , researchers showed that a plant-unique RAB5 could interfere with and counteract the activities of a canonical RAB5 . However , it remained ambiguous how these proteins could do this . To resolve this question , Ito et al . – who include several researchers from the 2011 study – set out to find proteins that interact with a plant-unique RAB5 from Arabidopsis thaliana . The experiments identified one partner of a plant-unique RAB5 , which was named PUF2 . Unexpectedly , further experiments revealed that PUF2 also regulates canonical RAB5 . PUF2 was found on the surface of the endosome together with RAB5s and a protein that activates RAB5s . Notably , PUF2 also interacted with the activating factor and the inactive form of canonical RAB5 . Based on these findings , Ito et al . propose that PUF2 acts as a landmark to bring inactive canonical RAB5 close to its activating factor , which helps to activate canonical RAB5 . They suggest that the plant-unique RAB5 also competitively binds to the landmark and blocks the canonical RAB5 . Membrane trafficking is a universal system for all living organisms , yet the system seems to be customized among different organisms . These new findings provide further evidence that land plants have evolved a unique mechanism to regulate the activities of their endosomes . The next step is to reconstruct how this system evolved and unravel its relevance to the evolution of plant-specific traits .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "cell", "biology" ]	2018	Integration of two RAB5 groups during endosomal transport in plants
To investigate the mechanisms by which β-subunits influence Nav channel function , we solved the crystal structure of the β2 extracellular domain at 1 . 35Å . We combined these data with known bacterial Nav channel structural insights and novel functional studies to determine the interactions of specific residues in β2 with Nav1 . 2 . We identified a flexible loop formed by 72Cys and 75Cys , a unique feature among the four β-subunit isoforms . Moreover , we found that 55Cys helps to determine the influence of β2 on Nav1 . 2 toxin susceptibility . Further mutagenesis combined with the use of spider toxins reveals that 55Cys forms a disulfide bond with 910Cys in the Nav1 . 2 domain II pore loop , thereby suggesting a 1:1 stoichiometry . Our results also provide clues as to which disulfide bonds are formed between adjacent Nav1 . 2 912/918Cys residues . The concepts emerging from this work will help to form a model reflecting the β-subunit location in a Nav channel complex . Voltage-gated sodium ( Nav ) channels are part of membrane-embedded signaling complexes that initiate the rising phase of action potentials , a crucial event in generating and propagating electrical signals throughout the human body ( Hille , 2001; Catterall , 2012 ) . As key components of these protein assemblies , β-subunits ( 1 ) modify Nav channel-gating properties; ( 2 ) regulate channel trafficking and localization to distinct surface compartments; and ( 3 ) influence channel oligomerization ( Calhoun and Isom , 2014; Namadurai et al . , 2015 ) . Moreover , β-subunits can alter the toxin pharmacology of Nav channels ( Gilchrist et al . , 2014 ) , a concept that has been exploited to detect their presence in heterologous expression systems or native tissues ( Wilson et al . , 2011; Zhang et al . , 2013; Wilson et al . , 2015; Gilchrist et al . , 2013 ) . Structurally , β-subunits are single-transmembrane segment glycoproteins with a short cytoplasmic C-terminal tail and a large V-type immunoglobulin ( Ig ) extracellular domain that may participate in homophilic and heterophilic interactions , cell adhesion , and cell migration ( Calhoun and Isom , 2014; Namadurai et al . , 2015; Brackenbury et al . , 2008 ) . Although all β-subunits belong to the Ig family , recent atomic resolution information for the β3 and β4 extracellular domain revealed substantial differences in their 3D structure ( Gilchrist et al . , 2013; Namadurai et al . , 2014 ) . Given their distinct features and functional roles , it has now become clear that each β-subunit structure should be obtained and assessed separately . Of the four known β-subunits and their splice variants ( β1–4; gene names Scn1b-Scn4b ) ( Isom et al . , 1992; Isom et al . , 1995a; Morgan et al . , 2000; Patino , 2011; Yu , 2003 ) , β2 and β4 form a disulfide bond with an unidentified Cys within particular Nav channel isoforms . In contrast , non-covalent interactions underlie β1 and β3 association with Nav channels as well as other members of the voltage-gated ion channel family ( Calhoun and Isom , 2014; Namadurai et al . , 2015; Marionneau , 2012; Nguyen et al . , 2012; Deschenes et al . , 2008 ) . Aberrant behavior of the ubiquitously expressed β2 and β4 subunits has been linked to disorders such as long-QT syndrome , atrial fibrillation , sudden infant death syndrome , and epilepsy , possibly through dysregulation of the Nav channel signaling complex ( Li et al . , 2013; Medeiros-Domingo et al . , 2007; Tan et al . , 2010; Baum et al . , 2014; Watanabe et al . , 2009 ) . Moreover , β2 has been implicated in neurodegenerative disorders and neuropathic pain and is therefore of potential interest for developing novel therapeutic strategies ( O'Malley et al . , 2009; Lopez-Santiago , 2006 ) . Finally , β2 is targeted by secretase enzymes , an observation that suggests a potential contribution to Alzheimer's disease ( Gersbacher et al . , 2010; Kim et al . , 2005 ) . Despite accruing evidence supporting their clinical relevance , fundamental questions on the causal relationship between β2 mutations and disorders remain unanswered . In contrast , auxiliary proteins are the topic of herculean research efforts in other fields where their role as vital contributors to cellular function or in forming drug receptor sites is well established ( Copits and Swanson , 2012; Gee et al . , 1996; Milstein and Nicoll , 2008; Dolphin , 2012 ) . To begin to appreciate β2 function and lay the foundations for constructing an interacting model with Nav channels , we first need to define the mechanism by which β2 regulates Nav channel function . In particular , knowledge of the relative orientation of both partners will help to assess whether a mutation modifies channel function or influences complex assembly . To this end , we solved the crystal structure of the extracellular β2 domain at 1 . 35Å and found a 55Cys-containing binding region that modulates Nav1 . 2 toxin susceptibility by β2 . Next , we exploited a combination of Nav channel mutagenesis , biochemistry , and β2-induced alterations in spider toxin pharmacology to uncover a disulfide bond between 55Cys and a partnering Cys located in the domain II S5-S6 pore loop of Nav1 . 2 . Remarkably , Nav1 . 5 and close relatives Nav1 . 8/Nav1 . 9 do not possess a corresponding Cys , which may explain a lack of β2 effect on Nav1 . 5 toxin pharmacology . In concert with the available structural information on bacterial Nav channels ( Payandeh et al . , 2012; 2011; Shaya et al . , 2014; Zhang et al . , 2012 ) , our results provide conceptual insights into the location of the β2-subunit in the Nav channel-signaling complex . To begin to understand the functional relationship between human ( h ) β2 and hNav channels at an atomic level , we solved the crystal structure of the extracellular hβ2 domain at a resolution of 1 . 35Å ( Figure 1a , Figure 1—source data 1 ) . The construct encompasses residues 30–153 ( Figure 2 ) , and contains a single Cys mutation ( C55A ) to facilitate crystallization [The C55A mutation is located on the protein surface and therefore unlikely to affect the overall structure ( Figure 1a , d ) . ] . The hβ2 configuration displays an Ig-like fold , consisting of eleven β-strands and three 310 helices . Other than the mutated Cys ( C55A ) , this domain contains four additional cysteines that are arranged in two bonds . The first , intra-subunit bond , is strictly conserved among all four β-subunit isoforms and is mediated by 50Cys and 127Cys buried within the core where it links two opposing faces of the protein . The second intra-subunit bond is located within a loop that spans residues 70–77 and connects strand β5 to β6 via 72Cys - 75Cys ( Figure 1a–b ) . This loop constitutes a unique feature of hβ2 since corresponding cysteines are absent in β1 , β3 , and β4 . Remarkably , this region displays a dual conformation in the crystal structure which indicates a high degree of flexibility ( Figure 1b ) . 10 . 7554/eLife . 10960 . 003Figure 1 . Crystal structure of hβ2 . ( a ) Cartoon representation of the hβ2 ( C55A ) extracellular domain crystal structure , showing β strands in gold and 310 helices in red . Cysteine side chains , as well as the 55Ala residue , are shown in stick representation . Positions of N-terminus ( N ) and C-terminus ( C ) are indicated . The loop containing the 72Cys- 75Cys disulfide bond is modeled in a dual conformation . ( b ) Detail of the dual conformation of the loop , showing all side chains in stick conformation . ( c ) Detail of the same loop in the C72A/C75A ( C55A ) mutant , shown from the same viewpoint as in panel ( b ) . ( d , e ) Comparison of the surfaces of hβ2 ( d ) and hβ4 ( e ) surrounding the reactive cysteines ( 55Cys and 58Cys , respectively ) . Side chains of hydrophobic residues are shown in green , negatively charged carboxyl groups in red , and positively charged amino and guanidinium groups in blue . The position of the cysteine ( which has been mutated to alanine to allow crystallization ) is shown in yellow . ( f ) Size exclusion chromatograms ( Preparative Superdex200 ) for hβ2 C55A and hβ2 C55/72/75A , which both elute as monomeric species . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 00310 . 7554/eLife . 10960 . 004Figure 1—source data 1 . X-ray data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 00410 . 7554/eLife . 10960 . 005Figure 2 . Structural comparison of hβ2 to hβ3 and hβ4 . ( a ) Superposition of the crystal structures of hβ2 C55A ( gold ) and hβ4 C58A ( blue ) in cartoon representation . Cysteines or their equivalent residues ( A55 and A58 , respectively ) are shown in sticks . The major differences are highlighted in the figure . ( b ) Superposition of crystal structures of hβ2 C55A ( gold ) and hβ3 ( red ) . ( c ) Root-mean-square-deviation ( RMSD ) plots showing the RMSD values per residue for hβ4 C58A ( left , blue plot ) and hβ3 ( right , red plot ) relative to hβ2 after a superposition . The residue numbers below correspond to the hβ2 numbers . Sections for which there are no corresponding residues at indicated with dotted lines . ( d , e ) Shown are sequence alignments of the extracellular domains of ( d ) hβ2 versus hβ4 and ( e ) hβ2 versus hβ3 . Conserved residues are highlighted in grey , and cysteines in yellow . Observed disulfide bonds are also labeled ( S-S ) . Secondary structure elements found in the corresponding crystal structures are also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 00510 . 7554/eLife . 10960 . 006Figure 2—figure supplement 1 . Amino acid sequence alignment of the β2 protein found in various organisms . Conserved cysteines 72Cys and 75Cys in hβ2 are indicated with a grey background . Residue numbers in hβ2 above sequences are shown as a reference . Protein accession numbers for sequences used are: H . sapiens ( NP_004579 . 1 ) , P . abelii ( XP_002822587 . 1 ) , M . davidii ( XP_006761624 . 1 ) , E . fuscus ( XP_008148335 . 1 ) , S . harrisii ( XP_003764204 . 1 ) , T . guttatus ( XP_010214560 . 1 ) , C . brachyrhynchos ( XP_008635161 . 1 ) , F . albicollis ( XP_005058672 . 1 ) , A . canadensis ( XP_011600162 . 1 ) , M . vitellinus ( XP_008927155 . 1 ) , A . carolinensis ( XP_003229016 . 2 ) , P . bivittatus ( XP_007424304 . 1 ) , X . tropicalis ( NP_001116903 . 2 ) , X . laevis ( NP_001088105 . 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 006 Although the overall hβ2 extracellular domain exhibits a similar fold compared to that of the previously reported hβ311 and hβ49 structures , considerable differences are observed . Aside from the distinct 72Cys - 75Cys disulfide bond , there are extensive variances in loop lengths and β strands , indicating that hβ2 and hβ4 have diverged substantially ( Figure 2a , c ) . Yet , at a protein sequence level , hβ2 is most closely related to hβ4 ( 26% identical; 48% conserved ) ( Figure 2d , e ) . Of note is that the N-terminal region of hβ2 is structured rather than disordered in hβ4 , adding an additional short β strand ( β1 ) which interacts with the novel β4 strand . The strand following the unique disulfide bond is shortened in hβ2 and two additional 310 helices can be seen , whereas one that is present in hβ4 , no longer exists in hβ2 . A comparison with the hβ3 structure also shows substantial divergence in loop lengths at particular locations ( Figure 2b , c ) . The hβ3 extracellular domain misses the additional N-terminal β-strand; instead , its N-terminus is anchored via a disulfide bridge between 26Cys and 48Cys , resulting in positional shifts in excess of 15Å . We previously found that 58Cys on the surface of β4 is crucial in modulating Nav1 . 2 susceptibility to toxins from spider and scorpion venom ( Gilchrist et al . , 2013 ) . The hβ2 subunit possesses a corresponding Cys at position 55 that is located in a longer loop which may result in an altered spatial position of this residue when compared to hβ4 ( Figure 2a , c , Figure 1d , e ) . In addition , the residues immediately N-terminal to 55Cys are stabilized through a β-strand interaction between strands β1 and β4 , both of which are absent in hβ4 . Taken together , these observations suggest that the environment of this key Cys may differ in both isoforms . To further examine the coupling of hβ2 to hNav1 . 2 , we next determined the functional role of 55Cys as well as that of 72Cys and 75Cys . Previously , we and others discovered that β-subunits can manipulate the pharmacological sensitivities of Nav channels ( Wilson et al . , 2011; Gilchrist et al . , 2013; Doeser et al . , 2014 ) . In particular , β4 is capable of dramatically altering animal toxin binding to rat ( r ) Nav1 . 2a ( Gilchrist et al . , 2013 ) . For example , the spider toxin ProTx-II ( Middleton et al . , 2002 ) binds to voltage-sensing domains ( VSDs ) I , II , and IV in rNav1 . 2a ( Bosmans et al . , 2008 ) , and is ∼five-fold less potent when β4 is present . [An intriguing concept is that ProTx-II may also bind directly to β4 ( Gee et al . , 1996 ) ; however , our isothermal calorimetry experiments do not support this notion ( Figure 3—figure supplement 1 ) . ] To examine whether this protective ability extends to hβ2 , we expressed hNav1 . 2 in Xenopus oocytes and measured ProTx-II susceptibility without or in the presence of the β-subunit ( Figure 3—source data 1 , Figure 3—figure supplement 2 ) . Similar to β4 , we observe that the hβ2 subunit expresses abundantly and traffics to the membrane ( Figure 3d ) where it is able to reduce the degree of hNav1 . 2 current inhibition by ProTx-II . Specifically , 100 nM ProTx-II reduces hNav1 . 2 conductance to ∼17% of peak whereas the current remaining in the presence of hβ2 is typically more than ∼64% of peak conductance , thereby demonstrating a protective effect . Other gating parameters such as conductance-voltage ( G–V ) and channel availability relationships are unaffected ( Figure 3—source data 1 ) . Next , we sought to determine if 55Cys in hβ2 is involved in reducing hNav1 . 2 susceptibility to ProTx-II by mutating this residue to an Ala . Indeed , the C55A mutant traffics to the membrane and causes ProTx-II inhibition of hNav1 . 2 to resemble that of the wild-type ( WT ) channel without hβ2 present ( Figure 3c , d , Figure 3—figure supplement 2 ) . Although Ala wields a small side chain and is often employed in mutagenesis studies , we also replaced 55Cys with a Ser , as it most closely resembles Cys in terms of size and electric properties . Similar to WT hβ2 and C55A , the C55S mutation impaired neither channel expression nor surface trafficking ( Figure 3d , Figure 3—figure supplement 2 ) . Moreover , the extent of ProTx-II-induced hNav1 . 2 inhibition in the presence of C55S is indistinguishable from that of the channel alone ( Figure 3c , Figure 3—source data 1 ) . Altogether , these results support the notion that hβ2 conveys ProTx-II protection to hNav1 . 2 via 55Cys and may relate to previous work in which the loss of the covalent link between rβ2 and hNav1 . 1 disrupts the targeting of rβ2 to nodes of Ranvier and to the axon initial segment in hippocampal neurons ( Chen et al . , 2012 ) . 10 . 7554/eLife . 10960 . 007Figure 3 . Effect of hβ2 on hNav1 . 2 toxin pharmacology . ( a ) Co-expression of hNav1 . 2 with hβ2 decreases the degree of inhibition by 100 nM ProTx-II . Left trace shows ProTx-II strongly inhibiting WT hNav1 . 2 whereas right trace displays attenuated inhibition in the presence of hβ2 . Black trace is control condition without toxin , red is in the presence of ProTx-II . Traces depict a 50 ms depolarization to -15 mV from -90 mV . Scale bar is 10 ms on horizontal axis and given nA vertically . ( b ) Normalized conductance-voltage ( G-V , filled circles ) and steady-state inactivation ( I-V , open circles ) relationships for hNav1 . 2 with and without hβ2 . Pre-toxin values are shown in black and post-toxin in red . Fit values can be found in Figure 3—source data 1 . ( c ) Dot plot comparing hβ2 mutations by ability to prevent ProTx-II inhibition of hNav1 . 2 . Black circles represent individual oocytes; vertical axis shows percent of inhibition by ProTx-II at peak conductance . Blue lines represent a 95% confidence interval . hβ2 mutations are presented underneath the horizontal axis and label the lanes below in ( d ) . Statistical significance ( p<0 . 01 ) is indicated by an asterisk . ( d ) Western blot against the C-terminal myc-tag of hβ2 . No signal is seen in the negative control but is observed for the WT hβ2 and all mutants , both in whole cell ( filled circle ) and surface ( open circle ) fractions . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 00710 . 7554/eLife . 10960 . 008Figure 3—source data 1 . Table providing values for fits of the data presented in Figure 3 and Figure 3—figure supplement 2 . G-V and SSI relationship data were fitted by a Boltzmann curve . V1/2 provides the midpoint voltage of the calculated curve ( in mV ) and Vc the unit-less slope , with standard error of the mean ( SEM ) . Right column shows peak conductance after toxin treatment as a fraction of untreated peak conductance with the upper and lower bounds of the 95% confidence interval in parentheses , reflecting the data displayed in the dot plots . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 00810 . 7554/eLife . 10960 . 009Figure 3—figure supplement 1 . ProTx-II does not bind directly to β4 . Isothermal calorimetry ( ITC ) experiments showing titration of 400 μM WT hβ4 extracellular domain into 40 μM ProTx-II ( left ) and into buffer ( right ) . No significant heat differences are detected . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 00910 . 7554/eLife . 10960 . 010Figure 3—figure supplement 2 . G-V and SSI relationships for hβ2 mutants . ( a ) Conductance-voltage ( G-V , filled circles ) and steady-state inactivation ( SSI , open circles ) relationships before 10nM ProTx-II addition are indicated in black and red in the presence of toxin . ( b ) Representative traces at -15 mV ( holding potential of -90 mV ) for the corresponding graphs seen in ( a ) . Scale bar is 10 ms along horizontal axis and displayed nA along the vertical axis . ( c ) Western blot showing that the C72A C75A hβ2 mutant is expressed in whole cell ( filled circle ) and surface ( open circle ) fractions either alone ( middle column ) or with hNav1 . 2 ( right column ) . The hβ2 subunit was detected using an antibody against a C-terminal myc-tag . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 010 In addition to 55Cys , the hβ2 crystal structure reveals a unique motif bearing a protruding disulfide-stabilized loop formed by 72Cys and 75Cys ( Figure 1a–b ) . Remarkably , this additional loop is highly conserved in almost all species that express β2 , suggesting an evolutionary conserved contribution to function ( see Figure 2 , Figure 2—figure supplement 1 ) . To determine if this loop regulates the gating or pharmacological influence of hβ2 on hNav1 . 2 , we mutated 72Cys and 75Cys to Ala ( C72A C75A ) but found that it is functionally indistinguishable from WT hβ2 ( Figure 3c , Figure 3—figure supplement 2 ) . The C72A C75A mutant localizes to the oocyte membrane surface without or with hNav1 . 2 co-expression . Moreover , typical gating parameters and hNav1 . 2 inhibition by 100 nM ProTx-II in the presence of C72A C75A is similar to that observed for the channel when co-expressed with WT hβ2 ( Figure 3—source data 1 ) . The lack of effect of the C72A C75A mutant on hNav1 . 2 function suggests that this disulfide bond is not essential for folding , and that its disruption may not significantly affect the position or environment of 55Cys . To verify this hypothesis , we produced recombinant hβ2 extracellular domain containing three Cys mutations: C55A , C72A , and C75A . Size exclusion chromatography demonstrates that the mutant produces monomeric protein , indicating that the bond between 72Cys and 75Cys is unessential for folding ( Figure 1f ) . Furthermore , we obtained a crystal structure of the triple mutant at 1 . 85Å which overlays well onto the C55A structure ( Figure 1c ) . The only significant difference is situated in the loop containing both cysteines , which now displays a single conformation . Although the spatial organization of this loop does not seem to impact the ability of hβ2 to modulate hNav1 . 2 gating or sensitivity to ProTx-II , this region may yet play a functional role in modulating other Nav channel isoforms . Although previous work has postulated the involvement of the domain II ( DII ) S5-S6 pore loop as the region responsible for forming an inter-subunit disulfide bond between particular Nav channel isoforms and β2 or β4 ( Chen et al . , 2012; Gajewiak et al . , 2014 ) , the precise residue has remained elusive . To explore the possibility of an hβ2 anchoring point in this region , we individually replaced each of the three cysteines found here ( 910Cys , 912Cys , and 918Cys ) with Ser . When expressed without or with WT hβ2 , the C910S mutant exhibits the same degree of inhibition by 100 nM ProTx-II , indicating that the protective effect of hβ2 is lost and that 910Cys is a critical residue for hβ2 binding ( Figure 4a–b , Figure 4—source data 1 and Figure 4—figure supplement 1 ) . In contrast , the C918S mutant retains protection from 100 nM ProTx-II by the β-subunit . The C912S mutant displays a split toxin-sensitive population that is consistently observed throughout multiple oocyte batches . One fraction of experiments reveals hNav1 . 2 current inhibition as though no hβ2 is present whereas another displays protection against 100 nM ProTx-II , similar to the WT channel and C918S mutant ( Figure 4a ) . To verify the presence of hβ2 , oocytes were collected after recording and checked for expression by Western blot ( see Figure 4b ) , thereby indicating that the observed loss-of-protection effects indeed relate to the hNav1 . 2 mutation . 10 . 7554/eLife . 10960 . 011Figure 4 . hβ2 forms a disulfide bond with 910Cys in hNav1 . 2 . ( a ) Dot plot showing degree of hNav1 . 2 mutant inhibition by 100 nM ProTx-II , with and without hβ2 . Black circles represent single oocytes expressing the indicated constructs and currents were measured at peak conductance . Blue bars indicate 95% confidence interval , and the red bars are 95% confidence intervals for both populations in the C912S mutant co-expressed with hβ2 . Statistical significance with p . 01 is shown by an asterisk . ( b ) Western blot against a myc-tag reveals the presence of hβ2 in whole cell fraction of oocytes co-injected with hβ2 mRNA . ( c ) Schematic depiction of the proposed disulfide arrangement in hNav1 . 2 mutants with hβ2 . hNav1 . 2 cysteines are displayed on the top arc with α ( pore-forming subunit ) , while 55Cys of hβ2 is on the bottom arc with β . Black lines represent putative disulfide bonds and dashed lines indicate alternate possibilities . Red X symbolizes a Cys to Ser mutation and an asterisk denotes mutants protected against 100 nM ProTx-II . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 01110 . 7554/eLife . 10960 . 012Figure 4—source data 1 . Table providing values for fits of the data presented in Figure 4 and Figure 4—figure supplement 1 . G-V and SSI relationship data were fitted by a Boltzmann curve . V1/2 provides the midpoint voltage of the calculated curve ( in mV ) and Vc the unit-less slope , with standard error of the mean ( SEM ) . Right column shows peak conductance after toxin treatment as a fraction of untreated peak conductance with the upper and lower bounds of the 95% confidence interval in parentheses , reflecting the data displayed in the dot plots . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 01210 . 7554/eLife . 10960 . 013Figure 4—source data 2 . Table providing values for fits of the data presented in Figure 4—figure supplement 2 . G-V and SSI relationship data were fitted by a Boltzmann curve . 1/2 provides the midpoint voltage of the calculated curve ( in mV ) and Vc the unit-less slope , with standard error of the mean ( SEM ) . Right column shows peak conductance after toxin treatment as a fraction of untreated peak conductance with the upper and lower bounds of the 95% confidence interval in parentheses , reflecting the data displayed in the dot plots . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 01310 . 7554/eLife . 10960 . 014Figure 4—figure supplement 1 . G-V and SSI relationships for hNav1 . 2 mutants . a ) G-V ( filled circles ) and SSI ( open circles ) curves are plotted for each of the hNav1 . 2 Cys to Ser mutants with and without hβ2 . Black indicates values before the addition of 100 nM ProTx-II , red shows values after toxin treatment . In the presence of hβ2 , the WT hNav1 . 2 , the C918S mutant , and the C912S C918S all show a diminished susceptibility to ProTx-II whereas the C912S mutant shows a mixed phenotype . b ) Sample traces illustrating the effect of 100 nM ProTx-II on hNav1 . 2 with and without hβ2 after depolarization to -15 mV from -90 mV . Degree of inhibition reflects the G-V and SSI data shown in a ) . Horizontal arm of scale bar is 10 ms and the vertical arm is in nA . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 01410 . 7554/eLife . 10960 . 015Figure 4—figure supplement 2 . Activity of the μO§-conotoxin GVIIJ is affected by hβ2 . ( a ) Representative traces of 5 μM GVIIJ affecting hNav1 . 2-mediated currents with and without hβ2 in Xenopus oocytes . Control traces are shown in black , red after treatment with 5 μM GVIIJ , and green after a 5-min washout with recording solution . Horizontal bar is 10 ms and vertical bar measures current in nA . The data illustrate that 5 M GVIIJ can partially block hNav1 . 2 only when hβ2 is not present . ( b ) G-V ( filled circles ) and SSI ( open circles ) relationships detailing the effect of 5 M GVIIJ on hNav1 . 2 with and without hβ2 across a wide voltage range . Color coding is the same as in ( a ) . ( c ) Dot plot showing the effect of GVIIJ on hNav1 . 2 Cys to Ser mutants . Red circles indicate fraction of peak control conductance after addition of 5 μM GVIIJ while green circles represent the same after a 5-min washout . Vertical bars show 95% confidence intervals for color-matched circles . The data clearly illustrate a reduced sensitivity to GVIIJ in the presence of hβ2; however , the results are variable as discussed in the text and led us to pursue ensuing experiments with ProTx-II . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 015 Altogether , these results hint towards a possibility of mutation-induced shifts in intra-subunit disulfide bond formations , a plausible scenario given the close vicinity of 910Cys , 912Cys , and 918Cys in hNav1 . 2 . To evaluate this hypothesis , we constructed a set of double Cys to Ser mutations ( C910S C912S , C910S C918S , and C912S C918S ) and examined the effects on ProTx-II susceptibility without and in the presence of hβ2 ( Figure 4a–b , Figure 4—source data 1 and Figure 4—figure supplement 1 ) . According to previous work with rNav1 . 2a , mutating 912Cys or 918Cys results in a bridge between 910Cys and the remaining intact cysteine ( Zhang et al . , 2015 ) . These experiments were carried out without a β2 subunit and with GVIIJ ( Gajewiak et al . , 2014 ) , a unique μO§-conotoxin that directly competes for binding to 910Cys; however , the rest of its binding site remains unexplored . In our hands , high concentrations of GVIIJ are needed to achieve an effect , and efficacy towards hNav1 . 2 is augmented in case of the C910S mutation ( Figure 4—source data 2 , Figure 4—figure supplement 2 ) , two complicating factors that made us revert to ProTx-II which is more potent and has clearly delineated binding sites within Nav1 . 2 . Our experiments on these double mutants show a much clearer picture wherein hβ2 only protects against ProTx-II when 910Cys is intact: the C912S C918S mutant still has lowered toxin susceptibility whereas C910S C912S and C910S C918S are completely inhibited by 100 nM ProTx-II , suggesting that these mutants no longer bind hβ2 ( Figure 4a–b , Figure 4—figure supplement 1 ) . Altogether , these results indicate that 910Cys is the disulfide bond partner of 55Cys in hβ2 , while 912Cys and 918Cys could form an intra-subunit bridge . At first sight , the C912S data which show the split population appear in conflict with this interpretation . However , it is conceivable that losing the 912Cys- 918Cys bond results in the formation of a non-native bond between 910Cys and 918Cys ( Figure 4c ) . Indeed , when 918Cys is mutated in addition to 912Cys , 918Cys again allows toxin protection by hβ2 similar to WT , suggesting it has become available again . The data also indicate that a non-native disulfide bond between 910Cys and 912Cys may not occur in the C918S mutant , likely because such a Cys-Val-Cys disulfide bond could be too constrained . Importantly , hβ2 cannot protect the C910S C918S or the C910S C912S mutants from 100 nM ProTx-II , indicating that neither 912Cys nor 918Cys can compensate for the loss of 910Cys ( Figure 4c ) . 918 To biochemically verify that Nav1 . 2 and β2 are covalently bound , we expressed the closely related and well-expressing rat variants ( ∼99% sequence identity ) in oocytes and immunoprecipitated the rNav1 . 2a/rβ2 complex . It is worth noting that WT myc-tagged rβ2 can form higher-order oligomers under non-reducing conditions , which may complicate the interpretation of immunoblots ( Figure 5—figure supplement 1 ) . However , these oligomers disappear upon removal of 55Cys , further highlighting the reactivity of this residue . Notwithstanding this phenomenon , probing crude cell lysate uncovers a clear co-migration of myc-tagged rβ2 with rNav1 . 2a under non-reducing conditions , since co-expression yields a distinct myc-stained band above the highest rβ2 oligomer ( Figure 5a ) . Substitution of either 55Cys in rβ2 or 910Cys in rNav1 . 2a with Ser results in the loss of this covalent complex . Furthermore , the presence of DTT completely abolishes the rβ2-rNav1 . 2a complex as well as the rβ2 oligomers ( Figure 5b ) . Thus , these results confirm our toxin experiments and suggest the formation of a disulfide bond between 910Cys in the DII S5-S6 linker of Nav1 . 2 and 55Cys in β2 . In addition to investigating the interaction of rNav1 . 2a and rβ2 in crude lysate from injected oocytes , we also pulled down rβ2 with an antibody directed against the C-terminal myc-tag and treated with DTT before loading onto a tris-acetate gel ( Figure 5c ) . In this experiment , we observe that WT rNav1 . 2a is present only when co-expressed with WT rβ2 . Moreover , both 910Cys within the channel and 55Cys in rβ2 are required for co-immunoprecipitation , thereby pointing to their role in forming a disulfide bond between both subunits ( Figure 3 , Figure 4 ) . 10 . 7554/eLife . 10960 . 016Figure 5 . Biochemical verification of the β2 and β4 disulfide bond with 910Cys in Nav1 . 2 . Top rows indicate particular Nav channel and β-subunit constructs with which oocytes were injected . WT proteins are indicated as such whereas mutants are noted with a residue number; a superscript letter symbolizes in which partner of the complex pair the mutation is found . rβ2 was loaded in each case except in the two rightmost lanes , where rβ4 was used . Labels on left column indicate which antibody was used to immunoblot the associated slice . ( a ) Western blot run under non-reducing conditions reveals a fraction of rβ2 migrating with WT rNav1 . 2a but not after selective cysteine substitution . Crude lysate from injected oocytes was run on a protein gel and probed for both the channel and the myc-tag of the β-subunit . The top PanNav slice and bottom myc slice show the expression of the respective proteins in lysates from oocytes injected with the indicated constructs . Even though equal quantities of protein was loaded in each lane ( 10 μg ) , the C910S channel mutant expresses less bountifully than the WT channel and as a result shows a weaker signal . The WT β-subunit can form redox-sensitive multimers that migrate at an apparent mass similar to that of the Nav channel . Mutation of 55Cys prevents multimer formation and disappearance of the high molecular weight band . Open arrows identify bands representing Nav channel-bound β-subunit and aid in distinguishing them from the multimeric β-subunit . Substituting 910Cys within rNav1 . 2a with Ser causes a loss of the channel-bound β-subunit band . rβ4 also binds to WT rNav1 . 2a , as evidenced by a second band , and no longer interacts with the C910S mutant . ( b ) Addition of DTT prevents both binding of rβ2 to the channel and formation of β-subunit multimers . The absence of myc signal at the same apparent weight as rNav1 . 2a indicates that binding to the channel is sensitive to reduction . In all cases , α-tubulin was used as a loading control . ( c ) WT rNav1 . 2a co-immunoprecipitates with rβ2 and rβ4 . The β-subunit was pulled down with an antibody directed against a C-terminal myc-tag and treated with DTT before loading onto the gel . The channel is present only when the WT channel is co-expressed with the WT β-subunit . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 01610 . 7554/eLife . 10960 . 017Figure 5—figure supplement 1 . Biochemical assessment of rβ2 oligomer formation . Figure represents extended data found in Figure 5 . WT myc-tagged rβ2 can form higher-order oligomers under non-reducing conditions ( indicated with an open arrow in the middle lane ) . These oligomers disappear upon removal of 55Cys ( right lane ) , further highlighting the reactivity of this residue . Left lane represents lysate from an uninjected oocyte . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 017 Next , we were curious to learn whether 910Cys also constitutes an inter-subunit anchoring point for 58Cys in the β4-subunit . To examine this notion , we measured ProTx-II susceptibility of rNav1 . 2a and its C910S mutant expressed in oocytes without or with the rβ4-subunit ( Figure 6a , b , Figure 6—source data 1 ) . Analogous to hβ2 , we observe rβ4 protein production in oocytes ( Figure 6c ) where it is able to influence the degree of rNav1 . 2a current inhibition by ProTx-II . In particular , 100 nM ProTx-II reduces rNav1 . 2a conductance to ∼22% of peak whereas the current remaining in the presence of rβ4 is ∼55% of peak conductance . Other gating parameters such as G–V and channel availability relationships are unaffected . In case of the C910S mutant , the presence of rβ4 no longer decreases ProTx-II efficacy ( from ∼22% current inhibition to ∼17% , respectively , Figure 6—source data 1 ) , thus illustrating the likely role of rNav1 . 2a 910Cys in forming a disulfide bond with rβ4 58Cys . In concert , ( co- ) immunoprecipitation experiments with rNav1 . 2a and rβ4 expressed in oocytes indicate that both partners are covalently bound and that mutating 910Cys in the channel indeed disrupts the disulfide bond ( Figure 5 ) . 10 . 7554/eLife . 10960 . 018Figure 6 . Mutation of 910Cys in rNav1 . 2a disrupts rβ4 influence on ProTx-II effect . ( a ) Replacement of 910Cys with Ser in rNav1 . 2a impairs the ability of rβ4 to protect against 100 nM ProTx-II inhibition in a manner similar to that of hβ2 . Co-expression of the rNav1 . 2a isoform with rβ4 minimizes ProTx-II inhibition compared to the channel expressed alone . Graphs compare G-V and SSI relationships , in filled and open circles , respectively . Black color is used for values before the addition of 100 nM ProTx-II and red after . ( b ) Dot plot comparing the degree of inhibition by 100 nM ProTx-II as a fraction of the pre-toxin peak current . rβ4 confers a high level of a protection against ProTx-II inhibition which is absent in the C910S mutant . ( c ) Western blot directed against the C-terminal myc-tag of rβ4 demonstrating its presence in the oocytes measured . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 01810 . 7554/eLife . 10960 . 019Figure 6—source data 1 . Table providing values for fits of the data presented in Figure 6 . G-V and SSI relationship data were fitted by a Boltzmann curve . V1/2 provides the midpoint voltage of the calculated curve ( in mV ) and Vc the unit-less slope , with standard error of the mean ( SEM ) . Right column shows peak conductance after toxin treatment as a fraction of untreated peak conductance with the upper and lower bounds of the 95% confidence interval in parentheses , reflecting the data displayed in the dot plots . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 019 Aligning the primary sequences of all tetrodotoxin ( TTX ) -sensitive hNavchannel isoforms ( Nav1 . 1–1 . 7 ) reveals that the cardiac hNav1 . 5 channel lacks the cysteine triad while the flanking sequences are highly conserved ( Figure 7a ) . Subsequently , it seems unlikely that hNav1 . 5 forms a covalent bond with hβ2 , unless it occurs via a distinct site . However , when applying 100 nM ProTx-II to hNav1 . 5 without and with the β-subunit , we observe no effects on gating or protection against the toxin , suggesting that hβ2 may not modulate this channel isoform ( Figure 7b–e , Figure 7—source data 1 ) . It is worth mentioning that hNav1 . 8 and hNav1 . 9 , two TTX-resistant isoforms that are evolutionary related to hNav1 . 5 , also lack the cysteine triad and may therefore not interact with hβ2 via a disulfide bond . 10 . 7554/eLife . 10960 . 020Figure 7 . ProTx-II inhibits hNav1 . 5 in the presence of hβ2 . ( a ) Sequence alignment comparing TTX-sensitive hNav channels in the beginning of the S5-S6 ( SS1 ) loop . Green bar indicates the C-terminal portion of the DII transmembrane segment 5 ( TM5 ) , and gray background highlights the conserved cysteine triad . The hNav1 . 2 amino acid sequence is shown in red and hNav1 . 5 in blue . Residue number of the N-terminal Val for each channel is superscripted . ( b ) Dot plot showing that hNav1 . 5 is not protected against inhibition by 100 nM ProTx-II upon co-expressing hβ2 . Black circles are individual oocytes of which sodium currents were measured at peak conductance and blue bars show 95% confidence interval . ( c ) Western blot probing for the C-terminal myc-tag of hβ2 reveals its presence in whole cell oocyte fractions . hNav1 . 5 sees no change in 100 nM ProTx-II effect in the presence of hβ2 . ( d , e ) G-V ( filled circles ) and SSI ( open circles ) relationships for hNav1 . 5 with and without hβ2 . Black is used for values before ProTx-II addition and red after toxin application . Also shown are representative traces illustrating the lack of effect of hβ2 on hNav1 . 5 susceptibility to 100 ProTx-II . Horizontal bar indicates 10 ms; vertical bar represents current in nA , with the provided magnitude . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 02010 . 7554/eLife . 10960 . 021Figure 7—source data 1 . Table providing values for fits of the data presented in Figure 7 . G-V and SSI relationship data were fitted by a Boltzmann curve . V1/2 provides the midpoint voltage of the calculated curve ( in mV ) and Vc the unit-less slope , with standard error of the mean ( SEM ) . Right column shows peak conductance after toxin treatment as a fraction of untreated peak conductance with the upper and lower bounds of the 95% confidence interval in parentheses , reflecting the data displayed in the dot plots . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 021 Having identified 910Cys within the DII S5-S6 loop of hNav1 . 2 and 55Cys within hβ2 as inter-subunit disulfide bond partners , we sought to find evidence for a potential locus of the β-subunit in relation to the channel . Although hβ2 seems to be anchored to the DII pore forming region , ProTx-II was previously shown to target VSDs I , II , and IV ( Bosmans et al . , 2008 ) , suggesting that the protective effect of hβ2 is through occlusion of the ProTx-II binding site in one or more of these regions . In order to determine which one , we assembled a set of animal toxins that , together , interact with VSDI , II , and IV in Nav1 . 2 ( Bosmans et al . , 2008 ) , and tested whether or not their function is influenced by hβ2 ( Figure 8a ) . A previous study in which rNav1 . 2a S3b-S4 paddle loops from each domain ( I-IV ) were transplanted into a homotetrameric Kv channel to identify the VSDs with which toxins interact ( Bosmans et al . , 2008 ) , found that the tarantula toxin PaurTx3 and scorpion toxin AahII ( Martin et al . , 1987 ) exclusively target VSDII and VSDIV , respectively . In addition , the tarantula toxins ProTx-I and ProTx-II both interact with the voltage sensor in DII and DIV , whereas ProTx-II also binds to DI with high affinity . Here , we tested these four toxins on hNav1 . 2 without and with hβ2 to determine if the presence of the subunit impacts toxin function . Aside from ProTx-II , we observe no significant difference in PaurTx3 , AahII , or ProTx-I effect which suggests that hβ2 does not impede binding of these toxins to VSDII and VSDIV ( Figure 8b–c , Figure 8—source data 1 and Figure 8—figure supplement 1 ) . Therefore , we speculate that hβ2 primarily influences VSDI and as such , is located near this region ( Figure 8d ) . Since μO§-conotoxin GVIIJ function is also influenced by hβ2 ( Gajewiak et al . , 2014 ) , it will be interesting to investigate a possible VSDI binding site for this toxin . While our data are limited by the absence of VSDI- and VSDIII-specific toxins and do not exclude the possibility of non-overlapping binding sites for a particular toxin and hβ2 on the same VSD , or of direct competition for binding to the DII S5-S6 loop , they may yet prove valuable as insights into hβ2 function accrue . 10 . 7554/eLife . 10960 . 022Figure 8 . hβ2 influences hNav1 . 2 VSDI toxin pharmacology . ( a ) Cartoon illustrating the binding site of AaHII , PaurTx3 , ProTx-I , and ProTx-II within the VSDs of Nav1 . 2 . Binding to a particular VSD is indicated by coloration . ( b ) Dot plot showing percent of change in peak conductance for hNav1 . 2 after treatment by each toxin without or in the presence of hβ2 . Blue bars represent 95% confidence interval and statistical significance ( p<0 . 01 ) is noted by an asterisk . c ) Western blot probing for the C-terminal myc-tag of hβ2 , showing its presence in each experimental condition except the negative control . ( d ) Shown is a Nav channel illustration consisting of the four VSDs I-IV ( S1-S4 ) and the corresponding pore-forming regions ( S5-S6 ) in grey ( DI , III , and IV ) or orange ( DII ) in a clockwise orientation which fits with the data reported here . Such an orientation places a S1-S4 voltage sensor of one domain adjacent to the S5-S6 region of the next . A/B ( shades of blue ) depicts a putative location of hβ2 in the complex where the subunit can interact with the S5-S6 pore region of DII as well as occlude the ProTx-II binding site on VSDI . In contrast , hβ1 ( green ) may position itself between VSDIV and VSDIII where it can interact with the S5-S6 region of DI and influence VSDIV movement to alter channel fast inactivation . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 02210 . 7554/eLife . 10960 . 023Figure 8—source data 1 . Table providing values for fits of the data presented in Figure 8 and Figure 8—figure supplement 1 . G-V and SSI relationship data were fitted by a Boltzmann curve . V1/2 provides the midpoint voltage of the calculated curve ( in mV ) and Vc the unit-less slope , with standard error of the mean ( SEM ) . Right column shows peak conductance after toxin treatment as a fraction of untreated peak conductance with the upper and lower bounds of the 95% confidence interval in parentheses , reflecting the data displayed in the dot plots . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 02310 . 7554/eLife . 10960 . 024Figure 8—figure supplement 1 . Activity of spider and scorpion toxins on hNav1 . 2 . G-V and SSI relationships for hNav1 . 2 , with or without hβ2 , treated with toxins ( 100 nM ) with known binding sites . ( a ) G-V and SSI curves are provided for each of the four toxins tested . Black indicates control values and red illustrates toxin effect . Filled circles are used for G-V curves and open circles for SSI . ( b ) Representative traces after depolarization to -15 mV from -90 mV for each toxin tested on hNav1 . 2 with black indicating pre-toxin currents and red showing currents after toxin application . Horizontal scale bar indicates 10 ms whereas vertical bar shows current magnitude in nA , with scale given by the attached number . DOI: http://dx . doi . org/10 . 7554/eLife . 10960 . 024 The goal of this study was to explore the interaction of β2 with Nav1 . 2 and identify anchoring residues in both partners which will help orient functional motifs within their extracellular domains . Although we obtained the first crystal structure of the hβ4 extracellular region ( Gilchrist et al . , 2013 ) , the second reported structure of the corresponding hβ3 domain ( Namadurai et al . , 2014 ) highlights the unique character of each β-subunit isoform . Since the ability to compare the distinct organizational features of β-subunits will provide valuable information as to their difference in action , we now obtained detailed structural information for the extracellular domain of hβ2 at a resolution of 1 . 35Å ( Figure 1a–b , Figure 1—source data 1 ) . We identified a flexible loop formed by 72Cys and 75Cys that is a unique feature among β-subunits but with a function that has yet to be elucidated ( Figure 1c , Figure 2 , and Figure 3 ) . Moreover , hβ2 contains a Cys at position 55 that , when mutated , disrupts the influence of hβ2 on hNav1 . 2 toxin pharmacology ( Figure 3b–d ) . Next , we combined mutagenesis and biochemical studies with spider and scorpion toxins that target specific VSDs within Nav1 . 2 ( Gilchrist et al . , 2014; Bosmans et al . , 2008 ) to probe the interaction with hβ2 . As a result , we found that 55Cys forms a distinct disulfide bond with 910Cys located in the domain II S5-S6 loop of hNav1 . 2 , thereby revealing a 1:1 stoichiometry ( Figure 4a–b , Figure 4—source data 1 and Figure 4—figure supplement 1 ) . We also exploited this toxin-reporter approach to investigate the possibility of intra-subunit disulfide bond formations between 910Cys , 912Cys , and 918Cys in hNav1 . 2 , three reactive residues that are in close vicinity to each other . The outcome of these experiments indicates that 912Cys and 918Cys form an intra-subunit bridge in WT hNav1 . 2 ( Figure 4 ) . When mutating 918Cys , hβ2 still interacts with 910Cys whereas substituting 912Cys suggests the possibility for bond formation between 910Cys and 918Cys ( Figure 4c ) . Interestingly , hNav1 . 5 lacks these three cysteines ( Figure 7a ) and as a result , ProTx-II is equipotent without or in the presence of hβ2 ( Figure 7b–e , and Figure 7—source data 1 ) This outcome provides evidence that this particular β-subunit may not modulate hNav1 . 5 in heterologous systems or that binding occurs through a different mechanism in native tissues . In concert , immunocytochemical studies in the heart and electrophysiological measurements in mammalian cell lines or oocytes disagree on whether hβ2 can modulate hNav1 . 5 function ( Johnson and Bennett , 2006; Zimmer and Benndorf , 2002 , 2007; Dhar Malhotra et al . , 2001; Maier et al . , 2004 ) . As opposed to native cardiomyocytes , hNav1 . 5 may not associate with hβ2 upon heterologous expression , a hypothesis that is supported by co-localization experiments in HEK293 cells where hNav1 . 5 and hβ2 were mainly found in the endoplasmatic reticulum or plasma membrane , respectively ( Zimmer et al . , 2002 ) . So far , none of the hβ2 mutations implicated in disorders have been found close to 55Cys , as they are located either in the signal peptide or near the transmembrane region ( Li et al . , 2013; Medeiros-Domingo et al . , 2007; Tan et al . , 2010; Baum et al . , 2014; Watanabe et al . , 2009 ) . In contrast , amino acid substitutions in the DII S5-S6 pore loop of particular Nav channel variants are linked to diseases . For example , more than 30 mutations have been identified within this region in hNav1 . 1 , several of which relate to Dravet syndrome ( Claes et al . , 2009 ) . This includes C927F ( corresponding to 918Cys in hNav1 . 2 ) , and other variants which introduce an additional Cys that may interfere with local disulfide bond pattern formation . Therefore , it is conceivable that DII S5-S6 linker mutations may affect channel interactions with hβ2 or hβ4 . Collectively , our results uncover the disulfide link between hβ2 and hNav1 . 2 which opens up the possibility to assign a potential orientation of this subunit in relation to the channel and provide an experimental basis for future docking efforts ( Figure 8d ) . hβ2 may position itself in the gaps between VSDs where it can interact with a voltage sensor as well as anchor to a pore-forming region . One important observation from bacterial Nav channel crystal structures ( Payandeh et al . , 2012; Payandeh et al . , 2011; Shaya et al . , 2014; Zhang et al . , 2012 ) is the domain-swapped architecture of the channel in which the S1-S4 VSD within one subunit is located adjacent to the S5-S6 segments of the next subunit . Such a structural clockwise arrangement has also been consistently observed in prokaryotic and mammalian Kv channels ( Long et al . , 2007; Jiang et al . , 2003 ) . Since hNav channels may be organized in a similar fashion , the determinants of hβ2-subunit sensitivity can be located in multiple Nav channel domains . For example , we found that hβ2 binds to 910Cys in the S5-S6 loop of DII and influences ProTx-II interaction with VSDI . Although it is challenging to pinpoint its precise location , our data suggest that hβ2 is positioned in the cleft between VSDIV and VSDI or VSDI and VSDII where it is presented with an opportunity to interact with both determinants to influence ProTx-II action ( Figure 8d ) . In contrast , the consistent picture emerging from the literature is that Nav channel fast inactivation and voltage-dependence of activation changes substantially when co-expressed with the β1-subunit in heterologous systems ( Chen and Cannon , 1995; McCormick et al . , 1998; Qu et al . , 1999; Isom et al . , 1995b ) . Given the pivotal role of VSDIV in channel fast inactivation ( Bosmans et al . , 2008; Capes et al . , 2013; Chanda and Bezanilla , 2002; Horn et al . , 2000; Sheets et al . , 1999 ) , β1 may be positioned close to VSDIV where it can also interact with the extracellular S5-S6 pore-loop within DI ( Figure 8d ) , a presumed interacting region ( Qu et al . , 1999; Makita et al . , 1996 ) . Altogether , this model suggests that particular Nav channel isoforms can interact simultaneously with a β1- and β2-subunit ( Calhoun and Isom , 2014 ) . Future experiments will further investigate the possible locations of β3 and β4 in this model as well as allow the incorporation of mutational effects which is essential to uncover the contribution of β-subunit mutations to human disorders or contribute to small molecule screening efforts geared towards disrupting or facilitating subunit interactions . Human β2 ( residues 30–153 ) was cloned into a modified pET28 vector ( pET28HMT ) ( Van Petegem et al . , 2004 ) . Mutations ( C55A and C55/72/75A ) were introduced using the Quikchange kit from Agilent Technologies ( USA ) according to the manufacturer’s instructions . Proteins were expressed at 18°C in E . coli Rosetta ( DE3 ) pLacI strains ( Novagen , USA ) , induced at an OD600 of ∼0 . 6 with 0 . 4 mM IPTG , and grown overnight prior to harvesting . Cells were lysed via sonication in buffer A ( 250 mM KCl and 10 mM HEPES at pH 7 . 4 ) , supplemented with 25 µg/ml DNaseI and 25 µg/ml lysozyme . After centrifugation , the supernatant was applied to a PorosMC column ( Tosoh Biosep , USA ) , washed with buffer A plus 10 mM imidazole , and eluted with buffer B ( 250 mM KCl plus 500 mM imidazole pH 7 . 4 ) . The protein was dialyzed overnight against buffer A and cleaved simultaneously with recombinant TEV protease . Next , the samples were run on another PorosMC column in buffer A , and the flowthrough was collected and dialyzed against buffer C ( 10 mM KCl plus 10 HEPES at pH 7 . 4 ) , applied to a HiloadQ column ( GE Healthcare , USA ) , and eluted with a gradient from 0% to 30% buffer D ( 1 M KCl plus 10 mM HEPES at pH 7 . 4 ) . Finally , the samples were run on a Superdex200 ( GE Healthcare , USA ) gel filtration column in buffer A . The protein samples were exchanged to 50 mM KCl plus 10 mM HEPES ( pH 7 . 4 ) , concentrated to ∼5 mg/ml using Amicon concentrators ( 3K MWCO; Millipore USA ) , and stored at −80°C . Crystals were grown using the hanging-drop method at 4°C . Both C55A and C55/72/75A were crystallized in 0 . 1 M Tris ( pH 8 ) , 15–20% ( w/v ) PEG 6000 . Crystals were ﬂash-frozen after transfer to the same solution supplemented with 30% glycerol . The data used to solve the final structures were collected at the Canadian Light Source ( Saskatoon ) beamline 08ID-1 and datasets were processed using XDS ( Kabsch , 2010 ) and HKL3000 ( Minor et al . , 2006 ) . A search model was created by using only β-strands from PDB 4 MZ2 , and with all side chains truncated to alanine . Molecular replacement was performed using Phaser ( McCoy et al . , 2007 ) , yielding poor initial phases which were improved via autobuilding in ARP/wARP ( Langer et al . , 2008 ) . The model was completed by successive rounds of manual model building in COOT ( Emsley et al . , 2010 ) and refinement using Phenix ( Adams et al . , 2010 ) . A simulated annealing composite omit map was calculated with CNS ( Brünger et al . , 1998 ) to verify the absence of residual model bias . 85Met was found to be in a disallowed region of the Ramachandran plot in both structures . All structure ﬁgures were prepared using PYMOL ( DeLano Scientiﬁc , San Carlos , USA ) . Coordinates and structure factors have been deposited in the Protein Data Bank under accession codes 5FEB ( C55A ) and 5FDY ( C55/72/75A ) . ProTx-I and ProTx-II were acquired from Peptides International ( USA ) , PaurTx3 from Alomone Labs ( Israel ) . AaHII from Androctonus australis hector venom was purified as described ( Martin et al . , 1987 ) . Toxins were kept at -20°C and aliquots were dissolved in appropriate solutions containing 0 . 1% BSA . The DNA sequence of hNav1 . 2 ( NM_021007 . 2 ) , rNav1 . 2a ( NM_012647 . 1 ) , rβ4 ( NM_001008880 ) and hβ2 ( NM_004588 . 4 ) ( acquired from Origene , USA ) , as well as their mutants was confirmed by automated DNA sequencing and cRNA was synthesized using T7 polymerase ( mMessage mMachine kit , Ambion ) after linearizing the DNA with appropriate restriction enzymes . Channels were expressed in Xenopus oocytes together with a β-subunit ( 1:5 molar ratio ) and studied following 1–2 days incubation after cRNA injection ( incubated at 17°C in 96 mM NaCl , 2 mM KCl , 5 mM HEPES , 1 mM MgCl2 and 1 . 8 mM CaCl2 , 50 μg/ml gentamycin , pH 7 . 6 with NaOH ) using two-electrode voltage-clamp recording techniques ( OC-725C , Warner Instruments ) with a 150 μl recording chamber . The data were filtered at 4 kHz and digitized at 20 kHz using pClamp 10 software ( Molecular Devices , USA ) . Microelectrode resistances were 0 . 5–1 MΩ when filled with 3 M KCl . The external recording solution contained 100 mM NaCl , 5 mM HEPES , 1 mM MgCl2 and 1 . 8 mM CaCl2 , pH 7 . 6 with NaOH . The experiments were performed at room temperature ( ∼22°C ) and leak and background conductances , identified by blocking the channel with tetrodotoxin ( Alomone Labs , Israel ) , have been subtracted for all Nav channel currents . All chemicals used were obtained from Sigma-Aldrich ( USA ) unless indicated otherwise . Voltage–activation relationships were obtained by measuring steady-state currents and calculating conductance ( G ) , anitted to the data according to: G/Gmax = 1 + e-zF ( V - V1/2 ) /RT-1 where G/Gmax is the normalized conductance , z is the equivalent charge , V1/2 is the half-activation voltage , F is Faraday's constant , R is the gas constant and T is temperature in Kelvin . Occupancy of closed or resting channels by ProTx-II and other toxins was examined using negative holding voltages where open probability was very low , and the fraction of uninhibited channels ( Fu ) was estimated using depolarizations that are too weak to open toxin-bound channels , as described previously . After addition of the toxin to the recording chamber , the equilibration between the toxin and the channel was monitored using weak depolarizations typically elicited at 5 s intervals . Off-line data analysis was performed using Clampfit 10 ( Molecular Devices , USA ) , and Origin 8 ( Originlab , USA ) . After each electrophysiological experiment , oocytes expressing hNav1 . 2 , hβ2 , and the described mutants were washed with ND100 and incubated with 0 . 5 mg/ml Sulfo-NHS-LC-biotin ( Pierce , USA ) for 30 min . Oocytes were thoroughly washed again in ND100 before lysis ( by pipetting up and down ) in 20 μl/oocyte buffer H ( 1% Triton X-100 , 100 mM NaCl , 20 mM Tris-HCl , pH 7 . 4 ) plus protease inhibitors ( Clontech , USA ) . All subsequent steps were performed at 4°C . Lysates were gently shaken for 15 min after which they were centrifuged at 16 , 200xg for 3 min . The pellet was discarded and the supernatant ( SN ) transferred to a fresh 1 . l Eppendorf tube . 40 μl of SN was stored at -80°C for later use as the whole cell protein aliquot . 200 μl of hydrophilic streptavidin magnetic beads ( New England Biolabs , USA ) were then added and the sample shaken gently at 4°C overnight . Beads were washed 6 times with buffer H and resuspended in 40 μl buffer H , after which the biotinylated protein was dissociated from the beads through the addition of 1X LDS loading buffer plus reducing agent ( 10% 2-ME , 50 mM DTT final conc . ) and boiling at 95°C for 5 min to generate the surface protein fraction . All samples were appropriately diluted in buffer H to give equal protein concentrations , as measured by a BCA assay ( Pierce , USA ) . 10 μg of the SN was run on a 10% Tris-Glycine Novex Mini-Gel ( Thermo Fisher Scientific , USA ) with Tris-Glycine running buffer and analyzed by Western analysis . Nitrocellulose membranes were probed overnight at 4°C with 1:1000 mouse anti-myc antibody as primary ( Cell Signaling Technologies , USA ) and for 45 min at room temperature with 1:10000 goat anti-mouse HRP-conjugated antibody as secondary ( Thermo-Fisher Scientific , USA ) . Membranes were incubated for 5 min with an enhanced chemiluminescent substrate before imaging . Xenopus oocytes were injected with 5 ng RNA of rβ2 , rNav1 . 2 , or both ( ∼1:5 molar ratio ) and incubated for 48 hr at 17°C . 30 oocytes were lysed for each condition , using 20 μl lysis buffer ( 1x PBS , 1% DDM , 10% glycerol , 1 mM EDTA , 1X protease inhibitor cocktail [Clontech , USA] ) per oocyte , and homogenized by passing through a 25-gauge syringe ( adapted from [Yu , 2012] ) . The lysate was rotated for 1 hr at 4°C in a 1 . 5 ml microfuge tube , after which it was spun for 30 min at 20 , 000xg at 4°C . Subsequent steps were performed at 4°C or on ice and during all rotations the tube was sealed with parafilm to prevent leakage . A fresh pipette tip was gently swirled in the supernatant to remove the bulk of the white goop by adhering to the tip and supernatant was transferred to a new tube , taking care not to disturb the pellet . The new tube was spun 3 min at 20 , 000xg and again was swirled with a fresh pipette tip to remove the white goop , after which the clear supernatant was transferred to a new tube . Protein concentration was assayed using a BCA protein concentration kit ( Pierce , USA ) . 150 μg protein was brought to 150 μl in lysis buffer and 1 . 5 μg ( 1 . 5 μl of 1 mg/ml ) anti-myc mouse antibody ( Thermo-Fisher Scientific , USA ) was added . The tube was rotated overnight at 4°C after which 30 μl protein G-coated magnetic Dynabeads ( Thermo-Fisher Scientific , USA ) were added and rotated for 4 hr at 4°C . The tube was spun for 1 min at 20 , 000xg and placed on a magnetic rack for 1 min to collect the beads . Supernatant was removed and stored for optimization and troubleshooting purposes , as were subsequent washes . The beads were washed with 200 μl wash buffer ( 1x PBS , 0 . 4% DDM , 10% glycerol , 1 mM EDTA ) by pipette mixing and then magnetized to collect beads . The wash process was repeated three times . After the third wash , beads were suspended in 100 μl wash buffer , transferred to another tube , spun for 1 min at 20 , 000xg , and magnetized to collect beads . The tube was then spun again for 1 min at 20 , 000xg and magnetized , followed by removal of the last residual fluid . Protein was eluted in 30 μl elution buffer ( 50 mM glycine , pH 2 . 8 ) by incubating for 3 min at room temperature , followed by magnetization and transfer to the final tube . For Western blotting of the immunoprecipitate , 17 μl of eluate was combined with 2 . 5 μl 10X reducing agent ( Thermo-Fisher Scientific , USA ) and 6 . 5 μl 4X LDS sample loading buffer ( Thermo-Fisher Scientific , USA ) , heated for 10 min at 37°C , then loaded onto a 1 . 0 mm 12-well 3–8% Novex Tris-Acetate pre-cast gel ( Thermo-Fisher Scientific , USA ) . A rabbit anti-PanNav antibody ( Alomone labs , Israel ) and rabbit anti-hβ2 antibody ( Cell Signaling Technologies , USA ) were chosen to avoid cross-reaction of the Western blot secondary antibody against the mouse antibody used for immunoprecipitation . For Western blotting , the primary antibodies were used at 1:200 and 1:1000 dilutions , respectively , applied for 1 hr at room temperature and incubated for 45 min at room temperature with a 1:10 , 000 goat anti-rabbit HRP-conjugated secondary antibody ( Thermo-Fisher Scientific , USA ) .	Our bodies run on electricity . The brain , heart and some other organs depend on small electrical signals that are generated by ions moving through specialized protein complexes that sit in the membrane surrounding a cell . One of these channels is a ‘sodium channel’ , through which positively charged sodium ions move . Tiny changes in the structure of the sodium channel can cause severe conditions such as epilepsy and heart arrhythmias , so it is crucial that we know how it works Sodium channels consist of different protein building blocks ( called α and β ) and it was not known exactly how these come together to form the full channel complex . However , previous studies hinted at which parts of the β building block make contact with the α protein . Now , Das , Gilchrist et al . have been able to visualize the three-dimensional structure of the β building block of the sodium channel in extremely high detail by using a technique called X-ray crystallography . The level of detail in the structure also allowed the amino acids that make up the β building block to be identified . Das , Gilchrist et al . then altered some of the amino acids in the sodium channel , and treated frog cells containing the mutant channel with a spider toxin that binds between the α and β building blocks . This revealed the location and identity of the exact contact points between the proteins . In the future , a full three-dimensional structure showing the α and β subunits bound together would yield invaluable information on how they cooperate to form the sodium channel complex and give insights into mutations that cause cardiac arrhythmias and epilepsy .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2016	Binary architecture of the Nav1.2-β2 signaling complex
The Drosophila protocadherin Fat ( Ft ) regulates growth , planar cell polarity ( PCP ) and proximodistal patterning . A key downstream component of Ft signaling is the atypical myosin Dachs ( D ) . Multiple regions of the intracellular domain of Ft have been implicated in regulating growth and PCP but how Ft regulates D is not known . Mutations in Fbxl7 , which encodes an F-box protein , result in tissue overgrowth and abnormalities in proximodistal patterning that phenocopy deleting a specific portion of the intracellular domain ( ICD ) of Ft that regulates both growth and PCP . Fbxl7 binds to this same portion of the Ft ICD , co-localizes with Ft to the proximal edge of cells and regulates the levels and asymmetry of D at the apical membrane . Fbxl7 can also regulate the trafficking of proteins between the apical membrane and intracellular vesicles . Thus Fbxl7 functions in a subset of pathways downstream of Ft and links Ft to D localization . An important goal for developmental biologists is to understand how organs achieve a predictable size and shape at the end of their development . The Hippo signaling pathway has emerged as a key regulator of organ size ( reviewed by Pan , 2010; Halder and Johnson , 2011; Tapon and Harvey , 2012 ) . While most components of this pathway were originally discovered using genetic screens in Drosophila , mammalian orthologs of those genes perform similar functions . Additionally , mutations in several components of the pathway have been described in human cancers . An exciting aspect of the Hippo pathway is that its growth-regulating activity can be modulated by cell-surface proteins that are capable of binding to ligands expressed on adjacent cells . Such interactions may be especially important for achieving precise control of growth at a local level that is necessary for generating the detailed features of an organ . Of the proteins that regulate the Hippo pathway , much research has focused on the protocadherin Fat ( Ft ) . In addition to regulating growth , Ft also regulates planar cell polarity ( PCP ) , oriented cell division and proximodistal patterning of appendages ( reviewed in Thomas and Strutt , 2012; Sharma and McNeill , 2013 ) and its regulated activity therefore impacts the size and shape of organs . The Ft protein localizes to the cell membrane just apical to the adherens junctions ( Ma et al . , 2003 ) . It has a large extracellular domain composed of 34 cadherin domains as well as 4 EGF-like domains and 2 laminin G domains ( Mahoney et al . , 1991 ) that binds to another large cadherin , Dachsous ( Ds ) ( Clark et al . , 1995 ) , on adjacent cells ( Matakatsu and Blair , 2004 ) . Ft–Ds interactions are modulated by the kinase Four-Jointed ( Fj ) , which resides in the Golgi and phosphorylates the extracellular domains of both Ft and Ds ( Ishikawa et al . , 2008; Brittle et al . , 2010; Simon et al . , 2010 ) . Both Ds and Fj are expressed in gradients in Drosophila imaginal discs where they function in patterning the disc along a major axis ( e . g . , equatorial to polar or proximodistal ) ( Yang et al . , 2002; Ma et al . , 2003 ) . While cadherins are known to have important functions in cell–cell adhesion , a key aspect of Ft function is its role as a signaling molecule ( Matakatsu and Blair , 2006 ) . Ft regulates the Hippo pathway in two ways . First , Ft influences the protein levels of Warts ( Wts ) , a kinase that regulates the activity and subcellular location of the pro-growth transcriptional co-activator Yorkie ( Yki ) ( Cho et al . , 2006; Rauskolb et al . , 2011 ) . Additionally , mutations in ft disrupt the localization of Expanded ( Ex ) , a FERM-domain protein that functions upstream of Hippo ( Hpo ) ( Bennett and Harvey , 2006; Silva et al . , 2006; Willecke et al . , 2006 ) , though other studies suggest Ft and Ex act in parallel ( Feng and Irvine , 2007 ) . A key downstream target of Ft is the atypical myosin Dachs ( D ) . The strong overgrowth elicited by ft mutations can be completely suppressed by loss of D function ( Cho et al . , 2006 ) . Additionally , PCP defects in ft mutants are partially rescued by loss of D ( Mao et al . , 2006 ) . D localizes to the apical membrane where , in cells of the wing disc , it localizes preferentially to the distal edge of the cell ( Mao et al . , 2006; Mao et al . , 2011; Ambegaonkar et al . , 2012; Bosveld et al . , 2012; Brittle et al . , 2012 ) . In ft mutants , increased levels of D are observed apically and D is redistributed around the entire perimeter of the cell ( Mao et al . , 2006; Brittle et al . , 2012 ) . However , the overall levels of D protein are not obviously changed ( Mao et al . , 2006 ) . It has been proposed that Ft restricts growth by negatively regulating the levels of D at the apical membrane and that it regulates the D-dependent PCP functions by maintaining D asymmetry ( Rogulja et al . , 2008 ) . An important gap in our current understanding of Ft function is how Ft regulates the levels and localization of D at the apical membrane . Ft does not bind to D itself , indicating that there must be one or more proteins that bind to Ft and mediate its regulation of D localization at the membrane . In an attempt to identify signaling pathways downstream of Ft , several recent studies have made systematic deletions in the intracellular domain ( ICD ) of Ft ( Matakatsu and Blair , 2012; Bossuyt et al . , 2013; Pan et al . , 2013; Zhao et al . , 2013 ) . These deletion studies implicate multiple non-overlapping regions in the ICD that differentially affect growth , PCP and organ shape , suggesting that Ft signals via multiple effector pathways . Additionally , several proteins have been shown to bind to the Ft ICD including the transcriptional repressor Atrophin/Grunge which regulates PCP ( Fanto et al . , 2003 ) , the novel protein Lowfat that regulates Ft protein levels ( Mao et al . , 2009 ) , and the casein kinase I protein Discs overgrown ( Dco ) that phosphorylates the Ft ICD ( Feng and Irvine , 2009; Sopko et al . , 2009 ) . Also , the palmitoyltransferase approximated ( App ) is needed for D localization to the membrane ( Matakatsu and Blair , 2008 ) . However , for each of these proteins , their role in mediating the regulation of D levels or asymmetry by Ft is not well understood . Here we describe the Drosophila ortholog of the Fbxl7 gene , which encodes an F-box protein and is a novel component of the Ft signaling pathway . Inactivation of Fbxl7 results in increased tissue growth via the Hippo pathway and abnormalities in wing shape and proximodistal patterning of appendages . Fbxl7 localizes preferentially to the proximal edge of cells in the wing pouch where it binds to and co-localizes with Ft . We find a role for Fbxl7 in one of the growth-suppressing signaling pathways downstream of Ft and also demonstrate a role for Fbxl7 in regulating the amount of D at the apical membrane as well as its distribution around the edge of the cell . In two different genetic screens , one for mutations that caused cells to outgrow their neighbors ( described in Tapon et al . , 2001 ) and another for mutations that enabled cells to promote the elimination of their slower-growing neighbors by cell competition ( Hafezi et al . , 2012 ) , we identified mutant alleles of the Drosophila Fbxl7 gene ( CG4221 ) , which encodes a protein with an F-box and 11 leucine-rich repeats ( LRRs ) ( Figure 1A , Figure 1—figure supplement 1A ) . Fbxl7 has a conserved human ortholog ( FBXL7 ) that shares 49% amino acid identity over the region spanning the F-box and the LRRs . Most proteins with these motifs function as part of an SCF-type ubiquitin ligase , a protein complex which polyubiquitylates substrate proteins and targets them for degradation by the proteasome ( Skaar et al . , 2013 ) . A third allele was identified fortuitously in an unrelated stock . Mutant clones of all three alleles were overrepresented in the adult eye when compared to clones of the parental FRT82B chromosome ( Figure 1B ) , suggesting that these Fbxl7 mutations cause increased tissue growth ( Figure 1C–E ) . Two of the mutations generate premature stop codons upstream of all conserved domains , while the third causes a cysteine-to-tyrosine change in a conserved residue in one of the LRRs ( Figure 1A , Figure 1—figure supplement 1B–C ) that likely interferes with the normal function of the protein , indicating that all three alleles reduce or eliminate Fbxl7 function . We also found that a Mi{MIC} minos insertion in the first intron of Fbxl7 ( Venken et al . , 2011 ) results in a strong loss-of-function phenotype similar to our other mutant alleles ( Figure 1—figure supplement 1A , Figure 1—figure supplement 2A–B ) . 10 . 7554/eLife . 03383 . 003Figure 1 . Fbxl7 negatively regulates growth through the Hippo pathway . ( A ) Protein model of Drosophila Fbxl7 and Human FBXL7 showing the three alleles identified ( red asterisks ) , F-box , and 11 Leucine Rich Repeat ( LRR ) domains . The two proteins have 49% amino acid identity throughout the F-box and LRR domains . ( B–E ) Mosaic adult eye assay . Heterozygous and wild-type cells have red pigment and homozygous mutant cells lack pigment . ( B ) Control mosaic eye . ( C ) Fbxl7C616Y , ( D ) , Fbxl7Q201X and ( E ) Fbxl7W389X mosaic eyes are composed of more mutant cells . ( F–K ) Adult wings with overlays . Arrows indicate anterior and posterior crossveins . Compared to ( F ) FRT82B control wings , ( G ) Fbxl7C616Y homozygous wings are larger and crossveins are closer . ( H ) Merge shows F in blue and G in red . Compared to ( I ) nubbin-Gal4 ( nb-Gal4 ) control wings , ( J ) nb>FLAG-Fbxl7 overexpressing wings are smaller and crossveins are closer . ( K ) Merge shows I in blue and J in red . ( L ) Quantification of wing area from Fbxl7 loss-of-function , RNAi ( JF01515 ) , and overexpression . n ≥ 20 wings , *p ≤ 0 . 001 , error bars show SD . ( M–M″ ) Cell competition assay in the mosaic eye imaginal disc . ( M ) Wild-type cells are marked by GFP ( green ) , while Fbxl7 mutant cells are GFP negative . ( M′ ) Activated caspase-3 ( red ) is detected in dying cells that are GFP positive ( arrows ) . ( M″ ) DAPI shows all nuclei . ( N–N″ ) Mosaic eye imaginal disc with diap1-GFP ( green ) reporter . ( N–N′ ) Wild-type cells are marked with RFP ( red ) and Fbxl7 mutant cells are RFP negative . ( N″ ) Mutant clones show higher levels of diap1-GFP ( arrows ) . ( O–O′ ) Mosaic wing imaginal disc with ex-lacZ reporter ( red ) . A clone overexpressing FLAG-Fbxl7 ( green , cells marked by EGFP ) has lower levels of ex-lacZ ( arrow ) . ( P–R ) Wing size genetic interaction assay . Compared to ( P ) nb>Fbxl7XP alone , ( Q ) reducing the dosage of wts partially rescues the small wing phenotype . ( R ) Merge shows P blue and Q in red . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 00310 . 7554/eLife . 03383 . 004Figure 1—figure supplement 1 . Fbxl7 gene and protein models . ( A ) Fbxl7 gene model . CG4221-RA is the only predicted isoform . Thin black lines are introns , gray boxes indicate non-coding exons , and blue boxes indicate coding exons . CG34276 is located in the first intron of Fbxl7 . P{XP}d08178 is inserted immediately upstream of the Fbxl7 transcript and contains UAS sequences to drive ectopic expression of Fbxl7 . Mi{MIC}MI04292 is inserted in the first intron of Fbxl7 . ( B ) Fbxl7 protein alignment of Drosophila Fbxl7 ( NP_650512 . 1 ) , Human FBXL7 ( NP_036436 . 1 ) , Mouse FBXL7 ( NP_795933 . 2 ) , and Zebrafish Fbxl7 ( NP_001073511 . 1 ) . Alignment performed with Clustal Omega and exported with Jalview software . Conserved F-box domains are shown in green and 11 Leucine Rich Repeat ( LRR ) domains are shown in yellow . Drosophila Fbxl7 mutations indicated in red . ( C ) Alignment of Cys-containing LRR subfamily domains from Drosophila Fbxl7 , Human FBXL7 , and S . cerevisiae Grr1 ( NP_012623 . 1 ) . LRRs identified using SMART software ( Letunic et al . , 2012 ) . Black box indicates the Drosophila Fbxl7C616Y amino acid mutation that affects a conserved cysteine . Structural model shows the predicted LRR β-sheet ( arrow ) and α-helix ( helix ) modeled after ( Hsiung et al . , 2001 ) . Consensus sequence definitions according to Bella et al . ( 2008 ) and Kobe and Deisenhofer ( 1994 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 00410 . 7554/eLife . 03383 . 005Figure 1—figure supplement 2 . Additional Fbxl7 mutant phenotypes . ( A–B ) Quantification of adult wing area and cross vein distance for different Fbxl7 loss of function and overexpression phenotypes . n ≥ 10 wings . Significance calculated with 1way ANOVA followed by Tukey's test . p ≤ 0 . 001 , p ≤ 0 . 05 . Error bars indicate SD . ( C–D ) Adult male prothoracic legs from control or overexpressing FLAG-Fbxl7 . Black bracket shows shortening of tibia and tarsus segments . ( E–J ) Wing hair polarity in the region between veins L3 and L4 and proximal to the anterior cross vein ( black box in E ) . ( J ) Overexpressing fat causes wing hairs to change direction ( yellow arrowhead ) . ( K–L ) Increased adult wing size from RNAi knockdown of Fbxl7 in the wing is enhanced by wts heterozygosity . Merge shows K in blue and L in red . ( M–O ) Confocal slices of imaginal discs with clones overexpressing FLAG-Fbxl7 and assessing Hippo pathway reporters ex-lacZ and Tub-EGFP . ban ( ‘bantam sensor’ ) . ex-lacZ positively reports Yki activity , whereas bantam-sensor inversely reports bantam activity ( bantam , a microRNA , is a transcriptional target of Yki ) . ( M–N ) Cells overexpressing FLAG-Fbxl7 ( marked by EGFP , green ) have lower levels of ex-lacZ ( red ) , and neighboring wild-type cells have higher levels of ex-lacZ ( arrowhead ) . ( O ) Cells overexpressing FLAG-Fbxl7 ( marked by myr-RFP , red ) have higher levels of bantam-sensor ( green ) , and neighboring wild-type cells have lower levels of bantam-sensor ( yellow arrowhead ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 005 Although clones of mutant cells display a clear growth advantage , flies homozygous for each of these Fbxl7 mutations are viable and fertile . However , the wings of Fbxl7 homozygotes or hemizygotes ( Fbxl7-/Deficiency ) are larger and more rounded than wild-type wings ( Figure 1F–H; quantified in Figure 1L and Figure 1—figure supplement 2A ) and the distance between the cross veins is reduced ( Figure 1F–G , Figure 1—figure supplement 2B ) . The same alterations in wing area and spacing between the cross veins were also observed when Fbxl7 function was reduced by RNAi ( Figure 1L , Figure 1—figure supplement 2A–B ) ( Dui et al . , 2012 ) . The combination of overgrowth and reduced spacing of the cross veins is especially reminiscent of mutations in the Ft branch of the Hippo signaling pathway ( Bryant et al . , 1988; Mahoney et al . , 1991; Clark et al . , 1995; Villano and Katz , 1995; Mao et al . , 2006; Matakatsu and Blair , 2008; Mao et al . , 2009 ) . Since we identified one of the Fbxl7 alleles in a screen for mutations that made cells capable of eliminating their neighbors ( Hafezi et al . , 2012 ) , we examined imaginal discs for evidence of cell death . We observed elevated levels of activated caspase-3 , a marker of apoptosis , especially in wild-type cells adjacent to Fbxl7 mutant clones ( Figure 1M–M″ ) . Thus Fbxl7 mutant cells do indeed behave as supercompetitors similar to loss-of-function mutations in ft or in core components of the Hippo pathway such as hpo or wts ( Tyler et al . , 2007 ) . Reduced signaling via the Hippo pathway results in increased activity of the transcriptional co-activator Yki . In Fbxl7 mutant clones in the eye imaginal disc , expression of a diap1-GFP reporter gene ( Zhang et al . , 2008 ) was increased , especially posterior to the morphogenetic furrow ( Figure 1N–N″ ) consistent with increased Yki activity . Additionally , the enlarged wing phenotype observed upon expression of Fbxl7RNAi was enhanced by heterozygosity of the wtsX1 allele ( Figure 1—figure supplement 2K–L ) . Together these results indicate that loss of Fbxl7 leads to increased growth via the Hippo pathway . When we overexpressed Fbxl7 in the wing imaginal disc , the adult wings were smaller and had a reduced distance between the cross veins ( Figure 1I–L , Figure 1—figure supplement 2A–B ) . Overexpressing Fbxl7 also reduced the length of distal leg segments ( Figure 1—figure supplement 2C–D ) . Results were similar using either a UAS-Fbxl7 transgene or a P[XP] transposon which contains UAS sequences upstream of the endogenous Fbxl7 transcriptional start site ( Figure 1—figure supplement 1A ) . This reduction in wing size was suppressed by heterozygosity of the wtsX1 allele ( Figure 1P–R , Figure 1—figure supplement 2A ) . When we overexpressed a form of Fbxl7 bearing the missense mutation identified in the screen , Fbxl7C616Y , there was , if at all , a slight increase in wing size ( Figure 1—figure supplement 2A ) suggesting that this mutation disrupts the normal function of the protein and likely functions as a dominant-negative mutation at least under conditions of overexpression . Since increased signaling via the Hippo pathway would reduce Yki activity , we examined expression of Yki reporters . Overexpression of Fbxl7 reduced expression of an ex-lacZ reporter ( Boedigheimer and Laughon , 1993; Hamaratoglu et al . , 2006 ) in a cell-autonomous manner ( Figure 1O–O′ ) . However , wild-type cells close to the Fbxl7-overexpressing cells had increased ex-lacZ reporter expression , especially in the notum of the wing disc and in the eye disc ( Figure 1—figure supplement 2M–N′ ) . A similar phenomenon was observed with the bantam sensor ( Figure 1—figure supplement 2O–O′ ) ( Brennecke et al . , 2003 ) , which is expressed at higher levels when Yki activity is reduced . This non-autonomous increase in Yki activity is similar to that seen when Ft is overexpressed ( Matakatsu and Blair , 2012 ) or at boundaries of differential Ds or Fj activity ( Willecke et al . , 2008 ) . Taken together , these results indicate that Fbxl7 functions as a negative regulator of growth via the Hippo pathway . Moreover , the multiple phenotypic similarities between alterations in Ft levels and Fbxl7 levels suggest that Fbxl7 functions in proximity to Ft . A polyclonal antibody to an N-terminal portion of Fbxl7 detects uniform Fbxl7 expression throughout the wing imaginal disc ( Figure 2A ) , with a slight enrichment at the dorsal-ventral boundary in the pouch as is also observed for Ft protein ( Mao et al . , 2009 ) . At the cellular level , punctate staining is observed outlining the apical profiles of cells , which is absent in homozygous mutant clones of the Fbxl7Q201X allele ( Figure 2B–B′ ) indicating that the truncated protein generated by this allele is likely unstable . In Fbxl7C616Y clones , apical puncta are absent but cytoplasmic staining is observed above background levels , indicating that the mutant protein is present but does not localize apically ( not shown ) . An Fbxl7 protein with an N-terminal FLAG epitope tag ( FLAG-Fbxl7 ) exhibits an apical localization that is very similar to that of the endogenous protein ( Figure 2C–F ) . Using either the anti-Fbxl7 antibody ( Figure 2C , E ) or FLAG-Fbxl7 ( Figure 2D , F ) , we found that Fbxl7 localizes to the subapical region of cells , apical to the adherens junctions marked by E-cadherin . FLAG-Fbxl7 is also found in intracellular puncta ( Figure 2G ) . In contrast , FLAG-Fbxl7 protein bearing the C616Y missense mutation displays only diffuse cytoplasmic localization ( Figure 2H ) suggesting that the normal function of Fbxl7 may be contingent upon its localization to the apical region or cytoplasmic puncta . In Drosophila S2 cells ( not shown ) or the flattened cells of the peripodial epithelium ( Figure 2I ) , confocal sections show puncta with diameters typically in the range of 400–500 nm ( some as large as 1000 nm ) with a hollow interior , consistent with the possibility that these might be vesicles . 10 . 7554/eLife . 03383 . 006Figure 2 . Fbxl7 is localized to apical membrane , cytoplasmic puncta , and the proximal side of planar polarized cells . Confocal slice of ( A ) endogenous Fbxl7 and ( A′ ) Armadillo ( Arm ) in the wing imaginal disc . Arrow indicates enrichment of Fbxl7 at the dorso-ventral boundary . ( B–B′ ) A confocal slice through the apical surface of wing disc cells . Fbxl7 ( red ) accumulates at the apical membrane and is lost from MARCM Fbxl7Q201X clones ( green ) . ( C–F ) Endogenous Fbxl7 and expressed FLAG-Fbxl7 ( green ) are localized to apical puncta aligned with cell edges marked by E-cadherin ( E-cad ) ( red ) . ( C–D ) Confocal slices through the apical surface of wing disc cells . ( E–F ) Confocal slice through folds in the wing disc . Fbxl7 is apical to E-cad ( arrowheads ) . ( F ) Asterisk indicates adjacent fold that does not express FLAG-Fbxl7 . ( G–H ) Confocal Z-slice through the wing disc with clones of cells expressing FLAG-Fbxl7 or FLAG-Fbxl7C616Y ( green ) . Nuclei are shown with DAPI ( blue ) . ( G ) FLAG-Fbxl7 localizes to apical membrane ( arrowhead ) and cytoplasmic puncta ( arrows ) , whereas ( H ) FLAG-Fbxl7C616Y shows diffuse cytoplasmic localization . ( I ) Confocal section through peripodial membrane showing FLAG-Fbxl7 localization to hollow puncta . Inset shows higher magnification of outlined box . ( J–K‴ ) Confocal slice of the wing disc pouch stained for E-cad ( red ) with clones expressing FLAG-Fbxl7 ( green ) . ( K–K″ ) Magnified region from box in J , showing FLAG-Fbxl7 enriched on proximal membrane ( arrowheads ) . ( K‴′’ ) Magnified region from box in K‴ . D = distal , P = proximal . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 006 In cells of the wing imaginal disc , Ft is preferentially expressed on the proximal side of cells and Ds and D on the distal surface ( Ambegaonkar et al . , 2012; Brittle et al . , 2012 ) . We generated small clones that expressed FLAG-Fbxl7 , which enabled us to examine the borders between FLAG-Fbxl7-expressing cells and wild-type cells . In the dorsal part of the wing pouch , where polarization of D is most evident ( Brittle et al . , 2012 ) , FLAG-Fbxl7 localizes preferentially to the proximal side of cells ( Figure 2J , K–K‴ ) . Since the localization of Fbxl7 is similar to that described for Ft , we examined whether the two proteins co-localize . Both anti-Fbxl7 and anti-Ft revealed apical staining in a punctate manner with a considerable degree of overlap ( Figure 3A–A″ ) . Additionally , we observed co-localization of FLAG-Fbxl7 and Ft at the apical membrane ( Figure 3B–B″ ) as well as in cytoplasmic puncta ( Figure 3B″–B‴ , Figure 3—figure supplement 1A ) , many of which were basally located . Higher gain settings were required to visualize the comparatively faint Ft staining in puncta ( Figure 3B‴′ ) . Because of a higher background level of cytoplasmic staining with anti-Fbxl7 , the FLAG-tagged Fbxl7 protein was necessary to observe co-localization in puncta . 10 . 7554/eLife . 03383 . 007Figure 3 . Fbxl7 physically interacts with Fat and regulates its apical localization . ( A–A″ ) Confocal slice through a wing disc fold showing endogenous Fbxl7 ( green ) and Fat ( red ) co-localize at apical membrane ( arrowhead ) . ( B–B‴′ ) Confocal Z-section showing FLAG-Fbxl7 ( green ) and Fat ( red ) co-localize at ( B–B″ ) apical membrane ( arrowhead ) and ( B″–B‴′ ) cytoplasmic puncta ( arrows ) . ( B″–B‴′ ) Inset shows magnification of puncta . B‴′ uses higher gain settings than B′ to visualize Fat in puncta . ( C ) Co-immunoprecipitation experiment in S2 cells . FatICD-V5 pulls down with FLAG-Fbxl7 , whereas pulldown is reduced with FLAG-Fbxl7C616Y . ( D–D‴ ) Confocal slice of the wing disc at the apical surface . Apical Fbxl7 ( red ) localization is lost from MARCM fatGrv clones ( green ) , whereas ( D′ ) E-cad ( blue ) localization is unchanged . ( D‴ ) shows magnification of the box in D . ( E–E‴ ) A basal confocal slice through the same clone in D , showing increased cytoplasmic levels of Fbxl7 . ( F–F″ ) Confocal slice through a fold showing a MARCM fatGrv clone ( GFP marker not shown ) which expresses FLAG-Fbxl7 ( anti-Flag , green ) . ( F′ ) E-cad ( red ) marks apical membrane . FLAG-Fbxl7 is not apically localized in fatGrv clones ( arrowhead ) , but does localize to cytoplasmic puncta ( arrows ) . ( G–G′ ) Confocal slice through the apical surface of a disc overexpressing FLAG-Fbxl7 ( green ) in clones . ( G′ ) Apical Fat ( red ) levels are elevated within the clone . ( H–H″ ) Confocal slice through the apical surface with a MARCM Fbxl7Q201X clone ( green ) showing ( H′ ) no change in levels of apical E-cad ( blue ) and ( H″ ) slightly elevated levels of apical Fat ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 00710 . 7554/eLife . 03383 . 008Figure 3—figure supplement 1 . Additional analysis of the relationship between Fbxl7 and Fat . ( A ) A confocal z-section through a wing disc with clones overexpressing FLAG-Fbxl7 and staining for FLAG ( green ) and Fat ( red ) . Apical membrane is towards the top of the image . Fat and FLAG-Fbxl7 colocalize at apical membrane ( yellow arrowhead ) . ( A‴–A‴′ ) Higher gain settings to observe Fat cytoplasmic punctae and heat map of fluorescence intensity . Fat localizes to punctae in both wild-type cells and FLAG-Fbxl7 expressing cells ( yellow arrows ) . Cytoplasmic Fat is slightly elevated in FLAG-Fbxl7 expressing cells . ( B ) Schematics of FLAG-Fbxl7 truncation constructs . ( C ) Western blots showing results of co-immunoprecipitation experiments from S2 cells expressing indicated transfected plasmids . FatICD co-immunoprecipitates with full length FLAG-Fbxl7 as well as the LRR domain . The Fbxl7C616Y protein reduces association with FatICD . The N-terminal domain of Fbxl7 can weakly associate with FatICD . FLAG-Fbxl7Δ3 protein is found at higher levels than other Fbxl7 proteins despite transfecting the same amount of plasmid and loading the same amount of total protein . ( D ) Western blots showing endogenous Fat protein from wing disc lysates of indicated genotypes . The asterisk indicates higher molecular weight Fat in discs overexpressing FLAG-Fbxl7 . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 008 To determine whether Ft and Fbxl7 can interact physically , we co-transfected S2 cells with tagged versions of Fbxl7 and a portion of Ft that includes the transmembrane domain and the entire intracellular domain ( FatICD ) . FatICD co-immunoprecipitates with FLAG-Fbxl7 , whereas association of FatICD with FLAG-Fbxl7C616Y is greatly reduced ( Figure 3C ) . We also examined the ability of truncated Fbxl7 proteins to interact with Ft and find that Fbxl7 interacts with Ft mostly via its LRRs ( Figure 3—figure supplement 1B–C ) . A weaker interaction is also observed between Ft and the N-terminal portion of Fbxl7 . Thus wild-type Fbxl7 can associate , either directly or indirectly , with the intracellular domain of Ft and this interaction mostly occurs via the LRRs of Fbxl7 . The apical localization of Fbxl7 was absent in ft clones ( Figure 3D–D‴ ) . However , an increase in diffuse cytoplasmic staining was observed ( Figure 3E–E‴ ) . Thus the localization of Fbxl7 to the apical region is dependent upon Ft and in the absence of Ft , Fbxl7 re-localizes to the cytoplasm . Since Ft and Fbxl7 also co-localize to cytoplasmic puncta or vesicles , we examined whether this localization of Fbxl7 also depends on Ft . Surprisingly , unlike the apical localization , punctate localization of FLAG-Fbxl7 was still observed in ft clones indicating that the localization of Fbxl7 in these cytoplasmic puncta is independent of Ft ( Figure 3F–F″ ) . Since proteins similar to Fbxl7 often bind to their substrates via their LRRs and promote their polyubiquitylation and degradation ( Skaar et al . , 2013 ) , we tested the effect of changes in Fbxl7 on the levels and localization of Ft . Increasing Fbxl7 levels resulted in clearly increased levels of apical Ft ( Figure 3G–G′ ) and slightly increased cytoplasmic staining of Ft ( Figure 3—figure supplement 1A ) . Surprisingly , a slight elevation of apical Ft levels was also observed in Fbxl7 mutant clones ( Figure 3H–H″ ) . The overall levels of Ft protein in imaginal discs , as assessed by Western blotting , were not obviously changed in either case ( Figure 3—figure supplement 1D ) . These results are inconsistent with Fbxl7 promoting Ft degradation and instead suggest that Fbxl7 regulates Ft localization . In support of this , we do not observe an obvious increase in Ft ubiquitylation from expressing Fbxl7 in S2 cells ( not shown ) . Since the phenotypic abnormalities of Fbxl7 mutants resemble those of hypomorphic alleles of ft , and the recruitment of Fbxl7 to the apical region of the cell is dependent upon Ft , we explored the relationship between Fbxl7 and proteins known to regulate Ft in more detail . In ds mutant clones , apical localization of Fbxl7 is no longer observed as discrete puncta at cell edges but is rather more diffuse ( Figure 4A–A″ ) . Moreover , in contrast to ft clones , we do not see an increase in cytoplasmic Fbxl7 in ds clones at more basal focal planes , indicating that Fbxl7 is still predominantly at an apical location ( Figure 4B–B″ ) . These changes in Fbxl7 localization could simply be a consequence of the more diffuse localization of Ft that is observed in ds clones ( Strutt and Strutt , 2002; Ma et al . , 2003; Mao et al . , 2009 ) . Fj is required for normal localization of Ds and Ft ( Strutt and Strutt , 2002; Ma et al . , 2003 ) . In agreement with this , we see subtle effects on Fbxl7 localization in fj clones , which appears similar to that seen in ds clones ( Figure 4—figure supplement 1A ) . 10 . 7554/eLife . 03383 . 009Figure 4 . Relationship between Fbxl7 and the Fat pathway proteins Ds and Dco . ( A–A″ ) Confocal slice through the apical surface of a disc with MARCM ds38k clones ( green ) showing disturbed localization of Fbxl7 ( red ) . E-cad staining is not altered ( blue ) ( B–B″ ) A basal confocal slice through the same clone in A , showing no change in Fbxl7 cytoplasmic levels . ( C–E″ ) Confocal slice through the apical surface of a disc with FLAG-Fbxl7 overexpressing clones ( green ) and stained for Ds ( red ) . ( C–D″ ) Apical Ds levels appear higher and more punctate in FLAG-Fbxl7 expressing clones . Wild-type cells immediately adjacent to the clone have reduced apical Ds ( arrowheads ) . ( E–E″ ) Ds and FLAG-Fbxl7 puncta are aligned on either side of the clone boundary ( arrowheads ) . ( F–F″ ) Apical confocal slice of a disc containing MARCM Fbxl7Q201X clones ( green ) and stained for Ds ( red ) and E-cad ( blue ) . Ds levels are normal or slightly elevated , in clones . ( G–G″ ) Confocal Z-section of a clone expressing FLAG-Fbxl7 ( green ) and stained for Ds ( red ) . Both are localized to apical membrane ( arrowhead ) and frequently co-localize in cytoplasmic puncta ( arrows ) . ( H–I″ ) Apical confocal slice of MARCM dco3 or dcole88 clones ( green ) and staining for Fbxl7 ( red ) and E-cad ( blue ) . Apical Fbxl7 levels are unchanged in ( H–H″ ) dco3 and ( I–I″ ) dcole88 clones . ( J ) Co-immunoprecipitation experiment in S2 cells . Dco-V5 pulls down with FLAG-Fbxl7 , whereas pulldown is reduced with FLAG-Fbxl7C616Y . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 00910 . 7554/eLife . 03383 . 010Figure 4—figure supplement 1 . Additional analysis of the relationship between Fbxl7 and Fat pathway proteins . ( A ) Confocal slice through the apical membrane of wing disc cells bearing MARCM fjN7 clones ( green ) and antibody stained for Fbxl7 ( red ) and E-cad ( blue ) . Fbxl7 apical localization is partially disrupted . ( B–C″ ) Confocal slices through the apical membrane of wing disc cells bearing MARCM clones ( green ) and antibody stained for Fbxl7 ( red ) and E-cad ( blue ) . Fbxl7 apical localization is normal in ( B–B″ ) dachsGC13 or ( C–C″ ) wtsX1 clones . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 010 When Fbxl7 is overexpressed in clones , cells have more prominent apical expression of Ds in puncta ( Figure 4C–C″ ) . Additionally , in wild-type cells bordering the Fbxl7-overexpressing clone , Ds staining is reduced and accumulates in prominent puncta at the surface that abuts the Fbxl7-overexpressing cells ( Figure 4D–E″ ) . Given that Ds can be drawn toward cells with greater levels of Ft ( Ma et al . , 2003 ) , Ds may be drawn toward Fbxl7-overexpressing cells due to the increased Ft levels . Furthermore , the puncta of Ds in adjacent wild-type cells are in register with Fbxl7 puncta , consistent with the coupling of Ds in wild-type cells to Fbxl7-bound Ft within the clone . In Fbxl7 mutant clones , there is , at best , a very slight elevation of Ds levels ( Figure 4F–F″ ) . Thus , the effects of Fbxl7 on Ds levels are minor compared to the effects on Ft levels . Additionally , we could not detect Ds in immunoprecipitates of Fbxl7 when the two proteins were co-expressed in S2 cells ( not shown ) . Together , these findings suggest that Fbxl7 binds to and functions with Ft rather than Ds . Despite this , we did observe co-localization of Fbxl7 and Ds at apical membranes and in more basally located cytoplasmic puncta ( Figure 4G–G″ ) . In the absence of evidence for direct interactions between Fbxl7 and Ds , their co-localization , at least at the cell surface , may result from Fbxl7 bound to Ft that is in turn bound to Ds . Ds binding to Ft induces the phosphorylation of the ICD of Ft , which requires the protein kinase , Dco ( Feng and Irvine , 2009; Sopko et al . , 2009 ) . Since some F-box proteins bind to phosphorylated proteins ( Skaar et al . , 2013 ) , we tested whether the apical localization of Fbxl7 was dependent upon Dco function . The apical localization of Fbxl7 was not obviously changed in clones of the dco3 allele that is unable to phosphorylate Ft ( Figure 4H–H″ ) ( Sopko et al . , 2009 ) . While Dco is capable of binding to Fbxl7 as assessed by co-immunoprecipitation from S2 cells ( Figure 4J ) , the apical localization of Fbxl7 was still observed in clones of the null dco allele , dcole88 ( Figure 4I–I″ ) , thus indicating that Dco function is altogether unnecessary for the apical localization of Fbxl7 . Furthermore , while changes in Fbxl7 alter Hippo signaling , changes in Hippo signaling do not regulate Fbxl7 levels or localization , as Fbxl7 localization is normal in clones mutant for dachs or wts ( Figure 4—figure supplement 1B–C″ ) . The primary amino acid sequence of the ICD of Ft does not predict any domains with enzymatic activity or known protein–protein interaction motifs . Hence , it has not been easy to understand how it functions in signal transmission . However , six blocks of sequence ( labeled A–F in Figure 5A based on the nomenclature of Pan et al . ( 2013 ) ) are conserved with the ICD of mammalian Fat4 . A region between the conserved blocks ‘B’ and ‘C’ seems necessary for the major growth-suppressive function of Ft ( Matakatsu and Blair , 2012; Bossuyt et al . , 2013; Zhao et al . , 2013 ) . In our screen , we identified an allele of ft , ft61 ( Figure 5A ) , which displays strong overgrowth ( Figure 5C , K ) and is caused by a single amino acid change ( T to I ) within this region . ft61 displays phenotypic abnormalities that are very similar to those described for ftsum , which also changes a single amino acid two residues N-terminal to ft61 ( Bossuyt et al . , 2013 ) . Additionally , in a ft null background , deletion of one of the conserved blocks ( block D in Figure 5A ) in a ft genomic rescue transgene was shown to cause overgrowth ( Pan et al . , 2013 ) albeit to a much lesser extent than for ft61 and ftsum; flies had slightly overgrown , rounder wings with decreased spacing between the crossveins ( Figure 5E ) . 10 . 7554/eLife . 03383 . 013Figure 5 . Fbxl7 functions in one of two growth-suppressing pathways downstream of Ft . ( A ) Protein model of the intracellular domain of Fat showing the transmembrane domain ( TM ) , regions conserved with mammalian Fat4 ( blue , A–F ) ( defined by Pan et al . , 2013 ) , regions associated with the major growth suppressive function of Fat ( red ) ( HM , Bossuyt et al . , 2013; Hippo-N , Hippo-C , Matakatsu and Blair , 2012; H2 , Zhao et al . , 2013 ) , region required for Dco binding ( green ) ( Sopko et al . , 2009 ) , mutV region ( orange ) ( Pan et al . , 2013 ) , Su ( DN ) region ( purple ) ( Matakatsu and Blair , 2012 ) , and two point mutations , ftsum ( Bossuyt et al . , 2013 ) and ft61 ( this study ) . Size and position of regions are drawn to scale relative to the ICD . ( B–C ) Mosaic adult eye assay . Heterozygous wild-type cells have red pigment and homozygous mutant cells lack pigment . Compared to ( B ) control FRT40A mosaic eyes , ( C ) ft61 mosaic eyes are larger and have more mutant tissue . ( D ) ftGrv/ft8; ft+ adult wing and ( D′–D‴ ) confocal slice of a wing disc showing that Fbxl7 ( red ) is localized to the apical membrane similar to E-cad ( green ) . ( E ) ftGrv/ft8; ftΔD adult wing and ( E′–E‴ ) confocal slice showing that Fbxl7 ( red ) apical localization is disrupted . ( F ) ftGrv/ft8; ftΔF adult wing and ( F′–F‴ ) confocal slice showing that Fbxl7 ( red ) apical localization is normal and similar to that in D′–D‴ . ( G ) Confocal slice of a disc containing a MARCM ft61 clone ( green ) and stained for Fbxl7 ( red ) and E-cad ( blue ) . Fbxl7 apical localization is normal in ft61 cells ( H ) Co-immunoprecipitation experiment in S2 cells . Fat-V5 , Fat61-V5 , and FatΔF-V5 pull down with FLAG-Fbxl7 , whereas pull down of FatΔD-V5 and FatmutV-V5 is reduced . Expressed Fat proteins contain only transmembrane and cytoplasmic regions ( ICD ) . ( I–M ) Wing imaginal discs ( and associated leg and haltere discs ) at low magnification . Compared to ( I ) control tub-Gal4 discs , ( J ) ftGrv/ft8 and ( K ) ft61/ft8 discs are larger and have more folds . ( L ) Ubiquitous expression of Fbxl7 does not rescue ftGrv/ft8 disc overgrowth . ( M ) Ubiquitous expression of Fbxl7 rescues disc overgrowth of ft61/ft8 . ( N–O ) Adult wing from ( N ) control tub-Gal4 and ( O ) ubiquitous expression of FLAG-Fbxl7 in an ft61/ft8 background . Animal lethality is rescued . ( P–Q′ ) Confocal slice through the eye imaginal disc showing MARCM clones ( green ) and anti-Diap1 staining ( red ) . ( P–P′ ) dco3 clones have elevated Diap1 levels and are overgrown , whereas ( Q–Q′ ) dco3 clones expressing FLAG-Fbxl7 have wild-type Diap1 levels and are reduced in size . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 01310 . 7554/eLife . 03383 . 014Figure 5—figure supplement 1 . Additional images of Fbxl7 localization in Fat deletion backgrounds . Confocal sections of wing disc cells stained with anti-Fbxl7 ( red ) and anti-E-cad ( green ) in different ft genetic backgrounds . ( A–A″ , C–C″ , E–E″ , G–G″ ) show apical sections and ( B–B″ , D–D″ , F–F″ , H–H″ ) show z-sections . ( A–B″ ) ftGrv/ft8 , ( C–D″ ) ftGrv/ft8; ft+ , ( E–F″ ) ftGrv/ft8; ftΔD , ( G–H″ ) ftGrv/ft8; ftΔF . Arrows indicate location of apical membrane . Fbxl7 apical localization is lost in ftGrv/ft8; ftΔD . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 01410 . 7554/eLife . 03383 . 015Figure 5—figure supplement 2 . Domain D of Ft is required for the effects of Fbxl7 on Ft localization . Confocal projection of apical membrane in wing disc cells stained for anti-V5 ( red ) with clones overexpressing FLAG-Fbxl7 ( green , marked by GFP ) in different genetic backgrounds . Ft proteins are N-terminally tagged with V5 . ( A–A′ ) V5-Ft+ , ( B–B′ ) V5-FtΔD , ( C–C′ ) V5-FtΔF . Arrows indicate edge of clones . Apical levels of V5-FtΔD protein are not increased upon Fbxl7 overexpression . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 01510 . 7554/eLife . 03383 . 016Figure 5—figure supplement 3 . Additional characterization of Fbxl7 rescue experiments . ( A–G ) Wing imaginal discs ( and associated leg and haltere discs ) at low magnification . Expanded set of panels compared to Figure I–M . Compared to ( A ) control tub-Gal4 discs , ( B ) ftGrv/ft8 and ( C ) ft61/ft8 discs are larger and have more folds . ( D ) Ubiquitous expression of Fat rescues ftGrv/ft8 disc overgrowth . ( E ) Ubiquitious expression of Fbxl7 does not obviously change disc size from wild type . ( F ) Ubiquitous expression of Fbxl7 does not rescue ftGrv/ft8 disc overgrowth . ( G ) Ubiquitous expression of Fbxl7 rescues disc overgrowth of ft61/ft8 . ( H–K ) Adult wings . ( H ) Control adult tub-Gal4 wing . ( I ) Ubiquitous expression of Fat can rescue ftGrv ft8 animal lethality . ( J ) Ubiquitous expression of Fbxl7 does not affect viability and wings are smaller . ( K ) Ubiquitous expression of Fbxl7 rescues animal lethality of ft61/ft8 . ( L–O′ ) Confocal slice through the eye imaginal disc showing MARCM clones ( green ) ( arrowheads ) and anti-Diap1 staining ( red ) . ( L–L′ ) wtsX1 clones have elevated Diap1 levels and are overgrown . ( M–M′ ) Expressing FLAG-Fbxl7 in wtsX1 clones does not effect the elevated Diap1 levels or overgrowth . ( N–N′ ) dco3 clones have elevated Diap1 levels and are overgrown , whereas ( O–O′ ) dco3 clones expressing FLAG-Fbxl7 have wild-type Diap1 levels are reduced in size . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 016 In contrast to null alleles of ft , which display strong overgrowth and cause lethality well before the adult stage , flies lacking Fbxl7 function are viable and fertile but have slightly overgrown wings that are rounded and have decreased spacing between the cross veins . Thus , their phenotypic abnormalities are very similar to those observed when the ft D region is deleted ( ftΔD ) . We therefore examined the localization of Fbxl7 in a ftΔD background . When a heteroallelic combination of null ft alleles , ftGrv/ft8 , is rescued by a wild-type version of ft ( ft+ ) , wings are normal ( Figure 5D ) and Fbxl7 displays normal apical localization ( Figure 5D′–D‴ , Figure 5—figure supplement 1 ) . However , apical localization of Fbxl7 is markedly reduced in ftGrv/ft8; ftΔD ( Figure 5E′–E‴ ) . We also examined a different deletion , ftΔF , in which wings from these flies are not enlarged but have greatly reduced spacing between the cross veins ( Figure 5F ) . In ftGrv/ft8; ftΔF imaginal discs , the apical localization of Fbxl7 is not disrupted ( Figure 5F′–F‴ ) . Similarly in ft61 clones , which display strong overgrowth , Fbxl7 localization was normal ( Figure 5G–G″ ) . Thus , the apical localization of Fbxl7 requires the Ft D domain but neither the F domain nor the motif that is disrupted by the ft61 allele . To examine whether the effects on Fbxl7 localization in vivo correlated with the ability of Fbxl7 to physically interact with Ft , we tested the ability of these mutant Ft proteins to co-immunoprecipitate with FLAG-Fbxl7 ( Figure 5H ) . Indeed , Ft61 and FtΔF proteins co-immunoprecipitated at levels comparable to wild-type Ft . However , the level of FtΔD in FLAG-Fbxl7 immunoprecipitates was greatly reduced , as was that of FtmutV , a mutant Ft protein in which a cluster of 10 serine/threonine residues overlapping the D domain was mutated to alanines . These sites were identified as candidates for phosphorylation by Dco ( Pan et al . , 2013 ) . However , since Fbxl7 localizes normally in dco mutant clones , the inability of Fbxl7 to bind to FtmutV might be caused by a change in its conformation that does not depend on phosphorylation by Dco . Indeed ftGrv/ft8; ftmutV flies also have phenotypic abnormalities that are very similar to those of Fbxl7 mutants ( Pan et al . , 2013 ) . To test for a functional relationship between the D domain of Fat and Fbxl7 , we monitored apical levels of Ft , FtΔD , and FtΔF under conditions of Fbxl7 overexpression . Ft and FtΔF levels are increased in cells overexpressing Fbxl7 , while FtΔD levels do not increase ( Figure 5—figure supplement 2 ) . This demonstrates that the D domain is required for Fbxl7 to physically interact with and exert its effects on Ft localization . If Ft61 protein is still capable of recruiting Fbxl7 to its apical location , then overexpression of Fbxl7 might suppress the overgrowth observed in mutant discs . The overgrowth and lethality of a ft null background ( ftGrv/ft8 ) can be rescued by ubiquitous expression of Ft ( Matakatsu and Blair , 2012; Bossuyt et al . , 2013; Pan et al . , 2013; Zhao et al . , 2013; Figure 5—figure supplement 3D , I ) . While ubiquitous Fbxl7 expression was unable to suppress ftGrv/ft8 phenotypes ( Figure 5L ) , the overgrowth and lethality of ft61/ft8 discs was indeed suppressed , resulting in viable adult flies ( Figure 5M , O ) . dco3 and wts mutant cells in the eye imaginal disc are overgrown and express higher Diap1 levels , an indicator of Yki activity ( Figure 5P–P′ , Figure 5—figure supplement 3L–N′ ) . Fbxl7 overexpression can rescue both clone size and Diap1 levels in dco3 mutant cells ( Figure 5Q–Q′ , Figure 5—figure supplement 1O–O′ ) , but not wts clones ( Figure 5—figure supplement 3M–M′ ) . Thus , mutant Ft61 protein , or Ft protein that cannot be phosphorylated by Dco , can still bind to Fbxl7 and facilitate the growth-suppressive functions of Fbxl7 . Taken together , these findings implicate Fbxl7 in one of two growth-suppressive pathways downstream of Ft and suggest that these two pathways might converge further downstream ( ‘Discussion’ ) . Since Ft and Fbxl7 localized preferentially to the proximal side of cells , we compared the localization of Fbxl7 with that of D . In confocal z-sections , D and Fbxl7 co-localize at the subapical membrane in puncta , apical to the adherens junction marker Armadillo ( Arm ) ( Figure 6A–A‴ ) . However , careful examination of these puncta in x-y sections shows that the Fbxl7 and D puncta are slightly offset in the proximodistal direction ( Figure 6B–B′ ) . D is localized at higher levels at the distal edge of the cell ( Mao et al . , 2006; Brittle et al . , 2012 ) where it is likely stabilized by physical interaction with the cadherin Ds ( Bosveld et al . , 2012 ) . Therefore , a likely explanation is that the formation of multimeric Ft–Ds complexes between cells results in the concomitant accumulation of Fbxl7 at the FatICD and D at the DsICD ( Figure 6C ) . 10 . 7554/eLife . 03383 . 011Figure 6 . Fbxl7 regulates the localization of Dachs . ( A–A″ ) Confocal slice through a bend in the wing disc showing ( A′ ) Dachs ( red ) and ( A″ ) Fbxl7 ( green ) localize at subapical membrane . ( A‴ ) Like Fbxl7 , Dachs is apical to the adherens junction marked by Arm ( blue ) . ( B–B′ ) Confocal slice through the apical surface of the wing disc , specifically the dorsal edge of the pouch , showing Dachs ( red ) and Fbxl7 ( green ) staining . Dachs and Fbxl7 puncta abut each other on either side of the cell boundary . Proximodistal axis indicated as P<−>D . ( C ) Diagram of polarized wing disc cells in which Dachs is enriched on the distal side and Fbxl7 is on the proximal side , linked by their association to Dachsous and Fat , respectively , which bind across cells . ( D–D″ ) Apical confocal slice of MARCM Fbxl7Q201X clones ( green ) and staining for Dachs ( red ) and Arm ( blue ) . Dachs levels are elevated in clones . ( E–G ) Apical confocal slice with staining for Dachs in ( E ) wild-type or ( F ) Fbxl7Q201X discs . Images are from the dorsal edge of the pouch and are aligned so the proximodistal axis is vertical . Dachs enrichment on P/D membrane , seen in ( E ) wild-type discs , is impaired in ( F ) Fbxl7Q201X discs . ( G ) Quantification of Dachs P/D enrichment in wing discs . Dachs is localized in a P/D direction , whereas Actin is not . Dachs P/D asymmetry is impaired in both Fbxl7C616Y and Fbxl7Q201X discs . Significance calculated with one-way ANOVA test . **p ≤ 0 . 001 , p ≤ 0 . 05 . Error bars indicate SD . ( H–H″ ) Apical confocal slice of FLAG-Fbxl7 overexpressing clones ( red , cells marked by RFP ) and staining for anti-GFP ( green , Dachs:GFP ) and E-cad ( blue ) . Apical Dachs levels within the clone are reduced , and Dachs is enriched at the edge of the clone . ( I–I‴ ) Confocal z-section of a wing disc with a FLAG-Fbxl7 expressing clone ( green ) and stained for Dachs ( red ) . FLAG-Fbxl7 and Dachs co-localize to apical membrane ( arrow ) and intracellular puncta ( arrowheads ) . ( I″–I‴’ ) Cytoplasmic levels of Dachs are slightly elevated within the clone . ( I‴ ) Heat map of I″ . ( J ) Western blots from wing disc lysates . Endogenous Dachs protein levels are not changed in Fbxl7 mutant wing discs compared to control . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 01110 . 7554/eLife . 03383 . 012Figure 6—figure supplement 1 . Additional Dachs tissue staining and Dachs levels in wing discs and S2 cells . ( A–A′ ) Confocal slice through apical membrane of a clone expressing FLAG-Fbxl7 ( green ) in the wing disc and staining for Dachs ( red ) . In wild-type cells bordering the clone , Dachs is enriched at the boundary . Puncta of Dachs in wild-type cells are apposed across the boundary with puncta of FLAG-Fbxl7 in overexpressing clones ( arrowheads ) . ( B–C ) Confocal slice through apical membrane of ft61 clones ( green , marked by GFP ) and stained for anti-Dachs ( red ) and anti-Arm ( blue ) . Arrowheads indicate edge of clone . ( B ) Dachs levels are higher in ft61 clones . ( C ) Overexpression of FLAG-Fbxl7 in ft61 clones reduces apical Dachs levels , to levels comparable to neighboring wild-type cells . Dachs levels are enriched at the edges of the clone . ( D ) Additional western of endogenous Dachs protein in dissected wing discs . Dachs protein levels do not change in Fbxl7 mutant , Fbxl7 overexpressing , or ft mutant backgrounds compared to control . ( E ) Western analysis of transfected Dachs-V5 protein levels in S2 cells . Transfecting increasing doses of FLAG-Fbxl7 does not change Dach-V5 levels . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 012 To investigate whether Fbxl7 can regulate the levels or localization of D , we first examined Fbxl7 mutant clones . The levels of apical D are increased throughout the clone ( Figure 6D–D″ ) although not to the extent that occurs in ft clones . Thus Fbxl7 negatively regulates the level of D at the apical membrane . To determine whether Fbxl7 has a role in generating or maintaining the asymmetrical distribution of D , we examined the distribution of D in Fbxl7 mutant wing discs . In these experiments , the distal edge of one cell cannot be distinguished from the proximal edge of its neighbor . However , in wild-type cells , endogenous D is preferentially observed on the proximal/distal edges and is found at lower levels at the other edges ( Brittle et al . , 2012; Figure 6E , G ) . In Fbxl7C201X and Fbxl7C616Y homozygotes , this bias in the distribution of D within the cells is reduced ( Figure 6F–G ) , indicating that Fbxl7 also has a role in regulating the asymmetric localization of D . We examined the localization of Dachs-GFP in clones that overexpressed Fbxl7 . In these clones there was reduction in the overall levels of apical D ( Figure 6H–H″ ) . In addition , Dachs-GFP puncta ( Figure 6H–H″ ) or endogenous D ( Figure 6—figure supplement 1A ) in neighboring wild-type cells are enriched against the border with Fbxl7 overexpressing cells , and are aligned with puncta containing FLAG-Fbxl7 , reminiscent of Ds staining in Figure 4E–E″ . This likely resulted from the elevated levels of Ft in Fbxl7-overexpressing clones , which would cause an enrichment of Ds ( and hence D ) on the surface of wild-type cells contacting the clone . In z-sections , we observed subtle changes in the localization of D within the clone itself ( Figure 6I–I‴ ) . There was a slight increase in D throughout the cell , possibly at the expense of some of the bright puncta that are normally observed at the apical region . Furthermore , overexpressing Fbxl7 can rescue the higher apical Dachs levels seen in ft61 clones ( Figure 6—figure supplement 1B–C ) . Thus , overexpression of Fbxl7 may cause a shift in the overall distribution of D from the apical region to the interior of the cell . To determine whether Fbxl7 functions as part of an SCF-type ubiquitin ligase , we first tested whether Fbxl7 was capable of interacting with either SkpA or Cul1 . In co-transfection experiments in S2 cells , robust interactions were observed in both cases indicating that Fbxl7 likely functions as part of an SCF complex ( Figure 7A ) . Furthermore , when Fbxl7 was cotransfected with HA-tagged ubiquitin , and ubiquitylated proteins immunoprecipitated with anti-HA , a high molecular weight smear above the size of wild-type Fbxl7 was observed indicating that Fbxl7 is ubiquitylated under these conditions ( Figure 7B ) . This is expected , as F-box proteins that function in SCF complexes are often themselves ubiquitylated ( Galan and Peter , 1999; Yen and Elledge , 2008 ) . Interestingly , Fbxl7C616Y , which is incapable of binding to Ft , is also ubiquitylated suggesting that the incorporation of Fbxl7 into an active SCF complex does not require Ft . 10 . 7554/eLife . 03383 . 017Figure 7 . Fbxl7 does not affect Dachs ubiquitylation , and Fbxl7 affects the localization of Cindr . ( A ) Co-immunoprecipitation assay from S2 cells . SkpA-HA and Cul1-HA immunoprecipitates with FLAG-Fbxl7 . ( B ) In-vivo Fbxl7 ubiquitylation assay in S2 cells . FLAG-Fbxl7 and FLAG-Fbxl7C616Y are ubiquitylated in vivo . ( C–D ) In-vivo Dachs ubiquitylation assay in S2 cells . Dachs-V5 is ubiquitylated under wild-type conditions , and does not change with ( C ) overexpression of FLAG-Fbxl7 or ( D ) knockdown of Fbxl7 with two different dsRNAs . ( E–E″ ) Confocal slice showing localization of FLAG-Fbxl7 ( red ) and GFP-Cindr ( green ) in puncta ( arrowheads ) . ( F ) Confocal slice through a bend in the wing disc . GFP-Cindr ( green ) localizes to subapical membrane , apical to E-cad ( red ) . Asterisk indicates an adjacent bend in the tissue . ( G ) Co-immunoprecipitation experiment in S2 cells . Cindr-V5 pulls down with full length FLAG-Fbxl7 , and FLAG-Fbxl7Δ2 , which contains only the LRR domains . ( H–I″ ) Confocal slice in a disc with clones overexpressing FLAG-Fbxl7 ( red , cells marked by myr-RFP ) in a GFP-Cindr background . ( H–H″ ) An apical plane shows loss of apical GFP-Cindr within the clone , and ( I–I″ ) a basal plane shows accumulation of GFP-Cindr in puncta . ( J ) Compared to ( J ) nb>FLAG-Fbxl7 alone , ( K ) overexpressing GFP-Cindr partially rescues the small wing phenotype . ( L ) Merge shows J blue and K in red . ( M ) Model of Fbxl7 as a component of Fat signaling . Not drawn to scale . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 01710 . 7554/eLife . 03383 . 018Figure 7—figure supplement 1 . Additional analysis of Dachs ubiquitylation , vesicle markers , and Cindr . ( A ) In-vivo ubiquitylation assay of Dachs-V5 from imaginal discs . Expressing FLAG-Fbxl7 does not alter ubiquitylation of Dachs-V5 . ( B ) Knockdown of endogenous Fbxl7 and ( C ) knockdown of transfected FLAG-Fbxl7 by two different dsRNAs ( DRSC15553 , DRSC38270 ) in S2 cells . ( D ) In-vivo ubiquitylation assay of Dachs-V5 from S2 cells . Expressing Cul1DN does not affect Dachs-V5 ubiquitylation . ( E–J″ ) Confocal slice showing localization of FLAG-Fbxl7 ( red ) and a vesicle marker ( green ) in the wing disc . Fbxl7 does not colocalize with a marker of the ( E ) ER ( GFP-KDEL ) , ( F ) trans-Golgi ( GFP-Galt ) , or ( G ) early endosomes ( GFP-FYVE ) . ( H–I″ ) Fbxl7 partially colocalizes with ( H–H″ ) Snx3-GFP and ( I–I″ ) Vps35-myc in vesicles . Insets shows magnification of vesicles with overlap . ( J–J″ ) Confocal section through the apical surface of a wing disc expressing UAS-Fat ( marked with mRFP ) in the posterior compartment ( engrailed-Gal4 ) in a GFP-Cindr ( green ) background and staining for E-cadherin ( blue ) . GFP-Cindr apical localization is unchanged with Fat overexpression . ( K ) Genetic interaction of Fbxl7 and Cindr experiment and quantification of adult wing size from expression of genes in the wing pouch ( nb-Gal4 ) . UAS-GFP-Cindr overexpression partially rescues the small wing due to UAS-FLAG-Fbxl7 overexpression . UAS-GFP-Cindr expressing wings are wild type in size . ( L–M ) Compared to ( L ) control nb-Gal4 wings , ( M ) overexpressing GFP-Cindr causes wings to be rounder and crossveins closer ( arrows ) . ( O ) Quantification of reduced crossveins of GFP-Cindr overexpressing wings . ( K and O ) Significance calculated with 1way ANOVA followed by Tukey's test . n ≥ 10 wings . ***p ≤ 0 . 001 . Error bars indicate SD . DOI: http://dx . doi . org/10 . 7554/eLife . 03383 . 018 Since Fbxl7 may function as a component of an E3 ubiquitin ligase , the most parsimonious explanation of its function would be that Fbxl7 ubiquitylates Dachs directly and promotes its degradation by the proteasome . However , the overall levels of D are unchanged in Fbxl7 mutant discs ( Figure 6J ) , discs that overexpress Fbxl7 , or ft mutant discs as assessed by Western blotting ( Figure 6—figure supplement 1D ) . In addition , increasing doses of transfected FLAG-Fbxl7 in S2 cells does not affect total levels of Dachs-V5 ( Figure 6—figure supplement 1E ) . If at all , a slight increase in Dachs-V5 levels was observed . Since Fbxl7 is localized apically and preferentially localizes to the proximal edge of the cell , Fbxl7 could promote D degradation locally and this may not be reflected in the overall levels of D . We therefore tested whether Fbxl7 was capable of promoting D ubiquitylation . These experiments were conducted in both S2 cells and imaginal discs . Ubiquitylated D was readily detected . However , the level of ubiquitylation was unchanged when Fbxl7 was increased ( Figure 7C , Figure 7—figure supplement 1A ) . Additionally , we reduced Fbxl7 in S2 cells by RNAi-mediated knockdown using two different dsRNAs and still observed no change in D ubiquitylation ( Figure 7D ) . Importantly , we observe knockdown of Fbxl7 protein in S2 cells using these dsRNAs ( Figure 7—figure supplement 1B–C ) . Furthermore , expressing a dominant negative version of Cul1 ( Cul1DN ) does not impair D ubiquitylation ( Figure 7—figure supplement 1D ) , implying that the SCF complex may not be involved . Thus we have no evidence that Fbxl7 influences D ubiquitylation . If Fbxl7 negatively regulates the accumulation of D at the apical membrane , it may do so by promoting the trafficking of D into intracellular vesicles . Indeed , we observed a population of intracellular puncta , likely vesicles , that contain both Fbxl7 and D ( Figure 6I–I″ ) . Moreover Fbxl7 overexpression can cause an overall shift in D from the apical membrane to the interior of the cell ( Figure 6I‴ ) . To further characterize the population of vesicles that contain Fbxl7 , we examined the localization of FLAG-Fbxl7 with 59 different markers that each labeled a subpopulation of vesicles and with several proteins that have been identified as interactors of Ft in proteomic studies ( Kwon et al . , 2013 ) ( Supplementary file 1 ) . No co-localization was observed with most of these markers ( Figure 7—figure supplement 1E–G as examples ) and partial co-localization was seen with two markers of the retromer pathway , Snx3 and Vps35 ( Figure 7—figure supplement 1H–I″ ) . However , strong co-localization was seen with a protein-trap insertion of GFP in the cindr locus . Cindr is thought to be an adapter protein that links membrane proteins to the actin cytoskeleton ( ‘Discussion’ ) . In basal sections , there is almost complete overlap between GFP-Cindr and FLAG-Fbxl7 in puncta ( Figure 7E–E″ ) . GFP-Cindr is normally localized to the subapical membrane , apical to E-cadherin , but its localization there is less punctate and more diffuse than that of FLAG-Fbxl7 ( Figure 7F ) . When tagged versions of both proteins were expressed in S2 cells , Cindr co-immunoprecipitated with full length Fbxl7 or with a version containing only the LRRs ( Fbxl7Δ2 ) ( Figure 7G ) . To determine whether Fbxl7 could influence the cellular localization of GFP-Cindr , we overexpressed FLAG-Fbxl7 in clones in GFP-Cindr animals . In these clones , we observed a dramatic re-localization of GFP-Cindr . GFP-Cindr is almost entirely eliminated from the apical membrane ( Figure 7H–H″ ) and increased numbers of basally located vesicles are observed ( Figure 7I–I″ ) . Thus , Fbxl7 is capable of displacing a protein associated with the apical membrane into intracellular vesicles . Importantly , this is unlikely to be a consequence of increased Ft in the apical membranes of Fbxl7-ovexpressing cells , since increasing Ft levels has no effect on the apical localization of GFP-Cindr ( Figure 7—figure supplement 1J–J″ ) . We next tested whether changes in Cindr levels are capable of modifying Fbxl7 phenotypes . Indeed we find that the reduction in wing size from overexpression of Fbxl7 was suppressed by co-expression of GFP-Cindr ( Figure 7J–L , Figure 7—figure supplement 1K ) . Overexpression of GFP-Cindr alone causes slightly rounder wings with closer crossveins , though these wings were not significantly overgrown ( Figure 7—figure supplement 1L–O ) . Under conditions of Fbxl7 overexpression , we did not observe any increase in Cindr ubiquitylation indicating that Cindr is unlikely to be a direct target of Fbxl7 ( not shown ) . Moreover , reducing Cindr levels by RNAi did not elicit phenotypic abnormalities in wings suggestive of defects in Ft or D ( not shown ) . However , the ability of Fbxl7 to cause changes in the localization of Cindr and Ft indicates that it can regulate trafficking of proteins between the apical membrane and the interior of the cell in either direction , and the pathways that regulate the trafficking of these proteins and D might share common components . Some of these shared components could potentially be direct targets of Fbxl7 ubiquitylation . Recent studies have revealed that Ft's effects on distinct pathways may be genetically separated , and that multiple effector domains can contribute to the same function . Indeed , the growth-suppressing function of Ft may occur via at least two regions of the Ft ICD . One or more regions between amino acids 4834 and 4899 in full-length Ft appear responsible for Ft’s ability to regulate Hippo signaling ( labeled HM in Figure 7M ) ( Matakatsu and Blair , 2012; Bossuyt et al . , 2013; Zhao et al . , 2013 ) . Several mutations within this region compromise this function of Ft and cause massive tissue overgrowth ( Bossuyt et al . , 2013 ) . Intriguingly , an allele of ft , ft61 , which harbors such a mutation , showed neither an effect on the recruitment of Fbxl7 to the apical membrane nor on the binding of Ft to Fbxl7 . Thus , signaling via this region of the ICD appears to be independent of Fbxl7 . A second , more C-terminal region of the Ft ICD ( Region D in Figure 7M ) that extends between amino acids 4975 and 4993 of full-length Ft , is removed by the ftΔD deletion and also has a growth-suppressive function albeit weaker than that of HM ( Pan et al . , 2013 ) . This second growth-suppressive pathway requires the function of Fbxl7 , as the protein generated by the ftΔD allele cannot bind to Fbxl7 nor can it localize Fbxl7 to the apical membrane . Additionally , the phenotypic abnormalities of null alleles of ft rescued by ftΔD are very similar , if not identical to those of Fbxl7 mutants . Furthermore , like ftΔD , Fbxl7 mutations do not display overt abnormalities of hair orientation in the wing ( Figure 1—figure supplement 2E–J ) , or abdomen ( not shown ) . We have shown that hyperactivation of the “weaker” Fbxl7-dependent pathway can overcome the absence of the ‘stronger’ Fbxl7-independent pathway; overexpression of Fbxl7 can suppress the overgrowth of ft61 . Thus , while these two pathways can be dissociated at the level of the Ft ICD , they nevertheless seem to converge further downstream . This point of convergence likely involves D since the overgrowth of ft mutant tissue can be suppressed completely by eliminating D function ( Cho et al . , 2006 ) . Indeed , it has previously been suggested that Ft regulates growth by restricting the levels of apical D , and regulates PCP by influencing the planar asymmetry of apical D ( Rogulja et al . , 2008; Pan et al . , 2013 ) . Another key finding in our experiments is that Fbxl7 mutations perturb the distribution of D around the perimeter of the apical region of the cell . D is normally biased towards the distal edge of the cell; in Fbxl7 mutants , D is more evenly distributed around the cell perimeter . The asymmetric localization of D depends on at least two different regions of Ft ( Pan et al . , 2013 ) . One is the region that binds to Fbxl7 ( Region D ) and the other is composed of the last three amino acids at the C-terminus of the protein ( Region F in Figure 7M ) , which is not necessary for Fbxl7 localization to the apical membrane . Thus , for the regulation of D asymmetry as well , there appears to be an Fbxl7-independent pathway . The existence of multiple downstream effector pathways that converge on common biological outcomes suggests that these pathways might function redundantly to some extent and thus provide robustness . This might also explain why the phenotypes elicited by overexpression of Fbxl7 are , in general , more severe than those observed in loss-of-function mutations . Previous observations of the localization of Ft , Ds , and D to vesicles are suggestive of trafficking events being involved in Ft signaling ( Ma et al . , 2003; Matakatsu and Blair , 2004; Mao et al . , 2006 ) . We have demonstrated that , in addition to the apical membrane , Fbxl7 localizes to vesicles . Moreover , FLAG-Fbxl7 vesicles can contain Ft , Ds and D , and these may be related to the apical puncta observed on cell edges . This localization is likely specific , since we do not see Fbxl7 co-localization with other cell surface proteins such as Crumbs , Notch , and E-cadherin ( not shown ) . Currently very little is known about the role of each of these proteins in vesicles . However , there is an increasing appreciation that most transmembrane proteins , and even proteins that are associated with the inner leaflet of the cell membrane are maintained at the plasma membrane by a dynamic process involving endocytosis and vesicle recycling ( e . g . , Schmick et al . , 2014 ) . We provide evidence that Fbxl7 regulates Ft apical localization , but how this regulation relates to the Fbxl7 phenotypes is not clear . Since Fbxl7 overexpression increases Fat signaling , and rescues the overgrowth-inducing Ft61 allele , perhaps this is due to the increased levels of Ft protein at the apical membrane . However , Ft levels are slightly elevated in Fbxl7 mutants , which display mild overgrowth . Therefore the mutant phenotype cannot be explained by the effect on Ft . Another known regulator of apical Ft levels is lowfat ( lft ) ( Mao et al . , 2009 ) . Fbxl7 and Lft appear to regulate Ft in different ways . Lft overexpression , like Fbxl7 , increases Ft levels . However , while Ft levels are decreased in lft mutant cells , Ft levels are increased in Fbxl7 mutant cells , though less so compared to Fbxl7 overexpression . Interestingly , for many proteins that regulate cellular trafficking , similar phenotypic abnormalities are observed with gain-of-function and loss-of-function mutations , since the normal execution of the process requires the protein to shuttle efficiently between two states ( Park et al . , 1993 ) . Thus dynamic aspects of the localization of Ft , Ds and D clearly merit more attention . The interactions we have observed between Fbxl7 and the adapter protein Cindr may provide clues for how Fbxl7 regulates D localization . Fbxl7-associated vesicles show almost complete overlap with GFP-Cindr and Fbxl7 can re-localize Cindr from the apical membrane to the interior of the cell . This finding , together with the observed increase in basal levels of D upon Fbxl7 overexpression ( Figure 6I–I‴ ) , suggests that Fbxl7 may function to regulate D trafficking in a similar manner . Cindr and its mammalian orthologues Cin85 and CD2AP are thought to regulate interactions between membrane proteins and actin cytoskeleton ( Haglund et al . , 2002; Petrelli et al . , 2002; Soubeyran et al . , 2002; Johnson et al . , 2011 , 2012 ) . D is an atypical myosin with a predicted actin binding domain in its conserved head domain . Therefore , the vesicles which Fbxl7 associates with D and Cindr may be linked to the actin cytoskeleton . In addition , our finding of partial colocalization of Fbxl7 with retromer components further supports the possibility that Fbxl7 may have a role in protein trafficking . Many F-box proteins associate with Skp1 and Cul1 to form an SCF E3 ubiquitin ligase complex ( reviewed in Skaar et al . , 2013 ) . Recruitment of specific substrates results in their poly-ubiquitylation and degradation , or mono-ubiquitylation , which can have non-degradative signaling roles . In addition , some F-box proteins have SCF-independent roles ( Nelson et al . , 2013 ) . Fbxl proteins are thought to recruit substrates to the SCF complex through the interaction with their LRR domains , and substrates have been identified for several Fbxls such as Skp2 ( Fbxl1 ) , which degrades p27 ( Carrano et al . , 1999; Sutterluty et al . , 1999 ) . However many , like Fbxl7 , are still uncharacterized as ‘orphan’ F-box proteins with no known substrates . Since we find that Fbxl7 associates with Skp1 and Cul1 , its potential substrates may be involved in Ft signaling . Fbxl7 has one described substrate in mice , Aurora A ( Coon et al . , 2012 ) . However we do not believe Aurora A is a relevant substrate in Drosophila , as we do not observe Ft signaling defects when Aurora A is knocked down or overexpressed ( not shown ) . The identification of F-box protein substrates has mainly been accomplished by unbiased approaches ( Skaar et al . , 2013 ) . Similarly , a combination of unbiased approaches , involving proteomics , genetic interaction screens , and identifying proteins that co-localize with Fbxl7 in vesicles could be used to identify Fbxl7 substrates . Fbxl7C616Y and Fbxl7Q201X alleles were isolated in two EMS screens , Fbxl7W389X was found fortuitously in a separate fly stock , and Fbxl7MI04292 ( BL37813 ) is a MI{MIC} insertion in the first intron of Fbxl7 . All Fbxl7 alleles are on chromosomes bearing a FRT82B insertion . Fbxl7 overexpression stocks used were UAS-FLAG-Fbxl7 ( this study , attP40 and attP2 ) , UAS-FLAG-Fbxl7C616Y ( this study , attP40 ) , and P{XP}CG4221d08178 ( BL19289 ) . Fbxl7 RNAi stocks used were UAS-Fbxl7RNAi ( JF01515 [BL31065] , VDRC108628 ) . All RNAi experiments performed in flies used UAS-Dcr2 , which increases knockdown . The fat61 allele was isolated in an EMS screen for supercompetitor mutations ( T4854I amino acid change ) . Additional stocks used were: FRT82B dcole88 ( Jursnich et al . , 1990 ) , P[acman]-Fat+; P[acman]-FatΔD , P[acman]-FatΔF ( Pan et al . , 2013 ) , Diap1 3 . 5-GFP ( Zhang et al . , 2008 ) , FRT42D fjN7 , FRT40A ftGrv , FRT40A ds38k , UAS-Fat ( Simon , 2004 ) , ykiB5 ( Huang et al . , 2005 ) , Tub-EGFP . ban ( ‘bantam sensor’ , Brennecke et al . , 2003 ) , FRT40A , FRT82B ( Xu and Rubin , 1993 ) , UAS-GFP-cindr-PC ( Johnson et al . , 2008 ) , Dachs-GFP ( Bosveld et al . , 2012 ) . Remaining stocks used were from , or derived from , the Bloomington Stock Center ( Bloomington , IN ) : UAS-dcr2; nub-Gal4 ( BL25754 ) , eyFLP; FRT82B ub-GFP ( BL5580 , BL5188 ) , FRT82B ub-RFPnls ( BL30555 ) hsFLP;; Act>CD2>Gal4 UAS-GFP ( BL26902 , BL4780 ) , FRT82B MARCM ( BL30036 ) , FRT40A MARCM ( BL5192 ) , FRT40A exe1 ( BL44249 ) , FRT82B wtsX1 ( BL44251 ) , FRT82B dco3 ( BL44250 ) , FRT40A dGC13 ( BL28289 ) , UAS-d:v5 ( BL28291 ) , UAS-zyx-ChRFP ( BL28875 ) , UAS-fj:V5 ( BL44252 ) , Df ( 3R ) BSC515 ( BL25019 ) , Df ( 3R ) BSC728 ( BL26580 ) , GFP-CindrCA06686 ( BL50802 ) , act-Gal4 ( BL3954 ) , tub-Gal4 ( BL5138 ) , FRT40A ft8 ( BL44257 ) , hs-Gal4 ( BL1799 ) , UAS-EGFP ( BL6658 ) , UAS-myr-mRFP ( BL7119 ) , UAS-GFP-KDEL ( BL9898 ) , UAS-Galt-GFP ( BL30902 ) , UAS-GFP-myc-2xFYVE ( BL42712 ) , eyFLP; 40A CL white+/CyO ( BL5622 ) , en-Gal4 UAS-RFP ( BL30557 ) . Stocks used for vesicle co-localization are listed in Supplementary file 1 and are listed with BL numbers if available . hsFLP-induced clones were generated by incubating larvae at 37°C at 48 hr after egg deposition ( AED ) . A 30-min incubation was used for experiments using Act>CD2>Gal4 and 2-hr incubation for experiments using MARCM . Immunostainings were performed by dissecting imaginal discs from wandering third instar larvae , fixing discs in 4% paraformaldehyde + PBS , followed by blocking in PBS + 0 . 1% Triton-X + 5% normal goat serum ( NGS ) , incubation with primary antibodies overnight at 4°C , and incubation with secondary antibodies overnight at 4°C . Immunostainings with anti-Fbxl7 antibodies required a separate optimized protocol: Larvae were dissected in 0 . 1 M NaPO4 , fixed in PLP-fixative ( 2% paraformaldehyde , 0 . 01 M NaIO4 , 0 . 075 M lysine , 0 . 037 M NaPO4 ) , washed with 0 . 1 M NaPO4 containing 0 . 1% saponin , blocked with 0 . 1 M NaPO4 containing 0 . 1% saponin and 5% NGS , primary and secondary antibodies were diluted in 0 . 1 M NaPO4 containing 0 . 1% saponin and 5% NGS . Samples were imaged on a Zeiss 700 confocal microscope ( Germany ) . The anti-Fbxl7 antibody was generated by immunizing guinea pigs ( Pocono Farms , Canadensis , PA ) with purified Fbxl7 ( amino acids 22–324 ) produced at the UC-Berkeley MacroLab ( His-Fbxl7 purified on a Nickel column ) , and used at 1:1000 for tissue staining . Other antibodies used: rat anti-Dachs ( 1:500 , Brittle et al . , 2012 ) , rat anti-Fat ( 1:1600 , Feng and Irvine , 2009 ) , rat anti-Dachsous ( 1:5000 , Yang et al . , 2002 ) rat anti-Ecad ( 1:100 , DCAD2 , DHSB , Iowa City , IA ) , mouse anti-FLAG ( 1:1000 , F3165; Sigma , St . Louis , MO ) , rabbit anti-FLAG ( 1:1000 , F7425; Sigma ) mouse anti-V5 ( 1:500 , R960-25; Invitrogen , Carlsbad , CA ) , mouse anti-Arm ( 1:100 , N2 7A1; DHSB ) , rabbit anti-LacZ ( 1:500 , #559762; MP Biomedicals , Santa Ana , CA ) , anti-Cleaved Caspase-3 ( 1:200 , 9661; Cell Signaling , Beverly , MA ) . Actin was visualized with Phalloidin-TRITC ( 1:500 , Sigma ) or Alexa Fluor 633 Phalloidin ( 1:500 , Invitrogen ) . Nuclei were visualized with DAPI ( 1:1000 ) . Plasmids were constructed using conventional ligation-based molecular cloning techniques . Oligonucleotide sequences are listed in a separate table in Supplemental file 2 . Fbxl7 was amplified from clone LD38495 ( DGRC , Bloomington , IN ) by designing oligonucleotides to amplify the single predicted coding sequence CG4221-RA and add Not1 and Xba1 restriction sites . The Not1-Fbxl7-Xba1 PCR fragment was digested and ligated into pUAS-FLAG attB ( adds an N-terminal FLAG tag ) to generate pUAS-FLAG-Fbxl7 attB . The C616Y amino acid change was introduced by site directed mutagenesis , generating pUAS-FLAG-Fbxl7C616Y attB . Transgenic flies were made from pUAS-FLAG-Fbxl7 attB and pUAS-FLAG-Fbxl7C616Y attB using PhiC31 integration ( BestGene , Chino Hills , CA ) , inserting into attP40 and attP2 landing sites . Fbxl7 truncation plasmids were generated by amplifying Fbxl7Δ1 ( 389-772aa ) , Fbxl7Δ2 ( 445-772aa ) , and Fbxl7Δ3 ( 1-388aa ) and ligating into pUAS-FLAG attB using Not1/Xba1 . pUAS-FLAG-EGFP attB was generated by amplifying EGFP from pEGFPattB ( K Basler ) and cloning into pUAS-FLAG attB using In-Fusion ( Clonetech , Mountain View , CA ) . SkpA and Cul1 coding sequence were amplified from genomic DNA and clone LD20253 ( DGRC ) , respectively . Not1/Xba1 sites were added to oligos that amplified SkpA , and Kpn1/Not1 was added for Cul1 . PCR fragments were digested and ligated into pMT-HA ( adds a C-terminal HA tag ) , generating pMT-SkpA-HA and pMT-Cul1-HA . dCul1DN is a C-terminal truncation ( 1-451aa ) which corresponds to 1-452aa of dominant negative human hCul1DN ( Wu et al . , 2000 ) and was cloned into pMT-HA as for full length dCul1 . pMT-FatICD-V5 was generated by amplifying FatICD coding sequence from pUAS-FatICD-V5 ( K . Irvine ) , adding Not1/Xba1 sites with oligos . PCR fragments were digested and ligated into pMT-V5/6xHis ( Invitrogen ) . pMT-FatICDΔD-V5 , pMT-FatICDΔF-V5 , and pMT-FatICDmutV-V5 were generated by using the same oligos to amplify from pUAS-FatICDΔD-V5 , pUAS-FatICDΔF-V5 , and pUAS-FatICDmutV-V5 ( Irvine ) , respectively . pMT-FatICD61-V5 was generated by site directed mutagenesis of pMT-FatICD-V5 to make the change T4854I . pUAS-HA-Ub attB was generated by amplifying Ubi-p5E coding sequence from genomic DNA , adding an N-terminal HA tag with primers , and inserting into pUAS attB ( K Basler ) . pMT-cindr-V5 was generated by amplifying the longest predicted isoform cindr-RC from S2R+ cell cDNA , adding Not1/Xba1 sites , and ligating into pMT-V5/6xHis ( Invitrogen ) . Other plasmids used are pMT-Dco-V5 ( Ko et al . , 2002 ) , pUAS-Dachs-V5 ( Mao et al . , 2006 ) . The sequence of oligos used are in Supplementary file 2 . S2R+ cells were cultured and transfected using conventional techniques . S2R+ cells were cultured in Schneiders medium containing 10% FBS at 27°C , transfected with Effectene ( Qiagen , Germany ) in six-well dishes , and harvested 72 hr later . 500 µM CuSO4 was added 24 hr before harvesting to induce expression from plasmids with metalothionein promoters . For Co-IP and in vivo ubiquitylation assays , 50 µM MG132 ( C2211; Sigma ) was added to transfected cells four hours before harvesting to inhibit proteasome activity . For experiments using dsRNA , S2R+ cells were transfected with dsRNA ± plasmids and were harvested as needed for protocols described above . Unless otherwise stated , wing discs or S2R+ cells were boiled in 1x or 2x SDS Sample buffer , run on 7 . 5% Mini-Protean TGX gels ( Bio-Rad , Hercules , CA ) , and transferred to nitrocellulose membrane . Protein bands were detected with primary antibodies and secondary antibodies conjugated to HRP , and imaged using ECL detection reagent ( RPN2232; Amersham , UK ) . For co-IP assays , 50 µM MG132 ( C2211; Sigma ) was added to transfected cells 4 hr before harvesting to inhibit proteasome activity . Cells were washed once with ice cold PBS , and lysed in lysis buffer ( 20 mM HEPES 7 . 5 pH , 5 mM KCl , 1 mM MgCl2 , 0 . 1% NP-40 , ‘Complete’ EDTA free protease inhibitor tablet [Roche , Switzerland] ) . Insoluble material and nuclei were removed by centrifugation at 13 , 000×g for 30 min at 4°C , and soluble cell lysate was incubated with anti-FLAG M2 affinity gel ( A2220; Sigma ) overnight at 4°C . Beads were washed twice in lysis buffer and denatured by boiling in SDS sample buffer for 10 min . For SkpA , Cul1 , and Cindr co-IP assays , to avoid detection of non-specific binding of transfected proteins to beads , FLAG-protein complexes were eluted off beads by incubating with 400 ng/µl 3x FLAG peptide ( F4799; Sigma ) for 30 min at 4°C . For in vivo ubiquitylation assays , 50 µM MG132 was added to transfected cells 4 hr before harvesting . Cells were washed once with ice cold PBS , and proteins denatured by boiling in 100 µl 1% SDS in PBS for 10 min 400 µl of 0 . 5% BSA , 1%Triton-X , in PBS was added , and samples were sonicated , then centrifuged at 13 , 000×g for 10 min . Supernatant was diluted to 1 ml with 5% BSA , 1%Triton-X and incubated with anti-HA agarose beads ( A2095; Sigma ) overnight at 4°C . Beads were washed twice with 1%Triton-X in PBS and boiled in SDS sample buffer for 10 min . For in vivo ubiquitylation of Dachs-V5 from larval tissue , 12 hr before dissection larvae were heat-shocked at 37°C for 1 hr to induce UAS transgenes by hs-Gal4 . 30 brain-eye-antennal complexes per genotype were dissected in Schneiders medium and incubated with 50 µM MG132 for 4 hr . Complexes were boiled , diluted , sonicated , and centrifuged as above . Supernatant was diluted to 1 ml with 5% BSA , 1%Triton-X and incubated with Protein G Sepharose ( P3296; Sigma ) for 1 hr at 4°C , replaced with Protein G Sepharose plus 1 µl mouse anti-V5 antibody ( R960-25; Invitrogen ) and incubated overnight at 4°C . Beads were washed twice with 1%Triton-X in PBS and boiled in SDS sample buffer for 10 min . For experiments using dsRNA , S2R+ cells were transfected with dsRNA ± plasmids and were harvested as needed for protocols described above . dsRNA was generated by PCR amplifying DRSC15513 and DRSC38270 from genomic DNA , and GFP coding sequence from pattB-EGFP ( K Basler ) , adding T7 sequence to forward and reverse primers , and in vitro transcribing dsRNA ( AM1333; Megascript T7 Transcription Kit , Invitrogen ) . For anti-Fat western blots from wing discs , 20 wing discs were dissected from third instar larvae in PBS and immediately boiled in 2x SDS Sample buffer . The amount loaded on gels was adjusted to load equivalent amounts of protein . For anti-Dachs westerns from wing discs , 20 wing discs were dissected from third instar larvae in PBS and lysed in 1x RIPA buffer . Total protein was quantified ( Micro BCA kit , 23235; Fisher , Hampton , NH ) and adjusted equally among samples . Secondary antibodies conjugated to LiCor fluorescent dyes were used to detect protein bands using a LiCor Odessey imager ( Lincoln , NE ) . Western blots were probed with the following antibodies: Guinea pig anti-Fbxl7 ( 1:1000 ) , rat anti-Fat ( 1:25 , 000 , K Irvine ) , rat anti-Dachs ( 1:5 , 000 , D Strutt ) , mouse anti-Tubulin ( 1:100 , E7; DHSB ) , mouse anti-FLAG ( 1:10 , 000 , F3165; Sigma ) , mouse anti-V5 ( 1:5 , 000; R960-25 , Invitrogen ) , rabbit anti-V5 ( 1:5000 , V8137; Sigma ) rabbit anti-HA ( 1:1 , 000 , 3724; Cell Signaling ) , rabbit anti-Ub ( 1:1 , 000 , Z0458; DakoCytomation , Carpinteria , CA ) , mouse anti-Lamin ( 1:100 , ADL67 . 10; DHSB ) , goat anti-rat-HRP ( 112-035-003; Jackson , West Grove , PA ) , goat anti-rabbit-HRP ( 111-035-003; Jackson ) , goat anti-mouse-HRP ( 172-1011; BioRad ) , goat anti-guinea pig-HRP ( 106-035-003; Jackson ) , goat anti-rat-IR680 ( 926-68 , 076; Licor ) , goat anti-mouse-IR800 ( 827-08 , 364; Licor ) . Wings or legs were mounted onto slides using Canadian Balsam medium ( Gary's Magic Mount ) and imaged on a Leica transmitted light microscope ( TL RCI , Germany ) . Wing area and cross vein distance was quantified in ImageJ . For cross veins , we measured the distance of a straight line drawn from intersection of the anterior cross vein and L4 to the intersection of the posterior cross vein and L4 . Statistical significance between groups was determined by one-way ANOVA using ( Tukey's or Dunnett's test ) . Quantifications were performed as in Brittle et al . ( 2012 ) using ImageJ . Wing discs were immunostained for Dachs and F-actin and imaged under identical settings at 20x to determine P-D orientation , and at 63x to image the dorsal portion of the wing pouch where Dachs asymmetry is highest . Images were rotated so that the P-D axis of the wing pouch oriented vertically ( 90° and 270° ) . A cropped 24 . 8 × 24 . 8um ( 500 × 500px ) square was used to quantify the mean fluorescence intensity of Dachs or actin along each cell edge while recording the angle of the cell edge relative to the P-D orientation . Cell edge data were measured using a 1 pixel width line . Mean fluorescence of cell edges oriented in the P-D orientation ( 45°–135° ) or the A-P orientation ( 0°–45°; 135°–180° ) was isolated into two different lists , which were each averaged . The ratio of mean fluorescence of the A-P orientation to P-D orientation gives the P-D/A-P localization . For example , asymmetric localization to the P-D sides of cells will give higher mean intensities on cell edges in the A-P orientation . Quantifications were performed on eight cropped boxes from different discs for each group . Statistical significance between groups was determined by one-way ANOVA using ( Tukey's test ) .	Multi-cellular organisms are made up of cells that are organized into tissues and organs that reach a predictable size and shape at the end of their development . To do this , cells must be able to sense their position and orientation within the body and know when to stop growing . Epithelial cells—which make up the outer surface of an animal's body and line the cavities of its internal organs—connect to each other to form flat sheets . These sheets of cells contain structures that are oriented along the plane of the sheet . However , how this so-called ‘planar cell polarity’ coordinates with cell growth in order to build complex tissues and organs remains to be discovered . A protein called Fat is a major player in both planar cell polarity and the Hippo signaling pathway , which controls cell growth . As such , the Fat protein appears to be crucial for controlling the size and shape of organs . Mutations in the Fat protein cause massive tissue overgrowth , prevent planar cell polarity being established correctly , and stop the legs and wings of fruit flies developing normally . The Fat protein also plays a role in distributing another protein called Dachs—which is also part of the Hippo signaling pathway . In epithelial cells of the developing wing , Dachs is mostly located on the side of the cell that is closest to the tip of the developing wing ( the so-called ‘distal surface’ ) . How Fat and Dachs work together is not understood , but it is known that they do not bind to each other directly . Now , Bosch et al . show that in the fruit fly Drosophila , the Fat protein binds to another protein called Fbxl7 . Flies that cannot produce working Fbxl7 have defects in some aspects of planar cell polarity and a modest increase in tissue growth . Fbxl7 seems to account for part , but not all , of the ability of Fat to restrict tissue growth . Furthermore , a lack of the Fbxl7 protein results in a spreading of Dachs protein across the apical surface—which faces out of the epithelial sheet—of epithelial cells . On the other hand , if Fbxl7 is over-expressed , Dachs is driven to the interior of each cell . Hence , a normal level of Fbxl7 protein restricts the Dachs protein to the correct parts of the cell surface . Together , the findings of Bosch et al . show that the Fbxl7 protein is a key link between the Fat and Dachs proteins . These results also provide an understanding of how growth and planar cell polarity—two processes that are essential for normal development of all multi-cellular organisms—are coordinated .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2014	The Drosophila F-box protein Fbxl7 binds to the protocadherin Fat and regulates Dachs localization and Hippo signaling
Mutations in the MECP2 gene cause the neurodevelopmental disorder Rett syndrome ( RTT ) . Previous studies have shown that altered MeCP2 levels result in aberrant neurite outgrowth and glutamatergic synapse formation . However , causal molecular mechanisms are not well understood since MeCP2 is known to regulate transcription of a wide range of target genes . Here , we describe a key role for a constitutive BDNF feed forward signaling pathway in regulating synaptic response , general growth and differentiation of glutamatergic neurons . Chronic block of TrkB receptors mimics the MeCP2 deficiency in wildtype glutamatergic neurons , while re-expression of BDNF quantitatively rescues MeCP2 deficiency . We show that BDNF acts cell autonomous and autocrine , as wildtype neurons are not capable of rescuing growth deficits in neighboring MeCP2 deficient neurons in vitro and in vivo . These findings are relevant for understanding RTT pathophysiology , wherein wildtype and mutant neurons are intermixed throughout the nervous system . Rett syndrome ( RTT ) is a severe progressive neurodevelopmental disorder , mainly caused by mutations in the X-linked gene encoding methyl-CpG binding protein 2 ( MeCP2 ) , a protein involved in transcriptional regulation ( Amir et al . , 1999; Wan et al . , 1999; Xiang et al . , 2000 ) . RTT patients show regression of head growth followed by various neurological symptoms including seizures , mental retardation , stereotypic hand-wringing movements , breathing irregularities , ataxia and autistic behavior ( Rett , 1966 ) . Mouse models with Mecp2 mutations display similar neurological phenotypes , and have been quite critical in defining the pathophysiology of RTT . Mecp2Null/y mice grow normally until 4–6 weeks of age , after which they display RTT-like symptoms such as reduced mobility , hindlimb clasping , abnormal breathing patterns and premature death ( Chen et al . , 2001; Guy et al . , 2001 ) . Similarly , mice engineered to express twice the endogenous levels of MeCP2 ( Mecp2Tg1 ) are characterized by seizures , forepaw clasping , hypoactivity , increased aggression , and around 30% die by one year of age ( Collins et al . , 2004; Jugloff et al . , 2008; Luikenhuis et al . , 2004 ) . Loss or doubling of MeCP2 in primary mouse hippocampal neurons results in reduction or enhancement of synaptic response respectively , primarily due to the number of glutamatergic synapses formed ( Chao et al . , 2007 ) . Furthermore , restoring MeCP2 levels rescued neurological defects associated with loss of MeCP2 the extent of which depends on timing of activation and dynamic variation in MeCP2 levels ( Giacometti et al . , 2007; Guy et al . , 2007 ) . Besides MeCP2 , manipulation of target genes regulated by MeCP2 by genetic means or by indirect and nonspecific mechanisms has resulted in alleviating various features of RTT ( Deogracias et al . , 2012; Johnson et al . , 2012; Kondo et al . , 2008; Kron et al . , 2014; Ogier et al . , 2007; Schmid et al . , 2012; Tropea et al . , 2009 ) . Functional interaction between MeCP2 and brain-derived neurotrophic factor ( BDNF ) has been reported in various studies and MeCP2 has been known to regulate expression of BDNF ( Chahrour et al . , 2008; Chang et al . , 2006; Martinowich et al . , 2003 ) . Bdnf deletion from postnatal forebrain excitatory neurons of MeCP2 mutant mice resulted in an earlier onset of RTT while conditional BDNF overexpression delayed RTT onset and normalized spontaneous activity ( Chang et al . , 2006 ) . Similarly , endogenous MeCP2 knockdown reduced dendritic length in E18 hippocampal neurons , which was fully rescued upon BDNF overexpression ( Larimore et al . , 2009 ) . Considering that BDNF is an important regulator of neurite outgrowth and synapse formation ( Cheng et al . , 2011; Finsterwald et al . , 2010; Gottmann et al . , 2009; McAllister et al . , 1999; Park and Poo , 2013; Poo , 2001; Tolwani et al . , 2002; Wang et al . , 2015 ) , it is essential to understand effects of BDNF modulation in Mecp2Null/y neurons specifically in steering neuronal growth and glutamatergic synapses . Early work has focused mainly on phenotypic rescue upon increasing BDNF levels in MeCP2-deficient mice but mechanistic interactions have not been resolved . In this study , we examine putative causal mechanisms of BDNF impacting cellular growth as well as the formation of excitatory synapses in hippocampal neurons lacking MeCP2 . We found that increased BDNF levels restored synaptic output and morphological phenotypes in MeCP2 deficient neurons in a cell autonomous and autocrine manner in vitro and in vivo . Importantly , blocking the BDNF pathway converted wildtype neurons to a phenotype that mimicked MeCP2-deficiency induced defects . These findings add to the current picture of BDNF signaling in stabilizing various features of RTT patients and are crucial for the profound understanding of BDNF-mediated therapeutic strategies and more generally , pathophysiology of the RTT disease . We previously reported that loss or doubling of MeCP2 results in altered glutamatergic synapse number in vitro and in vivo ( Chao et al . , 2007 ) underlining tightly controlled MeCP2 dosage as a prerequisite for optimal synapse formation . Why MeCP2 levels are rate limiting in glutamatergic synapse formation is not clear , and could include pre- or postsynaptic factors that regulate synapse formation as well as factors that regulate general neuronal development . Hence , we began this analysis by examining the role of neurite outgrowth . Single-cell hippocampal autaptic cultures were utilized and particularly beneficial due to the unambiguous assignment of axon and dendrite and the ability to restrict analysis to a cell autonomous condition . Axonal and dendritic outgrowths were measured during the first 12 days in vitro ( DIV ) from Mecp2Tg1 and Mecp2Null/yneurons as well as control neurons derived from their respective WT littermates ( Figure 1 ) . Neurons were labeled with Tau1 and MAP2 to mark for axons and dendrites respectively ( Figure 1A ) . Comparative analysis revealed that Mecp2Tg1 neurons appeared normal and did not display any morphological deficits , with axonal and dendritic outgrowth similar to WT neurons ( Figure 1C and E ) . However , Mecp2Null/y neurons showed a significant reduction of 23–41% and 27–41% in overall length of axons and dendrites respectively ( Figure 1B and D ) . Considering these deficits in general growth and differentiation in Mecp2Null/y neurons , we analyzed the size of neuronal somata and observed a 24–33% reduction in soma size while Mecp2Tg1 neuronal soma remained unchanged ( Figure 1F–H ) . These results do not explain the Tg1 gain-of-function phenotype suggesting that an alternate mechanism may be necessary for enhanced synapse number . Additionally , the morphological findings reveal a general growth deficiency in glutamatergic neurons lacking MeCP2 possibly accounting for decreased synapse number . 10 . 7554/eLife . 19374 . 003Figure 1 . Loss of MeCP2 alters neurite outgrowth and neuronal soma size . ( A ) Co-immunostaining of glutamatergic hippocampal neurons from WT and Mecp2Null/y mice at DIV 2 , 4 , 8 and 12 for Tau1 ( cyan ) and MAP2 ( green ) . Scale bars represent 20 µm . ( B–E ) Mean axonal length and dendritic length measured at different time points for Mecp2Null/y ( B , D ) and Mecp2Tg1 neurons ( C , E ) . ( F ) Somatic labeling of glutamatergic hippocampal neurons from WT and Mecp2Null/y mice at DIV 2 , 4 , 8 and 12 with MAP2 . Scale bar represents 20 µm . ( G , H ) Mean soma area measured at different time points for Mecp2Null/y ( G ) and Mecp2Tg1 neurons ( H ) . Data shown as mean ± SEM . p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 003 MeCP2 is a transcriptional regulator that activates or represses expression of several downstream genes based on cell type , origin , age and heterogeneity of brain region ( Chahrour et al . , 2008; Tudor et al . , 2002 ) . Given that BDNF ( i ) is a consistent neuronal target gene of MeCP2 ( Chahrour et al . , 2008; Martinowich et al . , 2003 ) , ( ii ) is an essential regulator of synapse formation and dendritic complexity ( Finsterwald et al . , 2010; McAllister et al . , 1999; Tolwani et al . , 2002 ) , and that ( iii ) levels are reduced in MeCP2 knockout mice ( Chang et al . , 2006 ) ; we investigated the mechanistic role of BDNF signaling in regulating synaptic output and synapse formation in Mecp2Null/y neurons at single-cell level . To address this question , we utilized the lentiviral system driven by a synapsin promoter to overexpress BDNF in Mecp2Null/y and WT neurons to attempt rescue of growth deficits ( Figure 2A ) . WT and Mecp2Null/y neurons expressing GFP were used as control . Indeed , BDNF overexpression in Mecp2Null/y neurons fully restored all morphological parameters analyzed . Particularly , glutamatergic synapse number , dendrite length and soma size in Mecp2Null/yneurons were reduced down to 51 , 58 and 72% of WT and rescued up to 105 , 89 and 102% of that of WT glutamatergic neurons respectively upon BDNF overexpression ( Figure 2B–E ) . Similarly , nucleus size was reduced by 26% in Mecp2Null/yneurons and rescued up to 95% in BDNF overexpressing Mecp2Null/y neurons ( Figure 2—figure supplement 1 ) . Overexpressing BDNF in WT neurons itself did not impact synapse number , dendrite length or soma size ( Figure 2L ) . Further , lack of enhancement in synapse number upon BDNF overexpression in WT neurons verified that the Mecp2Tg1 gain-of-function phenotype was not related to BDNF function and ascribed to potentially separate mechanism ( s ) . 10 . 7554/eLife . 19374 . 004Figure 2 . BDNF overexpression in Mecp2Null/y autaptic neurons normalizes cell morphology and restores synaptic output . ( A ) Experimental scheme of lentivirus-mediated BDNF overexpression in Mecp2Null/yneurons , showing neuronal membrane ( black ) , BDNF ( red ) and TrkB receptor ( blue ) . ( B ) Representative images of neuronal morphology under the following conditions ( from left to right ) : WT , Null , Null + BDNF , Null + BDNF + ANA12 , WT + ANA12 . Dendritic outgrowth ( top ) and glutamatergic synapses ( bottom ) indicated by MAP2 ( green ) and VGLUT1 ( red ) labeling . Scale bars represent 20 µm . ( C–E ) Bar graphs show mean dendrite length ( C ) , glutamatergic synapse number ( D ) and neuronal soma area ( E ) . ( F ) Representative traces of evoked EPSCs recorded from autapses under the following conditions: WT , Null , Null + BDNF , Null + BDNF + ANA12 , WT + ANA12 , Null + ANA12 . ( G ) Bar graph shows mean evoked EPSC amplitude . ( H ) Representative traces of average postsynaptic response to 5 s application of 500 mM sucrose . Experimental groups same as ( F ) . ( J–K ) Bar graphs show mean RRP size ( I ) , vesicular release probability Pvr ( J ) and paired-pulse ratio with 25 ms inter-stimulus interval ( K ) . ( L ) Bar graph shows EPSC amplitude , RRP size and Pvr , PPR and dendrite length , glutamatergic synapse number and soma area , normalized to WT ( dashed red line ) . Number of neurons ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . p<0 . 05; p<0 . 01; p<0 . 001; ns: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 00410 . 7554/eLife . 19374 . 005Figure 2—figure supplement 1 . BDNF overexpression in Mecp2Null/y autaptic neurons normalizes nuclei size . Bar graph shows mean neuronal nuclei size . Number of neurons ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . p<0 . 05; p<0 . 01; p<0 . 001; ns: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 005 We next examined functional implications of the observed morphological defects by measuring synaptic output in Mecp2Null/y neurons overexpressing BDNF . Mecp2Null/y neurons revealed 54% decrease in evoked EPSC amplitude ( 2 . 92 ± 0 . 42 nA , p<0 . 05 ) and 58% decrease in RRP charge ( 0 . 37 ± 0 . 05 nC , p<0 . 01 ) as compared to WT neurons ( EPSC: 6 . 38 ± 0 . 9 nA; RRP: 0 . 89 ± 0 . 14 nC ) , both of which were rescued back up to 113 and 87% of WT levels , respectively ( EPSC: 7 . 23 ± 0 . 97 nA , p<0 . 01; RRP: 0 . 77 ± 0 . 09 nC , p<0 . 01 ) ( Figure 2F–I ) . We did not observe any differences in short-term plasticity or release efficiency in either WT or Mecp2Null/y neurons with or without BDNF overexpression , as reflected by the paired-pulse ratio ( PPR ) and vesicular release probability ( Pvr ) measurements ( Figure 2J–L ) . These findings further illustrate that interaction of MeCP2 and BDNF was predominantly related to scaling of neuronal size , growth and synaptic response . Overall , synaptic and morphological deficits resulting from loss of MeCP2 function were fully re-established by restoring the ability of Mecp2Null/y neurons to synthesize increased levels of BDNF . We now asked if disruption of BDNF binding to TrkB could negate effects of BDNF overexpression in Mecp2Null/y neurons . The low molecular weight TrkB ligand , ANA-12 , selectively binds to the TrkB receptor thereby blocking BDNF-induced TrkB activation and inhibiting intracellular signaling cascades downstream of TrkB ( Cazorla et al . , 2011 ) . Application of ANA-12 to the BDNF overexpressing Mecp2Null/y neurons reverted both synaptic output ( evoked EPSC amplitude: 2 . 83 ± 0 . 47 nA , p<0 . 001 and RRP size: 0 . 19 ± 0 . 02 nC , p<0 . 001 ) ( Figure 2G and I ) and morphological phenotypes ( synapse number: 151 ± 10 , p<0 . 01; dendrite length: 748 ± 46 µm , p<0 . 05; soma size: 93 ± 4 µm2 , p<0 . 05 and nucleus size: 45 ± 2 µm2 , p<0 . 01 ) ( Figure 2C–E and Figure 2—figure supplement 1 ) back to Mecp2Null/y control levels . Control groups included WT and Mecp2Null/y neurons both with and without application of ANA-12 . Interestingly , Mecp2Null/y neurons treated with ANA-12 did not undergo any further loss-of-function effect and remained comparable to Mecp2Null/y control neurons ( Figure 2G and I ) indicating that the BDNF-TrkB canonical pathway was quantitatively disrupted in MeCP2-deficient neurons . On the other hand , WT neurons treated with ANA-12 displayed morphological as well as synaptic deficits and behaved similar to Mecp2Null/y neurons ( Figure 2G and I ) . As before , PPR and Pvr remained unchanged in all conditions ( Figure 2J–L ) . These results strongly suggest that BDNF synthesis and an active BDNF-TrkB pathway are essential for normal neuronal growth in WT as well as RTT-like hippocampal glutamatergic neurons . Consistent with these results , we show that treatment of BDNF overexpressing Mecp2Null/y neurons with an anti-BDNF neutralizing antibody ( Figure 3A ) decreased EPSC amplitude and RRP size by 50 and 53% respectively ( Figure 3B–E ) , thereby negating phenotype rescue seen via BDNF overexpression . Thus , we reveal that BDNF synthesis was clearly impaired in glutamatergic neurons lacking MeCP2 . To further test the specificity of phenotype rescue , an alternate neurotrophic factor , nerve growth factor ( NGF ) , was overexpressed in WT and Mecp2Null/y neurons ( Figure 3—figure supplement 1A ) . We found that NGF failed to restore normal synaptic transmission , emphasizing the specificity of the role of BDNF-TrkB in Mecp2Null/y neurons ( Figure 3—figure supplement 1B and C ) . PPR and Pvr measurements remained unchanged across all conditions in both BDNF neutralization ( Figure 3F and G ) and NGF overexpression experiments ( Figure 3—figure supplement 1D and E ) . These findings show that abolishing BDNF function as well as BDNF-induced TrkB activation affected synapse formation in Mecp2Null/y autaptic neurons , postulating the mechanistic role of BDNF in regulating synaptic function in RTT-like excitatory neurons . 10 . 7554/eLife . 19374 . 006Figure 3 . BDNF neutralization fails to rescue physiological phenotypes while exogenous application of BDNF restores physiological and morphological phenotypes , reaffirming specificity of BDNF-TrkB interaction in Mecp2Null/y glutamatergic neurons . ( A ) Experimental scheme of lentivirus-mediated BDNF overexpression and BDNF neutralization in Mecp2Null/yneurons , showing neuronal membrane ( black ) , BDNF ( red ) , TrkB receptor ( blue ) and BDNF neutralizing antibody ( cyan ) . ( B ) Representative traces of evoked EPSCs recorded from autapses under the following conditions: WT , Null , Null + BDNF , Null + BDNF + neutralization . ( C ) Bar graph shows mean evoked EPSC amplitude . ( D ) Representative traces of average current response to 5 s application of 500 mM sucrose . Experimental groups same as ( B ) . ( E–G ) Bar graphs show mean RRP size ( E ) , Pvr ( F ) and 25 ms ISI – PPR ( G ) . ( H ) Experimental scheme depicting exogenous application of BDNF in Mecp2Null/y neurons , showing neuronal membrane ( black ) , BDNF ( red ) and TrkB receptor ( blue ) . ( I–M ) Bar graphs show mean evoked EPSC amplitude ( I ) , RRP size ( J ) , dendrite length ( K ) , glutamatergic synapse number ( L ) and neuronal soma area ( M ) . Number of neurons ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . p<0 . 05; p<0 . 01; p<0 . 001; ns: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 00610 . 7554/eLife . 19374 . 007Figure 3—figure supplement 1 . NGF overexpression in Mecp2Null/y neurons does not rescue glutamatergic synaptic output . ( A ) Experimental scheme of lentivirus-mediated NGF overexpression in Mecp2Null/y neurons , showing neuronal membrane ( black ) , BDNF ( red ) , TrkB receptor ( blue ) , NGF ( dark blue ) and TrkA receptor ( yellow ) . ( B–E ) Bar graphs show mean evoked EPSC amplitude ( B ) , RRP size ( C ) , Pvr ( D ) and 25 ms ISI – PPR ( E ) . Number of neurons ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . p<0 . 05; *p<0 . 01; ns: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 00710 . 7554/eLife . 19374 . 008Figure 3—figure supplement 2 . Exogenous application of BDNF does not alter release efficiency or short-term plasticity in Mecp2Null/y hippocampal neurons . ( A , B ) Bar graphs show mean Pvr ( A ) and 25 ms ISI – PPR ( B ) . Number of neurons ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 008 If BDNF synthesis were indeed disrupted and led to general growth and synaptic deficits in Mecp2Null/y neurons , exogenous application of BDNF should be able to bypass this deficiency and rescue loss-of-function . Neurons were treated with 50 ng/ml of recombinant human BDNF at DIV 2 and replenished every 2–3 days until DIV 14 when they were either taken for EPSC and RRP measurements or fixed for morphology analysis ( Figure 3H ) . Exogenous BDNF application normalized evoked EPSC amplitude ( Figure 3I ) and RRP size ( Figure 3J ) as well as re-established normal dendritic outgrowth ( Figure 3K ) and glutamatergic synapse number ( Figure 3L ) in Mecp2Null/y neurons . All restored phenotypes were specific to BDNF-TrkB since application of ANA-12 to Mecp2Null/y neurons treated with exogenous BDNF reverted evoked EPSC amplitude ( 44% decrease ) and RRP size ( 66% decrease ) as well as synapse number ( 39% decrease ) , dendrite length ( 47% decrease ) and soma size ( 28% decrease ) back to Mecp2Null/y levels ( Figure 3I–M ) . No significant changes were seen either in PPR or Pvr measurements upon BDNF application ( Figure 3—figure supplement 2A and B ) . These findings confirmed that Mecp2Null/y neurons were deficient in BDNF synthesis and revealed disrupted BDNF-TrkB signaling , which was bypassed upon exogenous application of BDNF . To verify if determining glutamatergic synapse number by estimating presynaptic VGLUT1+ puncta was correlated to identifying both pre- and postsynaptic colocalized puncta , we labeled WT and Mecp2Null/y neurons with Homer1 , VGLUT1 and MAP2 under different experimental conditions ( Figure 4A ) . First , we evaluated the density of VGLUT1 and Homer1 synaptic markers and found that MeCP2-deficient neurons displayed significant reduction in density of both markers , which were rescued upon BDNF overexpression and exogenous BDNF application ( Figure 4B and C ) . We also observed that VGLUT1 and Homer1 densities were reverted back to Mecp2Null/y levels upon BDNF neutralization ( Figure 4B and C ) . Next , we assessed the density of functional synapses by estimating the density of VGLUT1-Homer1 colocalized synaptic puncta . We found 41% decrease in VGLUT1-Homer1 puncta density in Mecp2Null/y neurons , which was restored up to 93 and 87% upon BDNF overexpression and exogenous BDNF application , respectively ( Figure 4D ) . Additionally , the rate of colocalization of VGLUT1 with Homer1 was significantly reduced upon loss of MeCP2 ( 20% decrease ) and restored up to 99 and 95% specifically via BDNF overexpression and exogenous BDNF , respectively ( Figure 4E ) . Finally , we also examined the expression levels of VGLUT1 and Homer1 and found that intensities of both markers remained unchanged across all experimental conditions ( Figure 4—figure supplement 1A–C ) . 10 . 7554/eLife . 19374 . 009Figure 4 . BDNF overexpression in Mecp2Null/y glutamatergic neurons restores synapse density . ( A ) Representative images of WT and Mecp2Null/y neurons labeled with MAP2 , VGLUT1 and Homer1 under the following conditions ( from top to bottom ) : WT , Null , Null + BDNF . Scale bar represents 3 µm . ( B–E ) Bar graphs show mean normalized VGLUT1 synapse density ( B ) , Homer1 synapse density ( C ) , colocalized VGLUT1-Homer1 synapse density ( D ) , and normalized fraction of VGLUT1 puncta colocalized to Homer1 ( E ) . Number of neurons ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 00910 . 7554/eLife . 19374 . 010Figure 4—figure supplement 1 . BDNF overexpression in Mecp2Null/y glutamatergic neurons does not alter expression levels of synaptic markers . ( A–C ) Bar graphs show mean normalized VGLUT1 fluorescence intensities in autaptic ( A ) and continental WT and Mecp2Null/y neurons ( B ) , and Homer1 fluorescence intensities ( C ) . Number of synaptic puncta ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 010 We have observed so far that the loss-of-function phenotype seen in Mecp2Null/y neurons is reminiscent of that of conditional BDNF knockout mice ( Chang et al . , 2006 ) , and that physiological and morphological deficits are normalized upon BDNF overexpression as well as exogenous application , at single-cell level . If BDNF deficiency is responsible for the synaptic and growth deficits in Mecp2Null/y glutamatergic neurons , does paracrine signaling play a role in ameliorating loss of synapse number and function ? This is particularly relevant for RTT patients wherein neighboring WT neurons may be impaired in their ability to restore loss of synapse number in Mecp2Null/y neurons despite mosaic expression of WT and mutant MeCP2 . In order to address this question , we designed an in vitro RTT model to examine putative paracrine BDNF activity . We co-cultured WT and Mecp2Null/y hippocampal neurons and labeled neurons and corresponding synapses . For this purpose , we pre-incubated neurons of either genotype with distinct lentiviral constructs expressing Synaptophysin-GFP tagged to nucleus-localized RFP ( WT ) and Synaptophysin-mKate tagged to nucleus-localized GFP ( Mecp2Null/y ) to identify origin of synapses ( Figure 5A-left panel ) . Co-labeling for VGLUT1 puncta and MAP2 enabled identifying glutamatergic synapses and their dendritic localization respectively , and synapse density was analyzed by estimating the number of colocalized Syp+ VGLUT1+ puncta . ( Figure 5B ) . We found that Mecp2Null/y neurons showed 34% decrease in number of glutamatergic synapses ( Figure 5B-left panel and C ) , which was also consistent with the data from single cells ( Figures 2 and 3 ) and in vivo ( Chao et al . , 2007 ) . 10 . 7554/eLife . 19374 . 011Figure 5 . TrkB activation specifically in Mecp2Null/y neurons in an in vitro RTT model normalizes glutamatergic synapse number in a cell autonomous and autocrine manner . ( A ) Left panel: Neuronal pair depicting lentivirus-mediated labeling of nucleus ( NLS ) and synapses of WT ( NLS: red , Synaptophysin: green ) and Mecp2Null/y neurons ( NLS: green , Synaptophysin: red ) , co-stained for MAP2 to identify synapses localized on dendrites . Right panel: Experimental scheme of lentivirus-mediated BDNF overexpression in WT or Mecp2Null/y neurons , showing neuronal membrane ( black ) , BDNF ( red ) and TrkB receptor ( blue ) . ( B ) Representative images of co-cultured WT and Mecp2Null/y neurons labeled with MAP2 under the following conditions ( from left to right ) : WT/Null; WT+BDNF/Null and WT/Null+BDNF . Bottom panel shows WT ( green ) and Null ( red ) Synaptophysin+ synapses localized on a single dendrite . Scale bars on top and middle panel represent 20 µm . Scale bar on bottom panel represents 3 µm . ( C , D ) Bar graphs show mean glutamatergic synapse density ( C ) and nucleus size ( D ) for all co-cultured groups . Number of neurons ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . p<0 . 01; p<0 . 001; p<0 . 0001; ns: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 011 We next examined the impact of BDNF overexpression in Mecp2Null/y neurons to probe their ability to restore synaptic deficit ( Figure 5A-right panel ) . To achieve BDNF overexpression , we pre-incubated WT and Mecp2Null/y neurons with distinct lentiviral constructs , expressing both BDNF and Synaptophysin-GFP ( WT ) or Synaptophysin-mKate ( Mecp2Null/y ) tagged to nucleus-localized RFP or GFP respectively , via P2A and T2A self-cleaving peptides ( Refer Materials and methods ) . Overexpressing BDNF specifically in WT neurons did not have an impact on number of WT synapses or those formed by proximate Mecp2Null/y control neurons with the latter still displaying a 35% decrease in synapse number ( Figure 5B-middle panel and C ) . Only overexpression of BDNF specifically in Mecp2Null/y neurons when co-cultured with WT neurons enhanced number of Mecp2Null/y synapses up to 99% of WT levels ( Figure 5B-right panel and C ) ; reaffirming that increased BDNF levels specifically in RTT-like neurons enabled rescue of synaptic phenotype . We then asked if BDNF overexpression also impacted nuclei size in Mecp2Null/y neurons . Mecp2Null/y neuronal nuclei were 23% and 27% smaller in size when co-cultured with WT and BDNF overexpressing WT glutamatergic neurons , respectively ( Figure 5D ) . However , BDNF overexpression specifically in Mecp2Null/y neurons restored nucleus size up to 96% of that of co-cultured WT neurons ( Figure 5D ) . These findings led us to two major conclusions . ( i ) Decreased synapse number and nucleus size in Mecp2Null/y neurons per se indicate a cell autonomous role for MeCP2 in regulating normal growth in glutamatergic neurons ( Belichenko et al . , 2009; Blackman et al . , 2012; Kishi and Macklis , 2010 ) . ( ii ) Restoration of both glutamatergic synapse number and nucleus size upon BDNF overexpression specifically in Mecp2Null/y neurons is best explained by an autocrine and highly focal effect for BDNF in regulating glutamatergic synapse formation and neuronal size . Given that restoration of autocrine BDNF secretion in itself is sufficient for normal growth of Mecp2Null/y neurons and that neighboring WT neurons are unable to support TrkB signaling or normalize synapse number , it is still possible to exogenously activate TrkB receptors on Mecp2Null/y neurons and restore glutamatergic synapse number . To test this , we repeated the above experiment in the presence of a TrkB agonist 7 , 8-dihydroxyflavone ( 7 , 8-DHF ) ( Figure 6A ) that binds to its extracellular domain and activates TrkB-mediated downstream signaling ( Jang et al . , 2010 ) . Intriguingly , application of 500 nM 7 , 8-DHF at DIV 6 , 9 and 12 equalized synapse number in all experimental groups . Mecp2Null/y synapses were restored back up to 94% and 97% of WT numbers when co-cultured with either WT neurons or WT neurons overexpressing BDNF , respectively ( Figure 6B ) . As found in the single-cell system , 7 , 8-DHF did not have an effect on WT neurons overexpressing BDNF . Similarly , Mecp2Null/y synapses already overexpressing BDNF remained unaltered indicating that sufficient TrkB stimulation was already prevalent in these neurons ( Figure 6B ) . These data proved that 7 , 8-DHF was able to bypass BDNF synthesis deficit and exogenously trigger TrkB activation in Mecp2Null/y neurons and augment synapse formation in a cell autonomous manner . 10 . 7554/eLife . 19374 . 012Figure 6 . Cell autonomous BDNF-TrkB signaling regulates glutamatergic synapse number by functioning as a presynaptic rate-limiting factor . ( A ) Experimental scheme of lentivirus-mediated BDNF overexpression and exogenous application of TrkB agonist in WT or Mecp2Null/y neurons , showing neuronal membrane ( black ) , BDNF ( red ) , TrkB receptor ( blue ) and TrkB agonist ( green ) . ( B ) Bar graph shows glutamatergic synapse density for all co-cultured groups upon application of TrkB agonist , 7 , 8-DHF . ( C ) Example two-neuron scheme illustrating pre- and postsynaptic neurons ( WT or Mecp2Null/y ) forming synapses onto itself or onto the partner neuron , in a mixed WT/Null co-culture system . ( D ) Bar graphs show glutamatergic synapse density measured from a postsynaptic WT ( clear bars ) and Null ( filled bars ) neuron for all experimental groups . Number of neurons ( n ) shown in the bars for all graphs . Data shown as mean ± SEM . p<0 . 001; ns: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 01210 . 7554/eLife . 19374 . 013Figure 6—figure supplement 1 . WT and Mecp2Null/y hippocampal neurons reveal transient and persistent fusion events upon stimulation , show equivalent activity-dependent BDNF secretion and membrane-resident BDNF-SpH fraction . ( A ) Representative images of a WT hippocampal neuron expressing BDNF-superecliptic pHluorin perfused with standard extracellular solution , 60 mM KCl , 50 mM NH4Cl and washout with extracellular solution ( left to right ) . Scale bar represents 10 µm . ( B ) BDNF fluorescence ( ΔF/F0 ) over time reflecting fusion events upon application of KCl , MES and NH4Cl solutions . Black line indicates duration of KCl , MES and NH4Cl application . ( C ) Average of BDNF fluorescence ( ΔF/F0 ) after NH4Cl application ( WT , n = 1521 events , Null , n = 948 events ) . ( D ) Bar graph shows activity-dependent BDNF release of WT ( 1464 events ) and Mecp2Null/y neurons ( 895 events ) . ( E , F ) Representative traces of BDNF fluorescence ( ΔF/F0 ) indicating transient and persistent fusion events upon KCl-induced membrane depolarization in WT ( E ) and Mecp2Null/y neurons ( F ) . ( G ) Bar graph shows BDNF-SpH fraction in WT and Mecp2Null/y neurons indicating surface accumulation of BDNF upon acid wash ( MES , pH 5 . 5; WT , 1480 events; Mecp2Null/y , 900 events ) . Data shown as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 013 These findings reveal a crucial mechanism wherein BDNF has an autocrine and cell autonomous role , potentially activating TrkB receptors only on the same neuron via a positive feed forward system , thereby regulating excitatory synapse formation and normal growth of Mecp2Null/y neurons . We then asked if loss of BDNF activity affects glutamatergic synapse formation through impairment of pre- or postsynaptic function . To address this question , we analyzed WT and Mecp2Null/y glutamatergic synapse densities , as described before , from proximal dendrites of identified WT and Mecp2Null/y postsynaptic neurons ( Figure 6C ) . We reasoned that in case of a presynaptic mechanism , Mecp2Null/y glutamatergic synapse densities would remain unchanged whether formed onto a WT or Mecp2Null/y postsynaptic neuron . Indeed , we found that the densities of Mecp2Null/y glutamatergic synapses were similarly ( 56 and 45% ) decreased when formed onto WT as well as Mecp2Null/y postsynaptic neurons , respectively ( Figure 6D-left ) . By contrast , in case of presynaptic Mecp2Null/y neurons overexpressing BDNF , densities of Null glutamatergic synapses made onto WT and Mecp2Null/y postsynaptic neurons were restored up to 87 and 86% of WT levels ( Figure 6D-right ) . This analysis strongly suggests that BDNF deficiency in cultured glutamatergic neurons reduces overall synaptic output without impairing the ability to receive glutamatergic synaptic input . The autocrine role of BDNF secretion in an in vitro RTT model might argue that in heterozygous mice with mosaic MeCP2 expression patterns mimicking RTT , impaired BDNF synthesis may contribute to a persistent growth deficit specifically in neurons lacking MeCP2 in vivo . In addition to smaller somata in Mecp2Null/y neurons , reduced neuronal nuclei size has been reported across previous MeCP2 loss-of-function studies ( Chen et al . , 2001; Rietveld et al . , 2015 ) that has been shown to be rescued by administering 7 , 8-DHF to Mecp2Null/y mice ( Johnson et al . , 2012 ) . Hence , we looked at the soma area and nucleus size of CA1 neurons in two- and eight-week old Mecp2+/- mice to examine if there is a persistent phenotype in MeCP2-deficient neurons . For this purpose , we labeled the hippocampal CA1 for MeCP2 , MAP2 and DAPI to perform quantitative analysis of MeCP2-positive and –negative neuronal nuclei ( Figure 7A and B ) . Strikingly , we found that MeCP2-negative neuronal somata were 20 and 14% smaller than that of MeCP2-positive neurons in two- ( Figure 7C ) and eight-week old ( Figure 7D ) Mecp2+/- mice respectively . Similarly , MeCP2-negative neuronal nuclei were 18 and 21% smaller than MeCP2-positive neurons in two- ( Figure 7E ) and eight-week old ( Figure 7F ) Mecp2+/- mice respectively . These in vitro and in vivo findings together validate a cell autonomous effect and autocrine role for BDNF secretion in Mecp2Null/y neurons and lead to promising avenue for future investigations studying cell autonomous effects of BDNF and MeCP2 . 10 . 7554/eLife . 19374 . 014Figure 7 . MeCP2-deficient hippocampal CA1 neurons have smaller somata and nuclei in Mecp2+/- heterozygous mice in vivo . ( A ) Representative images of hippocampal CA1 from Mecp2+/- mice labeled for DAPI ( blue ) , MAP2 ( green ) and MeCP2 ( red ) at two- ( top ) and eight-weeks of age ( bottom ) . Scale bar represents 50 µm . ( B ) Representative images of hippocampal CA1 from Mecp2+/- mice labeled for DAPI ( blue ) and MeCP2 ( green ) at two- ( top ) and eight-weeks of age ( bottom ) . Scale bar represents 50 µm . ( C–F ) Scatter plots show mean soma area and nucleus size of MeCP2+ and MeCP2- neurons at two- ( C , E ) and eight-weeks of age ( D , F ) . Number of neurons ( n ) shown in all graphs . Data shown as mean ± SEM . p<0 . 001; **p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 19374 . 014 Previous MeCP2 loss-of-function studies have shown decrease in glutamatergic synapse number and synapse density as well as reduced dendritic complexity and arborization associated with loss of MeCP2 ( Belichenko et al . , 2009; Chao et al . , 2007; Jentarra et al . , 2010; Larimore et al . , 2009; Zhou et al . , 2006 ) but underlying mechanisms are not fully understood . In our study , we find a generalized growth and differentiation defect in glutamatergic Mecp2Null/y neurons with MeCP2-deficient neurons displaying reduced neurite outgrowth and soma area ( Figure 1 ) . In examining causal mechanisms , we find a critical role for BDNF . The specific loss in neuronal growth can be attributed to suppression of BDNF synthesis since both synaptic output and morphological phenotypes could be restored by lentivirus-mediated and exogenous application of BDNF , in a TrkB-dependent manner ( Figure 2 ) . Moreover , we show that blocking BDNF-TrkB signaling in wildtype neurons leads to a MeCP2-deficient state , strongly arguing for nearly complete suppression of the canonical BDNF-TrkB pathway in mutant neurons . Further , we demonstrate that the BDNF-mediated growth pathway is strictly cell autonomous since ( i ) neighboring wildtype neurons fail to rescue Mecp2Null/y neuronal morphological deficits in a RTT model in vitro ( Figure 5 ) , and ( ii ) MeCP2-deficient neurons have smaller somata as well as nuclei in both two- and eight-week old Mecp2+/- mice in vivo ( Figure 7 ) , given the reduced BDNF protein levels in Mecp2+/- mice ( Wang et al . , 2006 ) and reduced mRNA levels in postmortem brain samples from RTT patients ( Abuhatzira et al . , 2007; Deng et al . , 2007 ) . This cell autonomy further illustrates that a change in circuit activity is unlikely to explain synapse loss associated with MeCP2-negative neurons in RTT models . Much attention has been received towards manipulating either MeCP2 levels or protein levels of MeCP2-regulated genes predominantly because reactivation of the Mecp2 or target gene has been shown to rescue morphological phenotypes , behavior and overall neurological function ( Chang et al . , 2006; Chao et al . , 2007; Guy et al . , 2007; Kline et al . , 2010; Larimore et al . , 2009; McGraw et al . , 2011; Nguyen et al . , 2012 ) . The fact that several synaptic defects in Mecp2Null/y neurons might be fundamentally linked to simply BDNF synthesis despite MeCP2 regulating transcription of a myriad of target genes renders MeCP2-BDNF interaction extremely pivotal . Along these lines , endogenous MeCP2 knockdown reduced dendritic length in E18 hippocampal neurons , which was fully rescued upon BDNF overexpression ( Larimore et al . , 2009 ) , which is in agreement with our rescue data ( Figure 2C ) . Differential BDNF function based on brain region and neuronal subtypes has been emphasized across various studies . In particular , reduced BDNF levels in the brainstem of Mecp2Null mice increased the amplitude of spontaneous and evoked EPSCs in nTS ( nucleus tractus solitarius ) neurons , which were fully rescued upon exogenous BDNF application ( Kline et al . , 2010 ) . This is comparable to the BDNF-mediated rescue of synaptic function in glutamatergic Mecp2Null/y neurons in our study ( Figure 2F–I ) . That BDNF is a neurotrophin with poor blood-brain barrier penetration characteristics ( Poduslo and Curran , 1996 ) prompted the need for small molecule substitutes that act as TrkB agonists and activate TrkB signaling ( Jang et al . , 2010; Massa et al . , 2010 ) . Administration of TrkB agonists ( Johnson et al . , 2012; Kron et al . , 2014; Schmid et al . , 2012 ) and other candidates including CX546 ( Ogier et al . , 2007 ) , environmental enrichment ( Kondo et al . , 2008; Lonetti et al . , 2010 ) , insulin-like growth factor-1 ( Tropea et al . , 2009 ) , cysteamine ( Roux et al . , 2012 ) and fingolimod ( Deogracias et al . , 2012 ) to MeCP2 mutant mice ameliorated several behavioral and functional RTT phenotypes both in vitro and in vivo . In our unique WT/Null RTT-like model , 7 , 8-DHF application clearly substantiated the autocrine and cell autonomous function of BDNF as well as BDNF-TrkB feed-forward loop impairment in Mecp2Null/y neurons ( Figures 5C and 6B ) . Impaired BDNF-TrkB activity could be due to differential surface expression of postsynaptic TrkB receptors and therefore inadequate availability of surface TrkB for binding to BDNF . However , complete rescue of both physiological and morphological phenotypes upon BDNF overexpression ( Figure 2 ) eliminated this possibility and highlighted a crucial upstream deficit in BDNF synthesis and release . Further studies are required to elucidate how 7 , 8-DHF functions to act in vitro and in vivo to ameliorate RTT symptoms and enable normalization of glutamatergic synapse numbers . BDNF synthesis deficit in MeCP2 lacking neurons has resulted in different phenotypic effects across various studies . In particular , Wang and colleagues reported decrease in absolute amounts of BDNF released but also an increase in percentage of BDNF content available for spontaneous release in Mecp2Null/y nodose ganglion ( NG ) neurons ( Wang et al . , 2006 ) . This may argue for cell-type specific effects of BDNF . Alternatively in our study , we demonstrate that glutamatergic synaptic output is normalized in Mecp2Null/y hippocampal neurons upon restoring BDNF levels ( Figures 2 and 3 ) and that BDNF neutralization experiments indicate a defect in synthesis ( Figure 3 ) , which suggest decreased availability of BDNF for release . By contrast , BDNF could also be functioning through other parallel pathways . For example , a study based on hippocampal slices demonstrated impaired activity-dependent endogenous BDNF release from presynaptic mossy fibers onto CA3 pyramidal neurons of symptomatic MeCP2 mutant mice that was correlated to impaired TRPC3 signaling ( Li et al . , 2012 ) . BDNF and other neurotrophins bind to the cell surface due to their highly positive charge at physiological pH ( Blöchl and Thoenen , 1996; Brigadski et al . , 2005; de Wit et al . , 2009 ) , and locally modulate developing synapses by spatially restricting BDNF action ( Zhang and Poo , 2002 ) . Additionally , several peptidergic hormones and secretory molecules including Wnt ( Cadigan et al . , 1998; Papkoff and Schryver , 1990 ) , Semaphorin 3A ( Bouzioukh et al . , 2006; De Wit et al . , 2005; Sahay et al . , 2005 ) and BDNF ( de Wit et al . , 2009 ) have been shown to remain membrane resident after secretion . Indeed , we verified this hypothesis by examining the fate of BDNF secretion in WT and Mecp2Null/y hippocampal neurons using the BDNF-superecliptic pHluorin ( BDNF-SpH ) live-cell imaging assay ( Kolarow et al . , 2007 ) ( Figure 6—figure supplement 1 ) ( Refer Materials and methods ) . We measured activity-dependent BDNF release by 60 mM KCl-induced depolarization , and quantified surface fraction of BDNF-SpH by application of pH 5 . 5 MES-buffered acid solution . We found that a significant fraction of BDNF-SpH remained membrane-resident as evident from MES-buffered acid quenching experiments ( Figure 6—figure supplement 1B and G ) , without affecting activity-induced BDNF secretion in both WT and Mecp2Null/y neurons ( Figure 6—figure supplement 1D–F ) . This complements our previous findings and shows that stable membrane deposits of BDNF could potentiate TrkB receptor activation cell autonomously and drive glutamatergic synapse number by confining activity to its own neurites . Additionally , the fact that activity-dependent regulated BDNF release was unaffected in MeCP2-deficient glutamatergic neurons suggested that the observed synaptic phenotypes were accounted for by a deficit in constitutive BDNF release . Further studies specifically measuring pre- and postsynaptic release of BDNF from WT and Mecp2Null/y cultured glutamatergic neurons are essential to determine the locus of activity-dependent regulation of excitatory synapses . What is the functional purpose of this autocrine feed-forward signaling loop ? There seems to be a link between intrinsic activity and growth in WT neurons by allowing BDNF to work on its own neurons , coupling neuronal firing patterns to their basic growth properties . This further implies that reduced activity in Mecp2Null/y neurons ( Chang et al . , 2006; Chao et al . , 2007; Dani et al . , 2005; Tropea et al . , 2009 ) may disrupt activity-dependent neuronal growth . This deficiency could be repaired by supplying an additional source of BDNF ( or TrkB agonist ) that activates WT-like TrkB signaling . BDNF has been suggested to act as an autocrine factor mainly in regulating dendrite development in adult-born granule cells ( GCs ) ( Wang et al . , 2015 ) and promoting axon formation in embryonic hippocampal neurons ( Cheng et al . , 2011 ) . In RTT pathophysiology in particular , we show for the first time that this localized autocrine function and cell autonomy of BDNF-TrkB together offer a plausible mechanism and explain why MeCP2 mosaicism in RTT females is ineffective in restoring phenotypes at a cellular , network or behavioral level . It may be possible that BDNF released could be less efficient in reaching neighboring neurons in co-cultures; however , the cell autonomous effect observed in autaptic neurons in vitro , and in vivo , as well as the autocrine BDNF dependence seen in vitro , argue against putative BDNF paracrine effects . However , what remains to be understood is if the predominant impact of impaired BDNF signaling on circuitry stems from the pre- or postsynaptic site . Initial morphological analysis showed axonal as well as dendritic growth deficiency ( Figure 1 ) . Further analysis in mixed co-culture experiments demonstrated a presynaptic deficit wherein densities of Mecp2Null/y glutamatergic synapses onto the postsynaptic neuron remained independent of genotype ( Figure 6D ) . This points towards a major ‘propagation’ defect specifically in Mecp2Null/y neurons in a mosaic circuit wherein they receive synaptic input similar to WT neurons . However , MeCP2-deficient glutamatergic neurons show reduced capacity to communicate output signal to a postsynaptic neuron thereby reducing efficiency of innervation across specific brain regions and dampening synaptic output . In fact , one could postulate that glutamatergic neurons lacking MeCP2 are essentially not ‘deaf’ but ‘mute’ . Interestingly , there is evidence of non-cell autonomous mechanisms contributing to neuronal development ( Braunschweig et al . , 2004 ) and dendritic arborization of cortical pyramidal neurons ( Kishi and Macklis , 2010 ) . In a mosaic brain , MeCP2-deficient mute neurons may contribute to reduced overall activity , which in turn could reduce activity-dependent BDNF secretion and thereby result in putative non-cell autonomous effects as well . Hence , future studies comparing wildtype and heterozygous Mecp2+/- mice may help reveal additional non-cell autonomous effects in vivo . It will be important to determine how the proposed presynaptic BDNF-mediated mechanistic pathway relates to other cell types , as well as to investigate specific downstream signaling pathways that might be affected as a result of BDNF synthesis deficiency . For example , MeCP2-deficient GABAergic neurons show decrease in inhibitory quantal size with reduced Gad1 and Gad2 levels ( Chao et al . , 2010 ) while BDNF released from postsynaptic target neurons acts locally to modulate GABAergic synapse formation ( Kohara et al . , 2007 ) . More generally , besides providing a mechanistic role for BDNF in MeCP2 mutant neurons , our results substantiate the significance of manipulating BDNF-TrkB interaction as a potential therapeutic strategy in alleviating the course of the RTT syndrome . All procedures to maintain and use mice were approved by the Animal Welfare Committee of Charité Medical University and the Berlin State Government . Mecp2Null/y male mice on a C57BLJ6/N background and/or Mecp2Tg1 male mice on a FVB/N background were used for morphological and electrophysiological studies . Primary hippocampal neurons were prepared from P0-P2 newborn mice and cultured on astrocyte feeder layers . First , WT astrocytes derived from P0-P2 mouse cortices were plated on collagen/poly-D-lysine microislands made on agarose-coated coverslips using a custom-made rubber stamp . For autaptic neuron studies , Mecp2Null/y or Mecp2Tg1 neurons were cultured along with their littermate controls , plated at a density of 3000 neurons per 35 mm well and grown in Neurobasal-A media containing B-27 supplement and Glutamax ( Invitrogen , Germany ) . For experiments utilizing continental cultures , glial proliferation was arrested by adding the antimitotic agent 8 μM 5-fluoro-2-deoxyuridine and 20 μM uridine ( FUDR ) to the astrocyte media . For experiments involving co-culture of WT and Mecp2Null/y neurons , cells of either genotype were incubated with lentiviral constructs of interest for 1 . 5 hr at 37°C on a rotating wheel . Cells were then centrifuged at 1500 rpm for 5 min twice to remove any viral debris after which WT and Mecp2Null/y neurons were plated at a density of 14 , 000 neurons each per 22 mm well . This procedure enables identification of synapses formed by either genotype under desired conditions . For all experiments , neurons were incubated at 37°C for 12–14 d before subjecting them to further analysis . Mouse cDNA constructs of pre-proBDNF , NGF and BDNF-SpH rat cDNA ( kindly provided by Prof . Matthijs Verhage , Center for Neurogenomics and Cognitive Research , Amsterdam , The Netherlands ) were cloned into vectors under control of the neuron-specific synapsin promoter . For co-culture experiments , we used a synapsin promoter-driven lentiviral shuttle vector expressing BDNF , cloned N-terminally to a self-cleaving T2A peptide of an expression cassette , which harbors ( i ) a nuclear localization sequence-tagged green fluorescent protein ( NLS-GFP ) or red fluorescent protein ( NLS-RFP ) , which was fused C-terminally via a self-cleaving P2A peptide ( Kim et al . , 2011 ) to ( ii ) Synaptophysin-mKate or Synaptophysin-GFP respectively ( NLS-GFP-P2A-SypmKate-T2A-BDNF or NLS-RFP-P2A-SypGFP-T2A-BDNF ) . Control lentiviral vectors included NLS-GFP-P2A-SypmKate and NLS-RFP-P2A-SypGFP . Lentiviral vectors and production were based on previously published protocols ( Lois et al . , 2002 ) . The production was done by the Viral-Core-Facility of the Charité – Universitaetsmedizin Berlin . Briefly , HEK293T cells were cotransfected with 10 µg of shuttle vector and helper plasmids ( pCMVdR8 . 9 and pVSV-G - 5 µg each ) with X-tremeGENE 9 DNA transfection reagent ( Roche Diagnostic , Switzerland ) . Virus-containing cell culture supernatant was harvested 72 hr post transfection and purified by filtration to remove cellular debris . Filtrate aliquots were flash-frozen in liquid nitrogen and stored at −80°C . Viral titer for all rescue experiments was determined using WT hippocampal continental neuronal cultures . Autaptic and continental hippocampal neurons were infected with lentivirus expressing BDNF or NGF on DIV 2 . In case of co-culture experiments , WT and Mecp2Null/y neurons were incubated with lentivirus expressing tagged-BDNF and plated as described above . Recombinant human BDNF ( Promega , Madison , WI ) was exogenously applied to the autaptic neurons at DIV 2 and added every 2–3 days at 50 ng/ml . The TrkB receptor antagonist ANA-12 ( Tocris , Germany ) was applied to neurons through culture medium at DIV 6 and added every three days at 10 µM . The BDNF neutralizing antibody α-BDNF ( Millipore , Germany ) was applied to neurons through culture medium at DIV 6 and added every three days at a dilution of 1:100 . Neurons were treated with the TrkB agonist 7 , 8-DHF ( Sigma Aldrich , Saint Louis , MO ) at DIV 6 , 9 and 12 at a working concentration of 500 nM . Whole-cell voltage-clamp recordings were obtained from autaptic neurons at DIV 12–17 . Currents were recorded from neurons held at −70 mV with a Multiclamp 700B amplifier ( Molecular Devices , Sunnyvale , CA ) under the control of Clampex 9 . 2 ( Molecular Devices ) . Data were sampled at 10 kHz and low-pass Bessel filtered at 3 kHz . Series resistance was compensated at 70% and only cells with <12 MΩ resistance were included . In general , an approximately equal number of cells were recorded from all groups on a given experimental day and data from at least two independent cultures were analyzed per experiment . Neurons were placed in standard extracellular solution , 300 mOsm pH 7 . 4 , containing 140 mM NaCl , 2 . 4 mM KCl , 10 mM HEPES , 10 mM glucose , 2 mM CaCl2 and 4 mM MgCl2 . The patch pipette internal solution , 300 mOsm pH 7 . 4 , contained 136 mM KCl , 17 . 8 mM HEPES , 1 mM EGTA , 0 . 6 mM MgCl2 , 4 mM ATP-Mg , 0 . 3 mM GTP-Na , 12 mM phosphocreatine , and 50 U/ml phosphocreatine kinase . Hypertonic sucrose solution was prepared as 500 mM sucrose in standard extracellular solution ( Rosenmund and Stevens , 1996 ) . Excitatory postsynaptic currents ( EPSCs ) were evoked by briefly depolarizing neurons from −70 mV to 0 mV for 2 ms . Application of hypertonic sucrose solution for 5 s was facilitated using a fast-flow system triggering release of the readily releasable pool ( RRP ) characterized by a transient inward current . RRP charge was estimated by integrating the total area under the transient curve obtained . Vesicular release probability ( Pvr ) was determined by calculating ratio of EPSC charge over RRP charge . Short-term plasticity was determined by evoking 2 EPSCs with an inter-stimulus interval of 25 ms to measure paired-pulse ratio ( PPR ) . PPR was determined by calculating the ratio of EPSC amplitude of second over the first synaptic response . Electrophysiological data were analyzed offline using Axograph X ( Axograph Scientific , Berkeley , CA ) , Excel ( Microsoft , Redmond , WA ) and Prism ( GraphPad , La Jolla , CA ) . Unless specified otherwise , statistical significance was determined using one-way ANOVA with Tukey post hoc test ( for three or more groups with normal distribution ) or Student’s t test ( for two groups with normal distribution ) . In case of data not normally distributed according to D’Agostino-Pearson test , statistical significance was determined using non-parametric Kruskal-Wallis test with Dunn’s post hoc test ( for three or more groups ) or Mann-Whitney U test ( for two groups ) . At DIV 13–15 , cells were fixed in 4% paraformaldehyde for 15 min after which they were washed thrice in 1x PBS . After permeabilizing and blocking with 5% normal donkey serum ( NDS ) in 0 . 1% PBS-Tween ( PBST ) for 1 hr , cells were subsequently incubated with antibodies of interest overnight at 4°C . After washing coverslips thrice with 0 . 1% PBST for 15 min each , primary antibodies were labeled with secondary Alexa-Fluor 405 , 488 , 555 or 647 ( 1:500; Jackson , West Groove , PA ) antibodies for 1 . 5 hr at room temperature . Coverslips were then washed twice with 0 . 1% PBST and twice with 1x PBS for 15 min each after which they were mounted on glass slides with either Mowiol or ProLong Gold Antifade Reagent ( Invitrogen ) . Labeling was done with ( i ) guinea pig anti-VGLUT1 ( 1:4000; Synaptic Systems , Germany ) , ( ii ) rabbit anti-VGLUT1 ( 1:4000; Synaptic Systems ) , ( iii ) chicken anti-microtubule-associated protein 2 ( MAP2 ) ( 1:2000; Millipore ) , ( iv ) mouse anti-Tau1 ( 1:1000; Millipore ) and ( v ) guinea pig anti-Homer1 ( 1:500; Synaptic Systems ) . First , 16-bit images were acquired on an Olympus IX81 inverted fluorescence microscope at 20x optical magnification with a CCD camera ( Princeton MicroMax; Roper Scientific , Trenton , NJ ) . All images were analyzed using ImageJ software with relevant custom plugins . At least two independent cultures were imaged and analyzed for every experiment . All images were subject to uniform background subtraction and optimal threshold adjustment . Quantification of VGLUT1+ puncta in autaptic WT or Mecp2Null/y neurons was used as a measure of presynaptic differentiation and estimating glutamatergic synapse number . This was done by identifying and counting all VGLUT1+ fluorescent spots localized on dendritic branches of every neuron using a custom macro in ImageJ . Quantification of all MAP2-positive processes with NeuronJ plugin was used to determine total dendrite length while measuring cross-sectional area across the MAP2+ cell body enabled estimation of area of neuronal somata . Nucleus cross-sectional area was measured by tracing the outline of the nucleus using NLS+ identified neurons . Total axonal length analysis was done by quantification of all Tau1+ MAP2- processes per neuron using NeuronJ plugin since an overlap of Tau1 labeling was observed in proximal MAP2-positive dendritic branches in several neurons . Pre- and postsynaptic puncta were estimated by manually counting individual VGLUT1+ and Homer1+ as well as double positive synapses on selected dendritic regions of interest . Glutamatergic synapse density in co-cultures was analyzed by manually counting VGLUT1 and synaptophysin double positive synapses on selected dendritic regions of interest . Unless specified otherwise , statistical significance was determined using one-way ANOVA with Tukey post hoc test ( for three or more groups with normal distribution ) or Student’s t test ( for two groups with normal distribution ) . In case of data not normally distributed according to D’Agostino-Pearson test , statistical significance was determined using non-parametric Kruskal-Wallis test with Dunn’s post hoc test ( for three or more groups ) or Mann-Whitney U test ( for two groups ) . Two animals per time point were fixed by transcardial perfusion with phosphate buffered-4% paraformaldehyde . Sectioning was done on a Leica cryostat CM3050 S and 25 µm coronal sections were obtained for both two- and eight-week old Mecp2+/- heterozygous mice . Sections were permeabilized with 1% Triton X-100 for 30 min and blocked with 5% NDS , 2% glycine , 0 . 5% Triton X-100 in PBS for 30 min at room temperature after which they were incubated in mouse anti-MeCP2 ( 1:500; Sigma ) and chicken anti-MAP2 ( 1:2000; Millipore ) overnight at 4°C . The sections were washed thrice with 1x PBS for 15 min each and stained with secondary Alexa 488 ( 1:500; Jackson ) for 1 hr at RT . Sections were again washed thrice with 1x PBS for 15 min each and mounted on glass slides in ProLong Gold Antifade reagent with DAPI ( Invitrogen ) . Images were acquired on a Leica SP8 laser-scanning confocal microscope . Images were acquired as 8-bit images with 63x oil objective at 1024 × 1024 pixel resolution . Three to four sections were imaged per mouse per time point , images were acquired as z-stack with 15–20 optical sections and maximum intensity projections were created using Fiji . Hippocampal CA1 MeCP2+ and MeCP2- neurons were identified based on their nuclear punctate staining and outlines of corresponding neuronal DAPI+ nuclei were manually traced and cross-sectional area measured using Fiji . Neuronal somata were measured by manually tracing MAP2+ cell bodies across different optical sections and cross-sectional area was measured using Fiji . We utilized an assay to monitor BDNF exocytosis in hippocampal neurons using a construct expressing BDNF tagged with superecliptic pHluorin ( BDNF-SpH ) ( de Wit et al . , 2009 ) . BDNF-SpH was expressed in primary hippocampal neurons and characterized using the pHluorin live-cell imaging assay . Neurons were placed in standard extracellular solution as described in case of electrophysiological measurements and 256 × 256 pixel images were acquired on an Olympus IX71 inverted microscope equipped with an Andor iXon back-illuminated CCD camera and Polychrome V Illumination Unit ( Till Photonics , Germany ) at 60x magnification ( numerical aperture 1 . 2 ) and a sampling rate of 1 Hz . Neurons were stimulated by application of 60 mM KCl ( prepared in standard extracellular solution ) for 30 s using the fast-flow system as mentioned above . 60 mM KCl-induced membrane depolarization of BDNF-SpH-expressing neurons resulted in an increase in fluorescence ( ΔF ) that resembled a punctate-pattern distribution representative of BDNF exocytosis . Some fusion events decayed abruptly while many others decayed slowly or remained persistent . 50 mM NH4Cl solution was applied for 30 s to neutralize intracellular pH , causing an abrupt increase in fluorescence intensity enabling visualization of all BDNF-SpH containing compartments . Low pH ( pH 5 . 5 ) solution of 2- ( N-morpholino ) ethanesulfonic acid was applied for 30 s for acid wash experiments , causing acute dimming of all surface resident BDNF-SpH proteins . BDNF release events were analyzed from stacks acquired from time-lapse recordings of BDNF-SpH containing vesicle clusters upon application of KCl , NH4Cl , pH 5 . 5 solutions or standard extracellular solution . BDNF release was analyzed from both synaptic and extrasynaptic sites . Background subtraction was done using Rolling Ball ( 50 pixel radius , ImageJ ) for all image stacks and ROIs were marked to measure ΔF , using ImageJ . 4 × 4 pixel regions of interest ( ROIs ) were identified and peak change in fluorescence ( ΔF ) in response to KCl-induced stimulation normalized to initial fluorescence ( F0 = average of five frames immediately before onset of stimulus ) was measured as a function of time and averaged in Axograph X ( Axograph ) . Baseline subtraction was done for each ROI per cell in Axograph X , and ΔF and ΔF/F0 values for all groups were analyzed in Prism ( GraphPad ) . Release events were characterized by their abrupt increase in fluorescence upon application of KCl followed by a rapid or gradual decrease of fluorescence back to baseline levels , and normalized to their corresponding peak NH4Cl events . The same ROIs were probed in recordings upon application of NH4Cl or MES-buffered acid solutions . Acid quenched BDNF fraction was normalized to corresponding peak NH4Cl response , reflecting the non-internalized pool of BDNF vesicles that are either found on the neuronal surface or easily accessible from the extracellular space around the cell membrane . All statistical analyses were done using Excel ( Microsoft ) and Prism ( Graphpad ) . Sample size estimation was done as described previously ( Chao et al . , 2007 , 2010 ) . Detailed statistical data are reported in Supplementary file 1 .	Rett syndrome is a progressive brain disorder . Individuals with the condition ( who are typically girls ) grow normally until they are 6-18 months old and then developmentally regress , with symptoms including anxiety , impaired coordination , seizures and breathing problems . Rett syndrome is caused by mutations in the gene that encodes a protein called MeCP2 . Researchers know that MeCP2 is vital for “excitatory” neurons in the brain to communicate with ( and activate ) their neighbors . Neurons that lack MeCP2 tend to make fewer of the connections across which they communicate – called synapses – with others . Many researchers who study Rett syndrome use male mice that lack the MeCP2 protein . This mouse model mimics the symptoms seen in Rett patients , but at a faster and more severe rate . These studies have shown that restoring normal levels of the protein in neurons prevents the majority of Rett-like symptoms in these mice and reverses the disorder . MeCP2 controls the activity of a number of other genes . These include the gene that produces a protein called Brain-Derived Neurotrophic Factor ( BDNF ) , which helps neurons to grow . Sampathkumar et al . have now studied neurons from mouse models of Rett syndrome to investigate whether BDNF can overcome the defects seen in neurons that lack MeCP2 . Viewed under a high-powered microscope , the Rett-like neurons appear smaller than healthy neurons and form fewer synapses . However , increasing the amount of BDNF in the diseased neurons restores normal growth and enables the cells to form more synapses . Girls with Rett syndrome tend to have a mixture of healthy neurons and those that do not produce the right amount of MeCP2 . To mimic this , Sampathkumar et al . grew a mixture of normal and Rett-like mouse neurons in a culture dish . The healthy neurons did not help the diseased neurons to form the correct number of synapses . However , increasing the levels of BDNF in the Rett-like neurons enhanced their ability to form synapses , and increased their cell size to match their healthy counterparts . Further work is now required to uncover whether manipulating the gene that encodes BDNF – or other genes that MeCP2 controls the activity of – in the brain can reduce the symptoms and slow the progression of Rett syndrome .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Loss of MeCP2 disrupts cell autonomous and autocrine BDNF signaling in mouse glutamatergic neurons
Many RNAs , including pre-mRNAs and long non-coding RNAs , can be thousands of nucleotides long and undergo complex post-transcriptional processing . Multiple sites of alternative splicing within a single gene exponentially increase the number of possible spliced isoforms , with most human genes currently estimated to express at least ten . To understand the mechanisms underlying these complex isoform expression patterns , methods are needed that faithfully maintain long-range exon connectivity information in individual RNA molecules . In this study , we describe SeqZip , a methodology that uses RNA-templated DNA–DNA ligation to retain and compress connectivity between distant sequences within single RNA molecules . Using this assay , we test proposed coordination between distant sites of alternative exon utilization in mouse Fn1 , and we characterize the extraordinary exon diversity of Drosophila melanogaster Dscam1 . One of the most important drivers of metazoan gene expression is the ability to produce multiple mRNA isoforms from a single gene . Around 58% of Drosophila melanogaster genes and >95% of human genes produce more than one transcript ( Pan et al . , 2008; Wang et al . , 2008; Brown et al . , 2014 ) , with most human genes expressing 10 or more distinct isoforms ( Djebali et al . , 2012 ) . Alternative promoter use , alternative splicing , and alternative polyadenylation all contribute to isoform diversity . In genes with multiple alternative transcription start and/or pre-mRNA processing sites , their combinatorial potential exponentially increases the number of possible products , with some human genes predicted to express >100 mRNA isoforms . In D . melanogaster , the number of isoforms observed per gene correlates with open reading frame length , suggesting that isoform complexity is a function of transcript length ( Brown et al . , 2014 ) . The current record holder in this regard is Dscam1 , in which four regions of mutually exclusive cassette exons combine to generate a remarkable 38 , 016 distinct >7000 nt mRNAs , each encoding a unique protein isoform ( Schmucker et al . , 2000 ) . In Dscam1 , the four regions of mutually exclusive cassette exon splicing are separated by one to eight constitutive exons . This feature of multiple alternative splicing regions separated by constitutive exons is shared by more than a quarter of human genes ( Fededa et al . , 2005 ) . In many cases , these regions are separated by >500 nts , the current limit for contiguous sequence output on most deep sequencing platforms . Further , high-throughput sequencing of RNA ( RNA-Seq ) generally requires its reverse transcription , with the processivity of available reverse transcriptases ( RTs ) limiting even single molecule cDNA sequencing ( e . g . , Pacific Biosciences ) to <2500 nt ( Sharon et al . , 2013 ) . Thus , existing high-throughput technologies cannot readily retain connectivity information between very distant sequences within individual mRNA molecules . Instead , full-length transcripts must be inferred by piecing together multiple short overlapping reads ( Garber et al . , 2011; Grabherr et al . , 2011; Haas et al . , 2013; Boley et al . , 2014 ) . For widely separated regions of alternative exon use , this loss of connectivity significantly limits our abilities to catalog isoform abundance and understand the mechanisms underlying alternative isoform generation . Here , we describe SeqZip , a method for profiling multiple distant ( >1000 nt ) sites of alternative splicing within individual RNA molecules . SeqZip uses sets of DNA oligonucleotides termed ‘ligamers’ . Each ∼40–60 nt ligamer hybridizes to the 5′ and 3′ ends of a single alternatively spliced exon or the beginning and end of a large block of constitutively included exons , looping out the sequences in between . These loops can be hundreds to thousands of nucleotides long . Juxtaposed ligamers hybridized to single RNA molecules are then joined by enzymatic ligation with T4 RNA ligase 2 ( Rnl2 ) ( Ho and Shuman , 2002 ) . The resultant DNA ligation products both capture the intramolecular connectivity among exons of interest and compress the sequence space necessary to identify those exons . Exon connectivity is subsequently decoded by assessing the sizes or sequences of the ligation products . Because SeqZip does not include an RT step and is therefore not subject to RT processivity and template-switching limitations , it can be used to assess intramolecular connectivity between regions separated by thousands of nucleotides . Further , relative ligation product abundance accurately reports spliced isoform abundance in the original sample . As a proof-of-principle , we here used SeqZip to test proposed connectivity relationships among alternatively spliced exons in mouse Fibronectin ( Fn1 ) and to define the molecular diversity of fly Dscam1 . The general idea of SeqZip is schematized in Figure 1 . This method requires efficient ligation of multiple DNA oligonucleotides ( oligos ) hybridized to an RNA template with little or no non-templated ligation . Although many ligases can join DNA or RNA oligos hybridized to a DNA template ( Bullard and Bowater , 2006 ) , when we initiated this study , only T4 DNA ligase was reported to join DNA fragments templated by RNA ( Nilsson et al . , 2001 ) . While T4 DNA ligase is the basis of multiple RNA-templated DNA ligation methods ( Nilsson et al . , 2001; Yeakley et al . , 2002; Conze et al . , 2010; Li et al . , 2012 ) , it also catalyzes non-templated DNA ligation ( Kuhn and Frank-Kamenetskii , 2005 ) , which would reduce SeqZip fidelity . 10 . 7554/eLife . 03700 . 003Figure 1 . Principles of SeqZip . The target RNA is hybridized with a set of DNA oligonucleotides ( ‘ligamers’ ) . Ligamers targeting outermost sequences contain one region of complementarity and primer sequences for subsequent amplification . Internal ligamers contain two regions of complementarity separated by a spacer sequence . Hybridization of the internal ligamers causes the RNA between the hybridization sites to loop out . Hybridized ligamers are ligated , amplified , and analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 00310 . 7554/eLife . 03700 . 004Figure 1—figure supplement 1 . Ligamer design workflow . ( A ) Schematic of mouse FN1 characterization using SeqZip . A total of six ligamers ( 1–6 ) are required to profile alternative splicing at the EDA and variable exons . Internal ligamer-specific barcodes ( BC1–4 ) are variable length unique sequences inserted in between exon-specific complementarity regions . ( B ) Design of a terminal ligamer specific to the 3′ end of FN1 exon 32 . The complementary strand of the genomic sequence for this region was trimmed from its 3′ end , measuring the Tm after each nt is removed , until the Tm is ≤65°C . ( C ) Design of an internal ligamer specific to the ‘95’ variation of the variable exon of mouse FN1 . This ligamer hybridizes to the beginning and end of the exon . Genomic sequence complementary to the 5′ and 3′ ends of the exon is trimmed from the inside of the exon out until the sequences meet two criteria: ( 1 ) each region has a Tm of ≤65°C and ( 2 ) the sum of the two region's lengths plus that of the desired barcode is ≤60 nt . ( D ) After the regions of complementary are trimmed for Tm and length , the sequences are BLATted against the target genome to ensure correct transcript and exon targeting . ( E ) Once regions of hybridization have been tested for specificity , the different components of each ligamer are assembled into a continuous sequence and the ligamer synthesized . Individual ligamer components can include: ( 1 ) one or two regions of complementarity to target RNA , ( 2 ) PCR priming sequence or internal barcode , and ( 3 ) 5′ phosphosphate for ligation . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 00410 . 7554/eLife . 03700 . 005Figure 1—figure supplement 2 . Other proposed uses of SeqZip . Shown are various uses of SeqZip toward multi-site sequence investigation of RNA . ‘Product Length Adjustment’ has applications similar to those shown in Figure 3E , where isoform discrimination solely on the basis of size separation of RT-PCR products would be ambiguous; with SeqZip , the lengths of individual products can be adjusted through ligamer design . ‘RNA barcoding’ depicts the introduction of randomized rather than static barcodes , allowing for molecular indexing or amplification bias estimation . ‘Quantify RNA-integrity’ relies on the requirement of molecular continuity between sites of ligamer hybridization in order to obtain a SeqZip product ( check mark ) . If the intervening sequences are not intact , no product is obtained ( X ) . Thus , SeqZip can be used to monitor the integrity of long RNAs . ‘Multi-site SNP detection’ is described in the ‘Discussion’ section ‘SeqZip uses and limitations’ . The panel depicting ‘Introduction of destruction sequences’ illustrates how short DNA oligos targeting ligamer-specific barcodes between hybridization regions ( in this case ‘B’ ) could be useful in the selective cleavage and destruction of particular ligation products . In the example shown , the ABC ligamer product would be cleaved with a restriction enzyme targeting the double-stranded oligo:barcode , while DEF would be left intact for downstream applications . ‘Sequence discovery using combined SeqZip and Reverse Transcription’ illustrates 5′ end sequence discovery using Cap Analysis of Gene Expression combined with SeqZip ligamers . This allows one to investigate novel 5′ end sequence connections to distant 3′ sequences . ‘Multi-site AS QPCR analysis’ is also described in the ‘Discussion’ section ‘SeqZip uses and limitations’ . The essential benefit over a conventional QPCR workflow is that SeqZip compresses distant sequences into a QPCR-friendly amplicon size and reduces the number of required primers . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 005 To find a suitable ligase for SeqZip , we tested the ability of several other commercially available enzymes to ligate four or five 5′ 32P-radiolabeled 20-nt DNA oligos hybridized to adjacent positions on either DNA or RNA ( Figure 2A ) . Although all DNA ligases tested could efficiently join multiple oligos hybridized to the DNA template ( Figure 2A; Bullard and Bowater , 2006 ) , only T4 DNA ligase and RNA ligase 2 ( Rnl2 ) joined the DNA oligos when hybridized to the RNA template . Of the two , Rnl2 was more active for RNA-templated DNA ligation ( data not shown ) and produced <1/7 as much non-templated product as T4 DNA ligase ( Figure 2A ) . Moreover , Rnl2 could not ligate DNA oligos hybridized to the DNA template , eliminating the possibility of contaminating genomic DNA confounding SeqZip ( Figure 2A ) . We note that Chlorella virus DNA ligase was recently commercialized for the purpose of RNA-templated DNA–DNA ligation ( SplintR ligase; NEB ) ( Lohman et al . , 2014 ) . We found , however , that SplintR ligase produces more non-templated DNA–DNA ligation events than Rnl2 ( Figure 2—figure supplement 1 ) . Also , while our paper was under review , another group reported RNA-templated DNA–DNA ligation by Rnl2 ( Larman et al . , 2014 ) , further validating its use in SeqZip . 10 . 7554/eLife . 03700 . 006Figure 2 . T4 RNA Ligase 2 catalyzes RNA-templated DNA-to-DNA ligation . ( A ) Left panel: ligase screen for RNA-templated DNA–DNA ligation activity . Ligases were incubated with an unlabeled single-stranded DNA ( left ) or RNA ( right ) template hybridized to a common pool of 5′ end 32P-labeled ( circled P ) DNA oligonucleotides for 1 hr . Both T4 DNA ligase and T4 RNA ligase 2 ( Rnl2 ) catalyze RNA-templated DNA–DNA ligation . Also note the inability of Rnl2 to ligate >2 oligos on the DNA template . For both templates , ligases are left to right: Tth DNA ligase ( Thermo ) , Tsc DNA ligase ( Prokaria ) , Thermostable DNA ligase ( Bioline ) , T4 DNA ligase ( NEB ) , T4 Rnl2 ( NEB ) , E . coli DNA ligase ( NEB ) . The three rightmost lanes are 32P-oligos only , 32P-labeled RNA template , and a 32P-labeled low-molecular weight DNA ladder ( NEB , N3233S ) . Right panel: Rnl2 and T4 DNA ligase time course for oligos hybridized to the RNA template . Templated ligation products ( –x2 through –x5 ) ; non-templated ligation product ( –x6 ) . ( B ) Rnl2 can join multiple 32P-labeled ligamers each looping out sections of the template but only when they are adjacently hybridized . Gray or white square: ligamer present or absent , respectively . No template ( -T ) ; no enzyme ( -E ) . ( C ) Cis- and trans-transcript hybridization and ligation using a ligamer ( W ) spanning 1046 nt common to two RNAs ( XWY and VWZ ) . Template concentrations ( nM ) were as indicated above each lane ( ranging from 0 . 01 to 100 nM ) , ligamers were held constant at 10 nM . Left panel , phosphoimage; right panel , SybrGold stained . ( D ) The ability of SeqZip to accurately report on relative input RNA concentrations was investigated using various ratios of two RNAs ( XWZ and VWY ) and a six ligamer pool . Observed product ratios were calculated from radioactive PCR band intensities . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 00610 . 7554/eLife . 03700 . 007Figure 2—figure supplement 1 . Examination of SplintR ligase in the SeqZip assay . Various concentrations of SplintR ligase and ATP were used to generate ligation products using Dscam1 ligamers and S2 cell RNA . Dscam1 ligation products appear as a ∼400 nt band , non-templated products as a ∼120 nt band . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 007 The SeqZip design requires efficient ligation of multiple DNA oligos ( ligamers ) , some spanning loops in the RNA template ( Figure 1 ) . To test the ability of Rnl2 to ligate these species , we designed four different 26 nt ligamers to loop out various lengths of a 307 nt transcript ( Figure 2B ) . Each 26 nt ligamer contained 10 nt of complementarity on either side of the loop , with a 6 nt spacer opposite the loop . The four ligamers—individually , pairwise , in threes , or as a complete set—were annealed to the template RNA and incubated with Rnl2 . Ligation products were only observed when ligamers bound to adjacent RNA sequences; four-way ligation products were obtained only when all ligamers were present . Thus , ligamers designed to loop out various lengths of a template RNA can be used to condense by more than twofold the information required to assess RNA connectivity—244 nt of the target RNA was condensed to a 104 nt DNA . Subsequent ligamer designs condensed connectivity information by >49-fold . A ligamer designed to loop out the sequences in between widely spaced regions of complementarity has the potential to bridge two RNA molecules . Such intermolecular ( trans ) hybridization would interfere with measurement of intramolecular ( cis ) RNA connectivity , producing artifacts akin to template switching in RT-based methods ( Figure 2C; Houseley and Tollervey , 2010 ) . To test the frequency of such trans events , we mixed two RNAs , each comprising a common 1106 nt internal sequence flanked by unique 5′ and 3′ sequences , with a ligamer set in which a single internal ligamer ( W ) looped out 1046 nt of the shared internal sequence ( Figure 2C ) . Because the terminal ligamers ( X , Y , V , and Z ) varied in length , polymerase chain reaction ( PCR ) of SeqZip reactions yielded 177 and 143 nt cis-templated products and 165 and 155 nt trans-templated products . Trans hybridization of ligamer W , a tri-molecular interaction , should be much more sensitive to RNA concentration than bimolecular cis hybridization . Consistent with this , whereas cis products were detected by end point PCR at every target RNA concentration tested down to 0 . 01 nM , trans products were only detected when target RNAs were ≥10 nM , ( Figure 2C , lower half ) . But , even when both targets were present at 50 nM , semi-quantitative radioactive PCR revealed that the cis hybridization products predominated ( Figure 2C , lower left ) . Nonetheless , to disfavor trans hybridization , the general conditions for SeqZip described below use cellular RNA concentrations ( 10–40 ng/ml polyA+ RNA ) at which most individual mRNAs are present at <1 nM . To be useful as a quantitative method , SeqZip should accurately report on input RNA abundances . To test this , we mixed two target RNAs at ratios varying from 100:1 to 1:100 ( a 100-fold dynamic range ) . Radioactive PCR revealed that their respective SeqZip product ratios paralleled these input ratios over the entire series ( Figure 2D ) . As a first test of SeqZip with a biological sample , we used it to assess alternative exon inclusion in endogenous human CD45 ( PTPRC ) mRNA ( Zikherman and Weiss , 2008 ) . CD45 isoforms contain various combinations of exons 4 , 5 , and 6 ( Figure 3A ) . Jurkat cells ( resembling naïve , primary T cells ) predominantly express isoforms containing exons 5 and 6 ( R56 ) , only exon 5 ( R5 ) , or no cassette exon ( R0 ) . U-937 cells ( resembling activated T cells ) predominantly express the R56 isoform and one containing exons 4 , 5 , and 6 ( R456; Yeakley et al . , 2002 ) . The three adjacent cassette exons occupy only 585 nt , making this region amenable to analysis by both reverse transcription and SeqZip . Reverse transcription-PCR ( RT-PCR ) products ranged from 365 to 848 nt , while SeqZip products ranged from 132 to 260 nt ( Figure 3B ) , representing a ∼threefold compression of connectivity information . 10 . 7554/eLife . 03700 . 008Figure 3 . SeqZip assay to measure endogenous mRNA isoform expression . ( A ) The SeqZip strategy to detect human CD45 mRNA isoforms . ( B ) Denaturing PAGE gels showing products of reverse transcriptase ( RT ) ( top left ) or SeqZip ( bottom left ) CD45 mRNA obtained from two different human Jurkat and U-937 T-cell lines , or a 1:1 mixture of the two . Top right: quantified band intensities from gels at left . Bottom right: mirrored lane profiles from the mix lanes ( RT—left; SeqZip—right ) . ( C ) The six possible combinations of EDA ( blue; + or − ) and V ( light blue; 120 , 95 and 0 ) alternative splicing within mouse Fn1 transcripts . Filled boxes depict exons , diagonal lines indicate isoform sequences not shown , and straight lines show absence of exon ( s ) in the final mRNA . ( D ) Detailed schematic of ligamer pools used to analyze indicated regions of Fn1 RNA . ( E ) SeqZip ligation products from immortalized MEFs with indicated Fn1 genotypes . Radioactive PCR separated on a native acrylamide gel . ( F ) Fn1 isoform abundance measured by SeqZip and PacBio . Black bars indicate observed individual exon ( ‘Individual Pool’; EDA , V ) or combination frequencies ( ‘Combination A + V pool’ , [EDA , V] ) . Shown in light gray are expected combination isoform intensities , and where available , the frequency of PacBio reads ( mid-gray , lower bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 008 Using RT-PCR and SeqZip , we measured CD45 isoforms from Jurkat or U-937 poly ( A ) -selected RNA or a 1:1 mixture of the two . Both methods reported the expected isoform abundances ( Figure 3B ) . Importantly , even though SeqZip detection of R456 , R56 , R5 , and R0 required different numbers of ligation events , all relative abundances were accurately reported , even in the mixture containing all four isoforms ( Figure 3B , lower right ) . For a more complex splicing pattern , we next turned to fibronectin ( Fn ) . Mouse Fn1 contains three well-characterized regions of alternative splicing: ( 1 ) the EDB exon included in embryos and adult brain but not other adult tissues , ( 2 ) the EDA exon variably included or excluded across multiple developmental and adult tissue types , and ( 3 ) the variable ( V ) region in which use of three alternative 3′ splice sites leads to inclusion of 120 , 95 , or 0 additional amino acids in the FN1 protein ( Figure 3C ) . The original suggestion that an upstream splicing decision can affect a downstream splicing decision came from analysis of the EDA and V regions where it was reported that EDA exclusion promotes use of the promoter-proximal 3′ splice site ( ‘120’ ) in the V region ( Fededa et al . , 2005 ) . The EDA and V regions are separated by six constitutively included exons , comprising 813 nt; thus , RT-PCR products including the EDA and V regions range from 1 to 1 . 6 kbp ( Figure 3D ) . Both the overall length of the RT-PCR products and the extensive region of similar sequence identity in the middle that can promote template switching ( see below ) confound RT-PCR analysis of the six possible EDA and V exon combinations . In comparison , our SeqZip ligation products were >fivefold smaller ( 139–318 nt; Figure 3D , E ) , and they contained no intervening region of extensive nucleotide identity . Thus , SeqZip provided a new means to test the possibility of connectivity between Fn1 EDA and V splicing decisions . The effects of EDA inclusion or exclusion on V region splicing were previously tested by creating mice via homologous recombination with intronic splicing enhancers modified to favor either constitutive inclusion ( +/+ ) or exclusion ( −/− ) of the EDA exon ( Chauhan et al . , 2004 ) . That study also analyzed mice heterozygous for the modified locus ( +/− ) and the wild-type parental strain ( wt/wt ) . We obtained immortalized mouse embryonic fibroblasts generated from all four mouse lines and performed SeqZip analysis ( Figure 3E , F ) . Three different ligamer pools allowed us to analyze each region in isolation ( individual pools A and V ) or both regions together ( combination pool A + V ) ( Figure 3D ) . EDA and V isoform ratios determined from low cycle , radioactive PCR band intensities of the A and V pool ligation products ( SeqZip: Observed ) were used to calculate expected EDA:V isoform abundances , assuming no interdependence between the two regions ( SeqZip: Expected ) . We also generated cDNAs by low-cycle RT-PCR and sequenced them on a Pacific Biosciences RSII instrument ( PacBio:Observed ) , a single molecule platform with sufficient read length to maintain connectivity between the EDA and V regions ( Sharon et al . , 2013 ) . In both the SeqZip and PacBio data sets , constitutive EDA inclusion or exclusion was as expected in the +/+ and −/− cells , respectively . Unexpectedly , however , we could not detect any EDA inclusion in the +/− cells despite confirming the presence of both alleles in gDNA ( data not shown ) . Regardless , neither SeqZip nor PacBio yielded any evidence for an effect of EDA inclusion or exclusion on V region splice site choice . That is , in no case was the observed frequency of any A + V combination statistically different from the frequency expected for independent events . This was also our observation in primary mouse embryonic fibroblasts ( MEFs ) from wild-type mice ( Figure 3F ) . Our results thus support the view that the EDA and V regions of mouse Fn1 are spliced autonomously ( Chauhan et al . , 2004 ) . For the Drosophila Dscam1 gene , alternative splicing of four blocks of mutually exclusive cassette exons ( exons 4 , 6 , 9 , and 17 ) can potentially produce 38 , 016 possible mRNA isoforms ( Figure 4A ) . Previous studies suggest that all isoforms can be generated ( Neves et al . , 2004; Zhan et al . , 2004; Sun et al . , 2013 ) , with all 12 exon 4 variants being stochastically incorporated in individual neurons ( Miura et al . , 2013 ) . 10 . 7554/eLife . 03700 . 009Figure 4 . Analysis of Dscam1 isoforms via high-throughput sequencing . ( A ) Architecture of Dscam1 . Black: constitutively included exons; colors: variant exons . Only one cassette exon per variant region is included in the mRNA . ( B ) Sequence similarity between 1000 random isoforms of Dscam1 in cDNA , circularized cDNA , and SeqZip ligation product form . All lengths are shown to scale . ( C ) Strategy to measure Dscam1 isoform diversity using SeqZip on the MiSeq platform . ( D ) Strategy to measure Dscam1 isoform diversity by triple-read sequencing on the Illumina MiSeq platform . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 00910 . 7554/eLife . 03700 . 010Figure 4—figure supplement 1 . Dscam1 in vitro transcript measurement . ( A ) Workflow schematic of Dscam1 ligation product sequence and alignment procedure . ( B ) Rank-order abundances for different Dscam1 isoforms detected in control reactions . Blue: input in vitro isoforms; red: template-switched isoforms; yellow: ligation products containing near-cognate ligamers; green: in vivo isoforms detected in S2 cell total RNA . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 010 Previous high-throughput methods for examining Dscam1 exon connectivity relied on RT-PCR , a technique potentially confounded by long stretches of sequence identity in the constitutive exons separating each cluster and by sequence similarity among exon 4 , 6 , and 9 variants ( Figure 4B ) . Long regions of sequence homology promote template switching during both RT and PCR ( Judo et al . , 1998; Houseley and Tollervey , 2010 ) ; this can generate novel isoforms not originally present in the biological sample . SeqZip can both dramatically reduce these regions of sequence of identity ( Figure 4B ) and introduce new exon-specific codes ( Figure 4C ) . Thus , in addition to maintaining connectivity information , SeqZip both compresses sequence length and increases sequence heterogeneity , thereby greatly decreasing the potential for template switching compared to cDNAs created by standard RT or circularized cDNA approaches . Prior to our development of SeqZip , we had attempted to use a RT-PCR-based triple-read sequencing method to determine exon connectivity between Dscam1 alternative splicing regions 4 , 6 , and 9 ( Figure 4D , Figure 4—figure supplement 1B , ‘Materials and methods’ ) . To measure the extent of template switching , we generated four RNA transcripts corresponding to distinct isoforms . As expected , this RT-based method detected many novel transcript isoforms containing exon combinations not present in the four input transcripts ( Figure 4—figure supplement 1B ) . These template-switched isoforms represented 34–55% of the isoforms detected , with many being significantly more abundant than one or more of the input isoforms . A similar circularized cDNA method , CAMSeq , has also been used to assess Dscam1 exon connectivity ( Sun et al . , 2013 ) . In light of the high rate of template switching in our triple-read sequencing approach , we reexamined the published CAMSeq control data to assess the extent of template-switching events ( Figure 4—figure supplement 1B ) . Indeed , template-switched isoforms were present in the CAMSeq data , with many template-switched isoforms being more abundant than the low abundance input isoforms . Moreover , we detected 5386–5914 additional isoforms whose presence could not be explained by either the composition of the 8 input RNA isoforms or by template switching . Thus , while CAMSeq was a clear improvement over both previous linear RT-PCR-based approaches and our triple-read sequencing approach , template-switching artifacts remained a substantial problem . By eliminating RT and using exon-specific barcodes to ensure unambiguous isoform assignment ( Figure 4C ) , SeqZip should greatly reduce template switching . To measure this directly , we mixed together three different in vitro-transcribed Dscam1 isoforms in the presence of total RNA from a mouse hepatoma cell line ( Hepa 1–6c; Figure 5A ) . This mixture was then divided into two separate ligation reactions , each containing a complete 97 ligamer pool that differed only in the 7 nt ligamer barcode assigned to two exons in each cluster ( in Figure 5A ) . Following ligation , the differentially coded samples were mixed together , subjected to PCR , and sequenced on the MiSeq platform ( on which paired-end reads can cover a total of 500 nts ) . Of the 50 , 475 reads obtained in these control reactions , none were indicative of template switching ( i . e . , no ligation product contained both pool 1 and pool 2 barcodes; Figure 5B ) . Moreover , when the same differential coding approach was applied to Drosophila S2 cell poly ( A ) -selected RNA , just 17 of 111 , 242 reads ( 0 . 015% ) corresponded to template-switched isoforms ( Figure 5B , Figure 4—figure supplement 1B ) . Thus , the SeqZip design greatly diminishes template switching . 10 . 7554/eLife . 03700 . 011Figure 5 . SeqZip Dscam1 control experiments . ( A ) Three in vitro-transcribed cDNAs used as controls containing the exon variants indicated and mixed in a 100:10:1 relative ratio . Also shown are a schematic of the ligamer pool , with each ligamer targeting a different variant exon , the six ligamers ( * ) containing having different codes in pool 1 and pool 2 , and a workflow for identifying near-cognate ligation and template-switching events . ( B ) Schematic showing how template-switched isoforms were identified as an incorrect combination of barcodes unique to the differentially coded pools shown in ( A ) . Also shown are the observed numbers of un-switched and template-switched reads and isoforms for controls in ( A ) and S2 cellular RNA . ( C ) Quantification of in vitro-transcribed control cDNAs analyzed by SeqZip according to the workflow shown in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 01110 . 7554/eLife . 03700 . 012Figure 5—figure supplement 1 . Cognate and nearest near-cognate folding energies for Dscam1 Exon 6 variant ligamers . ( A ) Left: folding energies between all exon 6 variants and their cognate ligamer . Right: folding energies for all exon 6 variants and their closest near-cognate ligamer . Yellow line = −36 kcal/mol; blue line = −67 kcal/mol . ( B ) Comparison of 6 . 8 and 6 . 24 ligamer sequences to their closest near-cognate ligamer and folding energies between ligamers 6 . 9 and 6 . 24 to all exon 6 . X sequences . Yellow line = −36 kcal/mol; blue line = −67 kcal/mol . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 012 Sequences at the ends of target exons specify where ligamers bind ( Figure 1 ) . The high similarity among the cassette exon sequences within each cluster raised the possibility that ligamers would bind near-cognate as well as cognate sequences . To assess the potential for mis-pairing , we calculated the free energy of hybridization ( Reuter and Mathews , 2010 ) between each ligamer and all exon variants within its target cluster ( Figure 5—figure supplement 1A ) . Cognate ligamer–exon pairs had predicted hybridization energies lower than ΔG° = −67 kcal/mol; the closest near-cognate pair was ≥12 kcal/mole higher . In the control experiments containing just three Dscam1 isoforms , only 642 of 50 , 475 high-confidence alignments ( 1 . 3% ) contained ligamers for exons not present in any input transcript , with the majority of these species ( 221/236 ) represented by three or fewer reads ( Figure 4—figure supplement 1B ) . Nonetheless , two near-cognate hybridization products with >100 reads were detected . Although both were less abundant ( 2 . 4- and 372-fold lower ) than reads corresponding to cognate targets ( Figure 5C and Figure 4—figure supplement 1B ) , this does raise a cautionary note with regard to interpretation of extremely low abundance ligation products in experiments wherein near-cognate ligation is a possibility . On the other hand , SeqZip accurately reported input cognate isoform abundances over 3 orders of magnitude ( Figure 5C ) . Thus , as with CD45 and Fn1 isoforms , SeqZip proved highly quantitative for assessing the majority of Dscam1 isoforms . We next used SeqZip to measure Dscam1 isoform identity and abundance in S2 cells , as well as 4–6 hr and 14–16 hr D . melanogaster embryos . Ligamers targeting every exon variant in clusters 4 , 6 , and 9 together with ligamers for constitutive exons 3 , 5 , 7 , 8 , and 10 ( 97 ligamers in all ) reduced the median size of mRNA sequences analyzed from 1734 nt ( 1722–1751 nt for exons 3–10 ) to 356 nt for a seven-ligamer product formed by six ligation events . This approximately fivefold length reduction allowed the products to be fully sequenced using 250 bp , paired-end reads in a single Illumina MiSeq run ( Figure 4C ) . Between 449 , 113 and 946 , 110 , high-confidence alignments were obtained for each sample ( Supplementary file 2 ) . Across all three samples , SeqZip detected 8397 of the 18 , 612 possible isoforms ( Figure 6A ) . Individual isoform abundances were highly correlated between both technical and biological replicates ( r = 0 . 8–0 . 95 , p < 2 . 2 × 10−16 , Fisher z-transformation; Figure 6—figure supplement 1 ) . Of the 97 possible exons represented in our ligamer set , all were detected except exon 6 . 11 , which is generally thought to be an unused pseudo-exon ( Neves et al . , 2004; Zhan et al . , 2004; Watson et al . , 2005; Miura et al . , 2013; Sun et al . , 2013 ) . The absence of exon 6 . 11 reads from our libraries provides additional evidence for the specificity of SeqZip . Further , with two exceptions , the patterns of individual exon use in S2 cells were directly comparable between the SeqZip and CAMSeq data sets ( r = 0 . 87 , p < 2 . 2 × 10−16 , Fisher z-transformation; Figure 7—figure supplement 1 ) : exon 6 . 47 was well represented in the CAMSeq data but undetectable by SeqZip , and exon 9 . 31 was more abundantly represented in our data . 10 . 7554/eLife . 03700 . 013Figure 6 . SeqZip captures diverse Dscam1 isoform expression and exon use . ( A ) Rank-order of isoform expression by sample type ( S2 , 4–6 hr , 14–16 hr ) . ( B ) Individual exon usage per library for each replicate ( differently shaded bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 01310 . 7554/eLife . 03700 . 014Figure 6—figure supplement 1 . Technical and biological reproducibility of SeqZip Dscam1 isoform quantification . Technical and biological replicates for S2 cells and 4–6 hr embryos using SeqZip . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 014 Comparison of exon usage patterns across three different biological samples revealed increasing isoform diversity with tissue complexity: S2 cells were the least diverse , 4–6 hr embryos had intermediate isoform diversity , and 14–16 hr embryos showed the greatest isoform diversity ( Figure 6B ) . As previously shown , cluster 4 and 9 exon usage patterns change during development , whereas the cluster 6 pattern remains more static ( Celotto and Graveley , 2001; Neves et al . , 2004; Zhan et al . , 2004; Miura et al . , 2013; Sun et al . , 2013 ) . In S2 cells , Dscam1 mRNAs incorporate very little of exon 4 cassettes 2 and 9 and use almost exclusively exon 9 cassettes 6 , 9 , 13 , 30 , and 31 . This pattern is the characteristic of hemocytes ( Watson et al . , 2005 ) and consistent with the macrophage-like nature of S2 cells ( Schneider , 1972 ) . Whereas 4–6 hr embryos are similar to S2 cells in exon clusters 4 and 9 , 14–16 hr embryos show increased exon diversity , particularly in cluster 9 . Figure 7 shows that Dscam1 isoforms associated with hemocytes ( i . e . , those lacking exon 4 . 2 and 4 . 9 or containing exon 9 cassettes 6 , 9 , 13 , 30 , or 31 ) are the most abundant in all three samples , but other isoforms emerge as development proceeds . 10 . 7554/eLife . 03700 . 015Figure 7 . Observed vs expected Dscam1 isoform abundance . Two-way ( 4:6 , 6:9 , and 4:9 ) and three-way ( 4:6:9 ) expected isoform abundances , calculated from the individual inclusion frequency for each variant exon ( Figure 6B ) in indicated sample type ( S2 cells , 4–6 , or 14–16 hr embryos ) , plotted against observed isoform abundances in that sample type . Isoforms are colored according to hemocyte-indicative ( red ) or non-hemocyte-indicative ( blue ) exon variants . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 01510 . 7554/eLife . 03700 . 016Figure 7—figure supplement 1 . Comparison of RT-PCR and ligation-based Dscam1 isoform analysis techniques . ( A ) Individual exon usage in S2 cells as measured by SeqZip and CAMSeq . Gray rectangles indicate exon variants ( 6 . 47 and 9 . 31 ) whose apparent inclusion frequency was substantially different between SeqZip and CAMSeq . ( B ) Scatter plot of isoform expression measured by SeqZip and CAMSeq . Gray areas indicate the off-axis populations containing exon variants 6 . 47 and 9 . 31 . DOI: http://dx . doi . org/10 . 7554/eLife . 03700 . 016 To examine the possibility of coordinated splicing , we calculated expected pairwise and three-way exon combination frequencies for every transcript isoform observed in each sample ( S2 cells , 4–6 hr and 14–16 hr embryos ) , assuming a null hypothesis of no coordination ( see ‘Materials and methods’ ) . Comparison of expected and observed frequencies ( Figure 7 ) revealed no statistically significant differences ( q ≤ 0 . 05 ) between expectation and observation for S2 cells and 4–6 hr embryos . For 14–16 hr embryos , however , 17 of 371 observed exon 4:9 combination frequencies , and 14 of 2004 observed 4:6:9 combination frequencies , did fall outside of the expected range ( q < 0 . 05 , Supplementary file 4 ) . The only pattern we could deduce was that the variant 4:9 combinations were all non-hemocyte combinations ( Figure 7 ) . Because the majority of 4:6:9 combination frequencies ( 99 . 3% ) were consistent with the null hypothesis of no coordination , our data agree with previous studies ( Neves et al . , 2004; Sun et al . , 2013 ) that individual cassettes in Dscam clusters 4 , 6 , and 9 are chosen independently , with exon choice in one cluster having no detectable effect on subsequent exon choice in another cluster . The idea that promoter-proximal ( upstream ) gene regions can affect distal ( downstream ) alternative splicing was first reported for the mammalian Fn1 gene almost two decades ago ( Cramer et al . , 1997 ) . Expressed sequence tag and oligonucleotide microarray data were subsequently interpreted to suggest coordination among different regions of alternative splicing in numerous other genes ( Fededa et al . , 2005; Fagnani et al . , 2007 ) . One potential explanation for this effect is promoter-dependent polymerase speed: slow RNA synthesis favors inclusion of cassette exons with weak splicing signals , while fast RNA synthesis favors their exclusion ( Fededa et al . , 2005; Dujardin et al . , 2014 ) . This polymerase speed effect does not necessitate any dependence of a downstream splicing decision on an upstream decision—the multiple sites of alternative splicing can be independently but similarly influenced by polymerase speed . In this case , although alternative splicing decisions in different regions within a transcript might correlate with one another , they would not depend on one another . Neither oligonucleotide microarrays nor short high-throughput sequencing reads preserve long-range exon connectivity within individual mRNA molecules . Thus , neither approach is able to unambiguously distinguish between dependent coordination , wherein alternative processing at an upstream site causes alternative processing changes at a downstream site , and independent coordination , wherein multiple regulated exons are simply subject to similar external influences ( Calarco et al . , 2007 ) . SeqZip , however , does preserve single molecule connectivity , so is perfectly suited to investigate coordination mechanisms . Our proof-of-principle SeqZip experiments with the mouse Fn1 gene revealed no evidence of dependent coordination between the two regions of alternative splicing we examined ( Figure 3 ) . Therefore , consistent with other findings ( Chauhan et al . , 2004 ) , we conclude that any apparent coordination between the Fn1 EDA and V regions is due to their independent response to external influences . With 38 , 016 possible isoforms , D . melanogaster Dscam1 produces the greatest known isoform diversity of any single gene . Diverse Dscam1 isoforms enable the developing nervous and immune systems to discriminate between heterotypic and homotypic connections ( Wojtowicz et al . , 2004; Watson et al . , 2005; Zipursky and Grueber , 2013 ) . While flies engineered to produce only 4752 unique isoforms display neurite formation functionally equivalent to wild-type controls , flies expressing just 1152 isoforms display neuronal branching defects . This supports the view that what is essential for biological function is molecular diversity not any particular sequence ( Hattori et al . , 2009 ) . Both D . melanogaster and Anopheles gambiae ( AgDscam ) also express Dscam in hemocytes , where isoform diversity has been implicated in opsonizing invading pathogens ( Watson et al . , 2005; Dong et al . , 2006 ) . Complete characterization of Dscam1 isoform diversity presents an extreme technical challenge ( Hattori et al . , 2008 ) . The four regions of mutually exclusive cassette exons span >4300 nt in full-length mRNAs , so maintaining connectivity among all cassettes or even just cassettes 4 , 6 , and 9 , which span >1700 nt , is all but impossible when sequencing with current high-throughput technologies ( Black , 2000; LeGault and Dewey , 2013; Zipursky and Grueber , 2013 ) . Single molecule methods capable of longer reads ( e . g . , Pacific Biosciences ) have limited read depths , making it difficult to fully analyze transcripts expressed over many orders of magnitude . Finally , many Dscam1 exon variants arose from exon-duplication events , so their sequences are highly similar ( Lee et al . , 2010 ) . This high-sequence similarity , combined with the long stretches of identical constitutive exons separating the distant alternative splicing regions , strongly favors template switching by RT ( Judo et al . , 1998; Houseley and Tollervey , 2010 ) . SeqZip has no RT step , it eliminates long intervening regions of common sequence , and the unique exon-specific barcodes introduced during the ligation step further discourage template switching during subsequent amplification . Using pools of 97 individual ligamers targeting every exon in clusters 4 , 6 , and 9 , we analyzed Dscam1 diversity in S2 cells , 4–6 hr , and 14–16 hr embryos . In all three samples , we observed individual exon use frequencies similar to those observed with CAMSeq ( Figure 7—figure supplement 1; Sun et al . , 2013 ) . SeqZip and CAMSeq both detected significant exon usage changes in clusters 4 and 9 between S2 cells and embryos . Analysis of 4–6 hr and 14–16 hr embryos allowed the timing of exon 4 and 9 usage changes to be narrowed to >6 and <16 hr ( Figure 5A ) , a developmental window when neurogenesis is occurring ( Goodman et al . , 1993 ) . In S2 cells and 4–6 hr embryos , we found no evidence of inter-cluster connectivity with regard to exon choice ( Figure 7 ) . In 14–16 hr embryos , we found weak evidence for such connectivity ( Supplementary file 4 ) . Multiple cell types expressing characteristic , but different , cluster 4 and 9 exon variants , however , likely confound determination of coordination in 14–16 hr embryos . Therefore , consistent with previous reports ( Neves et al . , 2004; Miura et al . , 2013; Sun et al . , 2013 ) , we conclude that individual Dscam1 isoforms are produced via stochastic alternative splicing . In mammalian neuronal development , cells use tandem arrays of protocadherin and neurexin genes to distinguish their own neurites from those originating from different cells . Some tandem arrays are capable of generating >1000 different spliced isoforms ( Ushkaryov et al . , 1992; Wu and Maniatis , 1999; Rowen et al . , 2002 ) . A recent analysis of mouse neurexin genes using long reads ( Pacific Biosciences ) of individual cDNA molecules showed that while these genes do produce many different isoforms , there is also no coordination among their alternative processing choices ( Treutlein et al . , 2014 ) . A potentially routine and robust use of SeqZip is highlighted by our Fn1 analyses , where we simultaneously measured 12 different alternative splicing isoforms and determined their relative expression by simple gel electrophoresis without sequencing ( Figure 3E ) . This application is similar to the multiple-exon-skipping detection assay ( MESDA ) used to study SMN1 and SMN2 isoform expression in different Batten disease cell lines ( Singh et al . , 2012 ) . MESDA measured the relative expression of >6 SMN isoforms and even identified a novel isoform , providing a useful tool for researchers working on spinal muscular atrophy . Measurement of SMN isoforms could also be performed using SeqZip , with several advantages over MESDA including reduction in amplicon size , lower propensity for template switching during amplification , and no RT step . One limitation of SeqZip is the number of ligamers required to create a ligation product . To achieve necessary sequence specificity , ligamers need to be 40–60 nt . Current illumina-based sequencing platforms can read ∼500 nt of contiguous sequence . Thus , ∼8–12 ligamers is currently the upper limit for high-throughput sequencing analyses . Although our quantitative analysis of CD45 showed no difference in ligation efficiency for isoforms requiring two ligations compared to those requiring five ( Figure 3B ) , other transcripts could theoretically differ . Because exon-excluded isoforms require fewer ligations , as the number of sites being examined grows , it is possible that detection of shorter ( i . e . , exon-excluded ) transcripts will be favored . Thus , if a SeqZip experiment requires different numbers of ligation events for different RNA isoforms , it is crucial to perform the necessary controls to ensure quantitative detection of all desired isoforms . In our experiments , we were able to demonstrate accurate isoform abundance reporting over >4 orders of magnitude for Dscam ( Figure 5B ) . At the low end of this range , however , near-cognate ligation events began to be problematic . As for sensitivity , we have been able to obtain detectable ligation products from as few as ∼900 ( 5 × 10−17 M; 0 . 05 fM ) target RNA molecules ( data not shown ) . Because the limit of detection for SeqZip is likely more than a single molecule , lack of detection of a particular isoform should only be interpreted as that isoform being below the SeqZip detection limit . Nonetheless , when properly controlled , SeqZip is a sensitive quantitative method for assessing complex isoform abundance patterns over a wide dynamic range . The easiest ‘complex’ form of alternative splicing for SeqZip analysis was Dscam1 , where alternative processing is limited to mutually exclusive exons ( i . e . , all spliced isoforms contain the same number of exons , and therefore , ligamer ligation events ) . However , many mammalian transcripts have more varied alternative splicing , including alternative 5′ and 3′ splice site usage and intron retention . Ligamer design against these types of alternative splicing quickly becomes unwieldy . For example , characterization of alternative transcriptional start and polyadenylation sites requires different terminal ligamers for each different start or polyadenylation site . Thus , while we were able to simultaneously assess two different types of alternative splicing in Fn1 ( exon inclusion/exclusion and alternative 3′ splice sites ) , other mammalian genes displaying even more numerous forms of alternative processing ( e . g . , Kcnma1 ) would require significantly more complicated ligamer pools . Indeed , there may be genes with splicing patterns that cannot be readily interrogated with a single ligamer pool capable of generating a unique ligation product for every possible isoform . In such cases , analysis using multiple ligamer pools each targeting a select region may still be advantageous for estimating splicing frequencies compared to more traditional methods like quantitative RT-PCR . To assist readers in designing their own ligamer pools , we have included a schematic of our ligamer design process for mouse Fn1 ( Figure 1—figure supplement 1 and Supplementary file 3 ) . For highly complex pools , this process can be automated by writing a simple Python , Perl , or R script specific to the problem being addressed . One potential future application of SeqZip is the detection of multiple single-nucleotide polymorphisms ( SNPs ) on a single molecule of a long RNA . By placing the ligation sites over each SNP , one could take advantage of the requirement by Rnl2 for complete complementarity at a ligation junction; mismatches would inhibit efficient ligamer joining ( Landegren et al . , 1988; Chauleau and Shuman , 2013 ) . Further , any sequence can be placed in between the two regions of target complementarity within each ligamer . Therefore , sequences for custom priming , restriction digestion , recombination , etc , can be introduced , allowing for quantification or subsequent manipulation of ligation products . Analysis of ligation products can even be multiplexed , allowing for simultaneous generation and analysis using internal controls . These applications and others are shown in Figure 1—figure supplement 2 . As demonstrated by our investigation of Dscam1 , SeqZip ligation products can be analyzed by high-throughput sequencing via incorporation of platform-appropriate priming sequences in the terminal ligamers or PCR primers or in the spacer sequences of internal ligamers . SeqZip could also be used to assess the integrity of very long RNAs , such as piRNA-precursor transcripts ( Li et al . , 2013 ) or mRNAs with extended 3′ UTRs ( Wang and Yi , 2013 ) . Thus , SeqZip , which retains sequence connectivity and overcomes template-switching artifacts of RT-based methods , represents a useful and adaptable new tool for detecting and quantifying numerous features of individual molecules of long RNA . The template sequence for the initial ligase screen ( Figure 2A ) was a 307 nt section of mouse DDX1 mRNA ( NM_134040 . 1; see Supplementary file 1 ) ; ssDNA and RNA templates were a synthetic oligonucleotide and in vitro transcript , respectively . Enzymes tested were Tth DNA ligase ( AB-0325; Thermo , Waltham , MA ) , Tsc DNA ligase ( Dlig 119; Prokaria , Reykjavik , Iceland ) , thermostable DNA ligase ( BIO-27045; Bioline , Taunton , MA ) , T4 DNA ligase ( M0202S; NEB , Ipswich , MA ) , Escherichia coli DNA ligase ( M0205S; NEB ) , Rnl2 ( M0239; NEB ) , SplintR ligase ( M0375; NEB ) . Templates and oligos were hybridized by heating to 65°C for 1 min , followed by slow cooling to room temperature . After buffer and enzyme addition , reactions were incubated at the manufacturer-specified optimal ligation temperature ( 16–65°C depending on enzyme ) for time indicated; denaturing polyacrylamide gels were quantified by phosphorimaging . Final ligation conditions in Figure 2A were ( left panel ) 1 . 5 μM ssDNA or RNA template , 5′-32P-labeled oligos ( 10 μM each ) , and 1 μl of neat indicated enzyme ( specific units varied according to the manufacturer and enzyme ) in manufacturer's recommended buffer; ( right panel ) 250 nM RNA template , 5′-32P-labeled oligos ( 500 nM each ) , and 10 U/μl Rnl2 or 20 U/μl T4 DNA ligase . Reactions in Figure 2B contained 1 . 25 μM DDX1 RNA template , 5 μM each 5′-32P-labeled oligo , and 10 U/μl Rnl2 were incubated for 4 hr at 37°C and separated on a 11 . 25% denaturing polyacrylamide gel . Reaction conditions in Figure 2C , D were as described in the SeqZip section ( see below ) , using indicated RNA templates in a background of 10 ng/μl total mouse embryo fibroblast ( MEF ) RNA . RNA templates in Figure 2C , D were runoff transcripts from PCR products generated with different oligo combinations ( Supplementary file 1 ) having partial complementarity to human eIF4A3 cDNA ( RefSeq: NM_014740 ) . For radioactive PCR using Taq Polymerase ( PN-M712; Promega , GoTaq Green Master Mix , Madison , WI ) , one PCR oligo was 32P-5′-end-labeled , and cycle numbers were confined to a range predetermined to yield a per cycle log2 linear increase in signal intensity ( typically 15–23 cycles ) . After resolution on a denaturing polyacrylamide gel , bands were quantified using a Typhoon imager ( GE Healthcare , Chicago , IL ) and the ImageQuant software package ( GE Healthcare ) . End point PCR was typically 35 cycles at a hybridization temperature 5°C below the lowest primer TM . End point PCR products were resolved on native 29:1 ( acrylamide: bis-acrylamide ) polyacrylamide gels , visualized by staining with SybrGold ( Invitrogen , Grand Island , NY ) , and also imaged/quantified as above . For SeqZip of endogenous CD45 , Fn1 , and Dscam1 mRNAs , individual ligamers were designed as follows ( Figure 1—figure supplement 1 ) . The 5′- and 3′-termini of each target sequence ( e . g . , one or multiple exons ) were obtained from online databases ( Ensembl and UCSD ) . For terminal ligamers , the length of complementarity necessary to obtain a predicted hybridization Tm nearest but not exceeding 65°C was calculated using the BioPerl Bio::SeqFeature::Primer Tm module with default [Na+] and [oligo] settings ( Allawi and Santalucia , 1997 ) . This complementary sequence was then appended to the desired PCR-primer hybridization sequence . For internal ligamers , the length of complementarity at each end ( generally 12–25 nt ) was adjusted to achieve a Tm of 60 ± 5°C for each end separately in order to maintain an overall length of ≤60 nt . End sequences were joined via a short spacer that could include a barcode . Ligamers were ordered from Integrated DNA Technologies , with or without a 5′ phosphate as required , and used directly in SeqZip reactions . Total RNA ( 200–800 ng per ligation reaction ) isolated from cells using Tri Reagent ( MRC ) was bound to Poly ( A ) Purist MAG magnetic beads ( Ambion AM1922; 2 . 25 μl slurry per ligation reaction ) according to the manufacturer's instructions . After removal of unbound RNA , ligamers ( 10 nM ( f . c . ) ) were hybridized to bead-bound poly ( A ) RNA in hybridization buffer ( 60 mM Tris-HCl , pH 7 . 5 at 25°C , 1 . 2 mM DTT , 2 . 4 mM MgCl2 , 480 μM ATP ) by heating samples to 62°C for 5 min in a thermocycler and then slow cooling to 45°C via a 3°C drop every 10 min . After 1 hr at 45°C , the temperature was again decreased 3°C every 10 min to 37°C , where samples were held until T4 RNA ligase 2 ( 2 U/μl ( f . c . ) , NEB , M0239 ) addition . At this point , the samples were in 1× ligation buffer ( 51 mM Tris-HCl , pH 7 . 5 at 25°C , 2 mM DTT , 5 mM KCl , 2 mM MgCl2 , 400 μM ATP , 3 . 5 mM ( NH4 ) 2SO4 , 5% ( vol/vol ) glycerol ) . After 8–16 hr at 37°C , beads were used directly for PCR using a polymerase appropriate for the downstream application ( e . g . , Taq for gel analysis , Herculase for sequencing ) . RT reactions in Figure 3 used SuperScript III ( 10 U/μl , Invitrogen ) at 55°C , 200 ng poly ( A ) selected RNA , and either anchored oligo ( dT ) or a gene-specific antisense primer . Fn1 amplicons were prepared using 12 cycles of Herculase II Fusion DNA Polymerase and primers targeting the sequence between the EDB and V regions ( Supplementary file 1 ) . Amplicons were submitted for library construction using The DNA Template Prep Kit 2 . 0 ( Pacific Biosciences ) and sequenced on a PacBio RS II . Circular consensus reads were aligned to an index of FN1 isoforms using BLAT . RT was performed using 5 μg total RNA , Superscript II ( Invitrogen ) , and random hexamers at 42°C for 1 hr . Strand-switching control experiments were performed by mixing plasmids encoding Dscam isoforms 1 . 33 . 9 , 12 . 32 . 9 , 1 . 24 . 6 , and 7 . 9 . 6 at 3:3:1:1 , 1:1:1:5 , and 1:1:1:1 . PCR with Phusion polymerase ( NEB ) ( annealing temperature , 55°C; 1 min extension ) was used to amplify cDNA or plasmids containing the region encompassing exons 4 , 6 , and 9 with exon 3 ( Not1Ex3For: TAT CGG CGG CCG CGG ACG TCC ATG TGC GAG CCG ) and exon 10 ( Ex10RevNot1: ATA TCG CGG CCG CGA GGA TCC ATC TGG GAG GTA ) primers . Both primers contained a 5′ end NotI restriction site . PCR products were gel purified and digested with NotI for 2 hr at 37°C , followed by a heat inactivation at 65°C for 20 min . The digested PCR products ( 0 . 5 μg ) were circularized in 500 μl 1× T4 DNA ligase buffer ( NEB ) with 1 μl T4 DNA ligase ( 0 . 8 U/μl , ( f . c . ) , NEB , M0202 ) at 18°C overnight . Inverse PCR was then performed using Phusion polymerase ( annealing temperature , 55°C; 30 s extension ) with primers specific to exons 7 ( PEex7Rev: CAA GCA GAA GAC GGC ATA CGA GAT CGG TCT CGG CAT TCC TGC TGA ACC GCT CTT CCG ATC TAT GAA CTT GTA CCA T ) and 8 ( PEex8For: AAT GAT ACG GCG ACC ACC GAG ATC TAC ACT GTT TCC CTA CAC GAC GCT CTT CCG ATC TAA GTG CAA GTC ATG G ) that contained Illumina paired-end clustering sequences . Libraries were gel purified , quantified via Nanodrop ( Thermo ) , and clustered on a Genome Analyzer IIx ( GAIIx ) paired-end flow cell on an Illumina cluster station using the standard clustering protocol . Sequencing was performed on an Illumina GAIIx by modifying the protocol for paired-end sequencing with an index read . Briefly , read 1 was performed for 24 cycles with a primer complementary to the 5′ end of exon 8 ( Ex8For: ACG ACG CTC TTC CGA TCT AAG TGC AAG TCA TGG ) . The flow cell was denatured to remove the exon 9 sequencing products , a primer complimentary to exon 3 ( Ex3For: CCC GGG ACG TCC ATG TGC GAG CCG ) was annealed , and read 2 sequenced for 12 cycles . Next , the flow cell was re-clustered using the paired-end protocol , and read 3 performed for 20 cycles using a primer complementary to exon 7 ( Ex7Rev: GAA CCG CTC TTC CGA TCT ATG AAC TTG TAC CAT ) . Base calling was performed from the raw images using the Firecrest , Bustard , and Gerald software modules of GAPipeline-1 . 4 . 0 and a matrix . txt file for a PhiX lane from a previous flow cell for calibration . This generated a single FastQ file per lane containing the three catenated reads from each cluster . The reads within the FastQ files were parsed to separate the three reads , the identity of each exon determined , and then the full isoform determined by matching to a database of known exon sequences . SeqZip ligation reactions were amplified via PCR ( Agilent , Herculase II Fusion DNA Polymerase , Catalog Number—600675 ) for 12 cycles using common primers . Reactions were resolved on a 5% polyacrylamide native gel , and DNA in the size range appropriate for full-length ligation products quantified by fluorescence imaging , cut and eluted from the gel , and precipitated . Equal DNA quantities based on the gel imaging were amplified for another 22 cycles using primers containing Illumina priming sequences with integrated barcodes . PCR products were purified ( 28104; Qiagen , QIAquick PCR Purification Kit ) and quantified using a Bioanalyzer 2100 ( Agilent ) and High-Sensitivity DNA chip . Samples were mixed and submitted for sequencing on the MiSeq instrument using the paired-end 250 nt read option . Sequencing data are available at Short Read Archive accession SRP043516 . All Dscam1 SeqZip products were shorter than 400 nt; therefore , paired-end MiSeq 250 nt reads contained overlapping 3′ sequences . Using these overlapping sequences , paired reads were combined into one sequence using the Paired-End Assembler ( pear , v . 0 . 8 . 1 ) and default options ( Zhang et al . , 2014 ) . An index of all possible Dscam1 ligamer combinations was created using a single PERL script that permuted all possible ligamer combinations with correct 5′ to 3′ exons 4 , 6 , and 9 ligamer arrangements . Paired MiSeq reads were aligned against this index using Bowtie2 v . 2 . 1 . 0 ( Langmead and Salzberg , 2012 ) in the very-sensitive-local mode and constrained using no-discordant to only look for reads where both pairs aligned to the same isoform . Using the SAMtools ( v . 0 . 1 . 19 ) software package ( Li et al . , 2009 ) , alignments were further filtered for alignments containing quality 31 and above ( -q 31 ) and read counts per isoform extracted . Count analysis was performed and graphs generated using R ( R Development Core Team , 2008 ) . For differential expression analysis , we treated each Dscam1 isoform as though it was its own gene . The percent use of individual exons for each cluster ( 4 , 6 , and 9 ) was determined . Expected use of all possible combinations of 4:6 , 4:9 , 6:9 , and 4:6:9 was calculated by multiplying the percentages of individual exon use . Expected use was compared to observed use of the equivalent combination . The DESeq differential gene expression R package ( Anders and Huber , 2010 ) was used to identify isoforms whose observed and expected abundances were ‘differential’ . Endogenous Dscam1 sequences were obtained from genomic build DM3 using BEDTools ( Quinlan and Hall , 2010 ) . All possible Dscam1 isoform sequences between exons 4 and 10 were assembled using a PERL script . Five hundred random isoforms were obtained and aligned using TCOFFEE ( Di Tommaso et al . , 2011 ) in the Jalview package ( Waterhouse et al . , 2009 ) . Consensus scores of alignments were exported and graphed in R . The same analysis was performed on Dscam1 ligation products , except ligamer sequences were used in place of endogenous exonic sequences . Error bars represent the standard error of the mean of experimental replicates . Errors were propagated from individual standard deviations according to standard methods ( Goodman , 1960; Natrella , 2012 ) .	A flow chart can show how an outcome can be achieved from a particular start point by breaking down an activity into a list of possible steps . Often , a flow chart contains several alternative steps , not all of which are taken every time the flow chart is used . The same can be said of genes , which are biological instructions that often contain many options within their DNA sequences . Proteins—which perform many roles in cells—are built following the instructions contained in genes . First , the DNA sequence of the gene is copied . This produces a molecule of ribonucleic acid ( RNA ) , which is able to move around the cell to find the machinery that can use the genetic information to make a protein . Genes and their RNA copies contain instructions with more steps—called exons—than are necessary to make a working protein , so extra exons are removed ( ‘spliced’ ) from the RNA copies . Different combinations of exons can be removed , so splicing can make different versions of the RNA called isoforms . These allow a single gene to build many different proteins . In fruit flies , for example , the different exons of the gene Dscam1 can be spliced into one of 38 , 016 unique RNA isoforms . Current technology only allows researchers to deduce the sequence of RNA molecules by combining sequences recorded from short fragments of the molecule . However , before splicing , RNA molecules tend to be much longer than this , so this restricts our understanding of the RNA isoforms found in cells . Here , Roy et al . devised and tested a new method called SeqZip to solve this problem . SeqZip uses short fragments of DNA called ligamers that can only stick to the sections of RNA that will remain after the molecule has been spliced . After splicing , the ligamers can be stuck together to make a DNA replica of the spliced RNA . The end product is at least 49 times shorter than the original RNA , so it is easier to sequence . In addition , the combinations of the ligamers in the DNA replica show which exons of a specific gene are kept and which ones are spliced out . To test the method , Roy et al . studied a mouse gene that has six RNA isoforms . SeqZip reduced the length of the RNA by five times and made it possible to measure how frequently the different isoforms naturally arise . Roy et al . also used SeqZip to work out which isoforms of the Dscam1 gene are used at different stages in the life of fruit fly larvae . SeqZip can provide insights into how complex organisms like flies , mice , and humans have evolved with relatively few—a little over 20 , 000—genes in their genomes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2015	Assessing long-distance RNA sequence connectivity via RNA-templated DNA–DNA ligation
Rhythmic actions benefit from synchronization with external events . Auditory-paced finger tapping studies indicate the two cerebral hemispheres preferentially control different rhythms . It is unclear whether left-lateralized processing of faster rhythms and right-lateralized processing of slower rhythms bases upon hemispheric timing differences that arise in the motor or sensory system or whether asymmetry results from lateralized sensorimotor interactions . We measured fMRI and MEG during symmetric finger tapping , in which fast tapping was defined as auditory-motor synchronization at 2 . 5 Hz . Slow tapping corresponded to tapping to every fourth auditory beat ( 0 . 625 Hz ) . We demonstrate that the left auditory cortex preferentially represents the relative fast rhythm in an amplitude modulation of low beta oscillations while the right auditory cortex additionally represents the internally generated slower rhythm . We show coupling of auditory-motor beta oscillations supports building a metric structure . Our findings reveal a strong contribution of sensory cortices to hemispheric specialization in action control . Functional asymmetries between the two hemispheres are an intriguing principle of brain organization . On the behavioral level , these become most evident in the way humans use their hands . In tasks requiring movements of both hands , right-handers typically use the right hand for the faster , dynamic movements while the left hand is used for slower movements , or even static control of hand position ( Sainburg , 2002; Swinnen and Wenderoth , 2004; Serrien and Sovijärvi-Spapé , 2015 ) . Cutting bread or hammering a nail into the wall represents everyday examples for such functional asymmetries . In the lab , finger tapping can be used to detect hemispheric asymmetries related to this phenomenon . Typically , the right hand taps relative higher tapping frequencies more precisely than the left hand even in bimanual monofrequent finger tapping ( Repp , 2005; Ivry , 1996; Peters , 1980 ) . Conversely , the left hand taps relative lower tapping frequencies more precisely than the right hand ( Pflug et al . , 2017 ) . This suggests the left hemisphere preferentially controls relative higher tapping frequencies and the right hemisphere preferentially controls relative lower tapping frequencies , but the origins of such hemispheric asymmetries are not known . There are several proposals on the origins of functional differences between the hemispheres ranging from specialized processing in the sensory domain , lateralized sensorimotor interactions , asymmetric motor control , to domain-general frameworks on hemispheric dominance ( Kimura , 1993; Minagawa-Kawai et al . , 2007; Toga and Thompson , 2003; Kell and Keller , 2016 ) . Behavior could benefit from parallel processing of different aspects of complex stimuli and/or movement planning in the left and right hemisphere ( Serrien et al . , 2006 ) . Influential theories suggest differential sensory processing of relative frequencies either in the spectral or the temporal domain ( Ivry and Robertson , 1998; Flevaris and Robertson , 2016; Poeppel , 2003 ) as computational bases of hemispheric specialization . However , empirical studies in which spectral or temporal aspects of the sensory input were parameterized did not always support those theories ( Luo et al . , 2007; Giraud and Truy , 2002; Boemio et al . , 2005 ) . This could represent a consequence of the fact that brain activity is only subtly lateralized during perceptual tasks . However , functional lateralization is thought to be amplified once a motor output is required ( Ivry and Robertson , 1998; Keller and Kell , 2016 ) . To dissociate the specific contributions of the sensory and motor systems to functional lateralization of hand control , we performed two imaging studies using functional magnetic resonance imaging ( fMRI ) and magnetoencephalography ( MEG ) . The study design excluded that condition effects resulted from sensory stimulus features or differential effector use . In an auditory-paced finger tapping paradigm , participants were asked to tap bimanually to auditory beats . Tapping to every auditory beat ( 2 . 5 Hz ) was defined as the fast tapping condition while tapping to every fourth auditory beat ( beat position four ) represented slow tapping at 0 . 625 Hz ( see Figure 1 ) . Both frequencies fall into the natural range of finger movements but represent different ends of the spectrum ( Parncutt and Cohen , 1995; London , 2012; Drake and Palmer , 2000; Repp , 2003 ) . While in the fast tapping condition , the fast auditory beat was the only rhythm that was processed and used for auditory-motor synchronization , this faster rhythm served as a timing signal to generate a slower rhythm in the slow tapping condition . The slow tapping condition was of primary interest in our study , because during slow tapping two interrelated rhythms had to be represented in parallel , a condition that could potentially reveal hemispheric specialization for controlling rhythms of different relative frequencies ( Ivry and Robertson , 1998 ) . A prior behavioral study ( Pflug et al . , 2017 ) suggested that representing a relative slow rhythm in parallel to a faster one should reveal the contribution of the right hemisphere to hand control . While we used fMRI to detect whether auditory or motor regions show a more pronounced lateralization profile , which answers the question of different contributions of the sensory and motor systems to hemispheric specialization , we used MEG to identify hemispheric differences in brain rhythms associated with finger tapping in a time-resolved manner and to investigate time resolved directed connectivity between auditory and motor association cortices ( Figure 2 ) . Movement is known to suppress beta oscillations and to increase activity in the gamma range ( Muthuraman et al . , 2012; Tamás et al . , 2018; Pfurtscheller et al . , 2003; Pfurtscheller and Lopes da Silva , 1999; Engel and Fries , 2010 ) . Yet , neural oscillations are , particularly in the beta range , not only reflected current motor state but also implicated in internal timing , especially during rhythm processing , and are amplitude-modulated during rhythm perception and production not only in the motor and supplementary motor cortex , but also in the auditory and auditory association cortex ( Arnal and Giraud , 2012; Doelling and Poeppel , 2015; Nobre et al . , 2007; Fujioka et al . , 2015; Meijer et al . , 2016; Morillon et al . , 2014; Kilavik et al . , 2013; Kulashekhar et al . , 2016; Morillon and Baillet , 2017; Iversen et al . , 2009 ) . Comparing neural oscillations during slow and fast rhythmic finger tapping may reveal the way the brain represents the two different rhythms in parallel . Amplitude modulations of beta oscillations should differ between functional homologues in case there were hemispheric processing differences in timing of relative tapping frequencies . We hypothesized that motor and/or auditory cortices may not only differ in overall beta power but also in terms of their degree of representing the slow and fast rhythms in the temporal modulation of beta power ( Fujioka et al . , 2015; Morillon and Baillet , 2017 ) . If the predictions from the signal-driven hypotheses on hemispheric specialization ( Ivry and Robertson , 1998 ) hold true , we specifically expected the right auditory cortex to more strongly represent the slow rhythm and the left auditory cortex the fast rhythm during slow finger tapping , the condition that comprised both rhythms . A left dominance in hand motor control based on left-lateralized sequencing skills ( Kimura , 1993; Haaland et al . , 2004 ) , instead , would predict control of both rhythms by the left hemisphere . Functional specialization of the two hemispheres has not only been linked with lateralized regional activation , but also with the formation of lateralized functional networks of regions ( Stephan et al . , 2003; Keller and Kell , 2016 ) . We thus investigated whether auditory-motor interactions between the right and left auditory association cortex and the supplementary motor area ( SMA ) , a motor association area highly involved in the internal generation of sequences ( Kotz et al . , 2009; Merchant et al . , 2013; Merchant et al . , 2015; Crowe et al . , 2014 ) , were modulated differently in the two hemispheres when representing the slow in addition to the fast rhythm . We hypothesized that auditory-motor effective connectivity may differ between the two hemispheres in terms of connection strength in the beta range . Our results identify the left auditory association cortex as the primary cortical area that represents the relative fast auditory rhythm while the right auditory association cortex is recruited to represent the relative slow tapping rate in an amplitude modulation of low beta oscillations . In contrast , motor cortices and the cerebellum only represent the temporal regularities of the motor output . Representing the slow in addition to the fast rhythm increases low beta functional connectivity from the right auditory association cortex to the SMA in parallel to increased BOLD activation of these regions . Further , stronger and bidirectional low beta functional connectivity between the SMA and the left auditory association cortex may privilege the left hemisphere for hierarchical integration of interrelated rhythms in a Gestalt ( Iversen et al . , 2008; Swinnen and Wenderoth , 2004 ) . Timing variability was defined as standard deviation of the absolute distance between the actual and target inter-tap-intervals ( Pflug et al . , 2017 ) . This measure characterizes internal timing well ( Repp , 2005 ) . A two-factor repeated measures analysis of variance ( ANOVA ) on timing variability across condition ( slow and fast bimanual tapping ) and hand ( left and right ) revealed expectedly an interaction between condition and hand ( F ( 1 , 41 ) = 10 . 23 , p=0 . 003 ) . Fast tapping was more precise with the right than the left hand ( right hand: mean ( M ) =13 . 29 ms , standard deviation ( SD ) = 2 . 74 ms; left hand: M = 14 . 41 ms , SD = 2 . 78 ms ) and slow tapping was more precise with the left compared to the right hand ( right hand: M = 33 . 78 ms , SD = 18 . 42 ms; left hand: M = 31 . 70 ms , SD = 14 . 97 ms , see Figure 3 ) . No main effect of hand was observed ( F ( 1 , 41 ) = 0 . 760 , p=0 . 388 ) . There was a main effect of condition ( F ( 1 , 41 ) = 108 . 54 , p<0 . 001 ) with an overall higher precision in fast ( M = 13 . 85 ms , SD = 2 . 80 ms ) compared to slow tapping ( M = 32 . 74 ms , SD = 16 . 90 ms ) ( Repp , 2005 ) . In fMRI , compared to silent baseline , both slow and fast bimanual tapping showed comparable activation patterns of bilateral regions involved in auditory-paced finger tapping , including the primary hand motor cortex , the dorsal and ventral premotor cortex , SMA , the cingulate motor area , parietal operculum , superior temporal cortex including the auditory cortex , posterior superior temporal gyrus and sulcus , the putamen , thalamus , and the superior cerebellum ( p<0 . 05 , FWE cluster-level corrected , see Figure 4 ) . Activity in the auditory association cortex was right lateralized during slow compared to fast tapping ( p<0 . 001 , FWE cluster-level corrected , cluster size 395 voxels ) and this was the only cortical patch that showed lateralized activity ( all other p>0 . 05 , FWE cluster-level corrected ) . Generation of the slow rhythm activated additionally the bilateral fronto-mesial cortex including the SMA ( see Figure 5 and Table 1 ) . fMRI revealed a higher activation in the SMA and right auditory association cortex for slow compared to fast tapping . Thus , MEG power spectral densities of both sources , as well as left auditory association cortex , were tested for differences between simple auditory-motor synchronization and additional internal generation of the slow rhythm . In all three areas ( SMA and both auditory association cortices ) , slow compared with fast tapping increased power in the low [14–20 Hz] and high beta band [21–30 Hz] but not in the delta , theta , alpha , or gamma range ( see Table 2 ) . Condition differences were stronger in the low compared to the high beta band in the SMA and in the right A2 ( SMA: t ( 16 ) = 3 . 033 , p=0 . 002 , p=0 . 818 , right A2: t ( 16 ) = 1 . 907 , p=0 . 046 ) , but not in the left A2 ( left A2: t ( 16 ) = 0 . 228 ) , a region that did not show condition effects in the fMRI . This confirms a more pronounced role of the low compared to the high beta band in rhythm generation ( Gompf et al . , 2017; Fujioka et al . , 2015 ) . Further analyses were therefore focused on the low beta band . The internal generation of the slow rhythm during slow tapping increased low beta power compared to fast tapping , during which beta power was strongly suppressed , in both auditory association cortices ( main effect of condition F ( 1 , 16 ) = 7 . 267 , p=0 . 011 , permutation ANOVA on mean values over the low beta band ) . Notably , low beta power condition differences between slow and fast tapping were larger in the right compared to the left auditory cortex ( interaction between condition and hemisphere F ( 1 , 16 ) = 3 . 460 , p=0 . 045 ) possibly explaining the right-lateralized activation of this cortical region in fMRI . During slow tapping , low beta power was maximal at beat position one and decreased to maximal beta suppression at the tap on beat position four in both the left and right auditory association cortex ( red curves in Figure 6 ) . While this temporal modulation that reflected the rate of the internally generated slow rhythm was observed in both the left and the right auditory association cortex , the additional temporal modulation at the relative fast auditory beat frequency ( 2 . 5 Hz ) was stronger in the left than in the right auditory association cortex ( red curve in Figure 6 , upper left panel ) . In the spectral domain , this translated to a stronger temporal modulation at the fast auditory beat rate in the left compared to the right auditory cortex during slow tapping ( t ( 16 ) = 1 . 8956 , p=0 . 037 ) . In contrast , low beta power modulation at the slow tapping rate was stronger in the right compared to the left auditory association cortex ( t ( 16 ) = 1 . 636 , p=0 . 040 ) . In the fast tapping condition , during which participants actively tapped to every auditory beat , beta power was maximally suppressed during the entire sequence of four beats ( blue curves in Figure 6 , upper panels ) . Consequently , decreases at the single beat positions were less pronounced ( Kilavik et al . , 2013 ) . Power in the low beta band was also less suppressed in the SMA during slow compared to fast tapping ( t ( 16 ) = 2 . 0917 , p = 0 . 002; dependent sample permutation t-tests on mean values over the low beta band ) . In contrast to the auditory cortices , temporal modulation of the low beta power envelope reflected the actual tapping rates ( see Figure 7 ) . While auditory-motor synchronization in fast tapping decreased low beta power at every beat position , low beta power in the SMA decreased linearly from start of the sequence to the fourth beat position in the slow tapping condition . Consequently , the SMA spectrum contained a strong peak around 0 . 625 Hz ( amplitude = 0 . 357 a . u . ) , but no peak at 2 . 5 Hz during slow tapping . During fast tapping there was a strong modulation at 2 . 5 Hz ( amplitude = 0 . 695 a . u . ) and only a very weak modulation around 0 . 625 Hz ( amplitude = 0 . 058 a . u . ) . We further investigated whether the signal in the primary hand motor cortices and the cerebellum resembled the one observed in the SMA . Low beta amplitude modulation at auditory beat frequency during slow tapping did not differ between primary hand motor areas and the SMA ( left M1: t ( 16 ) = 1 . 217 , p=0 . 133 , right M1: t ( 16 ) = 0 . 910 , p=0 . 182 ) , between the left and right hand motor cortex ( t ( 16 ) = 0 . 899 , p=0 . 332 ) or between the cerebellum and the SMA ( left cerebellum t ( 16 ) = 1 . 386 , p=0 . 095 , right cerebellum t ( 16 ) = 1 . 223 , p=0 . 110 ) . Together , in contrast to the auditory association cortices , the primary hand motor cortices , cerebellum and the SMA coded solely the motor output in the amplitude modulation of low beta oscillations . If indeed the low beta power modulation reflects internal timing during slow tapping , it should predict timing variability in single trials . To investigate low beta band differences between short and long inter-tap-intervals during slow tapping , a permutation cluster statistic was used to check for effects of timing variability . Low beta power modulation in the left auditory cortex did not contribute to timing variability during slow tapping ( Figure 8 , leftmost panel ) . In the right auditory association cortex , the amplitude modulation during too long inter-tap-intervals was larger compared to the power modulation during too short inter-tap-intervals in the sense that low beta power was enhanced at beat position one when participants produced a too long inter-tap-interval ( Figure 8 , left middle panel , significant cluster at 560–660 ms , p=0 . 042 ) . In the SMA , low beta amplitude at beat position one did not influence performance during slow tapping significantly . Instead , too long inter-tap-intervals during slow tapping were associated with a longer low beta suppression at the end of the sequence coinciding with the delayed tap ( Figure 8 , right middle panel , significant cluster at 1400–1470 ms , p=0 . 033; significant cluster at 1560–1800 ms p=0 . 001 ) . During fast tapping , low beta amplitude coded performance in the SMA . A permutation analysis on fast tapping sequences revealed amplitude coding with enhanced beta power modulations for too long inter-tap-intervals and reduced beta power modulations for too short inter-tap-intervals ( Figure 8 , rightmost panel , p=0 . 002 ) . Due to the maximal low beta suppression in auditory cortices during fast tapping , no significant difference between too long and too short inter-tap-intervals was observed in these regions . To study the contribution of auditory-motor interactions to the right-lateralized processing of the slow rhythm during slow compared to fast tapping , time-resolved partial-directed coherence ( TPDC ) was calculated between the secondary auditory cortices and the SMA and vice versa on MEG source level data . This measure is insensitive to local power differences ( Kaminski et al . , 2016; Tsapeli and Musolesi , 2015; Nalatore et al . , 2007; Muthuraman et al . , 2018 ) and is ideally suited to investigate time-resolved directed functional connectivity . Both slow and fast tapping increased TPDC between the auditory cortices and the SMA in the low beta and mid gamma range with strongest effective connectivity from the left auditory association cortex to the SMA ( see Figure 9 ) . To reveal directed connectivity when representing two rhythms instead of one rhythm , we focused the connectivity analyses on the contrast between slow and fast tapping and restricted them again to the low beta band ( Gompf et al . , 2017; Fujioka et al . , 2015 ) . A two-factor repeated measures ANOVA on averaged connectivity in the low-beta band across hemisphere ( left and right ) and direction ( auditory to motor and motor to auditory ) revealed a main effect of hemisphere ( F ( 1 , 16 ) = 7 . 00 , p=0 . 018 ) with stronger condition differences between slow and fast tapping in the left ( M = 0 . 007 , SD = 0 . 0048 ) compared to the right ( M = 0 . 003 , SD = 0 . 0054 ) hemisphere . This surprising effect was accompanied by a close-to-threshold interaction between direction and hemisphere ( F ( 1 , 16 ) = 3 . 83 , p=0 . 068 ) . While the connections from left A2 to the SMA ( t ( 16 ) = 4 . 174 , p=0 . 002 ) , from the right A2 to the SMA ( t ( 16 ) = 2 . 988 , p=0 . 005 ) , and the one from the SMA to the left A2 ( t ( 16 ) = 3 . 385 , p=0 . 001 ) increased low-beta connectivity for slow compared to fast tapping , the connection from the SMA to the right A2 was not enhanced for slow compared to fast tapping ( t ( 16 ) = −0 . 882 , p=0 . 392 , see Figure 10 , left panel ) . In sum , slow compared with fast tapping increased interactions in the low beta band between both the left and right auditory association cortex and the SMA and between the SMA and the left auditory association cortex ( see Figure 10 , right panel ) with an overall stronger connectivity in the left compared to the right hemisphere . We investigated individual timing variability in the slow tapping condition for a correlation with directed connectivity contrasts for slow > fast tapping . An increased connection strength in the connection from the SMA to the left A2 during slow compared to fast tapping correlated twith timing variability in the sense that it reduced timing variability of the right hand when tapping slowly ( r = −0 . 490 . p=0 . 04 ) . All other correlations were not significant ( p>0 . 05 ) . Auditory-motor interactions were not only structured in frequency , but also in time ( see Figure 9 ) . The effective connectivity in the low beta range for slow compared to fast tapping was amplitude-modulated by a theta rhythm at 6 . 5 Hz in all connections except for the connection from the SMA to the right auditory association cortex ( for statistics see Table 3 ) , the connection that also did not show significant low beta band condition effects . There was an additional modulation of effective connectivity in the low beta range by an alpha rhythm at 10 . 5 Hz in all connections except for the connection from the right auditory cortex to the SMA . There was no other rhythmic modulation of low beta effective connectivity for slow compared to fast tapping ( all p>0 . 05 ) . A functional lateralization in terms of differences in activation of functional homologues was only observed in the auditory and not in the motor association cortices . Together with the cerebellum , motor cortices rather mirrored the actual motor output with a stronger BOLD signal in the bilateral SMA associated with reduced beta suppression during internal timing compared to auditory-motor synchronization ( Gompf et al . , 2017 ) . Beta suppression during slow tapping was not maximal such that ceiling levels cannot explain missing lateralization in the SMA . Lateralization in motor association cortices is often observed in the lateral dorsal premotor cortex , particularly during asymmetric or complex bimanual actions as compared to the symmetric finger taps used in this study ( Haslinger et al . , 2002; Hlustík et al . , 2002; Hardwick et al . , 2013 ) . The lateral dorsal premotor cortex is activated by polyrhythmic external stimuli while internally generated rhythms activate the SMA ( Swinnen and Wenderoth , 2004 ) . The lack of lateralization effects in the motor cortices in our study suggests that the observed functional lateralization on the behavioral level was timing-related and not related to bimanual motor coordination ( Serrien et al . , 2003 ) . Indeed , right lateralization of auditory association cortex activity was related to improved left hand timing despite a bilateral activation of the SMA in slow tapping . No other brain areas beyond the bilateral SMA and the right auditory association cortex activated significantly for slow compared with fast tapping , a condition that could have been associated with increased counting effort compared to the fast tapping condition . Counting during perceptual grouping activates the intraparietal sulcus , dorsolateral prefrontal and inferior frontal cortex ( Ansari , 2008 ) , none of which was activated in our study . Our results confirm a strong contribution of the sensory cortices to the lateralization of action control , as suggested by the sensory-driven hypotheses on hemispheric specialization ( Minagawa-Kawai et al . , 2011; Ivry and Robertson , 1998 ) . Because both auditory cortices receive the same auditory input ( the fast auditory beat rate ) in both slow and fast tapping , the nearly absent fast auditory beat rate representation in the right auditory association cortex likely constitutes the consequence of dynamic attention to every fourth auditory stimulus . Notably , this filtering is performed in the temporal domain , suggesting that the right auditory association cortex actively selects the relevant auditory beats for slow rhythm generation . This is reminiscent of the dynamic attending theory ( Jones , 1987 ) which proposes that during perception , auditory cortex oscillations are aligned to rhythmic auditory input to select behaviorally relevant input ( Schroeder and Lakatos , 2009 ) . Beta power in the right but not in the left auditory association cortex explained timing variability during slow tapping . The internal generation of a too slow rhythm was associated with an even larger amplitude modulation with enhanced beta power at beat position one . The same amplitude coding was observed in the SMA , although only during fast tapping . In our experiment , time information was coded in the amplitude of beta oscillations . Recently , time information during rhythmic finger tapping in the subsecond range has also been related to the amplitude of abstract representations of the SMA neural population dynamics in non-human primates ( Gámez et al . , 2019 ) . Neither did the neuronal population dynamics scale in time , nor was the slope of the beta power modulation in our study modulated by slow vs . fast finger tapping , which suggests time indeed is coded in amplitude , at least during rhythmic finger tapping ( see Fujioka et al . , 2015 and Wang et al . , 2018 for contrasting views ) . Assuming time information is coded accumulator-like ( Ivry and Richardson , 2002 ) in the power difference between minimal and maximal beta suppression we may state that amplitude coding identifies brain regions with different preferred time intervals . The association between timing variability and amplitude coding in the motor association cortex during fast tapping and the relationship between timing variability and amplitude coding in the right but not left auditory association cortex during slow tapping confirms that the brain uses the motor system for subsecond timing and non-motor cortices for suprasecond timing ( Morillon et al . , 2009 ) . During slow tapping the SMA coded solely the information on the actual timing of the tap in the latency of the maximal suppression at beat position four . This emphasizes the contribution of the right auditory association cortex to the internal generation of a slow , supra-second rhythm . Low beta power in the cerebellar sources did only mirror the motor output in our experiment . However , subcortical regions including the cerebellum , basal gangia and thalamus , make part of a dedicated neural timing system and likely provide more than motor timing information ( Kotz and Schwartze , 2011 ) . We cannot rule out that other cerebellar sources with less cortico-muscular coherence compared to the cerebellar sources identified here show internal timing-related profiles . Although the right auditory association cortex was more strongly activated by the slow than the fast tapping condition , also the left auditory association cortex represented the slow tapping rate in a beta power decrease from start to the end of a sequence of four auditory beats . This raises the question why the right auditory association cortex was additionally recruited for slow tapping and associated with performance if all necessary temporal information could be decoded from left auditory association cortex . Note that in our study participants tapped the slow rhythm on every fourth auditory beat , which represents a syncopated rhythm with a 270° phase delay in relation to the 4/4 standard meter that was introduced by the four priming auditory beats prior to each trial . In a previous behavioral experiment , we showed that the right hemisphere advantage for the control of slow tapping depended on syncopation , because it was not observed when participants tapped at auditory beat position one when tapping slowly ( at 0° phase difference relative to the standard meter; Pflug et al . , 2017 ) . Non-syncopated slow tapping at beat position one reflects the overlearned 4/4 meter that constitutes the standard meter in Western culture ( London , 2012 ) . In dynamic pattern theory , 0° phase angles represent more stable dynamical states compared to 270° phase differences ( Zanone and Kelso , 1992 ) . Perceiving syncopated compared to non-syncopated rhythms activates the right more than the left auditory association cortex ( Herdener et al . , 2014 ) . However , our results are not a consequence of stimulus features like accentuation or lengthening , because the auditory stream used in this experiment consisted of identical beats . They rather indicate that the involvement of the right auditory cortex is not a direct consequence of the increased complexity of rhythmic grouping during syncopated slow tapping compared to simple auditory-motor synchronization during fast tapping . Instead , the MEG and behavioral results demonstrate that the two rhythms are not represented randomly in the left and right hemispheres , but rather systematically with a stronger representation of the relative fast rhythm in the left and of the relative slow rhythm in the right hemisphere . We interpret our observation in such a way that syncopated rhythms are represented separately by the two hemispheres as long as they are not yet hierarchically integrated in a Gestalt based on experience . From a dynamic pattern theory perspective , the 270° phase angle during syncopated tapping induces competition between attractor states with a strong tendency to tap at 0° phase difference relative to the standard meter ( Swinnen , 2002 ) . This tendency could be reduced by increasing the energy needed for a phase transition from tapping the instructed 270° phase angle to the tapping along the standard meter . Note that in motor as well as auditory cortices , beta amplitude was minimal at beat position four and maximal at beat position one during slow tapping , which decreases tapping probability at beat position one . Competition between the standard meter and the phase-shifted slow tapping rhythm of same frequency could be reduced by increasing the physical distance of their representations . This would permit parallel representations of competing attractors . The brain could potentially solve this problem by representing the standard meter and its relationship to the fast auditory beat rate in the right hemisphere and the slow tapping rhythm in the left hemisphere . The fact that the brain does not follow this path and rather represents the internally generated phase-shifted slow rhythm in the right auditory association cortex and the fast auditory beat rate in the left auditory association cortex speaks in favor of different temporal filters in auditory association cortex as sources of hemispheric specialization . In addition , preferential binding of rhythms with different frequencies in the left hemisphere may contribute to functional lateralization . The left hemisphere outperforms the right hemisphere in local binding ( Flevaris et al . , 2010 ) , which is critical for beat and meter integration . Integrating fast rhythms and slow rhythms with 0° phase angle relative to the standard meter may facilitate hierarchical binding in a Gestalt ( Zanone and Kelso , 1992 ) which could bias meter processing to the left hemisphere . This may explain the numerous reports on an involvement of the left hemisphere in rhythm production in professional musicians ( Vuust et al . , 2006; Kunert et al . , 2015; Herdener et al . , 2014 ) and explain empirical findings that ostensibly support the motor-driven hypotheses of left hemispheric dominance ( Kimura , 1993; Haaland et al . , 2004 ) . Tapping rhythms were most strongly represented in the low beta band in both motor and auditory cortices . Effects in the beta band have often been found in tasks that require synchronization of large-scale brain networks ( Gehrig et al . , 2012; Roelfsema et al . , 1997 ) and more recently have been associated with top-down signals in hierarchically organized cortical networks ( Bastos et al . , 2015; Fontolan et al . , 2014 ) . Beta oscillations are particularly strong in the motor system including the basal ganglia , which also play an important role in rhythmic motor behavior ( Kotz et al . , 2009 ) . During finger tapping , spike-field coherence in the striatum is stronger for beta compared to gamma oscillations and beta oscillations are more strongly related to internal rhythm generation than with sensory processing during tapping ( Bartolo et al . , 2014 ) . Consequently , an important role of auditory-motor interactions in the beta range was expected and , indeed , interactions between the auditory and motor association cortices were strongest in the low beta band . However , the internal representation of the slow tapping rhythm in the right auditory association cortex was not associated with additional top-down signals in the low beta band from the SMA to the right auditory association cortex compared to fast tapping during which no additional rhythm was represented . In the right hemisphere , slow tapping increased information flow in the low beta band only from the auditory association cortex to the SMA , possibly to provide slow rhythm information . The SMA received also stronger low beta input from the left auditory association cortex during slow compared to fast tapping which could reflect the effort to integrate the slow with the fast rhythm that was more strongly represented in the left auditory association cortex . The SMA could thus be interpreted as the midline structure that integrates rhythm information from both auditory association cortices and times tapping accordingly . Yet , in contrast to the right hemisphere , slow tapping also strengthened the top-down connection from the SMA to the left auditory association cortex compared to fast tapping . This left-lateralized top-down connection was the only connectivity in our study that correlated with tapping accuracy . The stronger the connection was from the SMA to the left auditory association cortex during slow compared with fast tapping , the more precise participants tapped with their right hand in the slow tapping condition . This suggests that the right hand , that taps fast rhythms more precisely than slow rhythms , may benefit from bidirectional auditory-motor interactions in the left hemisphere when tapping slowly . Together with the overall stronger directed connectivity between the left auditory association cortex and the SMA , this auditory-motor loop may facilitate rhythm integration in the left hemisphere ( Nozaradan et al . , 2015 ) . Auditory-motor loops can also be used to facilitate perceptual timing in the absence of overt motor behavior . Such a motor facilitation is efficient when estimating time periods of below one to two seconds ( Morillon et al . , 2009; Rao et al . , 1997; Funk and Epstein , 2004 ) . Auditory rhythmic sampling without overt motor behavior involves beta signals from the left lateralized motor cortex to the auditory association cortex ( Morillon and Baillet , 2017 ) . This finding confirms the left-lateralized top-down connection in beta connectivity between the motor and auditory association cortex found in our study even in the absence of overt movement . Beta signals associated with slow compared to fast tapping between the auditory association cortices and the SMA and vice versa were modulated by a theta rhythm . Fronto-temporal theta oscillations have been associated with auditory-motor and multisensory integration ( van Atteveldt et al . , 2014 ) and more specifically support auditory working memory ( Albouy et al . , 2017 ) , speech perception ( Assaneo and Poeppel , 2018 ) , and speech production ( Behroozmand et al . , 2015 ) . The auditory-motor theta rhythm observed in this finger tapping study was observed at a peak frequency of 6 . 5 Hz , which corresponds to the frequency at which also the velocity of slow finger movements is modulated ( Gross et al . , 2002 ) . This is slightly higher than the auditory-motor theta rhythm associated with speech processing , which peaks at 4 . 5 Hz ( Assaneo and Poeppel , 2018 ) , potentially due to the higher natural resonance frequencies of the fingers compared to the jaw ( Junge et al . , 1998 ) . The observed functional differences between the hemispheres during auditory-paced finger tapping remind the asymmetries observed during speech processing . During speech perception , the syllable rate in the theta range serves as a strong acoustic cue that entrains oscillations in the bilateral auditory association cortex and induces auditory-motor interactions in this frequency range ( Assaneo and Poeppel , 2018 ) . Auditory-motor interactions are left-lateralized both during speech perception ( Mottonen et al . , 2014; Murakami et al . , 2015; Hickok , 2015 ) and speech production ( Kell et al . , 2011; Keller and Kell , 2016 ) suggesting left-lateralized auditory-motor loops . In both motor and auditory association cortices , binding of speech-relevant rhythms via cross-frequency coupling is left lateralized in fronto-temporal cortices during speech perception ( Gross et al . , 2013 ) . These observations suggest left-lateralized auditory-motor loops could improve rhythm integration by cross-frequency coupling both during speech perception and production . Indeed , reduced auditory-motor coupling in the left hemisphere of people who stutter is associated with overt deficits in controlling speech rhythm ( Neef et al . , 2015; Chang and Zhu , 2013; Kell et al . , 2018 ) . The deficit in rhythm integration in people who stutter is associated with an over-recruitment of the right hemisphere during speech production that is reduced upon recovery ( Kell et al . , 2009; Kell et al . , 2018 ) . The right over-activation during speaking may be regarded as a strategy to separate competing attractors that arise from insufficient auditory-motor mapping in the left hemisphere ( Hickok et al . , 2011 ) . It is interesting to note in this context that this speech disorder that has been associated with basal ganglia dysfunction ( Alm , 2004 ) shows abnormal beta oscillations associated with timing both in speech and non-speech tasks ( Etchell et al . , 2016; Etchell et al . , 2014 ) . Future research will need to elucidate the commonalities and differences between the functional lateralization of hand control and the lateralization of speech production . We show here that representing two rhythms during syncopation lateralizes processing of the relative faster rhythm to the left and the relative slower rhythm to the right hemisphere . Auditory association cortices filter adaptively the preferred temporal modulation rate and send this time signal to the SMA for motor output coordination . The filter is relative rather than absolute , meaning that the hemispheres do not lose the complementary information , but nevertheless represent preferentially different rhythms . An additional top-down communication from the SMA to the left auditory association cortex may privilege the left hemisphere in integrating multiple rhythms in a multiplexed Gestalt , which likely represents a prerequisite for more complex cognitive functions like speech processing . Twenty-five participants ( 10 males; aged 19–31 years; M = 24 years ) were included in the fMRI study; seventeen participants ( six males , aged 21–38 years; M = 26 years ) in the MEG study . Number of participants was chosen based on a literature research for finger tapping experiments in MEG/EEG studies and fMRI , respectively . Participants had normal or corrected-to-normal visual acuity , normal hearing , no neurological deficits and were right-handed according to self-reports and their laterality index based on the Edinburg inventory of manual preference ( fMRI: M = 86; MEG: M = 89; Oldfield , 1971 ) . Participants performed a test run of approximately five minutes before measurement to become familiar with the task . All participants gave written informed consent prior to the study and were paid for participation . Experimental procedures were approved by the ethics committee of the medical faculty of Goethe university ( GZ 12/14 ) , and are in accordance with the declaration of Helsinki . In every trial of this auditory paced finger tapping paradigm , 36 auditory beats ( 1600 Hz , 2 ms ) were presented with a constant inter-beat-interval of 400 ms ( 2 . 5 Hz , 210 bpm ) . Participants were asked to tap with their index fingers at two different rates to these beats . No auditory feedback was provided . Thus , auditory input did not differ between conditions . In the fast tapping condition , participants tapped to every auditory beat . In the slow tapping condition , participants were instructed to iteratively count four beats internally and tap only on beat position four ( Figure 1 ) . Therefore , tapping to the 36 auditory beats resulted in 36 taps when tapping the fast rate and nine taps when tapping the slow rate during each 15 second-long trial . Eight different tapping conditions were performed . In four unimanual conditions participants tapped with one hand ( left or right ) either the fast or the slow tapping rate . Four bimanual conditions were either performed monofrequent , in which both hands tapped the same rate or multifrequent , in which one hand tapped the fast rate and the other hand the slow rate . Here , we report the two bimanual monofrequent conditions during which both hands were engaged in the same motor output . While 'fast tapping’ represents simple auditory motor synchronization to the presented auditory beat , in 'slow tapping’ the single beats had to be cognitively grouped in sequences of four . This necessitates the generation of a slower rhythmic structure in addition to processing the same fast auditory beats as in fast tapping . The task was performed in runs with 24 trials each , in semi-randomized order . A trial started with a presentation of a visual instruction that indicated the upcoming condition . Before participants started tapping , four auditory beats of higher pitch primed the auditory beat rate . The inter-trial interval was jittered in both recordings ( fMRI: 10 . 1–13 . 6 s and MEG: 7 . 7–12 s ) to reduce temporal predictability during baseline . Timing variability was calculated using the standard deviation of absolute distance between the actual inter-tap-intervals and the target inter-tap-interval ( 400 ms for fast , 1600 ms for slow tapping rates ) of consecutive taps ( for a more detailed description see Pflug et al . , 2017 ) . For tap detection , the maximal tap pressure was used . Timing variability was calculated for every hand and condition independently . Values of both recording methods ( fMRI and MEG ) were used as dependent variables in a 2 ( condition [slow , fast] ) x 2 ( hands [left , right] ) mixed design repeated measure analysis . Significant effects were post-hoc tested using paired-sample t-tests . Significance level ( alpha ) was set at 0 . 05 . Statistical tests were conducted using SPSS Statistics 22 . 0 ( IBM Company , RRID:SCR_002865 ) .	If you watch a skilled pianist , you will see that their right hand moves quickly up and down the high notes while their left hand plays the lower notes more slowly . Given that about 90% of people are right-handed , it might not seem too surprising that most people can perform faster and more precise movements with their right hand than their left . Indeed , when right-handers perform any task , from hammering a nail to slicing bread , they use their right hand for the faster and more difficult action , and their left hand for slower or stabilizing actions . But why ? It could be that the left hand is simply less capable of performing skilled movements than the right . But another possibility is that the left hand is actually better than the right hand when it comes to slower movements . To test this idea , Pflug , Gompf et al . asked healthy volunteers to tap along to a metronome with both index fingers . On some trials , the volunteers had to tap along to every beat . On others , they had to tap in time with every fourth beat . While the volunteers performed the task , Pflug , Gompf et al . measured their brain activity . The results showed that the volunteers , who were all right-handed , followed the fast rhythm more precisely with their right hand than with their left . But they tapped the slow rhythm more accurately with their left hand than with their right . Areas of the brain that process sounds showed increased activity during the task . This increase was greater on the left side of the brain – which controls movement of the right side of the body – when the volunteers tracked the faster rhythm . By contrast , sound-processing areas on the right side of the brain – which controls the left side of the body – showed greater activity when participants tapped the slow rhythm . The findings thus suggest that the left half of the brain is better at controlling faster rates of movement , whereas the right half is better at controlling movements with slower rhythms . This could also help explain why in most people the left side of the brain controls speech . Speech requires rapid movements of the lips , tongue and jaw , and so it may be better controlled by the left hemisphere . Understanding how the two hemispheres control different actions could ultimately lead to new strategies for restoring skills lost as a result of brain injuries such as stroke .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	Differential contributions of the two human cerebral hemispheres to action timing
Many adult stem cell communities are maintained by population asymmetry , where stochastic behaviors of multiple individual cells collectively result in a balance between stem cell division and differentiation . We investigated how this is achieved for Drosophila Follicle Stem Cells ( FSCs ) by spatially-restricted niche signals . FSCs produce transit-amplifying Follicle Cells ( FCs ) from their posterior face and quiescent Escort Cells ( ECs ) to their anterior . We show that JAK-STAT pathway activity , which declines from posterior to anterior , dictates the pattern of divisions over the FSC domain , promotes more posterior FSC locations and conversion to FCs , while opposing EC production . Wnt pathway activity declines from the anterior , promotes anterior FSC locations and EC production , and opposes FC production . The pathways combine to define a stem cell domain through concerted effects on FSC differentiation to ECs and FCs at either end of opposing signaling gradients , and impose a pattern of proliferation that matches derivative production . The physiological role of each type of adult stem cell is to maintain appropriate production of a restricted set of cell types throughout life ( Clevers and Watt , 2018; Post and Clevers , 2019 ) . To accomplish this objective , a sufficient population of stem cells must itself be maintained . Consequently , there must be some mechanism that balances stem cell proliferation and differentiation . The balance need not be precise or without fluctuations , especially if the stem cell population is large and therefore not at risk of temporary insufficiency or extinction . However , if the number of stem cells is held roughly constant over time , then an unchanging anatomy can provide a constant environment for regulating and matching stem cell divisions and differentiation . The balance between stem cell division and differentiation can operate at the single-cell level or at the community level ( Jones , 2010; Mesa et al . , 2018; Snippert et al . , 2010 ) . If each stem cell repeatedly divides to produce a stem cell and a differentiated product ( ‘invariant single-cell asymmetry’ ) , the rate of division must simply be matched to the required supply of product cells . More commonly , however , a group of stem cells in a given location is maintained by ‘population asymmetry’ , where individual stem cells exhibit non-uniform , stochastic behaviors and differentiation is commonly not temporally or mechanistically linked to division of the same stem cell ( Reilein et al . , 2018; Ritsma et al . , 2014; Rompolas et al . , 2016; Simons and Clevers , 2011 ) . The behavior of such stem cells is likely guided substantially by extracellular signals and it is commonly presumed that regulation is achieved substantially by defining a compartment with fixed dimensions that can maintain stem cells . However , very little is known about how extracellular signals define niche space and the number of stem cells accommodated , how they affect stem cell division and differentiation , and whether they co-ordinate those two fundamental behaviors . Drosophila ovarian Follicle Stem Cells ( FSCs ) provide an outstanding paradigm to pursue these questions . FSCs were first defined as the source cells for the Follicle Cell ( FC ) epithelium that surrounds each egg chamber ( Margolis and Spradling , 1995 ) . An egg chamber buds from the germarium of each of a female’s thirty or more ovarioles ( Figure 1A–D ) every 12 hr under optimal conditions , requiring a high constitutive rate of FC production throughout adult life ( Duhart et al . , 2017; Margolis and Spradling , 1995 ) . An FC is defined by permanent association with a germline cyst and therefore passes inexorably out of the germarium within about two days and through the ovariole within five days under optimal conditions . An FSC can therefore be defined by lineage analyses as a cell that produces FCs but persists longer than an FC . However , in the original study identifying FSCs an implicit assumption was made , in accord with contemporary precedents , that each FSC is long-lived and maintained by invariant single-cell asymmetry ( Margolis and Spradling , 1995 ) . The consequent deductions of FSC number , location and behavior were largely re-stated as dogma over the following two decades despite some contrary observations ( Hartman et al . , 2015; Nystul and Spradling , 2007; Nystul and Spradling , 2010; Zhang and Kalderon , 2001 ) . A comprehensive re-evaluation , which included the analysis of all FSC lineages , without any prior assumptions about their behavior , showed that individual FSCs were frequently lost or duplicated ( Reilein et al . , 2017 ) and that FSC differentiation to an FC was not temporally coupled to , or dependent upon division of the same FSC ( Reilein et al . , 2018 ) . These characteristics of maintenance by population asymmetry , together with independent cell division and cell differentiation events and decisions , are shared by two very important and intensively studied types of mammalian epithelial stem cell , in the gut and in the epidermis ( Jones , 2010; Mesa et al . , 2018; Ritsma et al . , 2014; Rompolas et al . , 2016 ) . The re-evaluation of FSC lineages and appreciation of population asymmetry as the governing principle not only highlighted FSCs as a suitable model for many types of mammalian stem cells but also drastically revised evaluation of the number , location and behavior of FSCs ( Reilein et al . , 2017 ) , as summarized below . Production of 5–6 ‘founder’ FCs ( the first FCs to associate with a germline cyst to seed the FC epithelium ) per budding cycle is accomplished by 14–16 FSCs , arranged in three anterior-posterior ( AP ) rings anterior to the border of strong Fas3 protein expression , near the mid-point of the germarium ( Figure 1A–D; Hayashi et al . , 2020; Reilein et al . , 2017; Reilein et al . , 2018 ) . These FSCs also produce a second cell type known as an Escort Cell ( EC ) ( Hayashi et al . , 2020; Reilein et al . , 2017 ) . ECs are quiescent cells anterior to the FSC domain ( Figure 1A ) that envelop and support the differentiation of developing germline cysts ( Decotto and Spradling , 2005; Kirilly et al . , 2011 ) . FCs , which first encapsulate region 2b germline cysts and are defined by continued association with a single cyst , derive directly from the posterior ( ‘layer 1’ ) FSCs , whereas ECs derive directly from anterior FSCs in layers 2 or 3 ( Reilein et al . , 2017 ) . Each FSC lineage ( marked descendants of a single FSC ) exhibits stochastic behaviors , including extinction or amplification and production of FCs , ECs or both . A single FSC lineage can include both ECs and FCs because FSCs can divide and can exchange AP locations over time . FSCs also exhibit extensive radial movements tracked by live imaging , and no radial germarium asymmetries are known , suggesting that all FSCs within a layer are equivalent ( Reilein et al . , 2017 ) . Posterior FSCs divide faster than anterior FSCs , so that the roughly four-fold greater efflux of derivatives from the posterior face of the FSC domain is supported without significant net flow of FSCs from anterior to posterior locations ( Reilein et al . , 2017; Reilein et al . , 2018 ) . Thus , FSCs present a paradigm of dynamic heterogeneity ( Greulich and Simons , 2016 ) , where each stem cell within a fluid spatially-defined community exhibits distinctive instantaneous properties characteristic of its precise AP location but future behavior can change as a result of apparently stochastic changes in position or cell division . How are these heterogeneous individual cell behaviors marshaled into a defined stem cell domain that maintains a roughly constant number of stem cells and a continuous supply of an appropriate number of FC and EC products ? FSC maintenance and amplification have been found to depend on the activity of many of the major pathways initiated by extracellular signals . The earliest studies highlighted the role of Hedgehog ( Hh ) signaling ( Zhang and Kalderon , 2001 ) . Hh is produced in Terminal Filament and Cap cells , the anteriormost cells of the germarium , and its release is regulated by the Hedgehog binding protein Boi ( Forbes et al . , 1996a; Hartman et al . , 2010 ) . Hh was shown to promote FSC survival and amplification principally by regulating the rate of FSC division through transcriptional induction of the Hippo-pathway transcriptional co-activator Yorkie ( Yki ) ( Huang and Kalderon , 2014 ) . This constitutive role of Hh signaling in well-fed flies is also part of an environmental sensor , with nutrient deprivation leading to reduced Hh dispersal and consequent slowing of FC and egg chamber production ( Hartman et al . , 2013 ) . The key role of FSC division rate for FSC competition was highlighted by studies of Hh signaling and also by the discovery of several regulators of proliferation in a genetic screen for FSC maintenance factors ( Wang et al . , 2012; Wang and Kalderon , 2009 ) . A functional connection between stem cell division rate and competition is not expected for stem cells maintained by invariant single-cell asymmetry and was finally explained by the finding that FSC differentiation is independent of FSC division ( Reilein et al . , 2018 ) . Although the Hh signal is graded , declining from anterior to posterior , initial tests indicated that graded signaling was not important for continued normal FSC function ( Vied et al . , 2012 ) . BMP , EGF , integrin and insulin receptor initiated pathways have also been implicated in FSC function ( Castanieto et al . , 2014; Johnston et al . , 2016; Kirilly et al . , 2005; O'Reilly et al . , 2008; Vied et al . , 2012; Wang et al . , 2012 ) but the two pathways that have emerged so far as the strongest candidates for determining niche space and position-specific stem cell behaviors are the Wnt and JAK-STAT pathways because they both have graded activities in the AP dimension ( Figure 1A; Reilein et al . , 2017; Vied et al . , 2012; Wang and Page-McCaw , 2014 ) and they both have a very strong influence on FSC behavior ( Reilein et al . , 2017; Song and Xie , 2003; Vied et al . , 2012 ) . The Wg and Wnt6 ligands are produced in Cap Cells at the anterior of the germarium and are supplemented by the production of Wnt2 and Wnt4 in ECs ( Forbes et al . , 1996b; Luo et al . , 2015; Sahai-Hernandez and Nystul , 2013; Waghmare et al . , 2018; Wang and Page-McCaw , 2018 ) to produce high levels of pathway activity over the EC domain with a sharp decline over the FSC domain and little or no activity in FCs ( Figure 1A; Reilein et al . , 2017; Wang and Page-McCaw , 2014 ) . FSCs were lost from the niche cell autonomously when Wnt signaling was either genetically elevated or reduced ( Song and Xie , 2003; Vied et al . , 2012 ) . More recently it was shown that the primary effects of altering Wnt pathway activity were exerted on the AP location of FSCs and their conversion to differentiated products , with increased pathway activity favoring more anterior locations and EC production , while reducing FC production ( Reilein et al . , 2017 ) . Thus , relatively rapid loss of FSCs due to elevated Wnt pathway activity results from conversion of all FSCs over time to ECs . The JAK-STAT ligand Unpaired ( Upd ) is produced in specialized FCs called polar cells that are found at the anterior and posterior ends of developing egg chambers ( Figure 1A , D; McGregor et al . , 2002; Vied et al . , 2012 ) . Pathway activity is high in FCs in the germarium and decreases from posterior to anterior over the FSC domain with only low levels in ECs ( Figure 1A; Vied et al . , 2012 ) . When JAK-STAT activity was elevated in FSC lineages , it was shown that these FSCs out-competed wild-type FSCs and that proliferating FSC derivatives could accumulate in EC territory . Conversely , loss of STAT activity resulted in accelerated FSC loss ( Vied et al . , 2012 ) . Those studies suggested potential roles in both FSC division and location or differentiation but detailed analysis was not possible at that time , before understanding of the organization and behavior of FSCs was drastically revised ( Reilein et al . , 2017 ) . Here , we have dissected cell autonomous responses to genetic changes in Wnt and JAK-STAT signaling pathways to separate their influences on each potentially independently controlled and separately measured parameter of FSC behavior: ( 1 ) FSC division rates , ( 2 ) FSC AP location , ( 3 ) FSC conversion to FCs , and ( 4 ) FSC conversion to ECs , which combine to control FSC competitive status , measured by changes in FSC numbers over time ( Figure 1F ) . The results showed that these two graded pathways are substantially responsible for defining the FSC domain . The polarity and consequent magnitude of each graded pathway activity promoted differentiation to ECs at the anterior and differentiation to FCs at the posterior of the FSC domain , with especially sensitive responses to Wnt at the anterior and to JAK-STAT at the posterior . The magnitude of JAK-STAT signaling also substantially influenced the spatial pattern of cell divisions . Some co-ordination of FSC division and differentiation results from the dual role of the JAK-STAT pathway in promoting FSC division and FSC conversion to FCs but this coordination did not suffice in the artificial absence of Wnt signaling . Finally , the general correspondence between overall FSC competitive outcomes and the independent constituent behaviors ( division rate , AP location , and differentiation rate to FCs or ECs ) measured under a large variety of genetic conditions provides further support for the current view of FSC numbers , locations and behaviors ( Reilein et al . , 2017; Reilein et al . , 2018 ) . Prior to 2017 , when each ovariole was thought to harbor just two or three FSCs , the cell autonomous effects of altered genotypes on FSC biology were ascertained by measuring the frequency of surviving marked FSC clones , defined by the presence of labeled FCs and a putative FSC , at various times after clone induction relative to control genotypes tested in parallel ( Castanieto et al . , 2014; Kirilly et al . , 2005; O'Reilly et al . , 2008; Song and Xie , 2003; Vied et al . , 2012; Wang et al . , 2012 ) . Numerous genetic changes were found to reduce FSC clone survival severely . Occasionally , the normally low frequency of ovarioles containing only marked FSCs and FCs ( ‘all-marked’ ) was also elevated , indicating a genotype that markedly increased FSC competitiveness . Now that it is appreciated that there are 14–16 FSCs in distinct AP locations , associated with different instantaneous division rates and differentiation potential , and that FSCs produce EC as well as FCs ( Hayashi et al . , 2020; Reilein et al . , 2017; Waghmare et al . , 2018 ) , the results of clonal analysis can reveal far more about the effect of a specific genotype on different aspects of FSC behavior and the net effect on FSC survival and amplification can be measured more precisely by measuring FSC numbers . Correspondingly , labeled lineages must be scored in far more detail than before to reveal that information . We conducted an extensive series of experiments using a standard regime in order to extract comprehensive quantitative information about FSC behavior and to be able to compare results for a large number of altered genotypes among all experiments in the series . We induced GFP-labeled clones in dividing cells of young , well-fed adult females using the MARCM ( Mosaic Analysis with a Repressible Cell Marker ) system ( Lee and Luo , 2001 ) with constitutive drivers ( actin-GAL4 and tubulin-GAL4 together ) of UAS-GFP and , where relevant , additional transgene expression . Heat-shock induction of a hs-flp recombinase transgene elicited recombination at the base of the relevant chromosome arm ( using FRT recombination sites on 2L , 2R or 3R ) in a fraction of FSCs ( about 20% ) to create homozygous recessive mutations or activate expression of a transgene ( or both ) . After 6 or 12 days , ovarioles were dissected , labeled for 1 hr with the nucleotide analog EdU to measure cell division ( for the 6d test only ) , fixed and stained to label all nuclei and the cell surface protein , Fasciclin 3 ( Fas3 ) . Each experiment included a variety of altered genotypes and a control with the same FRT recombination site . For each sample , GFP-labeled FC locations along the ovariole were recorded ( Figure 1C , D ) and complete confocal z-section stacks of the germarium were archived and analyzed to count labeled FSCs in layer 1 , immediately anterior to the border of strong Fas3 staining , and in the next two anterior layers ( 2 and 3 ) , as well as labeled ECs ( anterior to FSCs ) ( Figure 1B ) , scoring also the number of labeled ECs and FSCs that had incorporated EdU ( at 6d ) ( Figure 1E ) . Our objective was to use the results to measure each distinguishable parameter of FSC behavior or decision-making separately ( Figure 1F ) . Cell division over the FSC domain was measured by EdU incorporation at the earlier time-point ( 6d ) so that sufficient FSCs of poorly competitive genotypes were still present in good numbers and hyper-competitive FSCs were not sufficiently abundant to potentially distort germarial morphology or induce secondary , non-autonomous responses . Genotype-dependent changes in the precise AP location of FSCs within the FSC domain were evident at 6d but were consistently most prominent at 12d , as were changes in the average number of FSCs present , so only the 12d results are presented . FC production was assessed quantitatively by a method we devised for measuring the probability of FC production per posterior FSC in one cycle of egg chamber budding ( ‘p’ in Figure 1F ) , using 6d samples to ensure a suitably low frequency of posterior FSCs for all genotypes . EC production was measured from 0-6d and 0-12d; it was normalized to the inferred number of anterior FSCs present during those periods . Normal FSC behavior was reported by controls from 31 separate MARCM experiments , scoring at least 50 ovarioles in almost every case . The results were extremely similar to those deduced previously from the more limited set of multicolor and MARCM experiments that formed the basis of our current perception of FSCs ( Reilein et al . , 2017 ) . The results for controls are summarized in Figure 1F and will be referenced individually later , in the context of genetic changes that alter those behaviors . Here we note that each germarium contained an average of 3 . 2 marked FSCs at 6d and 3 . 3 marked FSCs at 12d , counting all ovarioles , including those with no labeled cells . If marked FSCs of the control genotype have no competitive advantage or disadvantage over unmarked FSCs it is expected that the average number of labeled FSCs should remain constant , as observed , and these measurements should therefore report the average number of FSCs initially labeled in each germarium ( as 3 . 2–3 . 3 at 0d ) . When deducing FSC properties for the first time it was often important to assay ovarioles with lineages derived from a single FSC ( Fox et al . , 2008; Kretzschmar and Watt , 2012; Reilein et al . , 2017 ) . For example , FSC lineages with only a single candidate FSC at the time of examination were used to ascertain the location of FSCs as being in any radial location , most frequently in layer one or in layer two and occasionally in layer three ( Reilein et al . , 2017 ) . Similarly , the ability of an FSC to produce both FCs and ECs was demonstrated by generating FSC lineages at very low frequency , so that most lineages originated from a single cell ( Reilein et al . , 2017 ) . In the present studies we already can define FSCs by location and the labeling of over three FSCs per ovariole is advantageous because it effectively allows us to examine the fate of a larger number of FSCs for a given number of ovarioles . Initial labeling of each FSC in a germarium is theoretically independent and the chance of an FSC being labeled in any two germaria is theoretically equal , so the number of initially labeled FSCs per ovariole can be estimated to have a binomial distribution centered around 3 . 2–3 . 3 ( ‘0d’ , red in Figure 1G ) . The observed distribution at 6d was quite different from the assumed starting distribution , most obviously because more than a third of ovarioles no longer included any FSCs , while the proportion of ovarioles with six or more FSCs had increased ( Figure 1G ) , consistent with expectations for neutral competition ( Jones , 2010; Reilein et al . , 2017; Reilein et al . , 2018 ) . These changes were further exaggerated at 12d but full colonization of a germarium by marked cells generally takes longer ( Reilein et al . , 2017 ) , so even at 12d marked FSCs remain in a minority and are competing against unmarked wild-type cells in almost all ovarioles ( Figure 1G ) . That circumstance also applies to almost all variant genotypes investigated , ensuring that results reflect competition of marked FSCs with wild-type FSCs . In control clones 6d after induction , the average percentage of all GFP-marked FSCs that incorporated EdU was 25 . 1% ( n = 4753 ) with a pronounced gradient of labeling , declining from posterior to anterior ( 33 . 4% for layer 1 , 20 . 0% for layer 2 , and 8 . 2% for layer 3 ) ( Figure 2A ) , similar to previous observations ( Reilein et al . , 2017 ) . It had previously been observed that FSCs are rapidly lost in the absence of STAT activity , and that FSCs with excess JAK-STAT pathway activity became unusually numerous and included derivatives that incorporated EdU within the EC domain , suggesting that this pathway may affect FSC proliferation ( Vied et al . , 2012 ) . However , the quantitative effect of JAK-STAT signaling on FSC division rates has not been reported . We found that only 2 . 4% of FSCs with either of two homozygous null stat alleles incorporated EdU ( n = 336 ) , a ten-fold reduction compared to controls ( Figure 2A , B ) . To increase JAK-STAT activity , we expressed excess levels of the only Drosophila Janus Kinase , Hopscotch ( Hop ) using a UAS-Hop transgene in FSC clones ( Vied et al . , 2012; Xi et al . , 2003 ) , and found that 43 . 9% of UAS-Hop FSCs incorporated EdU ( n = 1379 ) , nearly double the rate of control FSC clones ( Figure 2A , C; Figure 2—figure supplement 1 ) . The pattern of EdU incorporation in these FSCs still showed a posterior bias , with 49 . 9% of layer 1 , 39 . 8% of layer 2 , and 32 . 1% of layer 3 UAS-Hop FSCs labeled by EdU , although the EdU indices of layer 1 and layer 3 FSCs relative to the whole FSC population were significantly different from controls ( Figure 2A ) . These experiments demonstrated that the JAK-STAT pathway has a very strong positive , dose-responsive , cell autonomous influence on the FSC cell cycle . We also saw that labeled cells in the EC region , which normally do not divide at all , sometimes incorporated EdU ( 12 . 4% ) when JAK-STAT pathway activity was elevated ( Figure 2A; Figure 2—figure supplement 1B ) , as noted previously without quantitation ( Vied et al . , 2012 ) . JAK-STAT pathway activity , reported by a ‘STAT-GFP’ transgene with ten tandem STAT binding sites ( Bach et al . , 2007 ) , is graded from posterior to anterior over the FSC domain ( Figure 2D , F ) , with a major ligand emanating from polar follicle cells ( Vied et al . , 2012 ) . Because the JAK-STAT activity gradient runs parallel to the graded pattern of EdU labeling we wished to test whether the two gradients were causally related . To do this , we took advantage of the C587-GAL4 driver , which is expressed strongly in the anterior of the germarium and decreases in strength towards the posterior with almost no detectable expression in FCs ( Reilein et al . , 2017; Song et al . , 2004 ) . This pattern is roughly a mirror-image of the normal JAK-STAT signaling pathway gradient . We expressed UAS-Hop from the C587-GAL4 driver ( C587>Hop ) , utilizing a temperature-sensitive GAL80 transgene ( Zeidler et al . , 2004 ) to restrict UAS-Hop expression temporally . After 3d at the restrictive temperature of 29C , we measured STAT-GFP fluorescence and found it to be similar in each of the three FSC layers and also over more anterior regions , indicating that the entire FSC domain now has roughly even JAK-STAT pathway activity ( Figure 2E , F ) . With roughly even JAK-STAT activity across the FSC region , we measured proliferation in the three FSC layers . We observed nearly identical frequencies of EdU labeling in each FSC layer of C587>Hop germaria; 38 . 2% of layer 1 , 37 . 4% of layer 2 , and 41 . 1% of layer 3 FSCs ( Figure 2G , H , J ) . Thus , synthetically making JAK-STAT pathway activity uniform , rather than graded , eliminated the normal posterior to anterior gradient of EdU labeling . Additionally , elevated JAK-STAT activity in the anterior of the germarium stimulated EdU incorporation in 10 . 4% of ECs ( Figure 2H , J ) . In this experiment , those cells were quiescent ECs prior to increasing JAK-STAT activity with C587-GAL4 and temperature elevation . In the MARCM studies , the GFP-marked dividing cells in the EC region originated instead from marked FSCs with elevated JAK-STAT signaling . Clearly , excess JAK-STAT pathway activity can suffice to initiate cell division in the EC domain , whether the target cells were recently derived from FSCs or not . The rate of division of those cells , indicated by their EdU index , was substantially lower than for cells in the FSC domain ( Figure 2J ) despite similar levels of JAK-STAT pathway activity in the C587-GAL4/UAS-Hop experiment ( Figure 2F ) , suggesting the presence of other factors restricting EC division or inertia due to prior quiescence ( Spencer et al . , 2013 ) . For comparison , we tested the effect of increasing CycE expression . In MARCM clones expressing UAS-CycE the FSC EdU index was greatly increased ( Figure 2K ) but we observed no EdU incorporation in the EC region . When UAS-CycE was expressed with the C587-GAL4 driver ( C587>CycE ) we found that the profile of EdU incorporation for FSCs remained graded with a posterior basis , contrasting with the response to UAS-Hop , although the gradient was flatter than in control FSCs , as revealed by statistically significant differences in the relative EdU index for layer one and for layer three relative to the overall EdU index ( Figure 2I , J ) . Also , ECs remained quiescent . We conclude that graded JAK-STAT pathway activity instructs graded proliferation within the FSC domain , and that the anterior range of sufficient JAK-STAT pathway activity appears to define the anterior boundary of this critical stem cell property . If JAK-STAT signaling were uniquely responsible for regulating the FSC proliferation gradient , then we would not expect there to be any bias in EdU incorporation , by layer , for FSC MARCM clones that have no JAK-STAT pathway activity . Though EdU incorporation was very low in stat FSC clones , it was graded; 6 . 5% of marked layer 1 FSCs incorporated EdU , compared to 1 . 5% for layer 2 and 0% for layer 3 ( Figure 2A ) . Thus , there appear to be other influences that pattern FSC proliferation . Their magnitude is , however , hard to assess reliably from this experiment alone because stat mutant FSC cycling is very infrequent . We therefore sought to introduce additional genetic changes onto a stat mutant background that would increase FSC proliferation in the marked MARCM lineages without themselves altering the normal graded pattern of proliferation . We found that expression of excess CycE and inactivation of upstream components of the Hippo/Yorkie pathway , Kibra and Warts ( Wts ) , previously shown to influence FSC proliferation ( Huang and Kalderon , 2014 ) , appear to fulfill this requirement because the gradient of EdU labeling was largely unaltered ( Figure 2K ) . When each of these three manipulations was paired with stat , the average frequency of EdU labeling roughly doubled compared to stat alone ( 2 . 4% ) , with 4 . 8% of kibra stat FSCs ( n = 461 ) , 5 . 2% of wts stat FSCs ( n = 231 ) , and 5 . 2% of stat UAS-CycE FSCs ( n = 97 ) incorporating EdU , while kibra and UAS-CycE together increased EdU incorporation to 15 . 9% ( n = 252 ) ( Figure 2K ) . In all of these experiments , more FSCs in layer one incorporated EdU compared to the anterior layers in a pattern resembling that of normal FSCs ( Figure 2K ) . Moreover , there were no statistically significant differences from controls in the relative EdU index of any one layer relative to the whole FSC population for these genotypes ( Figure 2K ) or stat alone ( Figure 2A ) , indicating that , in the absence of graded JAK-STAT pathway activity in the marked cells , there remains a robust mechanism for imposing graded FSC proliferation . Once the source of this mechanism is identified it will be possible to test whether it contributes to graded FSC proliferation under normal conditions or is effective only in the absence of JAK-STAT pathway activity . When scoring germaria with stat mutant clones , we observed that 96 . 5% of stat FSCs were found in the anterior layers of the germarium by 12d ( Figure 3A ) . Since loss of STAT drastically reduces FSC proliferation we considered whether the location of FSCs might depend on their division rate . For example , even though there is exchange of FSCs between layers ( Reilein et al . , 2017 ) , a proliferation-deficient FSC may compete less well in layer 1 . When we tested other mutants that had severely impaired proliferation , including cycE and cutlet ( Wang et al . , 2012; Wang and Kalderon , 2009 ) , we observed an altered distribution of FSCs with the proportion of FSCs in layer one reduced from a control value of 48 . 8% to 41 . 0% for cycEWX hypomorphs and 43 . 3% for cutlet FSCs but those changes were not statistically significant and were much smaller than observed for stat mutant FSCs ( Figure 3A ) . We also tested the consequences of increasing the division rate of stat mutant FSCs using the kibra , wts , and UAS-CycE manipulations . The proportion of layer 1 FSCs was 29 . 7% for kibra stat , wts stat , and UAS-CycE stat FSCs ( ‘kibra/wts/UAS-CycE stat’; average for the three aggregated genotypes ) , and 36 . 5% for kibra UAS-CycE stat FSCs , both significantly lower than for controls ( 49% ) but higher than for stat alone ( 3 . 5% ) ( Figure 3A , D , E ) . Thus , we observed a consistent anterior bias for all FSCs lacking STAT activity , even for genotypes that permitted EdU incorporation at frequencies approaching normal values . The observation that stat mutant FSCs had a reduced anterior bias when their proliferation was enhanced also supports the hypothesis that reduced division rates selectively deplete FSCs from the fastest-dividing , posterior layer . The signals and mechanisms that govern conversion of an FSC to an FC are largely unknown . In fact , only with recent insights into FSC organization can we measure this process independently of other factors , such as altered division rates , in order to attribute changes in FSC survival to changes in the frequency of differentiation to FCs ( Figure 1F ) . Importantly , an FSC can become an FC at any time relative to its last division , and FSC division and differentiation can therefore potentially be regulated independently ( Reilein et al . , 2018 ) . By correlating the location of labeled FSCs with recent production of FCs it was determined that all or most FCs derive directly from layer 1; in other words , only layer 1 FSCs associate with a passing germline cyst to become an FC ( Reilein et al . , 2017 ) . Previous studies also found that a single founder FC produced a patch occupying 17 . 8% of the monolayer of an egg chamber on average , which translates to an average of 5 . 6 founder FCs ( 1/0 . 178 ) produced per cycle of egg chamber budding ( Reilein et al . , 2018 ) . We devised a method to determine the probability that a single FSC becomes an FC in one cycle of FC recruitment from our MARCM data . The germarium generally includes one stage 2b cyst and one stage three cyst contacting Fas3-positive FCs ( Figure 1C , D ) . We call the first layer of Fas3-positive cells adjacent to the posterior face of the stage 2b germline cyst ‘immediate FCs’ to acknowledge that these cells very likely became FCs during the most recent cycle of FC allocation to a cyst ( Figure 1B–D ) . The designation of ‘immediate FCs’ reflects a location used for scoring , with no implication of specialized properties . We can reliably score if there is no labeled immediate FC in a germarium but we cannot reliably score the number of immediate FCs . We therefore scored the presence or absence of immediate FCs in germaria with 0–3 marked posterior FSCs in 6d samples . Germaria with higher numbers of marked layer 1 FSCs almost invariably include marked immediate FCs ( layer 1 FSCs become FCs at a high frequency ) and are therefore not informative for calculating the frequency of conversion of a single layer 1 FSC to an FC . From 556 control germaria , across 31 MARCM experiments , we found that a layer 1 FSC has , on average , a 61 . 6% likelihood ( p=0 . 616 ) of becoming an FC in one cycle ( see Materials and methods ) ( Figure 3C ) . The calculation method assumes that 7 of 16 FSCs divide in an average budding cycle , based on data from Reilein et al . , 2018 , so that the number of layer 1 FSCs available for conversion to FCs in one cycle is higher than the measured steady-state number by a factor of a half ( because new FSCs will only be present on average for half the cycle ) of 7/16 . The average number of layer 1 FSCs measured from 31 control experiments was 7 . 6 ( with 5 . 3 in layer 2 and 2 . 8 in layer 3; Figure 1F ) , so the expected yield of FCs per cycle is 7 . 6 ( 1+ 7/32 ) ( 0 . 616 ) =6 . 1 . The result is close to the value of 5 . 6 calculated from founder FC clone sizes , validating the method employed to calculate the probability of FC formation per FSC . When applying the same method to mutant genotypes , the calculations factored in measured changes in FSC division rate relative to controls because that influences the total number of FSCs available for conversion to FCs during a cycle ( see Materials and methods ) . Using this method , we found that a stat mutant layer 1 FSC was much less likely than controls ( 34 . 2% compared to 61 . 6% ) to produce an FC ( Figure 3C ) . To determine if reduced FC production was dependent on FSC division rate we looked at kibra stat , wts stat , and UAS-CycE stat genotypes , . The average likelihood of an FSC becoming an FC was 25 . 3% for these three genotypes ( aggregated ) , and it was 22 . 8% for kibra UAS-CycE stat mutant FSCs ( Figure 3C–E ) . These experiments demonstrated that FC production from its immediate precursor , a layer 1 FSC , is greatly impaired in the absence of STAT activity and that this reduction is not related to changes in the rate of FSC division . We also examined the consequences of increasing JAK-STAT pathway activity in a layer 1 FSC . We found that the probability of becoming an FC increased from 61 . 6% to 78 . 1% for FSCs expressing UAS-Hop ( Figure 3C ) . To test any contribution of altered FSC division we introduced a UAS-Dacapo ( UAS-Dap ) transgene , encoding a CycE/Cdk2 inhibitor ( Lane et al . , 1996; Lehner et al . , 1992 ) . We found that 23 . 5% of UAS-Dap UAS-Hop FSCs incorporated EdU , a frequency similar to controls ( Figure 3H ) . Layer 1 FSCs expressing both UAS-Hop and UAS-Dap also had a higher probability than controls of becoming FCs ( 76 . 7% vs 61 . 6% ) ( Figure 3C ) . Thus , both increased and decreased JAK-STAT pathway activity significantly affected the production of FCs from FSCs independent of FSC division rate , suggesting that the magnitude of pathway activity is an important factor in regulating this transition . The proportion of FSCs in layer one depends not only on movements between FSC layers but also on the rate of depletion from layer one to form FCs . Loss of STAT activity in multiple genotypes reduced layer one occupancy even though conversion of layer 1 FSCs to FCs was reduced , suggesting that the bias towards anterior movement within the FSC domain is even stronger than measured simply by steady-state AP distribution ( Figure 3A ) . Excess JAK-STAT pathway did not affect steady-state AP location but increased conversion of layer 1 FSCs to FCs , suggesting that there is in fact a bias towards posterior movement within the FSC domain that matches the increased conversion of layer 1 FSCs to FCs . Thus , both FSC flux into layer one and conversion of layer 1 FSCs to FCs are enhanced by increased JAK-STAT signaling and opposed by loss of JAK-STAT pathway activity . It was previously noted that strong expression of the surface adhesion molecule Fas3 , which is normally observed only in FCs , was induced in some derivatives of FSCs with elevated JAK-STAT signaling in the EC and FSC domains ( Vied et al . , 2012 ) . We confirmed these observations for UAS-Hop MARCM clones , finding that 66% of germaria with labeled cells in the anterior half of the germarium ( the FSC and EC domains ) showed ectopic Fas3 expression at 12d after clone induction ( Figure 3F ) . Furthermore , we observed that these cells sometimes appeared to form a crude epithelial monolayer surrounding developing germline cysts , indicative of FC behavior . Similar structures were observed in germaria where UAS-Hop was conditionally expressed using C587-GAL4 ( Figure 3G ) . Here , 53% of germaria included some cells with ectopic Fas3 expression by 3d , increasing to 72% by 6d and 94% by 10d . Thus , high JAK-STAT pathway alone can instruct at least some aspects of the FC phenotype even in locations where FCs do not normally form . To test whether Fas3 might be an important intermediate in the normal influence of JAK-STAT signaling on FC production we expressed excess Fas3 in kibra UAS-CycE stat mutant FSCs . We observed a doubling in layer 1 FSC to FC conversion ( from 22 . 8% to 45 . 4% ) ( Figure 3C ) , suggesting that increased Fas3 expression can partially restore FC production in the absence of JAK-STAT pathway activity . The mechanisms controlling Fas3 expression and its role in supporting FC production remain to be explored more fully . By measuring the impact of altered genotypes on each component of FSC behavior ( Figure 1F ) it should be possible to predict , or at least rationalize , the net effect on FSC competitive behavior in MARCM lineage analyses , measured by the proportion of ovarioles that retain marked FSCs over time , or measured more precisely by the average number of marked FSCs per ovariole ( counting all ovarioles ) . For FSCs and other stem cells governed by population asymmetry in which differentiation is independent of stem cell division , the rate of stem cell division is necessarily a major determinant of competitive success ( Reilein et al . , 2018 ) . Excess JAK-STAT pathway activity substantially increased the FSC EdU index . Accordingly , by 12d , there were an average of 10 . 4 UAS-Hop FSCs per germarium ( counting all ovarioles ) , significantly greater than the 3 . 3 FSCs per germarium observed in controls ( Figure 3H ) . The proportion of ovarioles containing a marked FSC was also increased with UAS-Hop expression to 75 . 2% compared to 62 . 1% in controls ( Figure 3H ) . When the increase in EdU index was suppressed by co-expressing UAS-Dap with UAS-Hop , the increase in FSC numbers was greatly reduced ( Figure 3H ) . FSC clones expressing UAS-CycE had a similar increase in the average EdU index to those expressing UAS-Hop ( 40 . 4% vs 43 . 9% ) ( Figure 2A , K ) and , again , FSC numbers were increased . However , the increase was more modest for CycE overexpression , with an average of 5 . 6 marked FSCs per ovariole by 12d and 64 . 3% of ovarioles containing a marked FSC ( Figure 3H ) . Neither UAS-CycE nor UAS-Hop significantly altered steady-state FSC AP location ( Figure 3A ) , and while UAS-Hop promoted conversion of FSCs to FCs ( Figure 3B ) , UAS-CycE did not ( data not shown ) . The larger impact of increased JAK-STAT pathway activity on FSC numbers is plausibly because increased division was promoted preferentially in anterior FSC layers ( Figure 2A , K ) , which normally do not divide as frequently and are lost directly to differentiation at a lower frequency than posterior FSCs , or because the domain of dividing cells has expanded into the EC region . It is also possible that the EdU index does not reflect division rates accurately and that the FSC division rates in response to excess CycE or excess Hop are not as similar as suggested by EdU incorporation . When STAT activity was eliminated in FSC clones , there was an average of 0 . 3 stat FSCs per germarium and 25 . 3% of germaria retained a marked FSC after 12d ( Figure 3H ) . These are large deficits compared to controls ( 3 . 2 FSCs , 62 . 1% of ovarioles ) , and similar to cycE partial loss of function mutants ( 0 . 5 FSCs , 29% of ovarioles ) , which also drastically reduce FSC proliferation . When we tested kibra stat , wts stat , and UAS-CycE stat genotypes , the average number of FSCs increased to 2 . 1 per germarium with 68 . 5% of germaria containing an FSC clone . An improved persistence of FSCs was expected but the magnitude of rescue was surprisingly large if considering only FSC division rates . This disparity was even more pronounced when examining FSC competition for kibra stat mutants expressing UAS-CycE , which had an average EdU index about 64% of wild-type . Here , the average number of marked FSCs was extremely high at 15 . 9 per germarium and 94% of ovarioles included marked FSCs ( Figure 3E , H ) . The remarkable persistence and amplification of these FSCs shows that factors other than division rate are also major determinants of FSC competition . Specifically , the dramatically increased competitive success of FSCs lacking STAT activity for a given division rate very likely results from reduced conversion to FCs . Reduced conversion to FCs results from both the infrequent presence of FSCs in layer 1 ( Figure 3A ) and the markedly lower conversion of layer 1 FSCs into FCs when FSCs lack STAT activity ( Figure 3C ) . The virtual sealing off of this conduit , which is normally the major route for FSC loss , allows marked FSCs to accumulate despite dividing at rates lower than their normal unmarked FSC neighbors . Although the survival and amplification of FSCs lacking STAT activity were increased towards , and then beyond normal by relatively modest restoration of division rates , those FSCs still had much reduced physiological activity , measured by continued production of FCs , with very few ovarioles containing both FSCs and FCs ( 3 . 5% for stat , 9 . 5% for stat with kibra , wts or UAS-CycE , 17 . 0% for kibra UAS-CycE stat , compared to 42 . 8% for controls ) ( Figure 3D , E , H ) . Thus , interfering with the normal coordination of FSC division and conversion to FCs in response to JAK-STAT pathway activity by adding genetic modifiers of division alone led to extensive amplification of unproductive FSCs . In many of these ovarioles , egg chambers were surrounded entirely by unmarked cells and had an abnormal , elongated morphology ( Figure 3E ) , perhaps suggestive of a deficiency in overall FC production . When excess Fas3 was expressed in kibra UAS-CycE stat FSCs , doubling conversion of layer 1 FSCs to FCs and restoring rates of FC production towards normal values ( Figure 3B ) , the average number of FSCs per germarium declined sharply from 15 . 9 to 2 . 4 ( Figure 3H ) . At the same time , the percentage of ovarioles with at least one FSC at 12d declined from 94% to 59% but the proportion of ovarioles with FSCs and FCs increased from 17% to 31% ( Figure 3H ) . The response to excess Fas3 provides further evidence of the large impact of the rate of FC production on both FSC numbers and the ability of FSCs to fulfill their physiological role . It also demonstrates that appropriate magnitudes of artificial stimulation of both FSC division and FSC differentiation to FCs can partially substitute for the normal coordination of these rates by JAK-STAT signaling to bring about roughly normal FSC behavior . The role of Wnt signaling in regulating FSC behavior has already been examined in the context of a revised model of FSC numbers , locations and properties . A Fz3-RFP reporter demonstrated that Wnt pathway activity decreases in strength across the FSC region , in the anterior to posterior direction ( Reilein et al . , 2017; Wang and Page-McCaw , 2014 ) . FSCs with a null arrow ( arr ) mutation to eliminate the Wnt pathway response and axin ( axn ) or Adenomatous Polyposis Coli ( apc ) mutations to constitutively activate the Wnt pathway in MARCM clones ( Reilein et al . , 2017 ) , all illustrated a strong effect of higher Wnt signaling activity favoring anterior FSC locations and greater conversion to ECs; 77 . 6% of arr FSCs but only 15–20% of axn and apc FSCs were observed in layer 1 , while 9 . 1 axn and apc ECs and 0 . 1 arr ECs were observed per germarium , compared to an average of 1 . 5 ECs for controls ( Reilein et al . , 2017 ) . Here we also tested the effect of reducing rather than eliminating Wnt pathway activity by expression of a UAS-dnTCF transgene ( van de Wetering et al . , 2002 ) in clones . We found that 63 . 0% of UAS-dnTCF FSCs were observed in layer 1 ( compared to 48 . 8% for controls ) , a significant change but less pronounced than for arr FSCs , which showed a layer 1 occupancy of 79 . 3% across three replicates ( n = 237 cells ) , including two additional tests not previously reported ( Figure 4A ) . We also tested an additional axn replicate and confirmed layer one occupancy to be greatly decreased , to 20 . 6% at 12d . Occupancy of layer one was slightly higher ( 31 . 0% ) for axn FSCs that also expressed excess CycE ( Figure 4A ) to increase division rates ( from 7 . 4% to 15 . 3% towards control 25 . 0% EdU frequency ( Figure 4E ) ) , consistent with the evidence presented earlier of low division rates favoring more anterior locations . We also used the ‘immediate FC’ method to measure FC production . We found that arr layer 1 FSC clones had a significantly elevated probability ( 76 . 9% compared to 61 . 6% in controls ) of becoming an FC , ( Figure 4B ) . By contrast , reducing Wnt pathway activity with UAS-dnTCF did not increase conversion of layer 1 FSCs to FCs ( 52 . 9% probability ) . We also observed that FSCs with increased Wnt activity showed only a 22 . 4% likelihood of becoming an FC , a roughly threefold decrease from control values ( Figure 4B ) . We therefore extend previous conclusions to surmise that the AP location of FSCs in a competitive environment of normal FSCs is altered by reduction , elimination or increases of Wnt pathway activity . By measuring the conversion of layer 1 FSCs to FCs as an independent parameter for the first time , we also found that increased Wnt pathway activity strongly reduced FC production from posterior FSCs and that only the most severe reductions in Wnt pathway activity enhanced FC production . These results suggest that the magnitude of Wnt pathway activity affects AP migration over the whole FSC domain , where Wnt signaling is graded , and that the decline of pathway activity to near zero values at the posterior margin of the FSC domain is a significant determinant of the FSC to FC transition . To evaluate EC production from anterior FSCs , we calculated the average ratio of marked ECs per marked anterior ( layer 2 or 3 ) FSC . We calculated this ratio for all germaria that retained at least one marked FSC , so that there was a possibility of EC production throughout the period scored . The average number of marked anterior FSCs ( aFSCs ) for control clones was slightly higher at 12d ( 2 . 7 ) than at 6d ( 2 . 2 ) , as expected because more ovarioles lack any FSCs at 12d and are not included ( Figure 1G ) . We took the number of marked anterior FSCs per germarium at 0d to be equal to those scored at 6d because almost all germaria have FSCs at 6d . We then estimated the average number of anterior FSCs present during the 0-12d period as the average of the number present at 0d ( and measured at 6d ) and the number measured at 12d . In controls , we found that the EC/aFSC ratio was 0 . 55 ( SE 0 . 30; SEM 0 . 01 ) for the period from 0-6d and 0 . 63 ( SE 0 . 33; SEM 0 . 01 ) for the period from 0-12d ( Figure 4D; median and other statistical measures are in Figure 4C ) . If ECs were produced by FSCs at a constant rate and all labeled ECs accumulated without loss , we would expect the 0-12d ratio to be double the 0-6d ratio . The observed percentage increase was much lower ( 15% ) across 31 control tests , suggesting that marked ECs must also be lost at a significant frequency . The same inference is apparent from looking at the number of marked ECs per germarium at 6d ( 1 . 2 ) and at 12d ( 1 . 5 ) without any correction for the slightly greater number of anterior FSCs in germaria that retain FSCs at 12d ( but still considering the same set of samples with at least one FSC ) . It appears that by 12d the number of marked ECs is at , or approaching a steady-state where the rate of production is matched by the rate of loss . This occurs when the average number of anterior FSCs ( 2 . 7 ) is almost double the number of marked ECs ( 1 . 5 ) , suggesting that the rate of loss of marked ECs is greater than the rate of conversion from anterior FSCs on a per cell basis . Clearly , if this rate of EC loss applied to the whole population of over 30 ECs , far outnumbering the total of about eight anterior FSCs , there would be a severe net loss of ECs over time . We therefore deduce that the relatively high turnover that we infer for marked ECs must apply only to ECs newly-produced from FSCs and not to the bulk EC population present at the time of adult eclosion . It is certainly plausible that an FSC that moves into EC territory might indeed often return to its former FSC position or be unable to survive for long in EC territory if it does not associate with a germline cyst to receive key survival signals ( Kirilly et al . , 2011 ) . Apoptosis is observed in normal germaria at a low frequency ( Pritchett et al . , 2009 ) with the fraction of ECs undergoing apoptosis at any instant reported as 1 . 5% ( Wang and Page-McCaw , 2018 ) or about 0 . 5% ( 19% of germaria , each with about 35 ECs ) ( Kirilly et al . , 2011 ) by TUNEL labeling . The majority of apoptosis is observed in the neighborhood of the EC/FSC boundary . We sought to test whether the turnover of marked ECs was due to apoptosis by expressing the inhibitor of apoptosis , DIAP1 in otherwise wild-type clones . If EC turnover were reduced we might expect to see a greater number of ECs accumulating at all time points , including continued significant accumulation beyond 6d . We observed a small increase in marked EC numbers , whether measured per germarium at 6d ( 1 . 5 ) or 12d ( 1 . 7 ) , or per anterior FSC at 6d ( 0 . 67 ) or 12d ( 0 . 78 ) ( Figure 4D ) . These results suggest that apoptosis does contribute to the turnover of marked ECs but it does not appear to be the major factor , with EC production over 12d far short of doubling EC production over the first 6d . Other forms of cell death may play a role but it is perhaps most likely that an FSC that moves into the EC region frequently returns to the FSC domain . We measured EC production from FSCs of various altered genotypes as described for controls , using the number of anterior FSCs at 6d for controls in each experiment as the best estimate of the number of anterior FSCs at 0d for all genotypes , so that the average number of anterior FSCs present during 0-6d and 0-12d could be calculated . We found that EC production was drastically reduced for arr mutant FSCs at 12d ( 0 . 11 ECs per anterior FSC compared to 0 . 63 for controls ) and less severely at 6d ( 0 . 30 vs 0 . 55 ) ( Figure 4D ) . The relatively high yield of ECs at 6d is likely because several ECs were produced in the first day or two before wild-type Arr protein was depleted and cell behavior was altered . Similarly , in FSCs expressing UAS-dnTCF there was a severe loss of marked ECs at 12d relative to controls ( 0 . 22 vs 0 . 63 ) but not at 6d ( Figure 4D ) ; here some time is likely required to accumulate sufficient dnTCF protein to inhibit Wnt pathway activity substantially . These data indicated that reduction , and especially loss of Wnt signaling greatly reduced the net conversion of marked anterior FSCs to ECs . This may be due to reduced conversion of anterior FSCs to ECs , increased turnover of ECs derived from FSCs , or both . Loss of Wnt pathway activity is known to elicit apoptosis in ECs ( Wang et al . , 2015; Wang and Page-McCaw , 2018 ) . We found that when we expressed UAS-DIAP1 in arr mutant FSCs , EC accumulation was increased relative to arr alone at both 6d ( 0 . 39 vs 0 . 30 ) and 12d ( 0 . 21 vs 0 . 11 , p=0 . 06 ) but remained much below control values ( Figure 4D ) . The continued deficit in EC accumulation for arr mutant cells expressing DIAP1 and that of cells expressing UAS-dnTCF suggests that the equilibrium of conversion between anterior FSCs and ECs is tilted strongly towards FSCs when Wnt pathway activity is reduced . When Wnt pathway activity was increased using the axn and apc mutations we found the average ratio of ECs per anterior FSC to be 2 . 0 from 0-6d and 7 . 3 from 0-12d , revealing a greatly elevated rate of EC accumulation ( Figure 4D ) . Again , perdurance of wild-type Axn or Apc proteins over the earliest period may account for the less dramatic rate of EC accumulation over the first 6d . The rapid addition of labeled ECs over 6-12d and the observation that FSC numbers eventually decline towards zero , leaving many labeled ECs suggest that there is little or no turnover of newly-produced ECs . Thus , any normal reversion of FSCs moving into EC locations back to FSC locations is strongly opposed by high levels of Wnt pathway activity . For all genotypes with altered Wnt pathway activity the marked cells in EC locations did not incorporate EdU and therefore do exhibit one of the key characteristics that distinguishes them as ECs rather than FSCs . We did not mark cellular processes and did not therefore ascertain whether these cells encapsulated germline cysts . We also investigated JAK-STAT signaling and EC production dynamics . When STAT activity was eliminated , we found that EC production increased to 0 . 88 ECs per anterior FSC from 0-6d and 1 . 2 ECs per anterior FSC from 0-12d ( Figure 3B ) , showing that the normal equilibrium between anterior FSCs and ECs was shifted significantly towards ECs , potentially by both increasing EC production and reducing EC loss . The number of ECs initially produced from FSCs cannot be measured accurately for FSCs expressing UAS-Hop because increased JAK-STAT pathway activity sometimes induces the subsequent division of cells in the EC domain . We found that adding UAS-Dap to UAS-Hop reduced ectopic division in the EC domain from an EdU index of 12 . 4% to 5 . 9% . EC production from 0-6d was lower for UAS-Hop UAS-Dap FSC lineages than controls , suggesting some reduction in conversion to ECs ( Figure 3B ) . The increased yield of marked cells in the EC region by 12d presumably results from division of some of those cells . These results , together with the unaltered AP distribution of FSCs with excess JAK-STAT activity and the anterior accumulation of stat mutant FSCs ( Figure 3A ) suggest that a certain minimal level of JAK-STAT pathway activity is required to prevent unbalanced anterior migration of FSCs and accelerated net conversion of anterior FSCs to ECs . It has previously been reported that loss of Wnt signaling reduced FSC division by a small amount , with most measured FSCs in layer 1 , while increased Wnt pathway activity , measured mostly in anterior FSCs , greatly decreased FSC proliferation , with results normalized in each case for FSC locations ( Reilein et al . , 2017 ) . To examine the effects of reduced signaling more comprehensively in anterior FSCs we looked at more arr mutant samples , including those additionally expressing DIAP1 , and four independent experiments where Wnt signaling was reduced by expression of dnTCF , which results in a less pronounced , but still significant , posterior shift of FSCs than complete inhibition of Wnt signaling ( Figure 4A ) . We found that EdU incorporation was slightly reduced in all layers for FSCs lacking arr activity ( 26 . 3% , 15 . 3% , 4 . 0% for layers 1–3 ) or expressing dnTCF ( 24 . 9% , 15 . 2% , 5 . 4% for layers 1–3 ) relative to controls ( 33 . 4% , 20 . 0% , 8 . 2% for layers 1–3 ) ; the differences in layer one were statistically significant ( Figure 4E ) . In all experiments where Wnt signaling was reduced or eliminated there was a clear posterior to anterior gradient of EdU labeling , as in normal FSCs with no statistically significant difference from controls in the relative EdU index of any one layer relative to the whole FSC population ( Figure 4E ) . In response to increased Wnt pathway activity , the EdU index was substantially reduced in all layers ( 13 . 7% , 4 . 8% , 2 . 8% for layers 1–3 ) but a posterior to anterior gradient was still evident ( Figure 4E ) . To test the effect on graded proliferation further we combined loss of axn with UAS-CycE in an attempt to restore overall FSC proliferation towards wild-type levels . Excess CycE indeed doubled EdU labeling frequency overall ( from 7 . 3% to 15 . 3%; control was 25% ) and a robust posterior to anterior gradient was still evident ( 23 . 8% , 18 . 1% , 7 . 3% for layers 1–3 ) with no statistically significant difference from controls in the relative EdU index of any one layer relative to the whole FSC population for axn/apc or axn UAS-CycE genotypes ( Figure 4E ) . Finally , to assess the contribution of graded Wnt signaling to the A/P pattern of graded proliferation , we reduced Wnt signaling globally by expressing the UAS-dnTCF transgene with the C587-GAL4 driver ( C587>dnTCF ) . As both normal Wnt pathway activity and C587-GAL4 expression decline from anterior to posterior across the FSC domain , this manipulation ought in theory to flatten or eliminate the normal Wnt gradient to produce a roughly even , low level of pathway activity . Fz3-RFP reporter expression showed that Wnt signaling was considerably reduced and was close to uniform across the three FSC layers ( Figure 4F–H ) . Under these conditions there was very little difference in EdU incorporation in any FSC layer when compared to controls ( Figure 4I ) . This result is consistent with the evidence from MARCM clones that graded FSC proliferation does not rely on graded Wnt signaling . As both the JAK-STAT and Wnt signaling pathways play important roles in the regulation of FSCs , we asked whether regulation by each pathway is accomplished independently . We measured whether genetic manipulation of the Wnt pathway influenced JAK-STAT pathway activity cell autonomously by inducing GFP-positive MARCM clones that also expressed a STAT-eRFP reporter , which has STAT-responsive promoter sequences from the Socs36E gene ( He et al . , 2019 ) . Reciprocal tests measured effects of JAK-STAT pathway alterations on Fz3-RFP reporter activity . In these tests , we measured the signal intensity of the reporters in marked cells relative to unmarked neighbors ( Figure 5—figure supplements 1 and 2 ) . Cells were considered neighbors if they would have been scored in the same FSC layer , and if they were captured within the same z-section during confocal imaging . Samples were examined 6d after clone induction so that all genotypes included marked cells in a full range of FSC locations . For genotypes affecting JAK-STAT signaling components , STAT-RFP intensity was significantly altered when compared to neighbors , as expected . Reporter activity was significantly lower for stat mutant cells ( 42 . 4% , p=0 . 001 ) and was not altered in control clones ( Figure 5—figure supplement 1A , B ) . Cells expressing UAS-Hop had STAT-RFP expression roughly twice that of normal neighbors . The stat genotype used is expected to prevent all stimulated JAK-STAT pathway activity . The measured residual RFP likely corresponds to a combination of basal reporter expression that is not dependent on JAK-STAT pathway activity and any perduring RFP that was induced before normal STAT protein and pathway activity were fully depleted during clone expansion . Taking RFP levels in stat mutant clones as the effective null condition , the change in STAT-RFP levels in UAS-Hop clones corresponds to an increase of more than two-fold in normal JAK-STAT pathway activity ( roughly 160%/60% = 2 . 7 ) . Neither FSCs expressing UAS-dnTCF nor those with axn mutations ( together with UAS-CycE in order to increase FSC numbers ) had significantly altered STAT-RFP expression suggesting that the magnitude of Wnt pathway activity does not cell autonomously affect JAK-STAT pathway signal transduction ( Figure 5—figure supplement 1A , B ) . For genotypes affecting Wnt signaling components , Fz3-RFP intensity was significantly altered when compared to neighbors , as expected . Reporter activity was significantly lower for arr mutant cells ( 39 . 4% , p=0 . 001 ) and for cells expressing UAS-dnTCF ( 63 . 1% , p=0 . 01 ) ( Fig . S2 ) . The null arr genotype used is expected to prevent all stimulated Wnt pathway activity ( Wehrli et al . , 2000 ) , so residual RFP likely corresponds to perduring RFP and basal reporter activity . If RFP levels in arr mutant clones are taken as the effective null condition , UAS-dnTCF expression reduced Wnt pathway activity to below 40% of normal levels ( 23 . 4% reduction in a range of 60 . 3% from null to normal ) in MARCM FSC lineages and the measured elevation of pathway-induced Fz3-RFP levels in axn clones ( 169 . 4% , p=0 . 02 ) corresponds to a roughly two-fold increase in normal Wnt pathway activity ( 129 . 5% change compared to a normal range of 60 . 3% ) ( Figure 5—figure supplement 2A ) . Neither decreased ( stat and kibra UAS-CycE stat ) nor increased ( UAS-Hop ) JAK-STAT pathway activity significantly altered Fz3-RFP expression , suggesting no direct , cell autonomous effect of JAK-STAT signaling on Wnt signal transduction ( Figure 5—figure supplement 2A , B ) . We then asked how manipulating both pathways within the same FSC would influence FSC behavior by measuring proliferation , position , and differentiation when both JAK-STAT and Wnt signaling activity were altered in FSC clones . Reduction or elimination of Wnt signaling alone caused a small reduction in EdU incorporation in MARCM clones ( Figure 4E ) . It was therefore surprising that expressing UAS-dnTCF increased EdU incorporation in FSCs lacking STAT activity ( from 2 . 4% to 4 . 5% ) . This effect was more striking in STAT-deficient genotypes with higher division rates , elevating the EdU index of FSCs with stat mutations together with kibra , wts , or UAS-CycE from an average of 4 . 9% to 16 . 2% ( Figure 5A ) , close to the value observed for otherwise normal FSCs expressing dn-TCF ( 19 . 0% ) . In other words , a five-fold reduction of EdU incorporation ( 5% vs 25% ) due to the stat kibra/wts/UAS-CycE genotype was largely suppressed when Wnt pathway was reduced by UAS-dnTCF . By contrast , the increased FSC division induced by increased JAK-STAT pathway activity ( 43 . 9% EdU index ) was not significantly altered by addition of dnTCF ( 46 . 3% ) and was only slightly reduced by loss of arr function ( 37 . 7% ) ( Figure 5B; Figure 5—figure supplement 3A; Figure 6 ) . Ectopic EdU labeling in the EC region was also largely unaltered by reducing Wnt pathway activity ( increasing to 16 . 7% from 12 . 4% ) ( Figure 5B ) . EdU labeling of FSCs with increased JAK-STAT pathway activity was also only slightly reduced by genetically increasing Wnt pathway activity ( from an EdU index of 43 . 9 to 38 . 3% for axn UAS-Hop ) to a level far above that observed for axn/apc mutant FSCs ( 7 . 3% ) ( Figure 5B; Figure 5—figure supplement 3B , C ) . Even at a lower temperature of 22C , where GAL4 and consequently UAS-Hop activities are lower , the EdU index for axn UAS-Hop FSCs was higher than control values ( 28 . 2% vs . 18 . 9% ) and not much lower than for UAS-Hop alone ( 35 . 1% ) ( Figure 5—figure supplement 3D ) . At both temperatures the EdU index of marked cells in the EC region was higher for axn UAS-Hop than for UAS-Hop lineages ( Figure 5B; Figure 5—figure supplement 3D ) . These results indicate that normal Wnt pathway activity inhibits FSC division under artificial conditions of removing STAT activity , even in locations ( layer 1 ) where Wnt pathway activity is quite low ( Figure 5A ) . Under otherwise normal conditions , genetically increasing Wnt pathway activity to a level that approximates or slightly exceeds the highest physiological levels observed in the germarium , strongly inhibited FSC division but this inhibition was robustly overridden by genetically increasing JAK-STAT pathway activity ( Figure 5B ) . Thus , inhibitory actions of the Wnt pathway can be strong and dose-dependent but are strongly suppressed by JAK-STAT pathway activity under conditions when both pathways are unaltered or both are artificially elevated . In all samples where FSCs lacked STAT activity and expressed dnTCF there was a robust posterior to anterior gradient of EdU labeling ( Figure 5A ) . This included genotypes with overall division rates close to controls ( kibra UAS-CycE stat plus UAS-dnTCF; 21 . 6% vs 25% control EdU index ) . Thus , provided overall FSC division is bolstered by excess CycE and Yki activation , there remains a roughly normal FSC proliferation gradient in the complete absence of one major graded signaling pathway ( JAK-STAT ) and reduction of another ( Wnt ) . Increased JAK-STAT activity favored conversion of layer 1 FSCs to FCs ( 78 . 1% probability ) and this was further increased by loss of arr activity ( 90 . 3% for arr UAS-Hop FSCs ) to a level higher than induced by Wnt signaling deficiency alone ( 76 . 9% for arr FSCs ) ( Figure 6A ) . Thus , the combination of eliminating the normally low levels of Wnt signaling and increasing the already high levels of JAK-STAT pathway in layer 1 FSCs potently accelerated and almost mandated conversion of layer 1 FSCs to FCs . Moreover , despite accelerated loss from layer one to become FCs , 87 . 1% of arr UAS-Hop FSCs were found in this layer , representing an enhancement of the bias seen for arr FSCs ( 79 . 3% ) ( Figure 6D ) , and further accelerating overall FC production . Expression of dnTCF did not increase the probability of layer 1 FSC to FC conversion in the presence of either increased JAK-STAT pathway activity ( Figure 6A ) or in FSCs lacking STAT activity ( Figure 6—figure supplement 1A ) . Artificially increasing Wnt pathway activity in FSCs strongly inhibited conversion to FCs ( 21 . 5% probability ) , but this inhibition was entirely negated by increasing JAK-STAT pathway activity in axn UAS-Hop FSCs , whether tested at 25°C ( Figure 6A ) or 22°C ( Figure 6—figure supplement 2 ) . Thus , reducing Wnt pathway activity to zero synergized with elevated JAK-STAT pathway activity to potently drive FSCs to become FCs , while smaller decreases achieved with dnTCF were without major consequence . High JAK-STAT activity also overcame the normally strong inhibition of FC production by elevated Wnt pathway activity . EC production was reduced for arr FSCs , increased dramatically for axn FSCs and increased less prominently for stat FSCs . Reduction of Wnt signaling with dnTCF reduced EC production alone by 12d ( Figure 4D ) but did not diminish the increased EC production of FSCs lacking STAT activity ( Figure 6—figure supplement 1E ) . The effects of UAS-Hop on EC production per anterior FSC cannot be quantified accurately because some of these marked cells in the EC region divide . Nevertheless , measurement of the total number of marked ECs provides some guidance . The average number of marked cells in the EC region per germarium at 12d , which was 1 . 3 for controls and 3 . 3 for UAS-Hop lineages , was greatly reduced for arr UAS-Hop ( 0 . 07 ) , reduced to a lesser degree for UAS-dnTCF UAS-Hop , and greatly increased for axn UAS-Hop ( 17 . 2 ) FSC lineages ( Figure 6E , G–I ) , showing that EC production is still highly responsive to changes in Wnt pathway activity in both directions even when JAK-STAT pathway activity is elevated . Thus , epistasis tests reveal a primary role of Wnt signaling for differentiation decisions in anterior regions and a primary role for JAK-STAT signaling for differentiation decisions in posterior regions . The relatively high number and persistence of FSCs lacking STAT activity despite low division rates has been noted earlier and attributed to diminished conversion to FCs . The addition of dnTCF increased EdU labeling without increasing conversion of layer 1 FSCs to FCs and might therefore be expected to increase FSC persistence . However , for stat alone or together with kibra , wts or UAS-CycE the average number of marked FSCs at 12d was not greatly altered by addition of dnTCF ( 0 . 48 vs 0 . 33 for stat alone , 2 . 0 vs 2 . 1 for others ) ( Figure 6—figure supplement 1F ) . The location of FSCs did , however , shift towards layer 1 ( 23 . 5% vs 3 . 5% for stat alone , 36 . 9% vs 29 . 7% for others ) ( Figure 6—figure supplement 1B–D ) and this would be expected to increase FC production even for a constant rate of conversion of layer 1 FSCs to FCs . Thus , increased division of stat mutant FSCs in response to dnTCF plausibly did not increase FSC numbers because increased division was offset by a modest increase in FC production . FSCs with kibra stat and UAS-CycE were already present at saturating normal numbers ( 15 . 9 ) by 12d and that number was not significantly altered by the presence of UAS-dnTCF ( 16 . 9 ) ( Figure 6—figure supplement 1F ) . Elevated JAK-STAT activity increased the average FSC number to 10 . 4 per germarium , largely by increasing division rates ( Figure 6J ) . EdU incorporation was only slightly reduced for arr UAS-Hop FSCs and unchanged for UAS-dnTCF UAS-Hop FSCs . However , complete Wnt pathway inhibition greatly increased FSC concentration in layer 1 ( Figure 6D ) and enhanced conversion from layer one to FCs ( Figure 6A , F ) , with the net effect of drastically reducing the 12d average FSC population to 0 . 3 per germarium , with only 17 . 8% of ovarioles retaining any marked FSCs ( Figure 6G , J ) . Wnt pathway reduction with dnTCF only modestly increased posterior accumulation of FSCs without accelerating FC production from posterior FSCs ( Figure 6A , D ) and resulted in 3 . 2 FSCs per germarium ( Figure 6J ) . Thus , even a greatly elevated FSC division rate cannot maintain a sufficient supply of marked FSCs when they are drained by posterior flux and conversion to FCs at the high rates promoted by a combination of high JAK-STAT pathway activity and elimination of Wnt pathway activity . With elevated Wnt signaling in UAS-Hop mutant lineages , FSC division rates remained abnormally high ( UAS-Hop was largely epistatic to axn ) ( Figure 5B ) , FC production from layer 1 FSCs was roughly normal ( Figure 6A ) but FSCs were predominantly in anterior locations ( Figure 6D ) , thereby reducing the overall rate of FC production relative to UAS-Hop FSC lineages . Accordingly , labeled FSCs accumulated to an even greater extent than for UAS-Hop alone and indeed exceeded the normal capacity of the germarium at 28 . 1 axn UAS-Hop FSCs per germarium ( Figure 6H–J ) . Thus , changes in Wnt pathway activity profoundly altered the competitive success of FSCs with elevated JAK-STAT pathway activity by either promoting ( loss of Wnt ) or reducing ( elevated Wnt ) FC production , without significantly altering FSC division rates in either case . The near-parallel gradients of FSC division frequency reported by EdU incorporation and JAK-STAT pathway activity , declining from posterior to anterior across the FSC domain suggested a potential causal link ( Figure 7 ) . Indeed , when the JAK-STAT pathway was globally manipulated to be uniform across the EC and FSC domains , at a level marginally higher than seen normally in posterior FSCs , all FSCs were seen to incorporate EdU at the same high frequency and even some cells in EC locations entered the cell cycle . Thus , the pattern of JAK-STAT pathway activity appears to be a key determinant of the pattern of FSC divisions ( Figure 7 ) . Moreover , the anterior border of dividing somatic cells , a key characteristic of active stem cells , appears to be set by JAK-STAT pathway activity dropping below a critical threshold . In our studies EdU was incubated with freshly dissected ovarioles for 1 hr , so the EdU index reports the fraction of cells in S phase at the time of dissection or shortly afterwards . A higher EdU index reports a higher proportion of cells in S phase , which commonly correlates with a higher rate of cycling of the cell population assayed , but it is only a quantitative measure of the rate of cell division if the length of S phase is unchanged . It is quite possible that the length of S phase may differ between FSCs in different AP locations or under different genetic conditions , so the EdU labeling indices we report provide an indication , rather than a definitive measure of FSC division rates . Thus , the uniform EdU incorporation observed when JAK-STAT signaling was made uniform may not report the true status of FSC division rates . Additional influences on the FSC division pattern were clearly revealed in the absence of JAK-STAT pathway activity by a robust AP gradient of EdU incorporation observed in five different genotype combinations covering a range of overall FSC division rates . We cannot yet tell whether the signals responsible for that polarized behavior normally augment JAK-STAT pathway action or serve only as a latent reserve system . Graded Hh signaling strongly promotes FSC division but it declines from anterior to posterior and cannot therefore underlie the converse gradient of FSC division ( Hartman et al . , 2013; Hartman et al . , 2010; Huang and Kalderon , 2014; Vied et al . , 2012 ) . Wnt pathway activity clearly intersects with the mechanisms that regulate FSC cell cycles but it does not appear to have a major influence on the AP pattern of FSC divisions . Under conditions of normal JAK-STAT pathway activity , increased Wnt signaling strongly inhibits FSC division but global reduction of Wnt pathway activity that drastically diminishes graded signaling , permitted a normal pattern of FSC EdU incorporation at roughly normal frequencies . Similarly , reduced Wnt pathway activity strongly increased FSC division when STAT was inactivated but those FSCs with diminished Wnt signaling still showed a normal AP gradient of EdU incorporation . A potential physiological focus of Wnt pathway influence is at the EC/FSC border where Wnt pathway activity is high and JAK-STAT pathway low . While we did not observe any ectopic cell division in the EC region in response to Wnt pathway reduction or elimination , those observations were limited by the low frequency of Wnt pathway deficient cells in EC locations and the possibility that such cells may be prone to apoptosis or other stresses . Hence , it remains possible that high Wnt pathway activity may contribute to defining the anterior border of dividing somatic cells . Through concerted actions on two separable parameters of behavior ( FSC AP location and conversion of posterior FSCs to FCs ) , FC production is promoted by JAK-STAT pathway activity and reduced by Wnt pathway activity , in keeping with the relative levels and gradients of these pathways in the posterior half of the FSC domain ( Figure 7 ) . Although the mechanisms for FSC to FC conversion are currently unknown , it involves association with germline cysts and the beginning of a transition to an epithelial phenotype , neither of which are apparent in the AP movements of FSCs between layers . The concerted actions of Wnt and JAK-STAT pathways on both FC production and FSC AP location are not therefore likely different manifestations of exactly the same molecular responses to signaling . The responses to loss of Wnt signaling and elevated JAK-STAT pathway activity on these behaviors were found to be additive , with that specific combination causing an extremely high rate of FC production , severely depleting the marked FSC pool . In effect , that result shows that the signaling environment posterior to layer 1 FSCs cannot support maintenance of an FSC . EC production from anterior FSCs was strongly favored by increasing Wnt pathway activity , as noted previously ( Reilein et al . , 2017 ) , and was increased by loss of JAK-STAT signaling . We also found evidence of significant turnover of wild-type ECs produced from FSCs during adulthood and the limited effects of blocking apoptosis suggest that a major component of turnover may be a return of those cells to the FSC domain . Our measurements did not separate EC production from EC turnover but it seems likely that the changes we observed in EC accumulation in response to altered Wnt and JAK-STAT signaling may be due to regulation of both processes . By examining twin-spot products of recombination in an FSC ( with complementary colors ) where one daughter lineage consisted of just a single FC patch , it was possible to measure the time between division of an FSC and acquisition of FC status , revealing that an FSC can become an FC at any time after its last division ( Reilein et al . , 2018 ) . Thus , for an individual FSC , division and differentiation to an FC are separate processes . That separation of fundamental stem cell activities allows the possibility of independent regulation of each process at the single-cell level . There may , nonetheless , be some systematic connections among individual , separable FSC behaviors that coordinate behavior at the single-cell level . Our results to date suggest that most FSC behaviors are largely independent . One exception was a potentially systematic connection between division rate and AP location . We found that cycE mutant FSCs , which may have reduced division as the only direct consequence of the mutation , had a small anterior bias and that the anterior bias of stat mutant FSCs was significantly reduced when division rate was increased by elevating CycE expression or Yki activity . Changes in AP location might plausibly be caused by poorer competition of slow-dividing FSCs in the posterior layer where normal FSCs divide and become FCs at a high rate . In contrast to the noted effect of division rate on AP location , the greatly reduced FSC to FC conversion of stat mutants was not altered by increasing division rate through additional genetic manipulations , and the increased conversion of FSCs to FCs due to elevated JAK-STAT pathway activity was not altered by reducing FSC division rates ( with a Cdk2 inhibitor ) . Thus , the mechanisms that regulate FSC division and FSC differentiation do not appear to be robustly coupled within single cells . However , the balance between these two processes might in theory still be achieved in single stem cells if a key regulatory signal affected both division and differentiation in appropriate proportions . The JAK-STAT pathway was found to promote both FSC division and conversion to FCs , while reducing net conversion to ECs . About 5–6 FSCs become FCs per 12 hr budding cycle , while only one fourth as many FSCs become ECs ( Reilein et al . , 2018 ) . Conversion to FCs is therefore the main drain on the FSC population that mandates compensatory FSC division . Because JAK-STAT signaling stimulates both FSC division and conversion to FCs in a dose-dependent manner , there will be a tendency to maintain each FSC , without loss or amplification , even when there are stochastic or systemic changes in the strength of this pathway . Also , by instructing posterior FSCs to divide faster than anterior FSCs the generation and loss of FSCs is roughly balanced for each layer , allowing for a roughly equal dynamic exchange of FSCs between layers , rather than , for example , a net anterior to posterior flow that would make anterior FSCs systematically longer-lived . The role of JAK-STAT signaling coordinating FSC division and differentiation was evident when pathway activity was eliminated and only the division rate of stat mutant FSCs was elevated by genetic alteration of other agents ( in kibra UAS-CycE stat FSCs ) . The consequence was a large accumulation of hyper-competitive marked FSCs that supported very little FC production ( Figure 7 ) . Graded Wnt pathway activity appears to be important for supplementing the cell autonomous coordinating activity of JAK-STAT signaling . When JAK-STAT pathway is elevated , increased FSC division is partially offset by increased FC production and marked FSCs amplify over time . FSC amplification was greatly accelerated when Wnt pathway activity was genetically increased ( in axn UAS-Hop FSCs ) because FSC division remained rapid but FC production was limited through FSCs adopting more anterior locations ( Figure 7 ) . By contrast , when Wnt pathway activity was eliminated ( in arr UAS-Hop FSCs ) the marked FSC population was drained rapidly because conversion to FCs was enhanced without markedly altering FSC division rates ( Figure 7 ) . Although dual JAK-STAT pathway responses , supported by Wnt pathway input to FSC AP location and conversion to FCs , contribute cell autonomously to balancing FSC division and differentiation , as described above , that balance is largely exercised at the community level when stem cells are maintained by population asymmetry . Several parameters must be balanced at the community level . The average FSC division rate must be appropriate to support production of a specific number of FCs and ECs every 12 hr but this also depends on the total number of FSCs . This , in turn , appears to be defined by a specific domain or space where FSCs can reside . A key general question is how a stem cell domain is spatially-defined . The FSC paradigm provides an example where this can be understood in terms of the distribution and influences of ligands for two major extracellular signaling pathways . Both graded JAK-STAT and Wnt pathways contribute to both borders . The anterior border is between non-dividing ECs and FSCs . Increasing Wnt pathway activity and decreased JAK-STAT pathway activity both promote FSC to EC conversion , while declining JAK-STAT signaling also limits the proliferative zone . The posterior border is between FSCs and FCs . Increased JAK-STAT and reduction of Wnt pathway activity ( to zero ) promote the FSC to FC transition . Thus , the graded nature of both pathways plays a crucial role in determining the A/P extent of the FSC domain and , consequently , the number of FSCs supported . The influences of JAK-STAT and Wnt pathways were examined here as cell autonomous responses in mosaic tissues ( where most cells have normal genotypes ) . Further experiments manipulating pathway activities globally ( in all cells ) may well result in a number of compensatory changes in behavior and signaling properties , and will have to be evaluated in detail to understand to what extent the size and location of the FSC domain depends on the normal magnitude and gradations of these and other pathways . The organization of mammalian intestinal stem cells is quite similar to FSCs and Wnt signaling also plays a prominent role . There , Wnt signals derive from mesenchymal cells surrounding the crypt base and additionally from Paneth cells in the small intestine , to form a gradient ( Gehart and Clevers , 2019 ) . In collaboration with R-Spondin signals , the magnitude of Wnt signaling appears to be translated into a variety of responses , including promotion of stem cell division and the expression of a key transcription factor , Ascl2 , and Ephrin signaling components , which contribute to regulating cell migration and the transition to transit-amplifying cells ( Batlle et al . , 2002; Gehart and Clevers , 2019; Schuijers et al . , 2015; Yan et al . , 2017 ) . Whether stem cell division is patterned and whether additional spatially-restricted signals guide the development of different stem cell products remain to be explored . In the FSC paradigm the dependence of stem cells on intermediate , rather than the highest levels of Wnt signaling as in the mammalian intestine , and the use of a second major graded patterning influence may be essential adaptations to differentiation occurring at two faces of the stem cell domain . FSCs ( originally termed SSCs ) were first identified by lineage studies , which noted that long-lasting lineages generally extended from roughly the mid-point of the germarium ( Margolis and Spradling , 1995 ) . The conclusions from the same study that there were just two FSCs per germarium , each with a half-life around two weeks were subsequently cited many times . However , this dogma was dramatically revised two decades later through an extensive series of lineage studies comprising three different approaches to assess the number of FSCs as 14–16 , the first definitive approach to identifying precise FSC locations and a demonstration that FSCs produced ECs as well as FSCs . A key aspect of this revision was the realization that earlier studies had implicitly assumed that FSCs are long-lived and maintained by single-cell asymmetry , leading to the exclusion of the majority of FSC lineages from all analyses , necessarily leading to substantial under-estimation of FSC numbers , over-estimation of average FSC longevity and obscuring the evidence that FSCs are in fact maintained by population asymmetry . Our current picture of FSCs is supported by extensive direct evidence; recent claims supporting the original conception of FSCs ( Fadiga and Nystul , 2019 ) were addressed , and in our opinion , refuted comprehensively ( Kalderon , 2020 ) . Nevertheless , additional evidence can always contradict , confirm or refine an existing model . Here we have presented extensive quantitative sets of data concerning FSC division , location and differentiation . A demanding test of the FSC model is whether the aggregation of these measured parameters fits well with the measured retention , loss or expansion of FSCs of a large range of mutant genotypes , which instruct a wide range of altered behaviors . We found that this was the case , as discussed above . One or two outcomes are especially noteworthy with regard to the number and location of FSCs . When FSCs lacked STAT activity but were stimulated by additional genetic changes to divide at rates approaching normal FSCs , those kibra UAS-CycE stat FSCs were present at an average of 15 . 9 FSCs per germarium , including multiple cells in each FSC layer ( Figure 3A , E , H ) . The layer 1 FSCs were clearly stem cells rather than FCs because this genotype rarely produced FCs captured in any location along the ovariole ( Figure 3E ) . Cells in layers 2 and 3 were also clearly stem cells rather than ECs because a large fraction incorporated EdU at a given time , while ECs are naturally quiescent . The phenotype of arr mutant FSC lineages , where almost all labeled cells are in layer one or further posterior ( Figure 4A ) provides further evidence that layer 1 cells are stem cells because these arr mutant lineages survive almost as well as control lineages ( Figure 6J ) . These observations provide further confirmation of our current FSC model and are clearly not consistent with either the original models postulating just two FSCs in a single layer ( Fadiga and Nystul , 2019; Margolis and Spradling , 1995; Nystul and Spradling , 2007; Nystul and Spradling , 2010 ) or recent postulates of an intermediate model where FSCs occupy the full circumference of the germarium but only in layer 2 ( Singh et al . , 2018 ) . Some of the mutant FSC phenotypes we observed also illustrate disease-relevant situations that could arise in other stem cells with a similar organization , such as mammalian intestinal stem cells . The relevance of mutations that allow amplification of specific stem cell genotypes to cancer origins has been described previously , both within the concept of field cancerization and specifically with regard to the effect of mutations conferring increased rates of stem cell division when differentiation is not coupled to stem cell division ( Frede et al . , 2014; Reilein et al . , 2018; Ritsma et al . , 2014; Rompolas et al . , 2016; Slaughter et al . , 1953; Vermeulen and Snippert , 2014 ) . In those situations , exemplified by ptc mutations activating the Hh pathway in FSCs , a mutant stem cell expands within the normal stem cell domain to provide a stable , expanded source of hyperproliferative stem cell derivatives ( Huang and Kalderon , 2014; Reilein et al . , 2018; Vied and Kalderon , 2009 ) . The properties of FSCs with increased JAK-STAT signaling provide an illustration of a more potent variant of the same principle because the genetic alteration also allows expansion of the proliferative domain into more anterior EC territory . Interestingly , for FSCs there is a genetic remedy; loss of Wnt signaling encouraged differentiation of these hyperproliferative stem cells and effectively extinguished the mutant lineages . Finally , the kibra UAS-CycE stat genotype provides a strikingly different paradigm . Here , the mutant stem cells also amplify but these stem cells do so because they rarely differentiate; they are not hyper-proliferative . If these differentiation-defective stem cells eventually out-compete all normal stem cells the ability of the stem cell community to support continued production of derivative cells will be severely compromised . 1-3d old adult D . melanogaster females with the appropriate genotypes were given a single 30 min ( for FRT40A and FRT42D ) or 45 min ( for FRT82B ) heat shock at 37C , with the heat shock duration determined by the observed relative rate of recombination for different FRT sites . Afterwards , flies were incubated at 25°C , 29°C or 22°C . Higher temperature increases GAL4 activity and was used for expression of UAS-Hop when using FRT40A or FRT42D , while 22°C was also used for the axn UAS-Hop genotype to moderate activity . Flies were maintained by frequent passage on normal rich food supplemented by fresh wet yeast during the 12d experimental period . Flies were dissected at 6d and 12d . We waited until 6d to ensure that all GAL80 present in cells , prior to clone induction , would be titrated out , permitting robust GAL4 induction of UAS-GFP and any additional transgenes . The 6d time point also ensured that any marked cells are derived from FSCs , as dividing FCs marked in the heat shock would have passed through the ovariole in less than 5d . Immediately after dissection , 6d ovaries underwent 1 hr of EdU labelling based on the protocol of the Click-iT Plus EdU Cell Proliferation Kit for Imaging ( Invitrogen ) . Both 6d and 12d ovaries were stained for Fasciclin III ( Fas3 ) and GFP . Ovaries were then manually separated into constituent ovarioles , and mounted using DAPI Fluoromount-G ( SouthernBiotech ) to stain nuclei . Ovarioles were imaged with a Zeiss LSM700 or LSM800 confocal microscope , operated in part by the Zeiss ZEN software . The entire germarium was captured in the images , as well as an average of 3–4 egg chambers . Collected images were saved as CZI files , and were later analyzed utilizing the ZEN Lite software . We aimed to image 50 germaria for every genotype in each experiment . Flies with alleles on an FRT40A , FRT42D , or FRT82B chromosomes were used in MARCM experiments using the following genotypes: FRT40A: yw hs-Flp , UAS-nGFP , tub-GAL4/yw; act-GAL80 FRT40A / ( X ) FRT40A; act >CD2>GAL4/UAS- ( Y ) – where X , Y combinations included: ( X ) – NM ( Nuclear Myc , Control ) , cycEWX , cutlet4 . 5 . 43 ( Y ) - UAS-Hop3W , UAS-Dap , UAS-dnTCF . FRT42D: yw hs-Flp , UAS-nGFP , tub-GAL4/yw; FRT42D act-GAL80 tub-GAL80/FRT42D ( X ) ; act >CD2>GAL4/UAS- ( Y ) – where X , Y combinations included: ( X ) – sha ( Control ) , ubi-GFP ( Control ) , arr2 , ( Y ) – UAS-Hop3W , UAS-DIAP1 , Fz3-RFP . FRT82B: yw hs-Flp , UAS-nGFP , tub-GAL4/yw; act >CD2>GAL4 UAS-GFP/UAS- ( Y ) ; FRT82B tub-GAL80/FRT82B ( X ) – where X , Y combinations included: ( X ) – NM ( control ) , stat85C9 , stat06346 , axnE77 , axnS044320 , apc1Q8apc2D40 , kibra32 , kibradel , wtsx1 , UAS-Hop3W , UAS-CycE , UAS-dnTCF , ( Y ) UAS-dnTCF , Fz3-RFP , STAT-RFP , including combinations of ( X ) elements . All tests not involving UAS-Hop were performed at 25°C . For UAS-Hop in experiments with FRT40A or FRT42D a temperature of 29°C was required to observe strong excess JAK-STAT phenotypes , as reported previously ( Vied et al . , 2012 ) ( at 29°C act-GAL80 [provided by T . Laverty , Janelia Farms] in place of tub-GAL80 on 2L , or in addition to tub-GAL80 on 2R in MARCM clone stocks were essential to suppress GFP expression in non-recombined cells ) . The UAS-Hop insertion is on 3R , so that clones made with FRT82B contain two copies of UAS-Hop , with tests performed at 25°C . The phenotypes due to two copies at 25°C were overlapping with those from one copy at 29°C and those data were aggregated in summary results presented . For FRT82B clones with axn and UAS-Hop on 3R the accumulation of marked cells was so high at 25°C that samples could not be scored reliably at 12d after heat shock . Consequently , additional tests for that genotype , FRT82B Hop and an FRT82B control were performed in parallel at 22°C , with results given separately from those obtained at 25°C . 1-3d old flies of the genotype C587-GAL4; UAS-X/ts-GAL80 , FRT42D tub-lacZ; ( Reporter ) /TM6B were chosen , where UAS-X was UAS-dnTCF , UAS-CycE , or UAS-Hop , and the reporter was either STAT-GFP or Fz3-RFP . Flies were incubated at 29°C for 3d , and UAS-Hop flies were also incubated for 6d and 10d . Dissected ovaries underwent the EdU and Immunohistochemistry protocols as above , without staining for GFP . For Fz3-RFP experiments with EdU , Alexa Fluor 488 dye was used instead of 594 to avoid spectral overlap . Ovaries were directly dissected into a solution of 15 µM EdU in Schneider’s Drosophila media ( 500 µl , Gibco ) and incubated for one hour at room temperature . These tubes were laid on their side and rocked manually , to ensure all dissected ovaries were fully submerged . Ovaries were then fixed in 3 . 7% paraformaldehyde in PBS for 10 min , treated with Triton in PBS ( 500 µl , 0 . 5% v/v ) for 20 min , and rinsed 2x with bovine serum albumin ( BSA ) in PBS ( 500 µl , 3% w/v ) for 5 min each rinse . Ovaries were exposed to the Click-iT Plus reaction cocktail ( 500 µl ) for EdU visualization , for 45 min . The reaction cocktail was freshly prepared prior to use , with reagents from the Invitrogen Click-iT Plus EdU Cell Proliferation Kit for Imaging , including the Alexa Fluor 594 dye . Ovaries were then rinsed 3x with BSA in PBS ( 500 µl , 3% w/v ) for 5 min each rinse . For experiments without EdU , ovaries were dissected directly into a fixation solution of 4% paraformaldehyde in PBS for 10 min at room temperature , rinsed 3x in PBS , and blocked in 10% normal goat serum ( NGS ) ( Jackson ImmunoResearch Laboratories ) in PBS with 0 . 1% Triton and 0 . 05% Tween-20 ( PBST ) for 1 hr . Monoclonal antibodies for Fas3 were 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 52242 . 7G10 anti-Fasciclin III was deposited to the DSHB by Goodman , C . and was used at 1:250 in PBST . Other primary antibodies used were anti-GFP ( A6455 , Molecular Probes ) at 1:1000 in PBST . Ovaries were incubated in primary antibodies overnight , rinsed three times in PBST , and incubated 1–2 hr in secondary antibodies Alexa-488 and Alexa-647 ( ThermoFisher ) at 1:1000 in PBST to label GFP and Fas3 , respectively . DAPI-Fluoromount-G ( Southern Biotech ) was used to mount ovaries . All germaria were imaged in three dimensions on an LSM700 or LSM800 confocal laser scanning microscope ( Zeiss ) and using a 63 × 1 . 4 N . A . lens . Zeiss ZEN software was used to operate the microscope and view images . Images were typically 700 × 700 pixels with a bit depth of 12 . The scaling per pixel was 0 . 21 x 0 . 21 x 2 . 5 µm . The range indicator in ZEN was used to determine the appropriate laser intensity and gain . ZEN was used to linearly adjust channel intensity for dim signals to improve brightness without photobleaching samples . Images were saved as CZI files and scored directly in ZEN . DAPI and Fas3 staining were used as landmarks to guide scoring . Marked cells were considered FSCs if they were within three cell diameters anterior of the Fas3 border . Cells immediately adjacent to the border were considered to be in Layer 1 , with Layers 2 and 3 in sequentially anterior positions . Anterior to the FSC niche , the EC region was roughly divided into two halves , with region 2a ECs immediately anterior to FSCs and region 1 ECs anterior to that . Germaria were also scored ( Y/N ) for the presence of marked FCs . For the ‘immediate FC' method tabulation , the presence of an FC immediately posterior to Layer one was also scored Y/N . For publication , images were digitally zoomed in ZEN and exported as tif files using the ‘Contents of Image Window’ function . Images were rotated in Abode Photoshop CS5 to uniformly orient the germaria . STAT-GFP , STAT-RFP and Fz3-RFP reporter activity was quantified within ZEN software . Using the Draw Spline Contour function , an outline of a DAPI cell nuclei was traced , and the fluorescence intensity within the outline was recorded . The outline of cells not expressing the reporter strongly ( anterior ECs for JAK-STAT reporters and FCs for Fz3-RFP ) were used to determine background intensity and were subtracted from calculated totals . For quantification of signaling pathway gradients in experiments using C587-GAL4 the signal in germline cells of the first egg chamber was used to determine background intensity . Also , the intensity of FCs from the second or third egg chamber of each sample was used as a reference , and all intensity measurements of cells within the germarium were divided by the reference to produce a relative intensity that could be compared between samples . For quantification of individual clones , the RFP intensity of a GFP-positive cell was divided by that of a GFP-negative cell in a similar position along the A/P axis , within the same or an adjacent Z plane and an average was calculated from many such pairs to derive the percentage intensity for labeled cells relative to unlabeled cells . All images shown are representative of at least ten examples . In most cases the number is much higher and is given explicitly where relevant for statistical analysis of outcomes . No statistical method was used to predetermine sample size but we used prior experience to establish minimal sample sizes . No samples were excluded from analysis , provided staining was of high quality . The experiments were not randomized; all samples presented as groups in the results were part of the same experiment and treated in exactly analogous ways without regard to the identity of the sample . Investigators were not blinded during outcome assessment , but had no pre-conception of what the outcomes might be . For EdU incorporation , FSC layers , Immediate FC probability tabulations , proportion of germaria with FSCs and/or FCs , the ‘N-1’ Chi-squared test method was used to calculate a Z score for determining significance between indicated genotypes , and error was reported as standard error of a proportion . To determine whether the EdU index distribution among the FSC layers of an altered genotype different from controls , we first calculated the average EdU index for all FSCs of the altered genotype , with each layer contribution weighted based on the normal distribution of FSC among layers measured in appropriate controls ( for MARCM or C587-GAL4 tests ) . This average EdU index was then multiplied by the control EdU index for each layer to derive expected EdU indexes for each layer of an altered genotype if the EdU pattern matched controls . Finally , a chi-squared test was applied to compare observed and expected EdU indexes for each layer to determine the statistical significance of differences . For average number of FSCs , Fz3-RFP reporter intensity comparison , and EC/FSC ratio , a t-test was used to determine significance between indicated genotypes , and error was reported as standard error of the mean . The immediate FC method was used to calculate the probability for any layer 1 FSC to become an FC in a given cycle of egg chamber budding . As layer 1 FSCs directly give rise to FC daughters ( Reilein et al . , 2017 ) , this was assessed by determining the proportion of germaria that contained a marked FC immediately posterior to the FSC region , which indicated recent FC production . This was only assessed in germaria with a small number of FSCs ( 1-3 ) to reasonably deduce the likelihood for an individual FSC . As the rate of proliferation would also influence this probability , this was accounted for in the immediate FC method equation . We assume that on average FSC division occurs halfway through a cycle , such that the probability of a newly-produced FSC becoming an FC is p/2 . Therefore , the total probability ( P ) of an FSC becoming an FC in one cycle is the sum of two probabilities: an FSC becoming an FC and an FSC dividing and the additional FSC becoming an FC . We calculate the probability of an FSC becoming an FC by tallying the proportion of germaria ( x ) that do not have immediate FC daughters when a single marked layer 1 FSC is present ( we assume that , on average , a single marked layer 1 FSC was present at the start of the prior cycle ) . Using this equation , we can solve for p , as x and q can be tabulated from scoring data . We also considered germaria with immediate FC daughters that have no FSCs present in layer 1 , as the only possibility is that a layer 1 FSC was present at the start of the last cycle and then became an FC . These instances were incorporated into the calculation of the proportion of germaria with a single layer 1 FSC but no immediate FCs . If 2 or 3 FSCs were present , the square or cube root of the x ratio was used , respectively . A weighted average of adjusted x ratios for germaria with 1 , 2 and 3 FSCs was calculated ( weighted according to the number of examples of 0–1 , 2 , and 3 layer 1 FSCs ) and used in the formula to calculate p . The q value was adjusted based on measured proliferation for mutants compared to controls , as well as predicted daughter cell production , which assumes that seven FSCs ( out of a total of 16 FSCs ) are dividing per cycle to produce 5 . 6 FCs and 1 . 4 ECs per cycle . Therefore: q = ‘Proportion of EdU incorporation for layer 1 FSCs ( mutant or control ) ’ / ‘average EdU incorporation of all control FSCs’ * 7/16 . The EC/aFSC ratio was calculated by dividing the number of marked ECs per germarium present at 6d or 12d by the inferred average number of marked anterior FSCs per germarium during the 0-6d or 0-12d period . The inferred average number of anterior FSCs per germarium was the average of the observed number of anterior FSCs per germarium ( at 6d or at 12d ) and the observed number of anterior FSCs per germarium in the control samples at 6d for that experiment ( representing our best estimate of the number of marked anterior FSCs per germarium at 0d in all samples of the same experiment ) . In all cases , only germaria that contained at least one marked FSC ( in any position ) were scored in order that there was an opportunity to produce new marked ECS throughout the experimental period . Results from multiple experiments for a particular genotype were aggregated by calculating the totals for marked ECs and for the inferred average number of anterior FSCs before deriving the overall EC per anterior FSC ratio .	Adult organisms contain a variety of cells that are routinely replaced using adult stem cells which can generate the cells of a specific tissue . These stem cells are often clustered into small groups , where combinations of chemical signals from nearby cells can encourage each stem cell to divide or ‘differentiate’ into another type of cell . These different signals must somehow balance stem cell division and differentiation to maintain the size and shape of the community . The ovary of an adult fruit fly contains a group of adult stem cells called follicle stem cells , or FSCs for short . FSCs support the continual production of eggs by supplying two types of cell from opposite faces of the stem cell cluster: dividing follicle cells emerge from the back of the cluster and guide late egg development , while non-dividing escort cells come from the front and guide early egg development . Two of the signals that control FSCs are graded over the cluster . JAK-STAT signaling is strongest in the follicle cell territory and gradually declines towards the front , while Wnt signaling is strongest in escort cells and absent from early follicle cells . However , it was unclear how the gradients of these two signals maintain the FSC population and control the formation of follicle and escort cells . To answer this question , Melamed and Kalderon used genetic engineering to modify the strength of these two signals . The experiments measured how this affected the rate at which FSCs divide and are converted into follicle or escort cells . Melamed and Kalderon found that the strength of JAK-STAT signaling dictated division rates , which may explain why the rate cells divide varies across the FSC cluster and escort cells do not divide at all . JAK-STAT signaling also stimulated FSCs to become follicle cells and opposed their conversion to escort cells . Conversely , stronger Wnt signaling favored the production of escort cells and inhibited FSCs from transitioning to follicle cells . This suggests that the relative strength of these two opposing signals helps maintain thecorrect number of FSCs while also balancing the formation of follicle and escort cells . JAK-STAT , Wnt and other signals guide the development of many organisms , including humans , and have also been linked to cancer . Therefore , the principles and mechanisms uncovered may apply to other types of stem cells . Furthermore , this work highlights genetic changes that can allow a mutant stem cell to amplify and take over an entire stem cell community , which may play a role in cancer and other illnesses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "developmental", "biology" ]	2020	Opposing JAK-STAT and Wnt signaling gradients define a stem cell domain by regulating differentiation at two borders
Nitric oxide ( NO ) is released into the air by NO-producing organisms; however , it is unclear if animals utilize NO as a sensory cue . We show that C . elegans avoids Pseudomonas aeruginosa ( PA14 ) in part by detecting PA14-produced NO . PA14 mutants deficient for NO production fail to elicit avoidance and NO donors repel worms . PA14 and NO avoidance are mediated by a chemosensory neuron ( ASJ ) and these responses require receptor guanylate cyclases and cyclic nucleotide gated ion channels . ASJ exhibits calcium increases at both the onset and removal of NO . These NO-evoked ON and OFF calcium transients are affected by a redox sensing protein , TRX-1/thioredoxin . TRX-1’s trans-nitrosylation activity inhibits the ON transient whereas TRX-1’s de-nitrosylation activity promotes the OFF transient . Thus , C . elegans exploits bacterially produced NO as a cue to mediate avoidance and TRX-1 endows ASJ with a bi-phasic response to NO exposure . Nitric oxide ( NO ) is an important signaling molecule in both prokaryotes and eukaryotes . In mammals , NO regulates key physiological events , such as vasodilation , inflammatory response , and neurotransmission ( Feelisch and Martin , 1995 ) . NO regulates innate immunity and life span in the nematode C . elegans ( Gusarov et al . , 2013 ) , as well as virulence and biofilm formation in different bacteria ( Cutruzzolà and Frankenberg-Dinkel , 2016; Shatalin et al . , 2008 ) . NO signaling is mediated by either of two biochemical mechanisms . As a reactive oxygen species , NO covalently modifies the thiol side chain of reactive cysteine residues ( forming S-nitrosylated adducts ) , thereby modulating the activity of these proteins ( Foster et al . , 2003 ) . NO can also bind to the heme co-factor associated with soluble guanylate cyclases ( sGCs ) , thereby stimulating cGMP production and activating downstream cGMP targets ( Denninger and Marletta , 1999 ) . Almost all living organisms , including bacteria , fungi , plants and animals , are able to produce NO with nitric oxide synthases ( NOS ) ( Ghosh and Salerno , 2003 ) . Due to its small molecular weight and gaseous nature , NO readily diffuses throughout the surrounding tissues to regulate cellular physiology . NO is also released into air , where it may function as an environmental cue . Lightning generates the major abiotic source of environmental NO ( Navarro-González et al . , 2001 ) . Despite its prevalence in the environment , it remains unclear if NO is utilized as a sensory cue by terrestrial animals to elicit behavioral responses . sGCs are the only described sensors for biosynthetically produced NO , mediating NO-evoked muscle relaxation and vasodilation ( Gow et al . , 2002; Stoll et al . , 2001 ) . However , it is unclear if sGCs also play a role in NO-evoked sensory responses . In vertebrates , NO modulates the activities of various ion channels , either directly through S-nitrosylation or indirectly through sGCs . NO regulation of ion channels alters neuron and muscle excitability ( Bolotina et al . , 1994; Broillet and Firestein , 1996 , 1997; Koh et al . , 1995; Wang et al . , 2012; Wilson and Garthwaite , 2010 ) . For example , in salamander olfactory sensory neurons , S-nitrosylation of a cysteine residue in cyclic nucleotide-gated ( CNG ) channels activates these channels , thereby directly altering odor-evoked responses in these cells ( Broillet and Firestein , 1996 , 1997 ) . CNG channels are highly conserved among invertebrates and vertebrates . Because both CNG channels and guanylate cyclases are essential for transducing responses for many sensory modalities , these results suggest that CNG channels and guanylate cyclases may also play a role in NO-evoked sensory responses . Unlike most metazoans , the nematode C . elegans lacks genes encoding NOS ( Gusarov et al . , 2013 ) and consequently cannot synthesize NO . Nonetheless , C . elegans is exposed to several potential environmental sources of NO , including NO produced by bacteria , which regulates C . elegans stress responses and aging ( Gusarov et al . , 2013 ) . C . elegans lives in rotting organic matter , where it feeds on diverse microbes , including the gram-negative bacteria from the Pseudomonas and the Bacillus genera ( Samuel et al . , 2016 ) . C . elegans exhibits a rich repertoire of behavioral interactions with Pseudomonas aeruginosa , Serratia marascens , and Bacillus subtilis ( Brandt and Ringstad , 2015; Garsin et al . , 2003; Reddy et al . , 2009; Styer et al . , 2008; Zhang et al . , 2005 ) . A few bacterially derived metabolites have been shown to mediate these behavioral responses ( Brandt and Ringstad , 2015; Meisel and Kim , 2014; Pradel et al . , 2007 ) . Here we test the idea that bacterially produced NO is an ecologically significant environmental cue for C . elegans-pathogen interactions . When cultured with a non-pathogenic Escherichia coli strain as the food source , C elegans spends most of its time foraging inside the bacterial lawn ( Bendesky et al . , 2011 ) . By contrast , when feeding on a pathogenic bacterial strain ( e . g . P . aeruginosa PA14 ) , C . elegans avoids the pathogen by foraging off the bacterial lawn ( Reddy et al . , 2009; Styer et al . , 2008 ) . Using a modified assay , we assessed PA14 avoidance by C . elegans young adults ( detailed in Materials and methods ) . Consistent with prior findings , over a time course of a few hours the majority of the wild-type adult animals remained off the PA14 lawn ( Figure 1A ) . We quantified PA14 avoidance as the percentage of animals inside the PA14 lawn after 8 hr of co-culture ( Figure 1B ) . PA14 avoidance is contingent on the virulence of the bacterial strain , as an isogenic PA14 gacA mutant , which is significantly impaired in its ability to kill C . elegans ( Tan et al . , 1999 ) , failed to elicit avoidance of the bacterial lawn ( Figure 1C ) . To test the potential role of nitric oxide ( NO ) in regulating the interaction of C . elegans and PA14 , we tested a PA14 mutant that was deficient for NO production . P . aeruginosa produces NO via a biosynthetic pathway that converts nitrite to NO with nitrite reductase ( nir ) ( Figure 2A ) . Prior studies reported that a PA14 mutant carrying a nirS mutation exhibits decreased ability to kill infected C . elegans , likely due to the decreased expression of virulence factors ( Van Alst et al . , 2007 , 2009 ) . Prompted by these results , we tested the idea that PA14-produced NO elicits avoidance by C . elegans . We found that avoidance of nirS mutants was dramatically reduced compared to wild-type PA14 controls ( Figure 2B ) . Loss of repulsion by the nirS mutant could result from either decreased NO levels or from changes in other virulence factors potentially activated by NO ( Van Alst et al . , 2007 , 2009 ) . To distinguish between these possibilities , we asked if chemical NO donors also elicit an avoidance response . When placed inside a non-pathogenic E . coli ( OP50 strain ) lawn , two different NO donors ( MAHMA NONOate and DPTA NONOate ) elicited C . elegans avoidance . After a 30 min exposure , the majority of animals remained out of the NO-tainted E . coli lawn , similarly to the avoidance elicited by the PA14 lawn ( Figure 2C ) . Collectively , these results suggest that bacterially produced NO is required for PA14 avoidance and suggests that C . elegans responds to NO as a repulsive chemosensory cue . Next , we tested the idea that CNG channels are required for PA14 and NO avoidance . CNG channels mediate many C . elegans chemosensory responses . The C . elegans genome contains several genes that encode CNG channel subunits , including TAX-4/CNGα and TAX-2/CNGβ , which form a heteromeric cGMP-gated cation channel ( Komatsu et al . , 1999 ) and are expressed in several classes of chemosensory neurons ( Coburn et al . , 1998 ) . PA14 and NO donor avoidance were abolished in tax-4 ( p678 ) mutants ( Figures 2B and 3C ) , which contain a putative null mutation in tax-4 . Similar PA14 and NO avoidance defects were also observed in the tax-2 ( p671 ) mutants ( Figure 2D and E ) , which contain a strong loss of function mutation in tax-2 . Thus , TAX-4/CNGα and TAX-2/CNGβ subunits were both required for proper avoidance behavior . Taken together , these results show that the cGMP-gated sensory channel TAX-4/TAX-2 is required for NO and PA14 avoidance . Because TAX-4/TAX-2 channels mediate several chemosensory responses , these results further support the idea that C . elegans responds to NO as a repulsive environmental odorant . To identify the sensory neurons mediating PA14 and NO avoidance , we determined which neurons require TAX-4 expression for these responses ( Figure 3A–C ) . Transgenes expressing a tax-4 cDNA with either the odr-3 promoter ( expressed in AWA , AWB and AWC olfactory neurons ) ( Roayaie et al . , 1998 ) , or the gcy-36 promoter ( expressed in oxygen sensing AQR , PQR , and URX neurons ) ( Cheung et al . , 2004; Gray et al . , 2004 ) both failed to rescue the PA14 avoidance defects of tax-4 mutants ( Figure 3B ) . By contrast , a transgene expressing tax-4 selectively in ASJ sensory neurons ( using the trx-1 promoter ) ( Miranda-Vizuete et al . , 2006 ) fully rescued PA14 avoidance and partially rescued NO donor avoidance defects exhibited by tax-4 ( p678 ) mutants ( Figure 3A and C ) . Thus , TAX-4 CNG channels act in the ASJ neurons to promote PA14 and NO donor avoidance . To confirm that ASJ neurons are required for PA14 and NO avoidance , we genetically ablated ASJ neurons by expressing the pro-apoptotic EGL-1 protein ( Conradt and Horvitz , 1998 ) using the trx-1 promoter ( Figure 3D ) . As expected , inducing ASJ cell death significantly decreased PA14 and NO donor avoidance ( Figure 3E–F ) . A prior study showed that ASJ neurons respond to two bacterial metabolites to mediate PA14 avoidance ( Meisel and Kim , 2014 ) . Here , we show that ASJ neurons also sense PA14-derived NO as a repulsive cue through the TAX-2/TAX-4 CNG channels , thereby promoting avoidance of virulent PA14 strains . To determine if ASJ neurons are activated by NO , we recorded and quantified intracellular calcium transients in ASJ , using a genetically encoded calcium indicator GCaMP6s ( Chen et al . , 2013 ) . Worms were confined in a microfluidic device with their nose ( and associated chemosensory endings ) exposed to fluidic streams of sensory stimuli delivered with precise temporal control ( Chronis et al . , 2007 ) . Exposure to NO donor evoked a significant increase in GCaMP6s fluorescence in the ASJ neurons , reaching peak intensity within a few seconds and gradually returning to baseline fluorescence despite the continuing exposure to NO ( Figure 4A ) . Removing the NO stimulus also evoked an ASJ calcium transient that lasted ~20 s ( Figure 4A ) . By contrast , switching between streams of control buffer solution did not alter the GCaMP6s signal in ASJ ( Figure 4B ) . To determine if ASJ neurons sense NO directly , we examined unc-13 ( s69 ) null mutants , in which synaptic transmission is nearly completely blocked ( Richmond et al . , 1999 ) . We found that the ON and OFF responses of ASJ to NO remained intact in unc-13 mutants ( Figure 4C , Figure 4—figure supplement 1 ) , suggesting that NO-evoked ASJ calcium transients are unlikely to result from indirect activation of ASJ by synaptic input . Collectively , these results indicate that the ASJ sensory neurons directly sense NO , and that ASJ neurons have a biphasic response to NO , exhibiting increased cytoplasmic calcium at both NO onset and removal ( hereafter indicated as ON and OFF responses ) . To determine if CNG channels are required for NO-activation of ASJ , we recorded the ASJ GCaMP6s signal in tax-4 mutants . The tax-4 ( p678 ) mutation , which abolished PA14 and NO donor avoidance ( Figures 2B and 3A–C ) , also eliminated the NO-evoked ON and OFF calcium transients in ASJ ( Figure 4D , Figure 4—figure supplement 1 ) . A transgene selectively restoring tax-4 expression in ASJ reinstated the NO-evoked ON calcium transients in ASJ ( Figure 4E , Figure 4—figure supplement 1 ) . These results indicate that activation of TAX-4 channels in the ASJ sensory neurons generates increased calcium transients in response to NO exposure , which results in NO avoidance behavior . The requirement for TAX-2/TAX-4 CNG channels for NO sensation suggests that NO stimulates cGMP synthesis in ASJ . Responses to extracellular gaseous ligands , such as O2 and NO , are often mediated by sGCs; however , none of the sGC encoding genes is expressed in ASJ neurons ( www . wormbase . org ) . Receptor guanylate cyclases ( rGCs ) mediate C . elegans responses to CO2 ( Hallem et al . , 2011 ) . Prompted by these results , we tested the idea that rGCs mediate NO responses . ASJ neurons express two rGCs ( GCY-27 and DAF-11 ) ( www . wormbase . org ) . We found that PA14 and NO donor avoidance were both abolished in daf-11 ( m47 ) mutants ( which contain a temperature sensitive loss of function mutation ) and both avoidance responses were reinstated by a transgene that expresses daf-11 in the ASJ neurons ( Figure 5A ) . Next , we analyzed ASJ GCaMP6s signals in daf-11 mutants . We found that the daf-11 ( m47 ) mutation completely abolished the ASJ ON and OFF response to NO donor ( Figure 5C , Figure 5—figure supplement 1 ) . Expressing the daf-11 cDNA specifically in ASJ partially rescued the ASJ ON response to NO donor ( Figure 5D , Figure 5—figure supplement 1 ) . In contrast to DAF-11 , mutations inactivating GCY-27 decreased but did not abolish PA14 avoidance and had no effect on the ASJ ON response , although they did eliminate the ASJ OFF response to NO donor ( Figure 5B and E , Figure 5—figure supplement 1 ) . Together , these results indicate that the rGCs DAF-11 and GCY-27 function in the ASJ neurons to mediate NO sensation . Next , we examined how DAF-11 mediated the sensory response to NO . The predicted DAF-11 protein does not contain a heme-NO-binding ( HNOB ) domain , although it contains a heme-NO-binding associated ( HNOBA ) domain ( www . wormbase . org ) . HNOB domains are found in proteins that directly bind heme cofactors whereas HNOBA domains are similar to PAS domains and are found in a subset of HNOB containing proteins ( Iyer et al . , 2003 ) . To test its functional importance , we used CRISPR to delete the daf-11 HNOBA domain . The resulting daf-11 ( nu629 ) mutation abolished the ASJ response to both the onset and the removal of NO ( Figure 5F , Figure 5—figure supplement 1 ) , suggesting that DAF-11 mediates NO sensing by interacting with other NO-binding proteins . These results do not conclusively demonstrate a requirement for the HNOBA domain because the nu629 mutation may prevent expression or trafficking of the DAF-11 protein . NO is freely diffusible and membrane permeable; consequently , PA14 produced NO could act as either an external sensory cue or by directly activating intracellular signaling pathways in ASJ . We did several experiments to distinguish between these possibilities . If NO acts as an external chemosensory cue , PA14 and NO donor avoidance should be diminished in mutants that have defective ciliated sensory endings . Previous studies identified mutations in genes encoding the components of ciliated sensory endings that disrupt the function of chemosensory neurons , including the ASJ neurons ( Perkins et al . , 1986 ) . Two cilia defective mutants , osm-12 ( n1606 ) null mutants and bbs-9 ( gk471 ) null mutants , were both defective in avoiding the lawn of PA14 and the NO donor ( Figure 6A and B ) . These results indicate that the normal function of sensory cilia is required for the NO sensation . To further address this question , we asked if DAF-11 rGC and TAX-2/4 CNGs must be localized to ASJ sensory endings to mediate NO responses . To test this idea , we analyzed daf-25 ( m362 ) null mutants , which lack a cargo adaptor that is required for rGC and CNG transport to ciliated sensory endings ( Fujiwara et al . , 2010; Jensen et al . , 2010; Wojtyniak et al . , 2013 ) . Avoidance of NO and the PA14 lawn were both defective in daf-25 ( m362 ) mutants ( Figure 6A and B ) . The defect in PA14 avoidance was partially rescued by a transgene that restores DAF-25 expression in ASJ neurons ( Figure 6A ) . As in daf-11 ( m47 ) mutants , the NO evoked ON and OFF calcium transients in ASJ were completely abolished in daf-25 ( m362 ) mutants ( Figure 6C , Figure 6—figure supplement 1 ) . Thus , the ability of ASJ neurons to detect NO requires DAF-11 and TAX-2/4 CNG channel localization to ciliated sensory endings . Taken together , these results show that NO is sensed by ASJ as an external sensory cue . How do ASJ neurons detect NO ? NO covalently modifies reactive cysteine residues by S-nitrosylation ( SNO ) . To determine if protein-SNO modifications are involved in NO detection , we analyzed mutants lacking protein de-nitrosylating enzymes . Protein-SNO modifications are reversed by two classes of enzymes , Thioredoxins ( TRX ) and nitrosoglutathione reductases ( GSNOR ) ( Benhar et al . , 2009 ) ; consequently , protein-SNO adducts should accumulate in mutants lacking TRX and GSNOR . The C . elegans genome encodes multiple thioredoxin genes . We focused on the trx-1 gene ( Figure 7A ) because it is exclusively expressed in the ASJ neurons ( Miranda-Vizuete et al . , 2006 ) . The amplitude and duration of the NO-evoked ON transient in ASJ were both significantly increased in trx-1 ( jh127 ) null mutants ( Figure 7A and B , Figure 7—figure supplement 1 ) and this defect was rescued by a single copy transgene restoring TRX-1 expression in ASJ neurons ( Figure 7C , Figure 7—figure supplement 1 ) . To determine the effect of de-nitrosylation in NO sensing , we used CRISPR to isolate a catalytically inactive trx-1 ( nu517 ) mutant , containing the C38S mutation in the active site for de-nitrosylation ( Figure 7A ) . In trx-1 ( nu517 C38S ) mutants , the NO evoked OFF calcium transient was eliminated , whereas the ON transient was unaltered ( Figure 7D , Figure 7—figure supplement 1 ) . These results suggest that TRX-1’s de-nitrosylating activity is required for ASJ to generate increased cytoplasmic calcium in response to removing NO . Interestingly , the ASJ ON response of trx-1 ( jh127 ) null mutants was significantly larger than wild type whereas the ON response was unaltered in trx-1 ( nu517 C38S ) mutants ( compare Figure 7B and D; Figure 7—figure supplement 1 ) , suggesting that the null phenotype cannot be explained by decreased de-nitrosylation activity . Cysteine residues outside of the catalytic domain are thought to promote other TRX functions . For example , TRX promotes nitrosylation of other proteins , and this trans-nitrosylation activity is eliminated by mutations altering Cys-72 ( Mitchell and Marletta , 2005 ) . To address the role of trans-nitrosylation , we rescued trx-1 null mutants with a transgene expressing TRX-1 ( C72S ) ( Figure 7A ) . Unlike the wild type TRX-1 transgene , a TRX-1 ( C72S ) transgene failed to rescue the larger and more prolonged NO evoked ASJ ON response observed in trx-1 null mutants ( compare Figure 7C and E; Figure 7—figure supplement 1 ) . These results suggest that the exaggerated amplitude and prolonged time course of the ASJ ON response to NO donor results from inactivation of TRX-1’s trans-nitrosylation activity . To further investigate the role of protein-SNO modifications in ASJ responses to NO donors , we analyzed mutants lacking a second de-nitrosylating enzyme GSNOR . The C . elegans genome encodes a single GSNOR gene , H24K24 . 3 ( hereafter designated gsnor-1 ) . We used CRISPR to isolate an early nonsense mutation in gsnor-1 ( nu518 ) . The NO evoked ASJ OFF response was significantly diminished in the gsnor-1 mutants , whereas the ON response remained ( Figure 7F , Figure 7—figure supplement 1 ) . Thus , analysis of trx-1 and gsnor-1 mutants both suggest that protein de-nitrosylation is required for the ASJ response evoked by removing NO . To determine if protein-SNO modifications also regulate behavioral responses , we measured PA14 avoidance in trx-1 and gsnor-1 mutants . PA14 avoidance was significantly reduced in trx-1 ( jh127 ) null mutants ( Figure 7G ) . This PA14 avoidance defect was rescued by a single copy transgene expressing wild type TRX-1 but was not rescued by the TRX-1 ( C72S ) transgene ( Figure 7G ) . By contrast , PA14 avoidance was unaltered in the two de-nitrosylation defective mutants , trx-1 ( nu517 C38S ) and gsnor-1 ( nu518 ) ( Figure 7G ) . These results suggest that TRX-1’s trans-nitrosylation activity is required for PA14 avoidance . C . elegans is among only a few organisms that do not synthesize NO and , likely , acquires NO from the environment . We show that NO is sensed by a chemosensory neuron ( ASJ ) , NO sensing requires functional ciliated sensory endings , and that NO sensing is defective in daf-25 mutants ( which lack DAF-11 and TAX-4 localization to cilia ) ( Fujiwara et al . , 2010; Jensen et al . , 2010; Wojtyniak et al . , 2013 ) . Collectively , these results indicate the role of NO as an external sensory cue that represents environmental information to worms . Several results suggest that ASJ neurons are the primary NO-sensing neurons . NO-avoidance is strongly defective following genetic ablation of ASJ neurons . Similarly , tax-4 CNG and daf-11 rGC mutant defects in NO avoidance are partially rescued by transgenes restoring expression of these genes selectively to ASJ neurons . Given the partial rescue observed for ASJ expressed TAX-4 and DAF-11 , it remains likely that other neurons also contribute to NO-evoked behaviors . Seven heme-containing sGCs ( gcy-31–37 ) are expressed in oxygen sensing neurons AQR , PQR , URX , and BAG ( Cheung et al . , 2004; Gray et al . , 2004; Zimmer et al . , 2009 ) . Several sGCs , including GCY-35 , mediate sensory responses to oxygen but also display a low affinity binding to NO ( Gray et al . , 2004 ) . Thus , these oxygen sensing neurons may also respond directly to NO . The only previously described NO sensors are sGCs that bind NO via their associated Heme co-factor ( Malinski and Taha , 1992 ) . Here we show that NO-sensing by ASJ neurons is mediated by two rGCs ( DAF-11 and GCY-27 ) . Interestingly , a prior study showed that another rGC ( GCY-9 ) mediates CO2 detection by the BAG neurons ( Hallem et al . , 2011 ) . Thus , rGCs mediate detection of two environmental gasses ( NO and CO2 ) by C . elegans . In addition , we show that NO-sensing requires the function of the cyclic nucleotide-gated ( CNG ) channel subunit TAX-4 in the ASJ neurons . Together , our results identify rGCs and CNG channels as one underlying mechanism for NO sensation . How does daf-11 mediate NO responses ? While DAF-11 is required for both the onset and removal response to NO , neither DAF-11 nor GCY-27 contains a Heme-NO-binding ( HNOB ) domain . Instead , DAF-11 contains a Heme-NO-binding associated ( HNOBA ) domain , suggesting that DAF-11 mediates NO sensing by interacting with other NO-binding proteins . Consistent with this possibility , removing the HNOBA domain from DAF-11 abolished the ASJ calcium response to NO . In addition to NO sensing , daf-11 mediates chemosensory responses to several volatile chemicals ( Birnby et al . , 2000 ) . DAF-11 may regulate sensory responses to different cues by acting together with different signaling molecules . The complete transcriptome of ASJ is not yet available , which would aid the characterization of other factors that regulate various sensory transduction pathways in ASJ . How do DAF-11 and GCY-27 detect NO ? NO could be detected by S-nitrosylation of cysteine residues in DAF-11 and GCY-27 ( or proteins associated with them ) , thereby activating the GC catalytic domain . Consistent with this idea , mutations inactivating the de-nitrosylating enzymes TRX-1/Thioredoxin and GSNOR-1 eliminate the NO-evoked OFF transient in ASJ , while having little effect on the ON transient . Thus , accumulation of protein SNO-adducts or SNO-Glutathione adducts ( in the de-nitrosylation defective mutants ) was associated with a loss of the ASJ response to removing NO . Alternatively , the transmembrane GCs may associate with other NO binding proteins , such as Heme-containing globins . In this regard , it is interesting that DAF-11 , like the mammalian atrial natriuretic peptide ( ANP ) receptors , contains a conserved cytoplasmic HNOBA domain , which is similar to PAS domains . HNOBA/PAS domains are thought to be directly bound by HSP90 , a chaperone that catalyzes incorporation of heme groups into sGCs ( Ghosh and Stuehr , 2012; Sarkar et al . , 2015 ) . Interestingly , mutations in daf-21/HSP90 mimic all of the phenotypes found in daf-11 mutants , indicating that DAF-11 function requires its interaction with HSP90 ( potentially through binding to DAF-11’s HNOBA/PAS domain ) ( Birnby et al . , 2000 ) . Consistent with this possibility , deleting the DAF-11 HNOBA domain abolished the NO response in ASJ . Thus , the coordinated action of DAF-21/HSP90 and DAF-11 in NO sensing could indicate that DAF-11 associates with other heme-binding proteins ( e . g . globins ) . GCY-27 lacks the HNOBA/PAS domain , and consequently would have to detect NO by a distinct mechanism . Our results suggest that DAF-11 and GCY-27 mediate different aspects of the NO response . GCY-27 is required for the NO-evoked OFF transient whereas DAF-11 is required for both the ON and OFF transients . The mechanism underlying this difference is not known but could reflect a differential GC activation by increasing ( DAF-11 ) and decreasing ( DAF-11 and GCY-27 ) NO concentration . A similar mechanism was previously proposed for detecting increasing ( GCY-35/36 ) and decreasing ( GCY-31/33 ) O2 concentrations by distinct cytoplasmic GCs ( Zimmer et al . , 2009 ) . In this case , we predict that co-expression of DAF-11 and GCY-27 allows ASJ to detect both increasing and decreasing NO concentration , thereby producing ON and OFF transients . TRX proteins are redox sensitive proteins that have been proposed to play several important roles in cellular responses to NO . TRX has an enzymatic activity that removes SNO-protein and SNO-glutathione adducts ( Benhar et al . , 2009 ) . This de-nitrosylation activity is mediated by a pair of active site cysteine residues ( C38 and C41 in TRX-1 ) . Thioredoxins have also been proposed to promote nitrosylation of other protein substrates , and this trans-nitrosylation activity requires a third cysteine residue ( C72 in TRX-1 ) ( Mitchell and Marletta , 2005 ) . Our results suggest that TRX-1 regulates the NO-evoked ASJ response via two distinct activities . TRX-1 inhibits and shortens the NO-evoked ON response , as indicated by a larger and more prolonged ON response in trx-1 null mutants . This inhibitory function of TRX-1 is eliminated in the C72S mutant , implying that inhibition is mediated by the TRX-1’s trans-nitrosylation activity ( Mitchell and Marletta , 2005 ) . Two results suggest that TRX-1’s de-nitrosylating activity promotes the NO-evoked OFF response in ASJ . The OFF response was eliminated in both trx-1 mutants containing a mutation in the de-nitrosylation active site ( nu517 C38S ) and in gsnor-1 mutants ( which lack a second de-nitrosylating enzyme ) ( Benhar et al . , 2009 ) . Collectively , our results suggest that TRX-1 shapes ASJ’s bi-phasic response to NO ( Figure 8 ) . Specifically , we propose that during NO exposure TRX-1’s active site cysteines ( C38/41 ) are oxidized , thereby decreasing de-nitrosylation ( Engelman et al . , 2016; Wang et al . , 2014 ) . Oxidation of C38 and 41 promotes C72 nitrosylation ( Barglow et al . , 2011 ) , which increases trans-nitrosylation activity . Thus , during NO exposure the net effect of TRX-1 would be increased trans-nitrosylation of protein substrates . We propose that these trans-nitrosylated proteins inhibit the ON response . Following NO-removal , TRX-1 active site cysteines are reduced , thereby enhancing TRX-1’s de-nitrosylation activity and inhibiting its trans-nitrosylation activity . As a result , following NO removal , inhibitory SNO-protein adducts formed during NO exposure could be removed by TRX-1’s de-nitrosylating activity , giving rise to the OFF response . Thus , the combined activities of TRX-1 produce ASJ’s bi-phasic ON and OFF responses to NO . How does ASJ’s bi-phasic response to NO shape behavioral responses to NO and PA14 ? Mutations that diminish both the ON and OFF responses ( i . e . daf-11 , daf-25 , and tax-4 mutations ) were also deficient for both PA14 and NO avoidance . By contrast , mutations that diminish ASJ’s OFF response but retain ON responses produced inconsistent behavioral results . For example , PA14 avoidance was decreased in gcy-27 mutants , supporting the idea that OFF transients ( which are deficient in gcy-27 mutants ) are required to promote avoidance behavior . On the other hand , PA14 avoidance was unaffected in the de-nitrosylation defective mutants , trx-1 ( nu517 C38S ) and gsnor-1 , which also lack the OFF response . Finally , PA14 avoidance was significantly reduced in mutants deficient for trans-nitrosylation , trx-1 ( C72S ) and trx-1 ( jh127 ) null mutants , which have a heightened and prolonged ON response . Collectively , these results support the idea that the temporal structure of ASJ’s NO response plays an important role in PA14 avoidance . However , it remains unclear how ( or if ) the bi-phasic ON and OFF responses are utilized to produce behavioral responses . To further address this question , we will need new behavioral assays . In our current assays , behavior is measured over many minutes ( NO donor avoidance ) or hours ( PA14 avoidance ) . By contrast , ASJ calcium responses to NO addition and removal occur in 10–20 s . Thus , assessing the behavioral impact of the ON and OFF responses will require recording behavioral responses that are rapidly evoked by NO exposure and removal ( for example in microfluidic chambers ) . In principle , ASJ’s biphasic response could have several beneficial effects . A current model for C . elegans chemotaxis proposes that chemosensory cues elicit avoidance by decreasing the rate of turning , thereby promoting dispersal away from the cue ( Pierce-Shimomura et al . , 1999 ) . In this scenario , ASJ’s biphasic response could promote avoidance behavior if the ON and OFF transients both resulted in decreased turning ( thereby promoting movement away from PA14 or NO donors ) . Alternatively , ASJ’s rapid responses to NO ( or other PA14 metabolites ) could promote transcriptional responses or release of neuromodulators , which could modulate behavior or innate immune responses over longer time scales . A prior study suggested that bacterially produced CO2 was utilized as a contextual cue to promote C . elegans avoidance of another pathogen , Serratia marascens ( Brandt and Ringstad , 2015 ) . These authors proposed that environmental CO2 indicates proximity to metabolically active bacteria , and that pathogen avoidance is mediated by the coincident detection of CO2 in conjunction with other virulence factors . Such a coincidence detection strategy is proposed to optimize the ability of C . elegans to forage for nutritional resources since it would allow worms to actively feed on avirulent bacteria while avoiding ingestion of pathogenic bacteria . Our results suggest that C . elegans avoidance of PA14 is mediated by a similar coincidence detection strategy . In particular , we propose that recent exposure to NO enhances the repulsive effects of other PA14 metabolites , for example , phenazine compounds ( Meisel and Kim , 2014 ) . Thus , our results combined with this prior study suggest that C . elegans discriminates between benign and pathogenic microbes by coincident detection of multiple bacterial metabolites , including both CO2 and NO . Strains were maintained at 20°C as described ( Brenner , 1974 ) . The wild-type reference strain was N2 Bristol . The mutant strains used were: LGI , tax-2 ( p671 ) , unc-13 ( s69 ) , bbs-9 ( gk471 ) , daf-25 ( m362 ) ; LGIII , tax-4 ( p678 ) , osm-12 ( n1606 ) ; LGIV , gcy-27 ( ok3653 ) ; LGV , daf-11 ( m47 ) . Descriptions of allele lesions can be found at http://www . wormbase . org . CRISPR gene editing was utilized to isolate daf-11 ( nu629 ΔHNOBA ) , trx-1 ( nu517 C38S ) , and gsnor-1 ( nu518 ) mutations , using dpy-10 as a co-CRISPR marker ( Ward , 2015 ) . The gsnor-1 gene corresponds to the H24K24 . 3 gene in wormbase . Predicted amino acid sequence of these alleles are as follows ( mutant residues underlined , * indicates stop codons ) : daf-11 ( nu629 ΔHNOBA ) : TQGLNETVKNEVGRIELLPKSVANDLKN trx-1 ( nu517 C38S ) : EKIIILDFYATWSGPCKAIAPLYKE gsnor-1 ( nu518 ) : KTNLCQKIRI**GNGFMPDGSSRFTCNG All plasmids were derivatives of pPD49 . 26 ( Fire , 1997 ) and constructed utilizing standard methods . A 1028 bp trx-1 promoter and a 543 bp ssu-1 promoter were used for expression in ASJ . A 3 kb odr-3 promoter was used for expression in AWA/AWB/AWC . The 1089 bp gcy-36 promoter and the tax-4 cDNA were derived from pEM4 ( Gift from Cori Bargmann ) . GCaMP6 . 0s was a gift from Jihong Bai . The cDNAs of egl-1 , daf-11 and daf-25 were cloned from a cDNA library using primers corresponding to the predicted start and stop codons of each gene . Full descriptions of all plasmids are provided in the Key Resources Table . Transgenic animals were generated by injecting wild type or mutants with the transgene ( 10~25 ng/ul ) mixed with the co-injection marker , KP#1480 ( pmyo-2::NLS-mCherry , 10 ng/μl ) , using standard methods ( Mello et al . , 1991 ) . nuIs556 was generated by injecting wild type with KP#3311 ( ptrx-1::GCaMP6 . 0s ) at 25 ng/ul mixed with the co-injection marker KP#2186 ( punc-122::mCherry ) at 70 ng/ul , followed by UV irradiation . The single copy transgenes nuSi196 [pssu-1::daf-11] , nuSi197 [ptrx-1::trx-1] , and nuSi198 [ptrx-1::trx-1 ( C72S ) ] were isolated by the miniMos method ( Frøkjær-Jensen et al . , 2014 ) . Images were taken using an Olympus PlanAPO with a 100 × 1 . 4 NA objective and an ORCA100 CCD camera ( Hamamatsu ) . Young adult worms were immobilized with 30 mg/ml BDM ( 2 , 3-Butanedione monoxide , Sigma ) . Image stacks were captured and the maximum intensity projections were obtained using Metamorph 7 . 1 software ( Molecular Devices ) . Calcium imaging was performed in a microfluidic device essentially as described ( Chronis et al . , 2007; Ha et al . , 2010 ) with minor modification . Fluorescence time-lapse imaging was collected on a Nikon Eclipse Ti-U inverted microscope with a 40X oil immersion objective and a Yokogawa CSU-X1 scanner unit and a Photometrics CoolSnap EZ camera at five frames second–1 . The GCaMP6 . 0s signal from the soma of the ASJ neurons was measured using Fiji . The change in the fluorescence intensity ( ∆F ) for each time point was the difference between its fluorescence intensity and the average intensity over the 20 s of the recording before stimulus onset ( Fbase ) : ∆F = F–Fbase . To analyze the response evoked by the onset of the NO donor for each genotype , the average ∆F/Fbase% within a 3 s window prior to the stimulus onset was compared with the average ∆F/Fbase% within a 10 s window after the switch; average ∆F/Fbase% 10 s after onset minus average ∆F/Fbase% 3 s before onset generates the ‘ON response’ ( Figure supplements for Figures 4–7 ) . To analyze the response evoked by the removal of the NO donor , the average ∆F/Fbase% within a 3 s window prior to the stimulus removal was compared with the average ∆F/Fbase% within a 20 s window after the switch; average ∆F/Fbase% 20 s after removal minus average ∆F/Fbase% 3 s before removal generates ‘OFF response’ ( Figure supplements for Figures 4–7 ) . Data was assessed for a normal distribution using the Shapiro-Wilk normality test . To assess whether a genotype generates an ON or OFF response , paired sample t-test was used for in group comparison for normally distributed data and a Wilcoxon signed rank test was used ( SPSS Statistics ) for data that were not normally distributed . The ON or OFF responses of mutants in Figures 4–7 were compared with a common set of wild-type controls using Kruskal–Wallis one-way analysis of variance , which analyzes multiple comparisons on data that are not entirely normally distributed ( Figure supplements for Figures 4–7 ) . Fresh NO donor solution was prepared before each recording session by dissolving 10 mg of DPTA NONOate ( Cayman Chemical , Item Number 82110 ) in 15 ml of nematode growth medium buffer ( 3 g/L NaCl , 1 mM CaCl2 , 1 mM MgSO4 , 25 mM potassium phosphate buffer , pH6 . 0 ) and kept at room temperature for 30 min to allow adequate NO release before use .	Nitric oxide is a colorless gas that contains one nitrogen atom and one oxygen atom . Found at very low levels in the air , this gas is produced by the intense heat of lightning strikes and by combustion engines . Almost all living organisms also produce nitric oxide . In animals , for example , nitric oxide regulates blood pressure and signaling between neurons . However , it was not known if animals could detect nitric oxide in their environment and respond to it . Caenorhabditis elegans is a worm that has been intensively studied in many fields of biology . Unlike most animals , it cannot make nitric oxide . Yet , living in the soil , C . elegans does come into contact with many microbes that can , including the bacterium Pseudomonas aeruginosa . These bacteria can infect and kill C . elegans , and so the worm typically avoids them . Hao , Yang et al . asked whether C . elegans does so by detecting the nitric oxide that these harmful bacteria release into their environment . First , worms were added to a petri dish where a small patch of P . aeruginosa was growing . Consistent with previous results , the worms had all moved away from the bacteria after a few hours . The experiments were then repeated with mutant bacteria that cannot produce nitric oxide . The worms were less likely to avoid these mutant bacteria , suggesting that C . elegans does indeed avoid infection by detecting bacterially produced nitric oxide . Next , using a range of techniques , Hao , Yang et al . showed that C . elegans avoids nitric oxide released into its environment by detecting the gas via a pair of sensory neurons . These neurons require several specific proteins to be able to detect nitric oxide and respond to it . In particular , a protein called Thioredoxin was found to determine the beginning and end of the worm’s sensory response to nitric oxide . All of these proteins are also found in many other animals , and so it is possible that these findings may be relevant to other species too . Further studies are now needed to confirm whether other organisms can sense nitric oxide from their environment and , if so , how their nervous systems equip them to do this .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience", "microbiology", "and", "infectious", "disease" ]	2018	Thioredoxin shapes the C. elegans sensory response to Pseudomonas produced nitric oxide
The first wave of transcriptional activation occurs after fertilisation in a species-specific pattern . Despite its importance to initial embryonic development , the characteristics of transcription following fertilisation are poorly understood in Aves . Here , we report detailed insights into the onset of genome activation in chickens . We established that two waves of transcriptional activation occurred , one shortly after fertilisation and another at Eyal-Giladi and Kochav Stage V . We found 1544 single nucleotide polymorphisms across 424 transcripts derived from parents that were expressed in offspring during the early embryonic stages . Surprisingly , only the maternal genome was activated in the zygote , and the paternal genome remained silent until the second-wave , regardless of the presence of a paternal pronucleus or supernumerary sperm in the egg . The identified maternal genes involved in cleavage that were replaced by bi-allelic expression . The results demonstrate that only maternal alleles are activated in the chicken zygote upon fertilisation , which could be essential for early embryogenesis and evolutionary outcomes in birds . The genetic events of early embryogenesis are initiated by zygotic genome activation ( ZGA ) ( Lee et al . , 2014; Tadros and Lipshitz , 2009 ) . The timing and mechanisms of ZGA have been investigated in various species ( Aanes et al . , 2011; Baugh et al . , 2003; Harvey et al . , 2013; Karr et al . , 1985; Lee et al . , 2013b; Leichsenring et al . , 2013; Liang et al . , 2008; Newport and Kirschner , 1982; Poccia et al . , 1985; Tan et al . , 2013 ) . In mammals , the first wave ( 1st wave ) of transcriptional activation ( also known as minor ZGA ) occurs after fertilisation , during pronucleus ( PN ) formation . The subsequent second wave ( 2nd wave ) of transcriptional activation ( major ZGA ) occurs during the two-cell stage of mice and the eight-cell stage of humans ( Aoki et al . , 1997; Braude et al . , 1988; Xue et al . , 2013 ) . In avian species , reports in chicken and quail embryos have described gene activation during early cell cleavage ( Nagai et al . , 2015; Olszańska et al . , 1984 ) , but transcriptional activation has not been investigated during fertilisation . Recent studies suggest that there are two waves of ZGA in chickens based on mRNA profile ( Hwang et al . , 2018aHwang et al . , 2018c ) . However , it is necessary to examine features such as de novo transcription in order to determine the timing and mechanisms of ZGA precisely . The 1st wave of ZGA exhibits numerous characteristics that are species-dependent . In mice , the most distinctive feature of the 1st wave in the PN stage is that transcription from the paternal PN is greater than that from the maternal PN , due to the epigenetic regulation of the latter ( Aoki et al . , 1997; Aoshima et al . , 2015; Bouniol et al . , 1995; Wu et al . , 2016; Zhang et al . , 2016 ) . In addition , the 1st wave is highly promiscuous , in that the expression of untranslatable mRNAs and intergenic regions is observed ( Abe et al . , 2015 ) . In zebrafish , the mitochondrial genome is activated in the one-cell embryo ( Heyn et al . , 2014 ) . In plants , the zygotic genome is activated soon after fertilisation , and rice zygotes show asymmetric activation of parental genomes ( Anderson et al . , 2017; Chen et al . , 2017 ) . As the earliest expressed genes in ZGA are species-specific ( Heyn et al . , 2014 ) , the patterns of transcription during the 1st wave should be examined so that we can understand early embryogenesis in each species . However , no detailed investigation of transcription at fertilisation in avian species has been reported . As polyspermy is a distinctive feature in birds ( Snook et al . , 2011; Iwao , 2012 ) , we hypothesised that the 1st wave derived from the parental genome would exhibit unique characteristics . Here , we conducted a genome-wide study of primary transcripts to clarify which genes undergo transcriptional activation during embryogenesis in chicken . We identified avian-specific expression patterns of the parental genome during the 1st wave . The results provide intriguing insights into initial the genome activation associated with physiological characteristics upon fertilisation in birds . Detection of de novo transcription after fertilisation is difficult because of the large number of mRNAs that are being processed in the oocyte . We examined primary transcripts toassess the existence and timing of transcriptional activation accurately , using previously generated bulked embryonic whole-transcriptome sequencing ( WTS ) data ( Hwang et al . , 2018aHwang et al . , 2018c ) ( Figure 1A ) . Hierarchical clustering of precursor mRNA ( pre-mRNA ) expression demonstrated that zygotes differed from oocytes , suggesting dynamic changes in primary transcripts after fertilisation ( Figure 1B ) . Phosphorylated RNA polymerase II C-terminal domain first appeared during the late EGK . II to early EGK . III ( Nagai et al . , 2015 ) , but the expression of pre-mRNA differed between EGK . III and EGK . VI ( Figure 1B ) . The number of upregulated pre-mRNAs that are found in the zygote when compared to the oocyte provides evidence of a 1st wave ( Figure 1C ) . In addition , a large number of pre-mRNAs were upregulated between EGK . III and EGK . VI , revealing the presence of a 2nd wave . This result is more direct evidence of the existence and timing of two waves of ZGA in chicken . A number of expressed regions exhibited significant differences between the oocyte and zygote and between EGK . III and EGK . VI ( Figure 1—figure supplement 1 ) . The number of expressed regions was reduced during EGK . I and EGK . III but increased after EGK . VI . Of all of the genomic regions that are expressed , the proportion of expressed intronic regions decreased after fertilisation and increased gradually after EGK . VI ( Figure 1—figure supplement 2 ) . Unlike the expression patterns seen during the minor ZGA in mammals ( Abe et al . , 2015 ) , the proportion of expressed intergenic regions was constant regardless of transcriptional activation , indicating no expression of these regions during the 1st wave in chickens . In genic regions , large numbers of up- and downregulated mRNAs and long intergenic noncoding RNAs ( lincRNAs ) were observed during the 1st wave , while other RNAs were mostly downregulated after fertilisation ( Figure 1—figure supplement 3 ) , suggesting a potential role for long transcripts in the early cleavage stages . All RNA types were significantly upregulated during the 2nd wave . We examined the candidate genes affected by the two waves using reverse transcription PCR ( RT-PCR ) . Six upregulated genes in each wave were selected as representative genes ( Supplementary file 1 ) : DLX6 , GATA2 , ZIC4 , LYPD2 , IFITM5 and NKX6-3 for the 1st wave , and WNT11 , WNT3A , C8ORF22 , NAT8L , PCOLCE2 and AKAP2 for the 2nd wave . We successfully demonstrated two waves of transcriptional activation for all of the selected genes ( Figure 2 and Figure 2—figure supplement 1 ) . The validated genes belonging to the 2nd wave of activation indicated a lack of transcriptional activity during rapid cellularisation in the cleavage period , and showed that the 2nd wave of transcriptional activation in chicken occurred not between EGK . II and EGK . III , but between EGK . IV and EGK . V . The existence and timing of the two distinct waves of transcriptional activation were also confirmed experimentally and were consistent with the results of the bulked embryonic WTS analyses . We hypothesised that the haploid nucleus of supernumerary sperm could be substantially induced during the 1st wave in addition to paternal and maternal PN activation because polyspermic fertilisation occurs in avian species . To assess this hypothesis , we generated multiomics data including whole-genome sequencing ( WGS ) and WTS . We completed WGS of six parents ( three male Korean Oge ( mKO ) and three female White Leghorn ( fWL ) chickens ) to identify breed-specific single-nucleotide polymorphisms ( SNPs ) ( Figure 3A ) . We also generated single embryonic WTS data from hybrid oocyte , zygote and EGK . X blastoderms derived from the WGS-sequenced parents to examine the characteristics of the 1st wave-activated transcripts and of allelic expression . After confirming hybrid embryo formation between mKO and fWL ( Figure 3—figure supplement 1 ) , we collected oocytes , zygotes and EGK . X blastoderms from hens on the same day ( Figure 3—figure supplement 2 ) . Each embryo contained an average of 2 . 1 µg of total RNA ( Supplementary file 2A ) . We performed the same analysis used in bulked embryonic WTS on single embryonic WTS to further establish the characteristics of the 1st wave . The WTS samples generated from the single embryos were clustered according to their respective stages ( Figure 3B ) . A total of 4275 differentially expressed mRNAs were detected ( Figure 3C; FDR-adjusted p<0 . 05 ) , among which 1883 were upregulated and 2392 were downregulated in the zygote stage compared to the oocyte . We also observed that 118 and 786 lincRNAs were up- and downregulated , respectively . Owing to the dramatic changes in early development between fertilisation and oviposition , 10 , 298 mRNAs and 2507 lincRNAs were differentially expressed between the zygote and EGK . X stages ( Figure 3C ) . We also observed a large number of primary transcripts that are upregulated in the zygote stage when compared to the oocyte stage ( Supplementary file 2B; FDR-adjusted p<0 . 05 ) . These results once again demonstrate that primary transcriptional activation occurs as developmental stage moves from oocytes to zygotes at single-embryo resolution , in terms of the numbers of differentially expressed pre-mRNAs and long transcripts . Next , we identified parental allele-specific expression patterns during the 1st wave of transcriptional activation . A total of 1544 parentally derived SNPs were detected , distributed across 424 transcripts including mRNAs and lincRNAs ( Supplementary file 3A ) . Interestingly , all of the transcripts that were identified in the zygote stage exhibited maternally derived expression during the 1st wave ( Figure 4A and Supplementary file 3A ) . Most of the maternally derived transcripts , except for seven mRNAs and two lincRNAs , were replaced as bi-allelic expression occurred in the EGK . X stage . These nine transcripts could be interpreted as residual maternal transcripts that were not activated during the 2nd wave , rather than as genomic-imprinted genes , which are not conserved in avian species ( Frésard et al . , 2014 ) . To verify this observation , we selected six pre-mRNAs ( MAP7D1 , ESCO1 , CCNB3 , SYTL1 , GRHL1 and LLGL1 ) that are upregulated during the 1st wave as representatives and validated the genotypes using Sanger sequencing ( Figure 4B and Supplementary file 3B ) . All of the selected genes showed maternal allelic expression in the zygote until the EGK . VI stage , except for the GRHL1 gene . These maternally derived genes converted to bi-allelic expression after the maternal-to-zygotic transition ( MZT ) at EGK . X . This phenomenon is distinguished from that in mammals , in which transcriptional activity in the paternal PN is two times greater than that in the maternal PN ( Aoki et al . , 1997 ) . These results indicate that there is no possibility that the activated transcripts are derived from the supernumerary sperm nuclei and paternal PN , in contrast to the data from mammals ( Aoki et al . , 1997; Bouniol et al . , 1995 ) . We examined the functional characteristics of the maternal genes that are activated during the 1st wave of transcriptional activation identified from the single embryonic WTS data . The analysis revealed that the 1st wave-activated maternal transcripts were enriched in the following pathways: cell cycle; Notch signalling pathway; Wnt signalling pathway; regulation of transcription , DNA-templated; and regulation of small GTPase-mediated signal transduction ( Figure 4—figure supplement 1 and Supplementary file 4 ) . These pathways were activated from the maternal genome and are involved in rapid asymmetric cellularisation during the cleavage period in chickens ( Hwang et al . , 2018c ) and other species ( Castanon et al . , 2013; Huang et al . , 2015; Priess , 2005; Tse et al . , 2012; Zhang et al . , 2014 ) . While the 1st wave in mice promotes the low-level expression of numerous non-functional genes ( Abe et al . , 2015 ) , the maternal genes activated during the 1st wave in chickens seem to be related to early cell division in embryogenesis . As demonstrated in previous studies , the characteristics of the 1st wave vary among species ( Abe et al . , 2015; Anderson et al . , 2017; Chen et al . , 2017; Heyn et al . , 2014 ) . Our results suggest the exclusive activation of maternal alleles after fertilisation in chicken ( Figure 4C ) . However , after MZT , most expressed genes were derived from both paternal and maternal genomes . Functionally , transcripts affected by the 1st wave were involved in asymmetric rapid cellularisation and in the fundamental regulation of further development ( Figure 4C ) . We speculate that this evolved by necessity in animals following physiological polyspermy ( Figure 4—figure supplement 2 ) . Polyspermic animals require a number of sperm to activate large eggs ( Iwao , 2012 ) . In addition to pathological mitosis ( Snook et al . , 2011 ) , polyspermic embryos of sea urchin demonstrated that transcriptional activation after fertilisation was greatly stimulated by the PN of supernumerary sperm ( Poccia et al . , 1985 ) . Such a disproportionate genome contribution could result in an excessive amount of transcription . The total polyspermy number reportedly varies ( Hemmings and Birkhead , 2015; Lee et al . , 2013a ) and is positively correlated with egg size ( Birkhead et al . , 1994 ) . Individual sperm provide genomic diversity ( Wang et al . , 2012 ) but could result in genomic instability if different types of transcripts are expressed by various sperm nuclei . Therefore , polyspermic animals may have evolved means of inhibiting the activation of the paternal PN to control gene expression levels from the 1st wave . Taken together , our results suggest that the maternally derived 1st wave is essential for early development and evolutionary outcomes in avian species . The experimental use of chickens was approved by the Institute of Laboratory Animal Resources , Seoul National University ( SNU-150827–1 ) . The experimental animals were cared for according to a standard management program at the University Animal Farm , Seoul National University , Korea . The procedures for animal management , reproduction and embryo manipulation adhered to the standard operating protocols of our laboratory . To detect de novo transcription , the analytical approach to primary transcripts used in previous studies of other species ( Abe et al . , 2015; Graf et al . , 2014; Lee et al . , 2013b; Paranjpe et al . , 2013 ) was followed . In the quantification step , four types of genomic regions were considered: transcripts , exons , introns and intergenic regions . Although quantification of the transcript and exon level can be achieved directly without any pre-processing steps by using the galGal4 gene annotation file ( GTF ) , the genomic position needs to be defined in order to estimate the expression levels of the intron and intergenic regions . When defining intron area , overlapping annotation of the exon within the associated gene makes it difficult to define intron regions from the reference genome . In addition , information from different strands should be considered when defining intron regions between each exon . To address these issues , intron region was defined using custom python script ( Source code 1 ) . As in the method used to define the intron region between exons within the associated gene , python script was used to define intergenic regions between genes within the same chromosome . After defining intronic and intergenic regions , a GTF was generated using the coordinate information . Expression levels were measured with HTSeq-count ( v 0 . 6 . 1 ) on the basis of the the GTFs ( Anders et al . , 2015b ) . To explore gene expression changes during early developmental stages , pre-existing bulked embryonic WTS data covering the oocyte , zygote , EGK . I , EGK . III , EGE . VI , EGK . VIII and EGK . X stages ( GSE86592 ) ( Hwang et al . , 2018bHwang et al . , 2018c ) were employed . Three types of matrix data were generated , and these data were employed in statistical analyses . Six statistical tests , oocyte vs . zygote , zygote vs . EGK . I , EGK . I vs . EGK . III , EGK . III vs . EGK . VI , EGK . VI vs . EGK . VIII and EGK . VIII vs . EGK . X , were performed using the edgeR package ( Robinson et al . , 2010 ) in the matrix data derived from intron and intergenic regions separately . More detailed contrast tests were performed on the generalised linear model . In this study , a result was considered significant at a FDR-adjusted p-value of p<0 . 05 ( Benjamini and Hochberg , 1995 ) . Genomic DNA was isolated from blood collected from the wing vein of six parental chickens ( three mKO and three fWL ) using 1 mL 30-gauge syringes ( Shina Corporation , Seoul , Korea ) . The blood samples were transferred into EDTA tubes ( BD Biosciences , San Jose , CA , USA ) immediately after collection . Blood ( 10 µL ) was used for isolation of genomic DNA using a DNeasy Mini Kit ( Qiagen , Valencia , CA , USA ) . The quality of the extracted genomic DNA was determined using the Trinean DropSense96 system ( Trinean , Gentbrugge , Belgium ) , RiboGreen ( Invitrogen , Carlsbad , CA , USA ) and an Agilent 2100 Bioanalyzer ( Agilent Technologies , Santa Clara , CA , USA ) . Genomic DNA was used for the construction of cDNA libraries using a TruSeq Nano DNA LT Library Preparation Kit ( Illumina Inc . , San Diego , CA , USA ) . The resulting libraries were subjected to chicken genome resequencing ( 30 × coverage ) using the Illumina Nextseq 500 platform to produce paired 150 bp reads . The raw sequencing data were deposited in BioProject under accession number PRJNA393895 . Before collecting early embryos , EGK . X blastoderms formed from crosses between mKO and fWL were incubated in a chamber at 37 . 5°C under 80% humidity for 18 hr . Genomic DNA was isolated from Hamburger and Hamilton stage 4 ( HH4 ) ( Hamburger and Hamilton , 1951 ) embryos using a DNeasy Mini Kit ( Qiagen ) . RT-PCR was performed to confirm hybridisation between KO and WL using breed-specific primers ( AS3554-I9/P5FWD WL F: 5′-AGC AGC GGC GAT GAG CGG TG-3′; WL R: 5′-CTG CCT CAA CGT CTC GTT GGC-3’; AS3554-WT/P5FWD KO F: 5’-AGC AGC GGC GAT GAG CAG CA-3′; KO R: 5′-CTG CCT CAA CGT CTC GTT GGC-3′ ) ( Choi et al . , 2007 ) , with an initial incubation at 95°C for 10 min , followed by 35 cycles of 95°C for 30 s , 69°C for 30 s and 72°C for 30 s . The reaction was terminated after a final incubation at 72°C for 10 min . The paired-end reads for six chickens ( three biological replications of mKO and fWL breeds ) were generated using the Illumina Nextseq 500 platform . In total , 8 . 38 billion reads or ~2 . 53 Gbp of sequences were generated . Paired-read sequences were selected for quality using Trimmomatic ( v0 . 33 ) ( Bolger et al . , 2014 ) . Using Bowtie 2 ( v2 . 2 . 5 ) ( Langmead and Salzberg , 2012 ) , reads were aligned to the reference genome sequence galGal4 ( Build v 4 . 82 ) with an average alignment rate of 91 . 61% . After potential PCR duplicates were filtered and misalignments resulting from the presence of insertions and deletions ( INDELs ) were corrected , SNPs were detected using GATK v3 . 4 . 46 ( McKenna et al . , 2010 ) . More detailed , potential PCR duplicates were filtered using the option ‘REMOVE_DUPLICATES = true’ in the ‘MarkDuplicates’ open-source tool of Picard ( v 1 . 138 ) ( https://broadinstitute . github . io/picard/ ) . SAMtools ( v1 . 2 ) ( Li et al . , 2009 ) was then employed to create index files for reference and Binary Alignment/Map ( BAM ) files . In the variant-calling step with GATK v3 . 1 , local realignment of reads to correct misalignments was performed because of the presence of INDELs ( ‘RealignerTargetCreator’ and ‘IndelRealigner’ arguments ) . In the GATK tool , two types of arguments , ‘UnifiedGenotyper’ and ‘SelectVariants’ were employed for variant calling . In addition , ‘VariantFiltration’ was applied to filter bad variants on the basis of the following criteria: ( 1 ) variants with a Phred-scaled quality score <30 were filtered; ( 2 ) SNPs with ‘mapping quality zero ( MQ0 ) >4’ , ‘quality depth <5’ and ‘ ( MQ0 / ( 1 . 0*DP ) ) >0 . 1’ were filtered; and ( 3 ) SNPs with ‘Phred-scaled P value using Fisher’s exact test >200’ were filtered . As a result , 10 , 529 , 469 variants were detected , of which 9 , 805 , 997 variants ( 93 . 129% ) were previously known variants ( Supplementary file 5A , B ) . The egg-laying times of three fWLs , which were mated with mKOs , were recorded . A single hybrid EGK . X blastoderm was collected from WL hens after oviposition . To collect single oocytes and hybrid zygotes , WL hens were sacrificed and their follicles were harvested . Oocytes and hybrid zygotes were collected simultaneously from one WL hen . Owing to the small transcriptomic differences between pre- and post-ovulatory oocytes observed in the previous study ( Elis et al . , 2008 ) and the infeasibility of simultaneous acquisition of post-ovulatory oocytes and zygotes from a single hen , only the pre-ovulatory large F1 oocyte was isolated . Only zygote embryos not showing cleavage and located in the magnum were collected within 1 . 5 hr after fertilisation , according to the recorded egg-laying times ( Figure 3—figure supplement 2 ) . All embryos were classified according to morphological criteria ( Figure 1A ) . Shortly after collection , the embryos were separated from the egg using sterile paper , and the shell membrane and albumen were detached from the yolk . A piece of filter paper ( Whatman , Maidstone , UK ) with a hole in the centre was placed over the germinal disc . After cutting around the paper containing the embryo , it was gently turned over and transferred to saline to further remove the yolk and vitelline membrane to allow embryo collection . Total RNA was isolated from early embryos using TRIzol reagent ( Invitrogen ) . The quality of the extracted total RNA was determined using the Trinean DropSense96 system ( Trinean ) , RiboGreen ( Invitrogen ) and an Agilent 2100 Bioanalyzer ( Agilent Technologies ) . Total RNA was used for construction of cDNA libraries using a TruSeq Stranded Total RNA Sample Preparation Kit ( Illumina , Inc . ) . The resulting libraries were subjected to whole-transcriptome analysis using the Illumina Nextseq 500 platform to produce paired 150 bp reads . The raw sequencing data were deposited in Gene Expression Omnibus ( GEO ) under accession number GSE100798 . Trimmomatic ( v 0 . 33 ) ( Bolger et al . , 2014 ) was used to generate clean reads . Per-base sequence qualities were checked using FastQC ( v 0 . 11 . 2 ) ( Andrews , 2010 ) and filtered fastq files . Trimmed reads were aligned to the galGal4 genome files using the HISAT2 alignment software ( v 2 . 0 . 0 ) ( Kim et al . , 2015 ) with the following alignment option: ‘--rna-strandness RF’ . Sequence Alignment/Map ( SAM ) files were converted into compressed and sorted BAM files using SAMtools ( v 1 . 4 . 1 ) ( Li et al . , 2009 ) . The mapped reads were quantified using HTSeq-count ( Anders et al . , 2015a ) with the merged GTF , with total RNAs and lincRNAs derived from Ensembl and ALDB ( Li et al . , 2015 ) , respectively . The quantification of mapped reads on intronic regions for single embryonic WTS data was performed using the procedure also used for bulked embryonic WTS data . Using the alignment file ( . BAM ) , potential PCR duplicates were removed using the Picard ( v 1 . 138 ) software with ‘REMOVE_DUPLICATES = true’ in the ‘MarkDuplicates’ option . After that , the SplitNCigarReads tool implemented in GATK was performed with the ‘-rf ReassignOneMappingQuality -RMQF 255 -RMQT 60 U ALLOW_N_CIGAR_READS’ option . In the variant-calling step with GATK , local realignment of reads was performed to correct misalignments ( using the ‘RealignerTargetCreator’ and ‘IndelRealigner’ options ) . Finally , base-recalibration was performed using BaseRecalibrator implemented in GATK with known variant sites in galGal4 . Using HaplotypeCaller in the GATK tool , variant calling was performed with the ‘-dontUseSoftClippedBases -stand_call_conf 20 . 0 -stand_emit_conf 20 . 0’ option . Finally , bad variants were filtered using the VariantFiltration tool with ‘-window 35 -cluster 3 -filterName FS -filter ‘FS >30 . 0’ -filterName QD -filter ‘QD <2 . 0’’ option . At the end of this process , 265 , 788 variants were detected , of which 248 , 030 variants ( 93 . 319% ) were previously known sites ( Supplementary file 5C , D ) . Maternal and paternal samples were genotyped using WGS , and their offspring , including maternal oocytes , were genotyped using WTS ( variant calling on the RNA-Seq data ) . After pre-processing , there were two types of genotype data ( DNA and RNA sequencing data ) available for the mother , father , oocyte , zygote and EGK . X . In two types of SNP data , 10 , 529 , 469 and 265 , 788 variants were detected in DNA and RNA sequencing data , respectively . First , breed-specific SNPs ( such as , first , SNPs ‘0/0’ and ‘1/1’ genotype for maternal and paternal groups , respectively; and second , SNPs ‘1/1’ and ‘0/0’ genotype for maternal and paternal groups , respectively ) were identified and annotated using SnpSift ( Cingolani et al . , 2012 ) in parental SNP data . As a result , 216 , 003 SNPs were identified as breed-specific SNPs . After that , two SNP datasets ( breed-specific SNPs and their offspring genotypes derived from the RNA-Seq data ) were combined to detect maternally and paternally expressed genes , and 14 , 817 SNPs were commonly identified in breed-specific SNPs and those derived from RNA-Seq data . Using these combined genotype data , three types of filtering steps were carried out . First , mismatched genotypes of the reference and alternative allele between breed-specific SNPs and SNPs derived from the RNA-Seq were removed; two variants were removed in this step . Second , different genotypes within the biological replicates were removed; 9 , 143 SNPs were removed in this step . Finally , mismatched genotypes between maternal samples and oocyte samples were removed; six SNPs were removed in this step . The remaining 5 , 666 SNPs were annotated using the SnpSift tool with galGal4 and ALDB GTFs . To find the most conservative evidence of parental expression , if a single SNP was found within the gene or genotype pattern that was not consistent among the SNPs , it was filtered out . In addition , unannotated SNPs in both databases , Ensembl and ALDB , were removed to facilitate biological interpretation . At the end of this process , 1 , 544 SNPs were detected as parental expression markers , all of which showed a maternal expression pattern ( Supplementary file 5E ) . On the basis of the biological process terms ( BP terms ) of the GO and KEGG pathways , functional enrichment tests using DAVID ( Dennis et al . , 2003 ) were performed on the differentially expressed genes . Total RNA ( 1 µg ) was used as the template for cDNA synthesis using the SuperScript III First-Strand Synthesis System ( Invitrogen ) . The cDNA was serially diluted 5-fold and equalised quantitatively for PCR amplification . To validate allelic expression , additional single hybrid embryos at EGK . III and VI were collected from parents with identical genotypes as confirmed by WGS , and their total RNA isolation and cDNA synthesis were performed as described above . Primers for exon–intron PCR of 12 genes and for allelic expression of six genes were designed using the program Primer3 ( Untergasser et al . , 2012 ) ( Supplementary file 6A , B ) . RT-PCR was performed with an initial incubation at 95°C for 5 min , followed by 35 cycles of 95°C for 30 s , 59°C for 30 s and 72°C for 30 s . The reaction was terminated after a final incubation at 72°C for 5 min . PCR products were cloned into the pGEM-T Easy Vector ( Promega , Madison , WI , USA ) for sequencing with an ABI 3730xl DNA Analyzer ( Applied Biosystems , Foster City , CA , USA ) .	The early stages of animal development involve a handover of genetic control . Initially , the egg cell is maintained by genetic information inherited from the mother , but soon after fertilization it starts to depend on its own genes instead . Activating genes inside the fertilized egg cell ( zygote ) so that they can take control of development is known as zygotic genome activation . Despite the fact that birds are often used to study how embryos develop , zygotic genome activation in birds is not well understood . Fertilization in birds , including chickens , is different to mammals in that it requires multiple sperm to fertilize an egg cell . As such , zygotic genome activation in birds is likely to differ from that in mammals . By examining gene expression in embryos from mixed-breed chickens , Hwang , Seo et al . showed that there are two stages of zygotic genome activation in chickens . The genes derived from the mother become active in the first stage , while genes from the father become active in the second stage . Genome activation in birds is therefore very different to the same process in mammals , which involves genome activation of both parents from the first stage . This extra level of control may help to prevent genetic complications resulting from the presence of multiple sperm , each of which carries a different set of genes from the father .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "short", "report", "genetics", "and", "genomics" ]	2018	Zygotic gene activation in the chicken occurs in two waves, the first involving only maternally derived genes
Human cytomegalovirus ( HCMV ) is endowed with multiple highly sophisticated immune evasion strategies . This includes the evasion from antibody mediated immune control by counteracting host Fc-gamma receptor ( FcγR ) mediated immune control mechanisms such as antibody-dependent cellular cytotoxicity ( ADCC ) . We have previously shown that HCMV avoids FcγR activation by concomitant expression of the viral Fc-gamma-binding glycoproteins ( vFcγRs ) gp34 and gp68 . We now show that gp34 and gp68 bind IgG simultaneously at topologically different Fcγ sites and achieve efficient antagonization of host FcγR activation by distinct but synergizing mechanisms . While gp34 enhances immune complex internalization , gp68 acts as inhibitor of host FcγR binding to immune complexes . In doing so , gp68 induces Fcγ accessibility to gp34 and simultaneously limits host FcγR recognition . The synergy of gp34 and gp68 is compelled by the interfering influence of excessive non-immune IgG ligands and highlights conformational changes within the IgG globular chains critical for antibody effector function . Human cytomegalovirus ( HCMV ) constitutes the prototypical human pathogenic β-herpesvirus found worldwide with high immunoglobulin G ( IgG ) sero-prevalence rates of 56–94% depending on the respective countries ( Zuhair et al . , 2019 ) . A hallmark of cytomegalovirus infection is the establishment of a lifelong persistence with recurring phases of latency and reactivation of productive infection and horizontal spread in presence of adaptive immune responses . HCMV encodes the largest known transcriptome of human viruses ( Stern-Ginossar et al . , 2012 ) , giving rise to an equally large antigenic proteome including a huge and varied arsenal of immunoevasins ( Berry et al . , 2020; Hengel et al . , 1998 ) that counteract immune recognition of infected cells facilitating virus persistence , shedding , and superinfection of sero-positive hosts . While primary HCMV infection of healthy individuals usually remains undetected , it can cause severe symptoms in the immunocompromised . Besides antiviral therapy involving nucleotide analogs , concentrated HCMV-immune IgG preparations ( e . g . Cytotect ) are used to prevent infection of immunocompromised patients including the prevention of congenital infection in primary infected pregnant women with varying degrees of success ( Kagan et al . , 2019; Revello et al . , 2014 ) . Generally , HCMV is observed to withstand a humoral immunity even replicating and disseminating in the presence of highly neutralizing immune sera in vitro and clinical isolates of HCMV tend to disseminate via cell-to-cell spread , thus avoiding an encounter with neutralizing antibodies ( Falk et al . , 2018 ) . This , however , cannot fully explain the resistance of HCMV to a humoral response leading to virus dissemination across several organs . Notably , HCMV encodes a set of Fcγ-binding glycoproteins ( viral FcγRs , vFcγRs ) that have been shown to antagonize host FcγR activation by immune IgG ( Corrales-Aguilar et al . , 2014b ) . In this previous study , we showed that HCMV gp34 , gp68 and HSV-1 gE/gI efficiently antagonize the activation of human FcγRs . By attacking the conserved Fc part of IgG , HSV-1 and HCMV are able to counteract potent antiviral immune responses including antibody-dependent cellular cytotoxicity ( ADCC ) . As it has become increasingly evident that Fcγ mediated immune control is essential not only for the effectiveness of non-neutralizing but also neutralizing IgG antibodies ( DiLillo et al . , 2014; Forthal et al . , 2013; Horwitz et al . , 2017; Van den Hoecke et al . , 2017 ) , a mechanistic and causal analysis of this evasion process is highly warranted in the pursuit of better antibody based intervention strategies targeting herpesviruses and HCMV in particular . While several herpesviruses encode vFcγRs , HCMV is the only virus known so far that encodes more than one individual molecule with the capacity to bind Fcγ . Specifically , HCMV encodes four distinct molecules which share this ability: gp68 ( UL119-118 ) , gp34 ( RL11 ) , gp95 ( RL12 ) , and gpRL13 ( RL13 ) ( Atalay et al . , 2002; Corrales-Aguilar et al . , 2014a; Cortese et al . , 2012; Sprague et al . , 2008 ) . Turning to other species , mouse CMV ( MCMV ) encoded m138 has been shown to bind Fcγ , yet it has more prominently been associated with a variety of unrelated functions proving m138 not to be a strict homolog of the HCMV counterparts ( Arapović et al . , 2009; Lenac et al . , 2006 ) . We recently found Rhesus CMV ( RhCMV ) to encode an Fcγ-binding protein in the Rh05 gene ( RL11 gene family ) seemingly more closely related to its HCMV analog ( Kolb et al . , 2019 ) . This is supported by the fact that gpRh05 , as HCMV vFcγRs gp34 and gp68 , is able to generically antagonize activation of all macaque FcγRs . While it is clear that by targeting the invariant part of the key molecule of the humoral immune response , vFcγRs have the potential to manipulate a multitude of antibody mediated immune functions , their role in vivo has yet to be determined . While the function of HCMV vFcγRs gp34 and gp68 as antagonists of host FcγRs has been established ( Corrales-Aguilar et al . , 2014b ) , the underlying mechanism ( s ) had not been addressed yet . In recent years it has been shown that gp68 and gp34 are able to engage in antibody bipolar bridging ( ABB ) forming ternary complexes consisting of antigen , antibody , and vFcγR ( Corrales-Aguilar et al . , 2014a; Corrales-Aguilar et al . , 2014b; Sprague et al . , 2008 ) . Moreover , gp68 has been shown to bind IgG in a 2:1 ratio and has the ability to internalize and translocate IgG to lysosomal compartments , while gp34 has been shown to form predominantly homo-dimeric structures ( Ndjamen et al . , 2016; Sprague et al . , 2008 ) . However , no studies have yet been performed in the context of HCMV infection investigating the coincident disposition of gp34 and gp68 at the plasma membrane and their functional interaction during the early and late phase of HCMV replication . Here , we show gp34 and gp68 to antagonize host FcγR activation by distinct but highly cooperative modes of Fcγ targeting , leading to efficient evasion from antibody mediated immune control by division of labor . As we observed gp34 and gp68 to work in concert to maximize FcγR inhibition , we set out to elucidate the mechanisms by which the vFcγRs achieve this synergistic effect . When comparing the amino acid sequences of the cytosolic domains of gp34 and gp68 it is revealed that both molecules encode distinct sorting motifs ( Figure 3A; Atalay et al . , 2002 ) ( source: AD169 pBAC-2 sequence ) . The cytosolic tail of gp68 harbors an YXXΦ motif seven aa downstream of the transmembrane domain . Such a motif has been described as a marker for lysosomal targeting ( Bonifacino and Traub , 2003 ) and is in line with the lysosomal translocation of gp68 shown in a previous study ( Sprague et al . , 2008 ) . In HCMV gB , a non-Fcγ-binding HCMV encoded envelope glycoprotein , the location of two such motifs is further away from the transmembrane domain indicating it being destined for general endocytosis rather than immediate degradation , which is shared by another HCMV vFcγR , gpRL13 , found primarily in intracellular compartments as well as HSV-1 gE , forming a heterodimeric gE/gI vFcγR described to internalize IgG and recycle back to the cell surface ( Figure 3—figure supplement 1; Bonifacino and Traub , 2003; Cortese et al . , 2012; Ndjamen et al . , 2016; Sprague et al . , 2004 ) . Conversely , the cytosolic tail of gp34 encodes a di-leucine [D/E]XXX[LL/LI] motif which we find to be unique among surface-resident vFcγRs . Therefore we suspected gp34 and gp68 to possess different dynamics regarding IC internalization . As internalization studies in the context of HCMV infection proved not to be feasible due to lacking gp34- and gp68-specific antibodies for tracing , we aimed to establish a gain-of-function cell-based experimental model . To this end , we characterized the surface dynamics of vFcγR expressing transfected 293 T cells stably expressing a model antigen ( hCD20 ) recognized by Rituximab ( RTX ) . In order to manipulate internalization , we chose to replace the transmembrane and cytosolic domains of gp34 and gp68 with the according sequences from human CD4 ( hCD4 ) as it contains a non-canonical di-leucine-based sorting motif ( SQIKRLL ) in its C-terminal cytoplasmic tail ( CD4-tailed ) . To be recognized by AP-2 the serine residue upstream of the di-leucine motif in hCD4 needs to be phosphorylated in order to mimic the negatively charged glutamate or aspartate residues of a classical di-leucine motif ( Pitcher et al . , 1999 ) . In a first experiment we ensured equal expression of the transfected constructs utilizing the polycistronic expression of GFP from a pIRES_eGFP expression vector to gate on transfected cells ( Figure 3B ) . GFP-positive cells were also detected for Fcγ binding by a surface stain using a PE-TexasRed conjugated human Fcγ fragment ( Figure 3C ) . This revealed that upon recombinant expression , surface exposition of Fcγ binding gp68 is drastically increased when CD4-tailed although overall protein expression was comparable and all constructs bear original signal peptides ( Figure 3B ) . This can be attributed to the abrogation of its internalization evidenced by the fact that internalization of complete IC was drastically reduced with CD4-tailed gp68 when tracking hCD20/Rtx ICs using pulse-chase flow cytometry ( Figure 3D ) . Conversely , while gp34 showed only mild differences in surface exposition when altered in the same way ( Figure 3C ) , it proved to rapidly internalize complete Rtx-CD20 IC only with its native cytosolic tail intact ( Figure 3D ) . Given that both molecules spontaneously internalize non-immune IgG ( Figure 3—figure supplement 2 ) , this could hint at different routes of trafficking following internalization . Specifically , as it has been shown that gp68 internalizes IgG-Fc translocating to the lysosomal compartment for degradation ( Ndjamen et al . , 2016 ) , the difference seen for gp34 here suggests a recycling route as described for HSV-1 gE/gI ( Ndjamen et al . , 2014 ) . Further , while we also observe gp34 to more efficiently internalize even non-immune IgG , the still rapid internalization of non-immune IgG by gp68 seems not to translate to the internalization of IC regarding efficiency ( Figure 3—figure supplement 2 ) , again in line with a previously published observation ( Ndjamen et al . , 2016 ) . As we observed significantly more efficient internalization of IC by gp34wt compared to gp68wt or HSV-1 gEwt ( Figure 3D , Figure 3—figure supplement 2 ) , we conclude that a major task of gp34 compared to gp68 is mediating the efficient internalization of IC , a feature likely linked to its particular di-leucine cytosolic motif . As previously reported ( Atalay et al . , 2002; Corrales-Aguilar et al . , 2014b ) and confirmed in this study , we find that in the context of HCMV infection gp68 is more surface resident compared to gp34 ( Figure 6A ) . Therefore , we next set out to test if the antagonizing effect of membrane-resident gp68 on FcγR activation can be attributed to a block of host FcγR binding to cell surface IC rather than IC internalization . To test this hypothesis , we established a flow cytometry based assay using vFcγR transfected 293 T-CD20 cells opsonized with Rtx and assayed for FcγR binding using soluble His-tagged ectodomains of human FcγRs , which are sequence identical between FcγRs IIIA and IIIB ( Figure 4B ) . In this assay we compared gp34 and gp68 regarding their ability to interfere with FcγR binding to cell surface IC . Advantageously , CD4-tailed vFcγRs , besides circumventing confounding effects of internalization on our binding assay , closely mimic the relative surface density of gp34 and gp68 found on the plasma membrane of an HCMV-infected cell judged by human Fcγ binding ( Figure 3C , Figure 5A; Corrales-Aguilar et al . , 2014b ) . To ensure equal density of antigen upon recombinant vFcγR expression , CD20 levels were directly measured and found only in the case of CD99 expression to be slightly reduced , which served as a non-Fcγ-binding control molecule ( Figure 4A ) . We found gp68 but not gp34 to significantly reduce binding of FcγRIIIA to cell surface immune complexes compared to CD99 ( Figure 4C ) . This finding can be explained with certain CH2–CH3 region residues of IgG playing a subordinate , yet significant role in FcγRIII binding to IgG ( Shields et al . , 2001 ) . Additionally , we found the effect of gp68 to be dependent on the accessibility of the CH2–CH3 interface region on IgG as pre-incubation with Protein G prevented the inhibition by gp68 and fully restored FcγRIIIA binding to Rtx ( Figure 4C ) . These observations further narrow down the binding region of gp68 to involve the above-mentioned residues as opposed to Protein G which has been shown to not interact with these residues ( Sauer-Eriksson et al . , 1995 ) . Finally , this conclusion is further substantiated by our observation that FcγRI , which binds to IgG independently of the above-mentioned residues in the CH2–CH3 interdomain region ( Shields et al . , 2001 ) , is not blocked by gp68 or gp34 ( Figure 4—figure supplement 1 ) . In order to test if our previous findings regarding the block of human FcγRIII binding to immune complexes in the presence of gp68 translate into a loss-of-function approach , MRC-5 cells were infected with AD169/pBAC2-derived HCMV mutants as described above . Uniformity of HCMV antigen expression and differential surface Fcγ binding indicating vFcγR expression between the different virus mutants was assured using a F ( ab’ ) 2 preparation of purified human IgG containing HCMV-specific IgG , Cytotect , or a conjugated human Fcγ fragment ( human Fcγ-Texas Red , TR ) , respectively ( Figure 5A ) . FcγR binding to opsonized-infected cells was detected by flow cytometry as described above and schematically depicted in Figure 5B . Expectedly , as we observed stronger Fcγ binding to cells infected with gp68 proficient viruses compared to a gp34 expressing virus ( Figure 5A ) , we also found more total IgG bound to the surface of cells infected with a gp68 expressing HCMV virus compared to a gp34 expressing virus ( Figure 5C , upper panel ) . Along these lines , we find FcγRIII to bind gp68-bound non-immune IgG ( Figure 5—figure supplement 1 ) , albeit at a markedly lower rate compared to immune IgG ( Figure 4 ) in line with FcγRs naturally showing higher affinity toward antigen-bound IgG ( Bruhns et al . , 2009 ) . As the Cytotect formulation contains an abundance of non-immune IgG , binding of host FcγR ectodomains as shown exemplarily in Figure 5C ( lower panel ) was normalized to levels of total surface bound IgG within each experiment ( Figure 5C , upper panel ) . Evaluating the results from three independent experiments performed with independent virus preparations confirmed a strong relative reduction in FcγRIIIA binding to opsonized HCMV-infected cells in the presence of gp68 , but not gp34 ( Figure 5D ) . This finding is in line with our previous gain-of-function experiments ( Figure 4C ) again showing an approximate 60% reduction in FcγRIII binding . While not further elaborated on in this study , we also measured binding of FcγRs I , IIA , and IIB/C in the same setup and observed that binding of FcγRI was only slightly reduced in the presence of either vFcγR , but clearly reduced in the presence of both molecules . Conversely , gp68 showed a similar effect on FcγRs IIA and IIB/C as observed for FcγRIII ( Figure 5—figure supplement 2 ) . This indicates a comparable antagonistic mechanism of gp68 on FcγRs IIA and IIB/C but shows again FcγRI to be more resistant to gp68 . While we delineated distinct mechanisms by which gp34 or gp68 are able to counteract FcγRII/III activation , we did not observe efficient antagonization of FcγR function by gp34 or gp68 individually in the context of HCMV deletion mutant infection ( Figure 2 ) . Therefore , we next wanted to test the antagonistic potential of gp68 or gp34 individually in the absence of non-immune IgG and in a gain-of-function setting . To this end , we conducted an FcγR activation assay with Hela cells co-transfected with Her2 as a model antigen and the indicated vFcγRs followed by incubation with titrated amounts of anti-Her2 IgG1 mAb ( herceptin , Hc ) ( Figure 6A ) . Using this approach , gp68 or gp34 individually confirmed to antagonize FcγRIII activation . However , the more membrane resident gp68-CD4 showed a markedly stronger antagonistic effect compared to gp68wt , indicating a more prominent membrane-residence of gp68 to be beneficial regarding evasion from FcγR recognition despite its lower capacity to internalize ICs ( Figure 3D ) . Conversely , when comparing unaltered gp34 to its less internalized gp34-CD4 variant we observed an opposite effect indicating internalization to be a major condition of gp34 driven antagonization of FcγR activation . Next , in an approach mimicking HCMV immune sera we tested if the addition of non-antigen-specific IgG impairs the antagonization of FcγR activation by gp34 or gp68 . To this end , we added 10 µg/ml TNFα-specific Infliximab ( Ifx ) IgG1 given concomitantly with the reporter cells and graded amounts of Hc IgG1 ( Figure 6B ) . This showed that indeed the presence of an excess of non-immune IgG interferes with both gp68 and gp34 antagonization in a dose-dependent manner . As ultrapure monomeric IgG does not activate the reporter cells on its own , shown by the samples that were treated only with Ifx but not Hc , this effect can be attributed to displacement of immune IgG from the vFcγRs resulting in restoration of FcγRIII activation . Synergism between gp34 and gp68 was optimal at a 100:1 ratio of non-immune Ifx over Hc ( Figure 6B , right panel ) indicating that vFcγRs are particularly adapted to work in the presence of excess non-immune IgG , supporting our previous observation with human hyperimmunoglobulin Cytotect ( see Figure 2 ) . On the other hand , host FcγRs have been shown to bind IgG immune complexes with a higher affinity compared to monomeric IgG ( Bruhns et al . , 2009 ) , limiting the efficacy of monomeric IgG in attenuating FcγR activation . Finally , we addressed cooperative antagonization by gp34 and gp68 in a loss-of-function approach in the context of HCMV infection . When using the humanized anti-HCMV mAb MSL-109 targeting HCMV gH , we found that low gH levels expressed on the surface of HCMV-AD169-infected cell are insufficient to detect FcγR triggering ( Figure 6—figure supplement 1 ) . To overcome this experimental problem , we generated Her2-expressing BJ fibroblasts exhibiting high levels of Her2 on the cell surface . Infected Her2 BJ fibroblasts were opsonized with titrated amounts of Herceptin before cells were analyzed by FcγRIIIA reporter cells ( Figure 6C ) . Although not statistically significant , this approach clearly demonstrated that compared to our observations with hyperimmunoglobulin Cytotect ( see Figure 2 ) , natively expressed gp34 as well as gp68 are able to antagonize FcγRIIIA activation individually . When both vFcγRs were expressed by HCMV-infected cells , a markedly stronger antagonistic effect was seen ( p=0 . 0252 ) . Following the reveal of synergistic modes of action , we next explored cooperativity between gp34 and gp68 in a setup of reduced complexity . Specifically , in order to elaborate on the cooperativity of gp34 and gp68 regarding the here highlighted main mechanisms of internalization ( gp34 ) and host FcγR-binding blockade ( gp68 ) we chose to co-express gp34 and Her2 antigen while adding soluble gp68 ( sgp68 , Figure 1—figure supplement 1 ) , or a soluble functional deficient gp34 point mutant as a control ( sgp34mtrp , Figure 1—figure supplement 1 and Figure 1—figure supplement 2 ) , together with the reporter cells after the removal of unbound Hc . Here we observe that the ectodomain of gp68 is sufficient to enhance the antagonistic effect of gp34 ( Figure 6C ) . In summary , we conclude that ( i ) gp34 is designed for internalization of IC while gp68 blocks FcγR binding to IC; ( ii ) gp34 and gp68 are able to antagonize FcγR activation individually when faced with titrated amounts of immune IgG , but non-immune IgG interferes with this inhibition; ( iii ) gp34 and gp68 show cooperativity in attenuating FcγR activation particularly under conditions of high excess of non-immune IgG . While we find both , gp34 and gp68 , able to internalize monomeric IgG as well as IC ( Figure 3 , Figure 3—figure supplement 2 ) , gp34 consistently shows more rapid internalization compared to gp68 ( Figure 3 ) . Consequently exchange of the cytosolic domain of gp34 leads to strongly reduced internalization and subsequently to a reduction in FcγR antagonization ( Figure 6 ) . Looking closely at the cytosolic domains of all identified cytomegalovirus vFcγRs reveals a unique [D/E]XXX[LL/LI] cytosolic motif present in gp34 . Conversely , the HCMV FcγRs gp68 , gp95 , and gpRL13 all share only a cytosolic YXXΦ motif , as the recently identified RhCMV encoded vFcγR gpRH05 ( RL11 gene family ) and its homologs which are conserved in Old World monkey CMV species ( Kolb et al . , 2019 ) . Further comparing the cytosolic sorting motifs between other membrane-resident glycoproteins of HCMV it seems that the YXXΦ motif of gp68 fulfills a more general purpose of limiting the overall surface exposure of concerned HCMV antigens ( examples listed in Figure 3—figure supplement 1; Corrales-Aguilar et al . , 2014b; Ndjamen et al . , 2016 ) . While the YXXΦ motif of gp68 lies within seven aa distance to the transmembrane domain , in line with it being translocated to lysosomal compartments ( Bonifacino and Traub , 2003; Ndjamen et al . , 2016 ) , the YXXΦ motif is additionally linked to specific targeting functions . For example , HSV-1 gE has been shown to bind in a pH-dependent manner not observed for any other HCMV encoded vFcγR ( Sprague et al . , 2004; Sprague et al . , 2008 ) which fits to it being destined for recycling rather than degradation ( Ndjamen et al . , 2014 ) . This is in line with its YXXΦ motif being more distant to its transmembrane domain within around seven amino acids ( Bonifacino and Traub , 2003; Figure 3—figure supplement 1 ) . Similar membrane-distal YXXΦ motifs also exist in the cytosolic domains of HCMV encoded gpRL13 and gB , but not the model antigen Her2 which we explored in Figure 6C , perhaps explaining the limited synergistic inhibition that was observed . These findings highlight the di-leucine motif found in gp34 to be unique among other surface-resident glycoproteins expressed by HCMV and all other vFcγRs known in CMV family members including RhCMV and MCMV ( Kolb et al . , 2019; Thäle et al . , 1994 ) . It remains to be explored as to what the further consequences of this seemingly unique internalization route is . However , as we also find gp34 to be incorporated into the virion ( Reinhard , 2010 ) it is tempting to speculate that the potent internalization of IC via gp34 has an additional role besides evasion from antibody-mediated attack of an infected cell . Although we find gp34 to efficiently internalize ICs from the cell surface , it does not manipulate host FcγR binding to ICs . This is remarkable given our data showing that gp34 binds to the hinge region of monomeric IgG , but implying that gp34 is not able to directly compete with host FcγRs for binding to IC on the plasma membrane when co-expressed with a target antigen . Conversely , gp68 binding to the CH2–CH3 interface domain also occupies a region on Fcγ that includes residues involved in FcγRII and FcγRIII binding to Fcγ ( Shields et al . , 2001; Sprague et al . , 2008 ) . This could explain the ability of gp68 to efficiently limit binding of FcγRs II and III to cell surface IC without directly competing for a shared binding region at the hinge region ( Figure 1 , Figures 4 and 5 ) . The subordinate role of the above-mentioned residues on IgG regarding FcγRIII binding further ties into the observation that the blocking effect of gp68 is not total but more in line with the reported reduction of FcγRIII binding to IgG being approximately 40–70% when these residues are manipulated ( Shields et al . , 2001 ) . Conversely , gp68 not showing a similar efficiency in blocking FcγRI binding to IC further supports this conclusion as FcγRI binding does not require the above mentioned CH2–CH3 region residues ( Figure 3—figure supplement 1; Shields et al . , 2001 ) . Taken together , we conclude that neither gp68 nor gp34 individually are able to efficiently antagonize FcγR activation by IgG-opsonized viral antigens in the physiological context of HCMV infection ( Figure 2 ) , that is , in the presence of non-immune IgG mitigating their inhibitory potential ( Figure 6B and Figure 6C ) . Our data provide evidence that gp68 rather functions to increase IC accessibility to gp34 leading to efficient internalization of IC , presumably by a repeating process of gp34 recycling . This is supported by our observation that the ectodomain of gp68 is sufficient to increase the antagonistic efficiency of gp34 regarding FcγR activation ( Figure 6 ) . Further , gp34 and gp68 are co-expressed at the cell surface throughout the protracted HCMV replication cycle and show simultaneous Fcγ binding ( Figure 1 ) . Our findings imply that gp68 binding , as it does not use the same region as host FcγRs , evolved to ensure a consistent and evolutionary less vulnerable way of shifting Fcγ accessibility in favor of gp34 . The idea that gp68 has not evolved to directly compete with host receptors for IgG binding is also supported by the finding demonstrating gp68 to possess a lower affinity to Fcγ when binding in a 2:1 ratio compared to host FcγRIIIA ( KD1:470 nM and KD2:1 . 600 nM vs 700 nM ) ( Li et al . , 2007; Sprague et al . , 2008 ) . The concept of gp68 driven accessibility shift also does not require gp68 to completely block FcγR binding to IC as subsequent gp34-driven internalization is a continuous and fast process . In this study we mainly focused on the mechanisms by which gp34 and gp68 cooperate to antagonize CD16/FcγRIII as the primary receptor on NK cells associated with ADCC , one of the most powerful mechanisms of antibody-mediated virus control . However , it is already known that HCMV vFcγRs also antagonize activation of FcγRIIA ( CD32 ) and FcγRI ( CD64 ) ( Corrales-Aguilar et al . , 2014b ) and we also consistently observed antagonization of the only inhibitory FcγR , FcγRIIB ( unpublished observation ) . Similarly , we could recently show antagonization of all canonical Rhesus FcγRs I , IIA , IIB , and III by the RhCMV encoded vFcγR gpRH05 ( RL11 gene family ) ( Kolb et al . , 2019 ) . Taken together with the data shown in this study that the underlying mechanism seems to be related between FcγRs IIA/B and III , we also speculate that more Fcγ-binding host factors might be antagonized by similar mechanisms involving vFcγRs . In support of this view , besides gp34 and gp68 , the HCMV RL11 gene family encodes additional vFcγRs . RL12 ( gp95 ) and RL13 ( gpRL13 ) likely are part of a larger arsenal of antibody targeting immunoevasins ( Corrales-Aguilar et al . , 2014a; Cortese et al . , 2012 ) . The fact that the interaction of gp34 and gp68 was analyzed in the absence of gp95 ( RL12 ) and gpRL13 ( RL13 ) is a major limitation of our study . Synergies between other or even all four vFcγRs are also conceivable and subject to further investigation . Altogether , our findings analyze the first mechanistic details of HCMV evasion from antibody-mediated control utilizing its vFcγR toolset . This deeper insight into the inner workings of such a process has consequences for the future evaluation and optimization of antibody-based treatment strategies targeting HCMV disease . In particular , our data will support the development of targeted intervention strategies that neutralize the function of gp34 and gp68 , to increase the efficiency of IVIg treatment in the future . All cells were cultured in a 5% CO2 atmosphere at 37°C . MRC-5 ( ECACC 05090501 ) , HFF ( HF-99/7 kindly provided by Dieter Neumann-Haefelin and Valeria Kapper-Falcone , Institute of Virology , Freiburg Germany ) , 293 T-CD20 ( kindly provided by Irvin Chen , UCLA USA [Morizono et al . , 2010] ) , BJ-Her2 ( BJ-5ta foreskin fibroblasts [ATCC CRL-4001] stably expressing Her2/Erbb2 ( NM_004448 ) , lentiviral transduction as in Halenius et al . , 2011 ) , and Hela cells ( ATCC CCL-2 ) were maintained in Dulbecco’s modified Eagle’s medium ( DMEM , Gibco ) supplemented with 10% ( vol/vol ) fetal calf serum ( FCS , Biochrom ) . BW5147 mouse thymoma cells ( kindly provided by Ofer Mandelboim , Hadassah Hospital , Jerusalem , Israel ) were maintained at 3 × 105 to 9 × 105 cells/ml in Roswell Park Memorial Institute medium ( RPMI GlutaMAX , Gibco ) supplemented with 10% ( vol/vol ) FCS , sodium pyruvate ( 1× , Gibco ) and β-mercaptoethanol ( 0 . 1 mM , Sigma ) . Cells were used when tested negative for mycoplasma contamination ( Eurofins ) . If positive , cells were treated using Plasmocure ( Invivogen ) or BM-Cyclin ( Roche ) as instructed by the supplier . Sequences for HCMV RL11 ( gp34 ) and HCMV UL119-118 ( gp68 ) were taken from according sequences in p_BAC2 AD169 ( MN900952 . 1 ) and synthesized as gBlocks ( IDT ) flanked with Nhe1 and BamH1 restriction sites suitable for insertion into pIRES_eGFP ( Addgene ) . Sequences for HSV-1 US8 ( gE ) and HSV-1 US7 ( gI ) were taken from HSV-1 ( strain F ) , synthesized as gBlocks and cloned as above . Human CD4 transmembrane and cytosolic domain ( GenBank: M35160 ) was used to substitute the transmembrane and cytosolic domain of respective vFcγR sequences . vFcγR-CD4 fusion constructs were synthesized as gBlocks and cloned as above . Her2 antigen was acquired from Addgene ( pCDNA3 ) . CD99 ( GenBank: BC010109 ) was cloned into pIRES_eGFP via Nhe1 and BamH1 restriction sites . Flanking restriction sites were introduced via PCR ( Primers: IDT ) . GOI-internal BamH1 sites were removed by silent single nucleotide substitutions during gBlock design . The construction of the rVACVs has been described before ( Atalay et al . , 2002; Sprague et al . , 2008; Staib et al . , 2004 ) . In brief , the gene of interest inserted in the vaccinia virus recombination vector p7 . 5k131a was transferred into the thymidine kinase open reading frame of the virus genome ( strain Copenhagen ) . rVACVs were selected with bromodeoxyuridine ( BrdU; 100 µg/ml ) using tk-143 cells ( ATCC CVCL_2270 ) . Metabolic labeling using Easytag Express [35S]-Met/Cys protein labeling , Perkin Elmer with 100 Ci/ml for 2 hr , and immunoprecipitation of Fcγ fragments using CNBr-Sepharose ( GE Healthcare ) was performed as described previously ( Sprague et al . , 2008 ) . Generation and purification of human Fcγ fragments wtFc and nbFc are described elsewhere ( Sprague et al . , 2004 ) . In brief , wild-type and mutant IgG Fc proteins were collected from CHO cell supernatants and purified using Ni2+-NTA affinity chromatography followed by pH sensitive FcRn-Sepharose column separation and subsequent size exclusion chromatography . B12 and B12-LALA were kind gifts from Ann Hessell ( Hessell et al . , 2007 ) . Direct IgG precipitation was performed using Protein G Sepharose ( Amersham ) . Samples were de-glycosylated using EndoH ( NEB , as suggested by the supplier ) . Expression constructs encoding tagged HCMV vFcγR constructs lacking their respective transmembrane and cytosolic domains were generated as described elsewhere ( Corrales-Aguilar et al . , 2014b ) . Soluble C-terminally V5-His-tagged vFcγR molecules were produced by transfection of 293 T cells in a 10 cm dish format and covered with 7 ml of DMEM/5% FCS . 3 days post transfection the supernatants were harvested , centrifuged ( 11 , 000 g , 30 min ) , and used directly or stored in 1 ml aliquots at −20°C . Production of soluble vFcγRs was controlled via α-tag immunoblot . Recombinant HCMV mutants were generated according to previously published procedures ( Tischer et al . , 2006; Wagner et al . , 2002 ) using pAD169-BAC2 ( MN900952 . 1 , Le-Trilling et al . , 2020 ) corresponding to AD169varL ( Le et al . , 2011 ) as parental genome . For the construction of the HCMV deletion mutants , a PCR fragment was generated using the plasmid pSLFRTKn ( Atalay et al . , 2002 ) as the template DNA . The PCR fragment containing a kanamycin resistance gene was inserted into the parental BAC by homologous recombination in E . coli . The inserted cassette replaces the target sequence which was defined by flanking sequences in the primers . This cassette is flanked by frt-sites which can be used to remove the kanamycin resistance gene by FLP-mediated recombination . The removal of the cassette results in a single remaining frt-site . The deletion of multiple non-adjacent genes was conducted in consecutive steps . The gene TRL11 was deleted by use of the primers KL-DeltaTRL11-Kana1 ( ACGACGAAGAGGACGAGGACGACAACGTCTGATAAGGAAGGCGAGAACGTGTTTTGCACCCCAGTGAATTCGAGCTCGGTAC ) and KL-DeltaTRL11-Kana2 ( TGTATACGCCGTATGCCTGTACGTGAGATGGTGAGGTCTTCGGCAGGCGACACGCATCTTGACCATGATTACGCCAAGCTCC ) . The gene TRL12 was deleted by use of the primers KL-DeltaTRL12-Kana1 ( CGGACGGACCTAGATACGGAACCTTTGTTGTTGACGGTGGACGGGGATTTACAGTAAAAGCCAGTGAATTCGAGCTCGGTAC ) and KL-DeltaTRL12-Kana2 ( CCTTACAGAATGTTTTAGTTTATTGTTCAGCTTCATAAGATGTCTGCCCGGAAACGTAGCGACCATGATTACGCCAAGCTCC ) . The gene UL119 was deleted by use of the primers KL-DeltaUL119-Kana1 ( TTGTTTATTTTGTTGGCAGGTTGGCGGGGGAGGAAAAGGGGTTGAACAGAAAGGTAGGTGCCAGTGAATTCGAGCTCGGTAC ) and KL-DeltaUL119-Kana2 ( AGGTGACGCGACCTCCTGCCACATATAGCTCGTCCACACGCCGTCTCGTCACACGGCAACGACCATGATTACGCCAAGCTCC ) . 10 ml soluble vFcγR supernatants were mixed 1:1 in the presence of 1 µg Rituximab . Antibody has to be given in a limiting amount to increase the probability of simultaneous binding to the same IgG molecule . Samples were incubated 1 hr at 4°C and then mixed with freshly prepared Ni2+-NTA Sepharose beads ( 2 µl BV , cOmplete His-Tag purification resin ) and incubated overnight at 4°C in the presence of 20 mM Imidazole ( rotate or shake ) . Beads were washed three times with PBS/20 mM Imidazole at 11 , 000 g and 4°C . After the final wash , beads were either resuspended in sample buffer ( Tris pH 6 . 8 , SDS , Glycerol , 2-mercaproethanol , bromophenol blue ) or subjected to a PNGase F digest before being supplemented with sample buffer ( NEB , performed as suggested by the supplier ) . Samples were then denatured at 95°C and analyzed via SDS-PAGE and subsequent immunoblot . We adapted a standard ELISA protocol to measure binding between soluble vFcγRs and target IgG . 96-well Nunc Maxisorp plates were coated with 1 µg of biotin in PBS . Plates were then blocked and incubated with titrated amounts of supernatants from soluble strep-tagged vFcγR producing cells ( 1° vFcγRs ) . Supernatants were diluted in PBS . After incubation with 100 ng/well of target antibody , plates were incubated with a 2° His-tagged soluble vFcγR ( 1:2 diluted ) followed by detection using a HRP-conjugated anti-His antibody . Plates were measured using a Tecan Genios Pro microtiter plate reader at 450 nm/630 nm . Plates were washed three times between steps using PBS/0 . 05% Tw-20 . Transfected or infected vFcγR expressing cells were harvested using Accutase ( Sigma-Aldrich ) to retain surface molecules upon detachment . Harvested cells were washed in PBS , equilibrated in staining buffer ( PBS , 3% FCS ) and sedimented at 1000 g and 10°C for 3 min . Cells were then incubated with staining buffer containing rituximab , Cytotect , or herceptin . Cells were then incubated in an adequate volume of staining buffer containing either mAbs , Fcγ fragment , or FcγR ectodomains pre-incubated with an αHis-PE antibody ( 30 min , 4°C ) . Human His-tagged FcγR ectodomains were used at a final concentration of 5 µg/ml ( 1:50 dilution from reconstituted 0 . 25 mg/ml stock solution; Sino Biological , 10389-H08H1 ) . Pre-incubation with Protein G ( Rockland , Biotin conjugated ) was performed prior to incubation with FcγR ectodomains ( diluted 1:100 ) . Further incubation steps were carried out at 4°C for 1 hr and followed by three washing steps in staining buffer . Dead cells were excluded via DAPI stain . Analysis was performed on a FACS Fortessa instrument ( BD Bioscience ) . Transfected 293 T-CD20 cells were harvested using Accutase ( Sigma-Aldrich ) to retain surface molecules upon detachment . Harvested cells were incubated with Rituximab ( 1 µg/well ) for 1 hr at 4°C in medium . Cells were then washed twice in medium containing 5% FCS and seeded into a 96-well plate at 2 × 104 cells/well . Each reaction was performed on cells from one 96-well . Cells were then incubated at 37°C in a 5% CO2 atmosphere until being harvested at different time points ensuring regular re-suspension to avoid cell attachment over longer periods of time . After harvesting , cells were directly stained with αhuman-IgG-PE for 1 hr at 4°C and fixed using 3% PFA . Analysis was performed on a FACS Fortessa instrument ( BD Bioscience ) . The assay was performed as described earlier ( Corrales-Aguilar et al . , 2013 ) . Briefly , target cells were incubated with titrated amounts of antibody ( 96-well format ) in medium ( DMEM ) supplemented with 10% FCS for 30 min at 37°C , 5% CO2 . Cells were washed with medium ( RPMI ) and co-cultured with BW5147-reporter cells ( ratio E:T 20:1 ) expressing individual host FcyR ectodomains or CD99 as control for 16 hr at 37°C in a 5% CO2 atmosphere . Reporter cell mIL-2 secretion was quantified by subsequent anti IL-2 sandwich ELISA as described previously ( Corrales-Aguilar et al . , 2013 ) . PBMCs were purified from healthy donor blood by centrifugation via Lymphoprep Medium according to the supplier instructions ( Anprotec ) . HCMV-infected MRC-5 or HFF cells ( MOI = 3 , 72hpi ) were incubated with titrated amounts of Cytotect at 37°C for 1 hr in a 5% CO2 atmosphere . After washing 2× with medium , 5 × 105 PBMCs were incubated on opsonized infected cells for 6 hr ( 100 µl per well in RPMI/10% FCS ) in the presence of αCD107a- , αCD56-BV650 , and Golgi-Plug/Golgi-Stop ( according to supplier , BD ) . After incubation , PBMCs were harvested , washed in staining buffer ( PBS/3% FCS ) , and incubated with αCD3-FITC 30 min at 4°C . After two final washing steps in staining buffer , analysis was performed on a FACS Fortessa instrument ( BD Bioscience ) . Antibodies were diluted 1:100 on 1 × 106 cells for flow cytometry; Cytotect ( Biotest ) ; αCD107a-APC ( BD FastImmune clone H4A3 ) ; αCD56-BV650 ( Biolegend clone 5 . 1H11 ) ; αCD3-FITC ( Biolegend clone UCHT1 ) ; αhuman-IgG-PE ( BD ) ; human Fcγ-TexasRed ( Rockland ) ; rituximab ( Rtx , Roche ) ; herceptin ( Hc , Roche ) ; αhuman-IgG-PE ( Miltenyi Biotec ) ; αHis-PE ( Miltenyi Biotec ) ; αCD20-PE ( Miltenyi Biotec ) ; polyclonal rabbit-αhuman IgG-FITC ( ThermoFisher ) ; THE Anti-His-HRP ( Genscript ) ; MSL-109 anti HCMV gH ( Absolute Antibody ) . Statistical analyses were performed using ANOVA or t-test ( Prism5 , Graphpad ) . Multiple comparison was corrected by Tukey test .	Human cytomegalovirus is a type of herpes virus that rarely causes symptoms in healthy people but can cause serious complications in unborn babies and in people with compromised immune systems , such as transplant recipients . The virus has found ways to successfully evade the immune system , and once infected , the body retains the virus for life . It deploys an arsenal of proteins that bind to antibodies , specialized proteins the immune system uses to flag virus-infected cells for destruction . This prevents certain cells of the immune system , the natural killer cells , from recognizing and destroying virus-infected cells . These immune-evading proteins are called viral Fc-gamma receptors , or vFcγRs . While it has been previously shown that these receptors are able to evade the immune system , it remained unknown how exactly they prevent natural killer cells from recognizing infected cells . Now , Kolb et al . show that the cytomegalovirus deploys two vFcγRs called gp34 and gp68 , which work together to block natural killer cells . The latter reduces the ability of natural killer cells to bind to antibodies on cytomegalovirus-infected cells . This paves the way for gp34 to pull virus proteins from the surface of the infected cell , making them inaccessible to the immune system . Neither protein fully protects virus-infected cells on its own , but together they are highly effective . The experiments reveal further details about how cytomegalovirus uses two defense mechanisms simultaneously to outmaneuver the immune system . Understanding this two-part viral evasion system may help scientists to develop vaccines or new treatments that can protect vulnerable people from diseases caused by the cytomegalovirus .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2021	Human cytomegalovirus antagonizes activation of Fcγ receptors by distinct and synergizing modes of IgG manipulation
mRNA transcription in dsRNA viruses is a highly regulated process but the mechanism of this regulation is not known . Here , by nucleoside triphosphatase ( NTPase ) assay and comparisons of six high-resolution ( 2 . 9–3 . 1 Å ) cryo-electron microscopy structures of cytoplasmic polyhedrosis virus with bound ligands , we show that the large sub-domain of the guanylyltransferase ( GTase ) domain of the turret protein ( TP ) also has an ATP-binding site and is likely an ATPase . S-adenosyl-L-methionine ( SAM ) acts as a signal and binds the methylase-2 domain of TP to induce conformational change of the viral capsid , which in turn activates the putative ATPase . ATP binding/hydrolysis leads to an enlarged capsid for efficient mRNA synthesis , an open GTase domain for His217-mediated guanylyl transfer , and an open methylase-1 domain for SAM binding and methyl transfer . Taken together , our data support a role of the putative ATPase in mediating the activation of mRNA transcription and capping within the confines of the virus . Viral transcription is highly regulated , as demonstrated biochemically in viruses of the Reoviridae ( Shatkin and Sipe , 1968; Furuichi , 1974 , 1978; Borsa et al . , 1981; Farsetta et al . , 2000 ) . mRNA transcription in these viruses is activated by external actions , for example , removal of their outer shell in multi-shelled reoviruses ( Shatkin and Sipe , 1968; Borsa et al . , 1981; Farsetta et al . , 2000 ) and binding of S-adenosyl-L-methionine ( SAM ) in the single-shelled cytoplasmic polyhedrosis virus ( CPV ) ( Furuichi , 1974 , 1978 ) . The outer shell and the binding sites of SAM are far away from the RNA-dependent RNA polymerases ( RdRPs ) inside the virus . How these external actions regulate viral mRNA transcription has been a mystery . Viruses in the Reoviridae contain 9–12 segments of dsRNA enclosed within an inner core that is a self-competent molecular machine fully capable of RNA transcription and processing ( Mertens et al . , 2004; Zhou , 2008 ) . Each of the 9–12 dsRNA segments wraps around an RdRP located underneath an icosahedral vertex and can undergo independent and simultaneous RNA transcription within an intact core ( i . e . , endogenous RNA transcription ) ( Smith and Furuichi , 1982 ) . The simplest of these , the single-shelled CPV ( Zhou , 2008 ) has been used as a model system for viral RNA transcription and high-resolution cryo-electron microscopy ( cryoEM ) studies , as highlighted by the discovery of mRNA cap structures ( Furuichi , 1974; Furuichi and Miura , 1975 ) and the demonstration of near atomic resolution cryoEM ( Yu et al . , 2008 ) . To find out how viral mRNA transcription is regulated , we set out to determine a series of structures of CPV in complex with different ligands at resolutions ranging from 2 . 9 to 3 . 1 Å . We discovered that the large sub-domain of guanylyltransferase ( GTase ) domain of CPV turret protein ( TP ) also has an ATP-binding site and is likely an ATPase that mediates the activation process of viral RNA transcription and capping . This process involves sensing the presence of the signal molecule SAM by methylase −2 ( MT-2 ) domain of CPV TP , activating the putative viral ATPase , enlarging the viral capsid for efficient mRNA syntheses , and opening the GTase and MT-1 to enable guanylyl and methyl transfer . To reveal the mechanisms of transcriptional regulation of viruses within the Reoviridae family , we determined the cryoEM structures of six CPV/ligand complexes in the presence of magnesium ion: CPV+SAM ( i . e . , ‘S-CPV’ ) , CPV+SAM+4 nucleoside triphosphates ( NTPs ) ( i . e . , transcribing , or ‘t-CPV’ ) , CPV+SAM+GTP+ATP ( i . e . , ‘SGA-CPV’ ) , CPV+SAM +GTP ( i . e . , ‘SG-CPV’ ) , CPV+GTP ( i . e . , ‘G-CPV’ ) , and CPV+ATP ( i . e . , ‘A-CPV’ ) at resolutions ranging from 2 . 9 to 3 . 1 Å ( Figure 1A–C , Table 1 , Video 1 and Figure 1—figure supplement 1 ) . This range of resolutions has permitted us to identify side chains of amino acid residues and to define conformations of bound ligands to build atomic models . Like the atomic model of unliganded CPV ( Yu et al . , 2011 ) , the atomic models of these liganded CPV all contain two conformers of the capsid shell proteins ( CSP-A and CSP-B ) , two conformers of the large protrusion proteins ( LPP-3 and LPP-5 ) , and one conformer of TP in each asymmetrical unit ( e . g . , Figure 1B ) . We show below that the large sub-domain of GTase domain of TP ( Zhou et al . , 2003; Yu et al . , 2008 ) also has an additional ATP-binding site and is likely an ATPase ( Figure 1D , E ) . Except for A-CPV , these atomic models also contain ligands revealed in our cryoEM maps ( Table 1 ) . In S-CPV , one SAM binds to the MT-2 domain of each TP . In G-CPV , one GTP binds to the GTase site of GTase domain . In SG-CPV , two SAM molecules bind to the MT-1 and MT-2 domains , one Mg2+-GTP to the GTase site and one GTP to the putative ATPase site . In t-CPV and SGA-CPV , two SAM molecules bind to the MT-1 and MT-2 domains ( Zhu et al . , 2014 ) , one Mg2+-GTP to the GTase site and one ATP to the putative ATPase site . As we will report in detail below , a comparison of these structures and their correlation with the accompanying biochemical results have led to our discovery of a putative ATPase-mediated regulation process for activating viral RNA transcription and capping . 10 . 7554/eLife . 07901 . 003Figure 1 . Structural overviews of cytoplasmic polyhedrosis virus ( CPV ) bound with different ligands involved in regulation and capping for viral RNA transcription . ( A ) Radially colored G-CPV reconstruction at 2 . 9 Å resolution as viewed along a fivefold axis . ( B ) Density map of an asymmetric unit of G-CPV is colored by protein subunit . ( C ) Density map ( mesh ) and atomic model ( stick ) of a selected region from CSP-A of G-CPV , showing characteristic side chains . ( D ) Structures of turret protein ( TP ) and ligands in t-CPV . TP is colored by domain . The Mg2+ and GTP in the guanylyltransferase ( GTase ) site are in green and orange , respectively; ATP in the putative ATPase site is in magenta; the two S-adenosyl-L-methionines ( SAMs ) in MT-1 and MT-2 are in green . ( E ) Schematic illustration of t-CPV TP structure . Secondary elements involved in hydrogen bonding or stacking interactions with GTP and ATP are highlighted in orange red and magenta , respectively . Secondary elements involved in interactions with SAM are highlighted in green . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 00310 . 7554/eLife . 07901 . 004Figure 1—figure supplement 1 . Resolution assessment of CPV particle reconstructions . ( A ) R-factors of the six different CPV particles and Fourier shell correlation coefficient ( FSC ) of G-CPV . ( B ) FSC curve between the SGA-CPV map and the SGA-CPV model ( red line ) and that between the SGA-CPV map and the t-CPV model ( blue line ) . ( C ) Density maps ( mesh ) and atomic models ( stick ) of a selected region from LPP-5 ( left ) and TP ( right ) of G-CPV at 2 . 9 Å resolution , showing characteristic side chains and main chain carbonyl oxygen . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 00410 . 7554/eLife . 07901 . 005Table 1 . CryoEM imaging and model refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 005Sample nameS-CPVt-CPVSGA-CPVSG-CPVG-CPVA-CPVCryoEM reconstruction Particles included in the final reconstruction44 , 90841 , 62440 , 89846 , 14771 , 94619 , 447 Resolution ( Å ) 3 . 133 . 13 . 12 . 93 . 1 Bound ligandsOne SAM bound to MT-2SAMs bound to MT-1 and MT-2; one Mg-GTP and one ATP bound to GTase domainIdentical to those of t-CPVSAMs bound to MT-1 and MT-2; one Mg-GTP bound to GTase site; one GTP to ATPase siteOne GTP bound to the GTase site of GTase domainNo ATP bound Structural changesStructure protein movements outwardsStructure protein movements outwards and local conformational changesIdentical to those of t-CPVIdentical local conformational changes; different global protein movementsNo changesNo changesModel refinement Resolution range ( Å ) 40–3 . 140–3 . 040–3 . 140–3 . 140–2 . 940–3 . 1 R-factor ( % ) 19 . 8519 . 7419 . 7818 . 2519 . 9319 . 5110 . 7554/eLife . 07901 . 006Video 1 . Radially colored G-CPV reconstruction at 2 . 9 Å resolution as viewed along a fivefold axis . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 006 Because SAM is required for efficient mRNA synthesis in CPV in addition to being the methyl donor for mRNA methylation ( Furuichi , 1974 , 1978 ) , we first asked whether the presence of SAM would have any effect on the structure of CPV . Superposition of the structures ( both at 3 . 1 Å resolution ) of S-CPV and the unliganded CPV ( Yu et al . , 2011 ) shows that the capsid shell of S-CPV is slightly expanded with a non-uniform outwards movement of all structure proteins ( Figure 2A , B , Video 2 and Figure 2—figure supplement 1A–C ) . ( In contrast , the A-CPV structure reported here does not have such capsid expansion and protein movements , see below . ) For example , the apical domain of CSP-A , located next to the fivefold axis , has the largest movement of ∼1 Å ( RMSD: 0 . 97 Å ) ; the dimerization domain , located near the twofold axis , has the smallest movement of ∼0 . 5 Å ( RMSD: 0 . 49 Å ) ; the CPV-unique small protrusion domain , which is located between the apical and the dimerization domain , moves outwards ∼0 . 8 Å ( RMSD: 0 . 83 Å ) ( Figure 2A and Video 3 ) . TP , residing on the apical domain of CSP-A , moves outwards ∼1 Å ( RMSD: 0 . 99 Å ) , which is the same as the displacement of the apical domain of CSP-A ( Figure 2B ) . In CSP-B , located around the threefold axis , the outwards movements of the apical , small protrusion , and dimerization domains are ∼0 . 85 Å ( RMSD: 0 . 85 Å ) , 0 . 6 Å ( RMSD: 0 . 57 Å ) , and 0 . 5 Å ( RMSD: 0 . 49 Å ) , respectively ( Figure 2—figure supplement 1A ) . Accordingly , the movement ( RMSD: 0 . 67 Å ) of LPP-5 is slightly larger than that of LPP-3 ( RMSD: 0 . 43 Å ) ( Figure 2—figure supplement 1B , C ) . 10 . 7554/eLife . 07901 . 007Figure 2 . SAM alone binds to MT-2 of TP and triggers global movement of all capsid proteins . ( A ) Superimposition of CSP-A between unliganded CPV ( gray ) and S-CPV ( colored by domain ) . Insets: zoom-in views of the boxed regions . The twofold and fivefold axes are indicated by a pentagon and an oval , respectively . ( B ) Superimposition of TP between unliganded CPV ( gray ) and S-CPV ( colored by domain as in Figure 1D ) . Insets: zoom-in views of the boxed regions from GTase and MT-1 domains , respectively . ( C ) Structure of MT-2 ( purple ) and SAM ( green ) . Left , view as the guide map ( inset ) . Right , view rotated as indicated . ( D ) Active site of MT-2 . SAM is colored by element: carbon in green , nitrogen in blue , oxygen in red , and sulfur in yellow . Side chains of those amino acids interacting with SAM are shown . ( E ) Superimposition of MT-2 between unliganded CPV ( gray ) and S-CPV ( purple ) before ( left ) and after ( right ) domain alignment using Cα positions . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 00710 . 7554/eLife . 07901 . 008Figure 2—figure supplement 1 . Global movement of viral capsid proteins caused by SAM bound to the externally located MT-2 . ( A ) Superimposition of CSP-B between unliganded CPV ( gray ) and S-CPV ( colored by domain ) . Left inset , zoom-in view of a boxed region from apical domain . Right inset , zoom-in view of a boxed region from dimerization domain . ( B ) Superimposition of LPP-3 between unliganded CPV ( gray ) and S-CPV ( blue ) . Upper: viewed from outside . Lower: view rotated as indicated . Inset: zoom-in view of the boxed region . ( C ) Superimposition of LPP-5 between unliganded CPV ( gray ) and S-CPV ( blue ) . Upper: viewed from outside . Lower: view rotated as indicated . Inset: zoom-in view of the boxed region . ( D ) Structure of MT-2 active site and the bound SAM in S-CPV . MT-2 is in purple . SAM is colored as in Figure 2D . Side chains of amino acids involved in interactions with SAM are shown . Density map of bound SAM is contoured at 1 . 4σ ( upper ) and 3 . 0σ ( lower ) above the means , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 00810 . 7554/eLife . 07901 . 009Video 2 . Conformational changes from unliganded CPV to S-CPV . Atomic model of an asymmetric unit is colored by protein subunit as in Figure 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 00910 . 7554/eLife . 07901 . 010Video 3 . Global movements of CSP-A caused by SAM bound to the externally located MT-2 . Superimposition of CSP-A between unliganded CPV and S-CPV . CSP-A from unliganded CPV is in gray . CSP-A from S-CPV is colored by domain as in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 010 Both MT-1 and MT-2 domains of TP have the typical structural motif of SAM-dependent methyltransferases with a seven-stranded β-sheet sandwiched by α-helices ( Schluckebier et al . , 1995; Hodel et al . , 1996; Reinisch et al . , 2000; Sutton et al . , 2007 ) . Unexpectedly , only the MT-2 domain in S-CPV bound SAM ( Figure 2C , D and Figure 2—figure supplement 1D ) . Except for slight displacement due to the outwards movement of TP as described above , the MT-2 domain structure of S-CPV is indistinguishable from that of the unliganded CPV at the current resolution of 3 . 1 Å ( Figure 2E ) . The above observed SAM-triggered conformational change correlates with previous biochemical data establishing a role of SAM in inducing mRNA synthesis ( Furuichi , 1974 , 1978 ) . In order to find out how SAM does this , we obtained a structure of t-CPV at 3 . 0 Å resolution , that is , virions incubated with SAM , 4 NTPs , and Mg2+ . In contrast to that of S-CPV ( Figure 3A ) , the cryoEM image of t-CPV shows string-like densities emanating from the viral particles , which we attribute to newly synthesized mRNA molecules in the process of release from the actively transcribing virions ( arrows in Figure 3B ) . However , no mRNA densities are visible in our icosahedral reconstruction because these RNA molecules are transcripts of different genomic segments at different stages of the dynamic transcription process and are smeared by averaging . 10 . 7554/eLife . 07901 . 011Figure 3 . Comparison of S-CPV and t-CPV reveals global protein movements and local conformational changes . ( A , B ) Cryo-electron microscopy ( cryoEM ) images of S-CPV and t-CPV . Unlike that of S-CPV ( A ) , the cryoEM image of t-CPV ( B ) shows characteristic string-like densities emanating from virus particles ( arrows ) . Scale bars , 50 nm . ( C ) Superimposition of TP between S-CPV ( gray ) and t-CPV ( colored by domain as in Figure 1D ) . Upper , domains that show global movements are indicated by dashed ellipses . Lower , GTase domain of t-CPV was aligned to that of S-CPV using Cα positions for residues in small sub-domain . Each of other three domains in t-CPV was aligned to its counterpart in S-CPV using Cα positions for residues in each domain . Regions that undergo local conformational changes are indicated by dotted ellipses . ( D ) Superimposition of CSP-A between S-CPV ( gray ) and t-CPV ( colored as in Figure 2A ) . Upper , domains that show global movements are indicated by dashed ellipses . Inset , density maps of S-CPV ( gray ) and t-CPV ( pink ) from the boxed region . Lower , molecules were aligned using Ca positions for residues in small protrusion , middle and dimerization domains . Region that undergoes local conformational change is indicated by dotted ellipse . Part ( 470–472 ) of a helix ( residues 460–472 ) in S-CPV becomes a loop in t-CPV ( inset ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 01110 . 7554/eLife . 07901 . 012Figure 3—figure supplement 1 . Global movements and local conformational changes of capsid proteins observed in t-CPV . ( A ) Superimposition of LPP-3 between S-CPV ( gray ) and t-CPV ( blue ) . Upper: viewed from outside . Lower , view rotated as indicated . ( B ) Superimposition of LPP-5 between S-CPV ( gray ) and t-CPV ( blue ) . Upper: viewed from outside . Lower: view rotated as indicated . ( C ) Superimposition of CSP-B between S-CPV ( gray ) and t-CPV ( colored by domain ) . Upper: domains that show global movements are indicated by dashed ellipses . Lower , the CSP-B molecules were aligned using Ca positions for residues in small protrusion , middle and dimerization domains . Region that undergoes local conformational change is indicated by dotted ellipse . ( D ) Structure of GTase domain with bound ligands in t-CPV . Density map and atomic model of GTase domain are in transparent gray and sky blue , respectively . Density map is contoured at 3σ above the means . GTP and ATP models are in orange red and magenta , respectively . Left inset , zoom-in view rotated from the boxed region showing the GTP density in the GTase site . Right inset , zoom-in view rotated from the boxed region showing the density of a ligand bound to the large sub-domain of GTase domain . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 012 Structural comparison between t-CPV and S-CPV reveals conformational changes of the capsid proteins in t-CPV ( Figure 3C , D , Video 4 and Figure 3—figure supplement 1A–C ) . Among the five protein molecules within each asymmetric unit of CPV , only LPP-3 remains unchanged in both the location and structure ( Figure 3—figure supplement 1A ) and the other four protein molecules exhibited changes in their locations , their structures , or both . The locations of LPP-5 molecules in t-CPV and S-CPV differ although their structures are the same ( Figure 3—figure supplement 1B ) , indicating a rigid-body type of movement , likely effected by changes of the underlying CSP molecules . By contrast , the other molecules undergo both global domain movements and local conformation changes from S-CPV to t-CPV ( Figure 3C , D and Figure 3—figure supplement 1C ) . All domains of TP undergo global outwards movements ( ∼9 Å ) , while only the MT-1 domain and the large sub-domain of GTase domain of TP exhibit local conformation changes ( Figure 3C ) . CSP-A not only rotates outwards ( up to 9 Å ) around a pivot point near the twofold axis ( global domain movements indicated by dashed ellipses in Figure 3D and Video 5 ) , but it also changes conformation in its apical domain ( dotted ellipse in Figure 3D ) . CSP-B also undergoes similar but less obvious changes than CSP-A ( Figure 3—figure supplement 1C ) . These structural changes in CSP molecules result in an enlarged , yet stable capsid of the transcribing CPV . Since viral mRNA synthesis takes place within the confines of intact virus core , an enlarged capsid would facilitate dsRNA template movement , enabling efficient mRNA synthesis . 10 . 7554/eLife . 07901 . 013Video 4 . Conformational changes from S-CPV to t-CPV . Atomic model of an asymmetric unit is colored by protein subunit as in Figure 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 01310 . 7554/eLife . 07901 . 014Video 5 . Global movements and local conformational changes of CSP-A from S-CPV to t-CPV . Density map superimposition of CSP-A between S-CPV and t-CPV . CSP-A of S-CPV is in gray . CSP-A of t-CPV is in orange red . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 014 The t-CPV structure contained two ligands bound to the GTase domain of each TP ( Figure 1D , Video 6 and Figure 3—figure supplement 1D ) . The first is the expected GTP molecule involved in transfer of a guanylyl group catalyzed by GTase and is located at a cleft of the GTase active site ( Video 6 and Figure 3—figure supplement 1D ) . The second is unexpected ligand , bound to the large sub-domain , away from the cleft ( Video 6 and Figure 3—figure supplement 1D ) . 10 . 7554/eLife . 07901 . 015Video 6 . GTase domain of t-CPV contains two ligands . The density map and atomic model of GTase domain in t-CPV are in transparent gray and sky blue , respectively . The density map is contoured at 3 . 0σ above the means . The atomic models of GTP and ATP are in orange and magenta , respectively . Mg2+ is in green . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 015 The density of the unexpected ligand is as strong as that of the GTP bound to the GTase site and the surrounding amino acid residues , and it fits very well with the atomic model of an ATP molecule , suggesting that the large sub-domain of GTase domain of TP could also be a viral ATP-binding site ( Figure 4A and Video 7 ) . Located outside the turret chamber ( Figures 1D , 4A and Figure 4—figure supplement 1 ) , the putative ATP-binding site is inaccessible to both the dsRNA genome and the nascent mRNA , thus rendering it unable to directly participate in the mRNA transcription and the capping reactions but may function as a regulatory protein or enzyme . 10 . 7554/eLife . 07901 . 016Figure 4 . Discovery of the viral ATP-binding site . ( A ) Structure of GTase domain and ATP in t-CPV . Left , view rotated from the guide map ( inset ) as indicated . GTase domain is in sky blue . ATP is in magenta . Middle , zoom-in view of the putative ATP-binding site . ATP is colored by element: carbon atoms are magenta , nitrogen atoms are blue , and oxygen atoms are red . The hydrogen bonds are indicated by black lines . Side chains of Tyr305 and Arg271 form pi–pi and cation–pi interactions with the adenine ring of ATP , respectively . Right , same view as the middle . The density map of bound ATP ( gray mesh ) is contoured at 3σ above the means . ( B ) Structure of GTase domain and ATP in SGA-CPV . Molecules are viewed and colored as in A . ( C ) Structure of GTase domain and GTP in SG-CPV . Molecules are viewed and colored as in A . GTP is colored analogously . The density map of bound GTP ( gray mesh ) is contoured at 1 . 4σ above the means . ( D ) Superimposition of GTase domain between SG-CPV ( gray ) and t-CPV ( sky blue ) . Inset: zoom-in view of the boxed region . Density maps from t-CPV ( sky blue ) and SG-CPV ( gray ) are contoured at 3 . 0σ above the means . ( E ) Superimposition of the large sub-domain of GTase domain between S-CPV ( gray ) and t-CPV ( sky blue ) . Molecules were aligned using Ca positions for residues in small sub-domain . The bound ATP of t-CPV is in magenta . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 01610 . 7554/eLife . 07901 . 017Figure 4—figure supplement 1 . The pentameric turret complex of t-CPV . ( A , B ) The turret viewed from the side and top , respectively . Four monomers are in gray . One monomer is colored by domain as in Figure 1D . The GTP and ATP ligands bound to the GTase domain ( sky blue ) are in orange red and magenta , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 01710 . 7554/eLife . 07901 . 018Figure 4—figure supplement 2 . CryoEM of SGA-CPV . ( A ) CryoEM image of SGA-CPV . Scale bar , 50 nm . ( B ) Superimposition of CSP-A between SGA-CPV ( gray ) and t-CPV ( colored by domain ) . ( C ) Stereo view of ATP-binding site and ATP in SGA-CPV . GTase domain is in sky blue . ATP is colored by element as in Figure 4B . The density map ( gray mesh ) of protein and bound ATP is contoured at 3σ above the means . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 01810 . 7554/eLife . 07901 . 019Figure 4—figure supplement 3 . CryoEM of SG-CPV . ( A ) Superimposition of CSP-A between SG-CPV ( gray ) and t-CPV ( colored by domain ) . ( B ) Stereo view of ATP-binding site and GTP in SG-CPV . GTase domain is in sky blue . GTP is colored by element as in Figure 4C . The density map ( gray mesh ) of protein and bound GTP is contoured at 3σ above the means . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 01910 . 7554/eLife . 07901 . 020Video 7 . Structure of ATP-binding site and the bound ATP in t-CPV . The atomic model of GTase domain is in sky blue . ATP is colored by element as in Figure 4A . Side chains of amino acid involved in hydrogen bonding ( black lines ) or stacking with ATP are shown . The density map of the bound ATP is contoured at 3 . 0σ above the means . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 020 To establish the identity of the aforementioned ligand and the function of the putative ATP-binding site , we subsequently performed the following two structural studies . First , through incubating CPV capsids with SAM , GTP , ATP , and Mg2+ , we obtained the SGA-CPV particle which , lacking of UTP and CTP , is incapable of mRNA transcription . Indeed , under cryoEM , SGA-CPV ( Figure 4—figure supplement 2A ) , and S-CPV ( Figure 3A ) particles look similar and differ from actively transcribing t-CPV particles ( Figure 3B ) . However , the 3 . 1 Å structure of SGA-CPV shows global movements and local conformational changes of its structure proteins that are indistinguishable from those of t-CPV ( Figure 4A , B and Figure 4—figure supplement 2B ) . Furthermore , the large sub-domain of SGA-CPV GTase domain also contains a ligand similar to that in the t-CPV ( Figure 4A , B and Figure 4—figure supplement 2C ) . Second , to eliminate the possibility of GTP as the ligand bound to the putative ATP-binding site in t-CPV , we obtained a 3D reconstruction of SG-CPV at 3 . 1 Å resolution . While the local conformational changes of SG-CPV structure proteins are identical to those of t-CPV ( Figure 4C ) , the global movements of structure proteins of SG-CPV are slightly less than those in t-CPV or SGA-CPV ( Figure 4—figure supplement 3A ) . For example , the movement of GTase domain in SG-CPV is ∼1 Å less than that of t-CPV or SGA-CPV ( Figure 4D and Video 8 ) . Most importantly , the density of the bound GTP is not as strong as that of the ligand in t-CPV or SGA-CPV , and its triphosphate group becomes invisible when displayed at the same threshold of 3σ ( Figure 4C and Figure 4—figure supplement 3B ) . 10 . 7554/eLife . 07901 . 021Video 8 . Structure comparison of GTase domain between SG-CPV ( gray ) and t-CPV ( sky blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 021 These results indicate that ( 1 ) the global movements and local conformational changes of structural proteins observed in t-CPV are not a consequence , but rather a trigger of RNA transcription; ( 2 ) the large sub-domain of the GTase domain binds ATP to mediate the conformational changes observed in t-CPV . Consistent with this assignment , only the large sub-domain ( the one containing the ATP-binding site ) of the GTase domain undergoes significant conformational changes between S-CPV and t-CPV ( Figures 3C , 4E ) . Accompanying these conformational changes , part of the loop connecting α13 and α14 in S-CPV became a helix ( αC ) in SGA-CPV , SG-CPV , and t-CPV ( Figure 4 ) . Previous biochemical studies have shown that the hydrolysis of ATP is required for mRNA synthesis ( Furuichi , 1978 ) and that efficient synthesis of CPV mRNA depended on the concentrations of SAM and ATP in a synergistic manner ( Furuichi , 1981 ) . We reason that the large sub-domain with the ATP-binding site is possibly an ATPase , and the synergy between SAM and ATP reflects a dependence of its activity on the presence of SAM . To test this hypothesis , we first obtained the 3D reconstructions of G-CPV and A-CPV at 2 . 9 Å and 3 . 1 Å resolutions , respectively ( Figure 1A–C , Table 1 , Video 1 and Figure 1—figure supplement 1 ) . Our structures of G-CPV and A-CPV show that , in the absence of SAM , neither GTP nor ATP induced any conformational change ( Table 1 ) . While a GTP bound to the GTase active site in G-CPV ( Figure 5A and Figure 5—figure supplement 1 ) , neither ATP nor GTP was observed at the newly discovered ATP-binding site in A-CPV and G-CPV ( Table 1 ) , indicating that ATP/GTP binding to the large sub-domain of TP GTase domain is directly regulated by SAM , most likely via binding to the MT-2 domain , since the structure of S-CPV revealed only MT-2 domain contained SAM ( Figure 2C , D ) . 10 . 7554/eLife . 07901 . 022Figure 5 . ATP binding and hydrolysis by the viral ATPase is SAM-dependent . ( A ) Structure of GTase domain and GTP in G-CPV . Left , view rotated from the guide map ( inset ) as indicated . GTase domain is in sky blue . GTP is in orange red . Middle , active site of GTase . GTP is colored by element: carbon atoms are orange red , nitrogen atoms are blue , and oxygen atoms are red . The hydrogen bonds are indicated by black lines . Side chain of Tyr59 also forms pi–pi stacking interaction with the guanylyl ring of GTP . Right , same view as the middle . The density map of bound GTP ( gray mesh ) is contoured at 3σ above the means . ( B ) Nucleotide substrates specificity by CPV nucleoside triphosphatase . Values are means derived from duplicate experiments . Standard deviations are indicated by error bar . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 02210 . 7554/eLife . 07901 . 023Figure 5—figure supplement 1 . Stereo view of GTPase site and GTP in G-CPV . GTase domain is in sky blue . GTP is colored by element as in Figure 5A . The density map ( gray mesh ) of protein and bound GTP is contoured at 3σ above the means . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 023 We then determined the rates of NTP hydrolysis of CPV in the presence and absence of SAM . Phosphate released upon hydrolysis of NTPs indeed depended on the presence of SAM and the most favorable NTP substrate was ATP , with decreasing rate of hydrolysis of other substrates in the order of GTP > CTP > UTP in the presence of SAM ( Figure 5B ) . Previous biochemical studies have shown that the mRNA transcription of CPV is SAM-dependent and is specifically coupled to ATP hydrolysis ( Furuichi , 1978 , 1981 ) . Additionally , it has been shown that the removal of turret in orthoreovirus leads loss of mRNA transcription activity ( Luongo et al . , 2002 ) . Our structural results , when combined with these biochemical data , suggest that the ATP-binding site of TP GTase domain is possibly a SAM-dependent ATPase that mediates the activation of mRNA transcription . The density for the bound GTP in the GTase site is strong in t-CPV ( Figure 6A , Video 6 and Figure 6—figure supplement 1A ) but weak in G-CPV , particularly at the triphosphate group ( Figure 5A and Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 07901 . 024Figure 6 . The catalytic activity of viral GTase is regulated by the viral ATPase through allosteric effect . ( A ) Structure of GTase domain and the bound Mg2+-GTP in t-CPV . Left , view rotated from the guide map ( inset ) as indicated . GTase domain is in sky blue . GTP is in orange red . Mg2+ is in green . Middle , active site of GTase with bound Mg2+-GTP . GTP is colored by element as in Figure 5A . The hydrogen bonds are indicated by black lines . Side chains of the two conserved His208 and His217 are shown . Right , same view as the middle . The density map of bound GTP ( gray mesh ) is contoured at 3σ above the means . ( B ) Superimposition of GTase domain between G-CPV ( gray ) and t-CPV ( sky blue ) . Molecules were aligned using Ca positions for residues in small sub-domain . GTPs bound to the GTase sites of G-CPV and t-CPV are in purple and orange red , respectively . Inset , zoom-in view of the boxed region . ( C ) Structure of GTase domain and the bound Mg2+-GTP in SGA-CPV . Molecules and Mg2+ are viewed and colored as in A . Side chain of the conserved His217 is shown . ( D ) Structure of GTase domain and the bound Mg2+-GTP in SG-CPV . Molecules and Mg2+ are shown as in A . Side chain of the conserved His217 is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 02410 . 7554/eLife . 07901 . 025Figure 6—figure supplement 1 . Structures of GTase sites and bound GTPs . ( A ) Stereo view of GTPase site and GTP in t-CPV . Molecules and density map are colored and displayed as in Figure 6A . ( B ) Stereo view of GTase site and bound GTP in SGA-CPV . Molecules and density map are colored and displayed as in A . ( C ) Structure of GTase site and bound GTP in SG-CPV . Molecules and density map are colored and displayed as in A . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 02510 . 7554/eLife . 07901 . 026Figure 6—figure supplement 2 . The conserved His217 is the catalytic amino acid for guanylylation of GTase in CPV . ( A ) Sequence alignment showing the conserved histidines of GTase domains among 3 different members in Reoviridae family . The conserved histidines are highlighted by green boxes . ( B ) Superimposition of GTase domains of CPV TP , orthoreovirus λ2 , and aquareovirus VP1 . Molecules were aligned using Cα positions for residues in domains . The GTase domains of CPV , orthoreovirus , and aquareovirus are in sky blue , yellow , and red , respectively . The GTP and Mg2+ bound to CPV GTase site are colored in gray . Inset: zoom-in view of the boxed region . Side chains of K234 , H208 , and H217 of CPV , K190 , H223 , and H232 of orthoreovirus and K196 , H229 , and H238 of aquareovirus are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 026 Compared to that of G-CPV , the opening ( or ‘gate’ ) leading to the GTase site and coupled to the putative mRNA releasing hole ( Yu et al . , 2011 ) is widened from 13 to 15 Å in t-CPV , likely to accommodate nascent mRNA ( Figures 5A , 6A , B ) . Following the nomenclature of PBCV-1 GTase ( Hakansson et al . , 1997 ) , we designate the states of GTase domain in G-CPV and t-CPV as closed and open states , respectively . In the open state of GTase domain in t-CPV , we observed extra density next to the β , γ phosphates of the bound GTP ( Figure 6A and Video 6 ) , which we interpret as coordinated Mg2+ for two reasons . First , Mg2+ was the only divalent cation in our reaction mixture . Second , the GTase domains in SGA-CPV and SG-CPV are also in the open state with prominent densities attributable to Mg2+ ( Figure 6C , D and Video 9 ) . Biochemical data have shown that the Mg2+ is required for GTase activity of viruses in Reoviridae ( Yamakawa et al . , 1982; Le Blois et al . , 1992; Martinez-Costas et al . , 1995; Qiu and Luongo , 2003; Mohd Jaafar et al . , 2005 ) . Therefore , only the open state ( with the putative Mg2+ ) is catalytically active . Remarkably , the gate opening is achieved through the displacement of α14 ( i . e . , the gate helix ) , one of the three helices comprising the active site of the putative viral ATPase ( Figures 4A–C , 6B ) . However , even though the gate helix α14 controls the open and closed state , it is not part of the GTase active site . In fact , the active sites of the putative ATPase and GTase do not share any amino acids or secondary elements ( Figure 1E ) . Therefore , the putative ATPase regulates GTase activity allosterically . 10 . 7554/eLife . 07901 . 027Video 9 . Structure of GTase active site and the bound Mg-GTP in SGA-CPV . Color coding: sky blue—atomic model of GTase domain; green—Mg2+; GTP—colored by element as in Figure 6C . Side chains of amino acids of the conserved His217 , and those involved in hydrogen bonding or in stacking with GTP are shown . The density map of the bound GTP is shown as mesh at a contour level of 3 . 0σ above the means . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 027 The bound GTP molecules in the GTase open and closed states exhibit differences in their conformations and interactions with the GTase active site ( Figures 5A , 6 ) . The triphosphate moiety forms only one hydrogen bond ( between the β phosphate and Tyr59 ) in the closed conformation ( Figure 5A ) but forms three or four more hydrogen bonds in the open conformation including the two formed by the α phosphate with His212 and Arg255 ( Figure 6A , C , D ) . The more extensive hydrogen bonds observed in the open conformation is consistent with our assignment of it as the active state . Catalysis of guanylyl transfer occurs in two steps: reaction with GTP to form a covalent enzyme–GMP intermediate ( enzyme guanylylation ) and transfer of GMP onto the 5′-diphosphorylated acceptor . Previous loss-of-function mutagenesis study of lysine residues in mammalian reovirus suggested that Lys190 of GTase domain is responsible for guanylylation of GTases ( Luongo et al . , 2000 ) . Lys190 is located in a 28-aa segment ( residues 168–195 ) that connects two structurally conserved β strands . Surprisingly , the conserved β strands ( β3 and β4 ) in CPV GTPase is connected by a segment of only 13 aa ( a loop from residues 166–178 ) , which contains no lysine ( Figure 1D , E ) . The connecting segments do not have sequence or structural similarities . Within the vicinity of the bound GTP , the only lysine residue in CPV GTase domain is Lys234 , which maps to a non-conserved residue ( Ser259 ) in mammalian reovirus GTase domain . Moreover , during GTase transition from its closed to open state , the α phosphorus of the GTP moves towards a histidine-rich segment and away from Lys234 ( Figures 6 , 7 and Figure 6—figure supplement 1 ) . Therefore , our structures indicate that Lys234 cannot directly participate in guanylylation of GTase , a conclusion that is contrary to a previous suggestion based on a likely incorrect placement of a GMP molecule in the active site in a poorer resolution map ( Yang et al . , 2012 ) . 10 . 7554/eLife . 07901 . 028Figure 7 . The α-phosphorus of GTP bound to the GTase site moves towards His217 and away from Lys234 accompanying the activation of GTase . ( A ) The distance between the Nε2 of His217 and the α-phosphorus of GTP in G-CPV is ∼6 . 5 Å . The distance between the Nε of Lys234 and the α-phosphorus is ∼5 . 4 Å . Molecules and Mg2+ are colored as in Figure 5B . ( B ) The distance between Nε2 of His217 and the α-phosphorus of GTP in t-CPV is ∼4 . 8 Å . The distance between the Nε of Lys234 and the α-phosphorus is ∼7 . 6 Å . ( C ) The distance between Nε2 of His217 and the α-phosphorus of GTP in SGA-CPV is ∼4 . 5 Å . The distance between the Nε of Lys234 and the α-phosphorus is ∼7 . 8 Å . ( D ) The distance between Nε2 of His217 and the α-phosphorus of GTP in SG-CPV is ∼4 . 7 Å . The distance between the Nε of Lys234 and the α-phosphorus is ∼7 . 4 Å . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 028 By contrast , the histidine-rich segment contains two histidines ( His208 and His217 ) that are either hydrogen bonded to or in proximity with the α- or β-phosphate of the bound GTP in the open state of CPV GTase ( Figure 6A , C , D ) and conserved in both the orthoreovirus ( Reinisch et al . , 2000 ) and aquareovirus ( Zhang et al . , 2010b ) ( Figure 6—figure supplement 2 ) . Located on the same side of the leaving group ( i . e . , β , γ-diphosphate ) , His208 is hydrogen bonded to the β-phosphate of the GTP in SGA-CPV and SG-CPV . Therefore , His208 is not the active site residue; rather , it stabilizes the charge built up on the β-phosphate in the transition state during the catalysis process . Instead , His217 is likely the active site residue . His217 and the leaving group are on opposite sides of the α phosphorus , a geometry suitable for in-line nucleophilic attack of the α phosphorus by His217 ( Figures 6 , 7 and Figure 6—figure supplement 2B ) . Indeed , these two conserved histidine residues in orthoreovirus are required for the GTase activities ( Qiu and Luongo , 2003 ) . Furthermore , unlike KxDG GTases that have maximum activity at high pH , GTases of viruses in the Reoviridae family have maximum activity at pH about or lower than the pKa value ( ∼6 . 0 ) of histidine ( Qiu and Luongo , 2003 ) . Because we observed only the pre guanylylation state of GTase in all three CPV structures ( t-CPV , SGA-CPV , and SG-CPV ) , we reason that enzyme guanylylation mediated by His217 is likely the rate-limiting step in the process of guanylyl transfer . Our structures also indicate that the putative viral ATPase regulates the methyl transfer activity of MT-1 through a long-range allosteric effect ( Figures 1D , 8 and Figure 8—figure supplement 1 ) . First of all , active sites of MT-1 and the putative ATPase are spatially separated from each other ( Figure 1D and Figure 8—figure supplement 1 ) as the distance from the putative ATPase site to the MT-1 in the same molecule is ∼80 Å , while that to a neighboring MT-1 is ∼40 Å ( Figure 8—figure supplement 1A ) . Second , even if the putative ATPase is activated by SAM but lacks ATP for binding/hydrolysis , MT-1 remains incapable of SAM binding as was observed in S-CPV . In t-CPV where ATP is available , MT-1 becomes SAM bound ( Figure 8 and Figure 8—figure supplement 2A ) . The structures of MT-2 in S-CPV and t-CPV are essentially identical ( Figure 8A , B ) , but their MT-1 structures differ . In particular , two loops lining one side of the un-occupied MT-1 active site in S-CPV shifted up to 4 Å in t-CPV , resulting in an enlarged active site to accommodate the SAM molecule required for methyl transfer ( Figure 8C , D ) . In SGA-CPV and SG-CPV , the structures of MT-1 domains with bound SAM are essentially identical to those in t-CPV ( Figure 8—figure supplement 2 ) , though the outwards movement of MT-1 in SG-CPV is ∼1 . 1 Å less ( Figure 8E ) , likely due to the lower rate of GTP hydrolysis by the ATPase ( Figure 5B ) . 10 . 7554/eLife . 07901 . 029Figure 8 . The catalytic activity of MT-1 is also regulated by the viral ATPase through allosteric effect . ( A ) Structure of MT-2 domain and the bound SAM in t-CPV . MT-2 domain is in purple . SAM is in green . Left , viewed as in Figure 2C . Right , view rotated as indicated . ( B ) Superimposition of MT-2 between S-CPV ( gray ) and t-CPV ( purple ) . Molecules were aligned using Ca positions for residues in domain . ( C ) Structure of MT-1 domain and the bound SAM in t-CPV . MT-1 domain is in magenta . SAM is in green . Left , view as the guide map ( inset ) . Middle , view rotated as indicated . Right , active site of MT-1 domain . SAM is colored as in Figure 2D . Side chains of amino acids involved in interactions with SAM are shown . ( D ) Superimposition of MT-1 between S-CPV ( gray ) and t-CPV ( magenta ) . Molecules were aligned using Ca positions for residues in MT-1 domain . The bound SAM of t-CPV is in green . Left , viewed as the guide map in C . Right , view rotated as indicated . Inset: zoom-in view of MT-1 active site . ( E ) Superimposition of MT-1 active site between SG-CPV ( gray ) and t-CPV ( magenta ) . The SAM molecules bound to the active sites of SG-CPV and t-CPV are colored in coral and green , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 02910 . 7554/eLife . 07901 . 030Figure 8—figure supplement 1 . The putative viral ATPase regulates the methyl transfer activity of MT-1 . ( A ) One TP monomer ( domain colored ) and its neighboring MT-1 domain ( red ) of t-CPV . SAM , GTP , and ATP are in green , orange red , and magenta , respectively . ( B ) One GTase domain ( sky blue ) and one MT-1 domain ( red ) from its neighboring TP molecule . ( C ) One GTase domain ( sky blue ) and one MT-1 domain ( red ) from its neighboring TP molecule in t-CPV and one GTase domain of S-CPV ( gray ) . The GTase domain of S-CPV was aligned into the GTase domain of t-CPV using Ca positions for residues in domain . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 03010 . 7554/eLife . 07901 . 031Figure 8—figure supplement 2 . Structures of MT-1 active sites and SAMs . ( A ) Structure of MT-1 active site and SAM in t-CPV . MT-1 domain is in magenta . SAM is colored by element as in Figure 2D . Side chains of amino acids interacting with SAM are shown . Left: density map ( gray mesh ) of SAM is contoured at 1 . 4σ above the means . Right: density map of SAM is contoured at 3 . 0σ above the means . ( B ) Structure of MT-1 active site and SAM in SGA-CPV . Models and density map are colored and contoured as in A . ( C ) The structure of MT-1 active site and SAM in SG-CPV . Models and density map are colored and contoured as in A . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 031 In this study , we discovered that the large sub-domain of CPV GTase domain has an ATP-binding site and is likely an ATPase . This putative viral ATPase has the conserved structural motif for recognition of the adenine base of ATP: hydrogen bonds between the side chains of Ser301 , Asp302 and Tyr316 and the adenine ring , pi–pi stacking between the side chain of Tyr305 with the adenine base , and the cation–pi stacking between the side chain of Arg271 and the adenine base ( Figure 4A , B ) . However , this putative viral ATPase lacks the structural motifs of canonical nucleoside triphosphatases ( NTPases ) ( including cellular kinases ) , most notably the P loop , for binding the phosphoryl moiety of NTP ( Saraste et al . , 1990; Smith and Rayment , 1996; Snider and Houry , 2008 ) . Instead , the phosphoryl group of the bound ATP is stabilized through hydrogen bonds with the side chain of Arg271 or Tyr268 from α13 helix ( Figure 4A , B ) . Although the large sub-domain has an α helices/β sheet fold , the putative viral ATPase active site is composed of three consecutive helices of α13 , α14 , and α15 ( Figure 4A , B ) . Structurally , the putative ATPase is different from all ATPase known to date , and it may thus represent a new type of ATPase . Although both GTP and ATP can bind at the putative viral ATPase site , their interactions with the protein have some differences ( Figure 4A–C ) . The base and ribose of the bound GTP are less hydrogen bounded to the active site than those of the bound ATP . More importantly , while the triphosphate group of the bound GTP forms only one hydrogen bond through its β phosphate with Tyr268 of the active site ( Figure 4C ) , the triphosphate moiety of the bound ATP is better stabilized by forming two more hydrogen bonds with the protein ( Figure 4A , B ) . Conceivably , the more hydrogen bonded ATP is a more efficient substrate of the putative viral ATPase for hydrolysis than GTP , consistent with the colorimetric assay of CPV NTPase activity ( Figure 5B ) . By integrating the atomic structures of six different CPV particles and correlation with NTPase assay results , we propose a viral ATPase-mediated activation of mRNA transcription and capping , as illustrated in Figure 9 . As a virus must rely on host cell for replication , it is to the best interest of the virus to remain quiescent outside host cells ( Figure 9A ) . CPV senses the entrance into host cytoplasm by detecting the presence of SAM . SAM , acting as a signal and binding to its receptor of MT-2 , causes initial conformational change of the virus capsid , which activates the putative viral ATPase ( Figure 9B ) . The activated viral ATPase then binds and hydrolyzes ATP to cause three major structural transformations , leading to mRNA transcription and capping ( Figure 9C ) . First , as a result of the translocation of CSP , the viral capsid is enlarged , facilitating dsRNA template movement and enabling efficient mRNA synthesis ( Figure 9C ) . Second , the GTase domain transforms from its closed to open state ( Figure 9C ) . Although GTP binds to the GTase active site in both states of GTase domain , only in the open state can the GTase bind Mg2+ and catalyze His217-mediated guanylyl transfer . Third , the MT-1 domain transforms from its closed to open conformation ( Figure 9C ) . Only in its open conformation can MT-1 bind SAM . While the MT-1 and GTase domains from the same molecule are separated by the bridge domain and have no direct contact with each other , the MT-1 from a neighboring TP sits atop the putative ATPase sub-domain of GTase domain ( Figure 8—figure supplement 1 ) . We , therefore propose that the putative ATPase regulates the activity of MT-1 in a neighboring TP , probably through the conformational changes of the putative ATPase sub-domain upon ATP binding/hydrolysis . Notably , from S-CPV to t-CPV , the C-terminal loop of GTase domain exhibits significant movement towards the MT-1 domain of its neighboring TP molecule , presumably contributing to open the active site of the MT-1 ( Figure 8—figure supplement 1 ) . 10 . 7554/eLife . 07901 . 032Figure 9 . Schematic illustration of the putative viral ATPase-mediated activation of mRNA transcription and capping . In this illustration , the active open states of enzymes are shown in filled colors and the inactive closed states of enzymes are shown in dotted color lines . ( A ) CSP-A ( red ) and GTase , bridge and MT-2 domains from the same TP molecule , and a neighboring MT-1 ( colored as in Figure 1D ) of unliganded CPV . The inactive ATPase site is indicated by three empty cylinders . ( B ) CSP-A and GTase , bridge and MT-2 domains from the same TP molecule , and a neighboring MT-1of S-CPV . SAM alone can only bind to MT-2 domain ( purple ) to cause conformational change and activate the putative viral ATPase . The activated ATPase site is indicated by three colored cylinders . ( C ) CSP-A and GTase , bridge and MT-2 domains from the same molecule , and a neighboring MT-1 of t-CPV . DOI: http://dx . doi . org/10 . 7554/eLife . 07901 . 032 Several viruses within the Reoviridae , such as rotaviruses and blue-tongue viruses , cause wide spread diseases in human and live stocks . Some of these multi-shelled viruses in the Reoviridae , including the animal reovirus and blue-tongue virus , have been shown to have ATPase activity ( Noble and Nibert , 1997; Ramadevi and Roy , 1998 ) . The remarkable parallel between the ATPase activity and the transcription activity indicated that they may have also employed the ATPase-mediated activation for mRNA transcription and capping identified here . For example , reovirus cores had high level of ATPase activity ( Noble and Nibert , 1997 ) and could synthesize mRNA ( Shatkin and Sipe , 1968; Banerjee and Shatkin , 1970; Drayna and Fields , 1982; Farsetta et al . , 2000 ) , but virions had little ATPase activity ( Noble and Nibert , 1997 ) and could not synthesize mRNA ( Shatkin and Sipe , 1968; Farsetta et al . , 2000 ) . Thus , it is the removal of outer shell other than SAM that triggers the activation process in multi-shelled reoviruses . CPV virions were purified as described ( Yu et al . , 2008 ) . Briefly , purified polyhedra were treated with an alkaline solution of 0 . 2 M Na2CO3-NaHCO3 ( pH 10 . 8 ) for 1 hr . The suspension was centrifuged at 10 , 000×g for 40 min . The resulting supernatant was collected and then centrifuged again at 80 , 000×g for 60 min at 4°C to pellet the CPV virions . The final pellet was re-suspended in a reaction buffer ( 70 mM pH 8 . 0 Tris-Cl , 10 mM MgCl2 , and 100 mM NaCl ) and used for the following experiments . We prepared six different CPV samples using a protocol modified from a previously described CPV transcription essay ( Smith and Furuichi , 1980 ) . Reaction mixtures ( 30 μl ) contained purified CPV , 70 mM Tris-Cl ( pH 8 . 0 ) , 10 mM MgCl2 , 100 mM NaCl , and 1 mM SAM ( S-CPV ) , or 1 mM SAM+2 mM GTP+2 mM UTP+2 mM CTP+4 mM ATP ( t-CPV ) , or 1 mM SAM+2 mM GTP+ 2 mM ATP ( SGA-CPV ) , or 1 mM SAM+2 mM GTP ( SG-CPV ) , or 2 mM GTP ( G-CPV ) , or 2 mM ATP ( A-CPV ) . All reactions were incubated at 31°C for 15 min and stopped by quenching the reaction tubes on ice . Each of the six different CPV particles mentioned above was embedded in a thin layer of vitreous ice suspended across the holes of holey carbon films by plunge-freezing into liquid ethane . Before data collection , beam tilt was carefully minimized by coma-free alignment . Viral particle samples were kept at liquid-nitrogen temperature . CryoEM images were recorded on Kodak SO163 films at a dosage of ∼25 electrons/Å2 on an FEI Titan Krios cryo-electron microscope operated at 300 kV and 59 , 000× nominal magnification with parallel beam illumination . The films were digitized with a Nikon scanner at a step size of 6 . 35 µm/pixel , corresponding to 1 . 076 Å/pixel at the sample level . Individual particle images ( 960 × 960 pixels ) were first boxed out automatically by the autoBox program in the IMIRS package ( Liang et al . , 2002 ) and then followed by manual screening using the EMAN boxer program ( Ludtke et al . , 1999 ) to keep only the well-separated , contamination-free , intact RNA-containing particles . The program CTFFIND ( Mindell and Grigorieff , 2003 ) was used to determine the defocus value and astigmatism parameters for each micrograph . We determined particle orientation , center parameters with the IMIRS package running in MPI-enabled Windows workstations ( Liang et al . , 2002 ) . 3D reconstruction was performed by eLite3D using graphical processing units ( Zhang et al . , 2010a ) . We considered astigmatism during CTF correction in the orientation/center refinement and 3D reconstruction steps . Effective resolutions of the final reconstructions were estimated to be 2 . 9–3 . 1 Å ( Figure 1A–C , Table 1 and Figure 1—figure supplement 1 ) , based on the structural features revealed in the cryoEM density maps , R-factors ( Wolf et al . , 2010 ) , and Fourier shell correlation coefficient ( FSC ) criterion as defined by Rosenthal and Henderson ( Rosenthal and Henderson , 2003 ) . We have previously shown that our common-lines-based programs do not suffer from the problems of over-fitting or model bias ( Zhou et al . , 2014 ) . To provide further validation , we took advantage of the existence of identical structures of t-CPV and SGA-CPV ( both independently determined ) and calculated the FSC curves between the SGA-CPV map and SGA-CPV model and that between the SGA-CPV map and t-CPV model . Because these structures were independently determined , they are essentially the same as ‘gold-standard’ FSC ( Scheres and Chen , 2012 ) . These analyses further support our conclusion that our reconstructions do not have over-fitting ( Figure 1—figure supplement 1B ) . Rebuilding the model to fit the EM map was done manually with COOT ( Emsley and Cowtan , 2004 ) with the help of REMO ( Li and Zhang , 2009 ) . The ‘regularize zone’ utility of COOT was used to improve model stereochemistry . These coarse full-atom models were then refined in a pseudocrystallographic manner using Phenix ( Adams et al . , 2010 ) . This procedure only improves atomic models and does not modify the cryoEM density map . Densities for individual proteins were segmented , put in artificial crystal lattices , and then used to calculate their structural factors . The amplitudes and phases of these structural factors were used as pseudo-experimental diffraction data for model refinement in Phenix . To improve the areas of interaction between different protein subunits , we put the refined structures of all five subunits from an asymmetric unit into a single coordinate file and pseudo-crystallographically refined them simultaneously with their non-crystallographic symmetry . This refinement process uses pseudo-experimental diffraction data generated from the cryoEM map of an asymmetric unit . CryoEM reconstruction was visualized and segmented using Chimera ( Pettersen et al . , 2004 ) . All figures were prepared with Chimera and COOT . The NTPase reactions and colorimetric assay were performed as described by Noble and Nibert ( 1997 ) in 1 . 5-ml eppendorf tubes . NTPase reaction mixtures contained 100 mM Tris-Cl ( pH 8 . 0 ) , 100 mM NaCl , 10 mM MgCl2 , without or with 1 mM SAM , 6 × 1011 CPV particles per ml , and 1 mM of one of the 4 NTPs in a total volume of 60 µl . Reaction components were mixed on ice , incubated at 31°C for 30 min and then returned to ice . Termination of each reaction was ensured by the addition of an equal volume of 10% trichloroacetic acid . To measure the amount of phosphate ion in each sample , the stopped reaction mixture was mixed with an equal volume of colorimetric reagent ( 3 vol of 0 . 8% ammonium molybdate , 1 vol of 6 N sulfuric acid , 1 vol of 10% [wt/vol] ascorbic acid ) . After all samples in the experiment were added , the eppendorf tubes were incubated in a water bath at 31°C for 30 min . During development , a reduced phosphomolybdate complex was formed , which was blue in color and quantifiable by A655 nm . In each experiment , samples containing NTP but no CPV were included to permit correction for background . The cryoEM density maps and atomic coordinates reported here are deposited in the EM Data Bank and the Protein Data Bank with accession codes EMD-6371 ( A-CPV ) , EMD-6374 ( G-CPV ) , EMD-6375 ( S-CPV ) , EMD-6376 ( SG-CPV ) , EMD-6377 ( SGA-CPV ) , EMD-6378 ( t-CPV ) ( Yu et al . , 2015a; 2015b; 2015c; 2015d; 2015e; 2015f ) and 3JAZ ( A-CPV ) , 3JB0 ( G-CPV ) , 3JB1 ( S-CPV ) , 3JB2 ( SG-CPV ) , 3JB3 ( SGA-CPV ) , 3JAY ( t-CPV ) ( Yu et al . , 2015g; 2015h; 2015i; 2015j; 2015k; 2015l ) , respectively .	Viruses can only replicate by invading the cells of other organisms , such as plants and animals . Each virus carries genetic material in the form of molecules of DNA or ribonucleic acid ( RNA ) , which are packaged in a shell made of proteins . The cytoplasmic polyhedrosis virus has a genome made of a type of RNA called double-stranded RNA . Once inside a host cell , sections of the virus genome are copied to make molecules of ‘messenger RNA’ in a process called transcription . Small chemical groups called guanylyl and methyl groups are added to the messenger RNAs before they are used as templates to make the virus proteins . A small molecule called S-adenosyl-L-methionine ( SAM ) can activate transcription of the virus genome by binding to a protein called turret in the shell of the virus . The turret protein is involved in adding the guanylyl and methyl groups to the messenger RNA molecules , but it is not clear how the protein activates transcription . Here , Yu et al . used a technique called cryo electron microscopy to study how the virus binds SAM to activate transcription . The experiments show that the binding of SAM to one region or ‘domain’ of the turret protein leads to changes in the virus shell . This enables another domain of the turret protein to bind a small molecule called ATP and break it down . The energy released from breaking down ATP causes further changes of the shell of the virus to activate transcription and the addition of guanylyl and methyl groups to the newly made messenger RNAs . In the future , experiments that directly observe the RNA inside each virus shall offer fresh insights as to how the genomes of cytoplasmic polyhedrosis virus and other similar viruses are transcribed .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	A putative ATPase mediates RNA transcription and capping in a dsRNA virus
Multicellular organisms maintain structure and function of tissues/organs through emergent , self-organizing behavior . In this report , we demonstrate a critical role for lung mesenchymal stromal cell ( L-MSC ) aging in determining the capacity to form three-dimensional organoids or ‘alveolospheres’ with type 2 alveolar epithelial cells ( AEC2s ) . In contrast to L-MSCs from aged mice , young L-MSCs support the efficient formation of alveolospheres when co-cultured with young or aged AEC2s . Aged L-MSCs demonstrated features of cellular senescence , altered bioenergetics , and a senescence-associated secretory profile ( SASP ) . The reactive oxygen species generating enzyme , NADPH oxidase 4 ( Nox4 ) , was highly activated in aged L-MSCs and Nox4 downregulation was sufficient to , at least partially , reverse this age-related energy deficit , while restoring the self-organizing capacity of alveolospheres . Together , these data indicate a critical role for cellular bioenergetics and redox homeostasis in an organoid model of self-organization and support the concept of thermodynamic entropy in aging biology . Substantial progress has been made in our understanding of the biology of aging , and these advances have the potential to improve both healthspan and lifespan , while alleviating the burden of age-related diseases . Self-organization in biological systems is a process by which cells reduce their internal entropy and maintain order within these dynamic , self-renewing systems ( Chatterjee et al . , 2017; Kiss et al . , 2009 ) . Exhaustion of stem/progenitor cells , cellular senescence , and altered intercellular communication have been proposed as aging hallmarks that increase susceptibility to age-related disorders ( Schultz and Sinclair , 2016 ) . However , the interactions between these hallmarks and whether cellular bioenergetics associated with cellular senescence may account for age-associated stem cell dysfunction and altered cell-cell communication have not been well defined . In this study , we utilized an organoid model to study stem cell behavior and intercellular communication that may account for age-related phenotypes; through these studies , we identify cellular bioenergetics and redox imbalance as critical drivers of these inter-dependent aging hallmarks . The mammalian lung serves an essential role in organismal metabolism , uniquely by serving as the primary organ for systemic exchange of oxygen for carbon dioxide . Regeneration and maintenance of structure-function of the lung are dependent on adult , tissue-resident stem cells that reside in unique niches along the airways ( Basil et al . , 2020; Hogan et al . , 2014 ) . Type 2 alveolar epithelial cells ( AEC2s ) serve as facultative stem/progenitor cells in adult mammalian lungs and differentiate into type 1 alveolar epithelial cells ( AEC1s ) in response to diverse injuries to reconstitute and reestablish the alveolar gas exchange surface ( Barkauskas et al . , 2013 ) . AEC2 regenerative capacity declines with age , resulting in impaired lung injury repair responses , thus increasing susceptibility to various lung diseases ( Schulte et al . , 2019; Watson et al . , 2020 ) . To further explore the relationship between AECs and lung mesenchymal stromal cells ( L-MSCs ) , we developed an alveolosphere assay system that has been traditionally used to assess AEC2 stemness or regenerative potential ( Barkauskas et al . , 2013 ) . In this assay system , L-MSCs and AEC2s are mixed together in a ratio of 100 , 000–5000 , , respectively , and seeded in Matrigel ( Figure 1A ) ; alveolospheres with a single layer of epithelial cells composed of both AEC2s ( surfactant protein C [SFTPC]-positive ) and AEC1s ( lung type I integral membrane glycoprotein [T1α]-positive ) surrounding a hollow sphere typically form 9–12 days following co-culture ( Figure 1B and C ) . L-MSCs expressing platelet-derived growth factor receptor-α ( PDFGRα ) , with a fewer number expressing α-smooth muscle actin ( α-SMA ) , were found primarily in cells lining the outer edges of alveolospheres ( Figure 1D and E ) . AEC2s within alveolospheres stained for both the cell proliferation marker , Ki-67 , and the double-strand DNA damage repair marker , histone H2A . X , suggesting ongoing AEC2 turnover within these 3D organoids ( Figure 1F and G ) . This self-organizing behavior of L-MSCs and AEC2s is critically dependent on the presence of both cell types as AEC2s in the absence of AEC2s do not self-organize to form alveolospheres ( Figure 1—figure supplement 1A ) . Together , these data support the crosstalk between L-MSCs and AEC2s that permit formation of distinct alveolar organoid-like structures; this intercellular communication may be perturbed during aging accounting for age-associated loss of AEC2 regeneration/maintenance . To determine the effect of age on cellular self-organization and alveolosphere formation , L-MSCs and AEC2s were isolated from the lungs of young ( 2–3 months ) and aged ( 22–24 months ) mice and seeded in various combinations ( Figure 1H , upper panel ) . L-MSCs and AEC2s from young mice generated normal alveolospheres , whereas L-MSCs and AEC2s from the lungs of aged mice showed impaired self-organization with condensed , poorly formed organoids . Interestingly , combining aged AEC2s with young L-MSCs resulted in relatively normal-appearing alveolospheres , while the reverse combination ( young AEC2s and old L-MSCs ) did not ( Figure 1H , middle and lower panels; Figure 1—figure supplement 1B ) . Quantitative analyses revealed that , while the number of alveolospheres formed was not significantly different between groups ( Figure 1I ) , the size of alveolospheres was critically dependent on the age of L-MSCs , and not that of AEC2s ( Figure 1J ) . These findings indicate that secreted factor ( s ) from young L-MSCs are capable of supporting the self-organizing behavior of alveolospheres , while aged L-MSCs do not . Cellular senescence accumulates in tissues with advancing age ( Krishnamurthy et al . , 2004 ) and has been proposed as a key driver of aging and aging-related disease phenotypes ( Baker et al . , 2016; Kennedy et al . , 2014 ) . Based on our observation that aged L-MSCs were incapable of supporting alveolosphere formation , we explored whether the emergence of cellular senescence may account for this finding . First , we confirmed the senescent features of L-MSCs from aged mice in monolayer 2D cell culture ( Figure 2A , Figure 2—figure supplement 1 ) and in 3D alveolospheres ( Figure 2B ) by staining for β-galactosidase ( β-gal ) ( Dimri et al . , 1995 ) and lipofuscin ( Georgakopoulou et al . , 2013 ) , respectively . To characterize secreted factors that may account for the age-associated dysregulation of cell-cell communication , we first measured cytokines/growth factors secreted from both young and aged L-MSCs using antibody arrays ( Figure 2C ) . A number of cytokines that have been associated with the senescence-associated secretory phenotype ( SASP ) were found to be released at higher levels by aged L-MSCs , including IL-6 , CCL12/MCP-5 , CCL11/Eotaxin , and WISP-1/CCN4; in contrast , IGFBP1 was the only secreted protein that was statistically more elevated in young L-MSCs in this cytokine array ( Figure 2D and E ) . In addition to predefined cytokine array analyses , we employed an unbiased approach to identification of secreted proteins from young and aged L-MSCs by mass spectrometry-based discovery proteomics ( Figure 2F ) . After adjustments for false discovery rates , 503 high-confidence proteins were identified; of these , 235 proteins were found to have non-zero quantifiable values in at least three of four experimental repeats per group for subsequent statistical analysis . The 36 proteins that passed both a single pairwise statistical test ( p<0 . 05 ) in addition to a fold change of ±1 . 5 ( Supplementary file 1A ) were subjected to principal component analysis ( PCA ) ( Figure 2G ) and heat map analysis ( Figure 2H; quantitation of heat map proteins , Figure 2I ) . Gene ontology localization analyses revealed enrichment in secreted proteins , including extracellular vesicles and extracellular matrix ( Supplementary file 1B ) ; interestingly , gene ontology processes analysis indicated the top cellular processes as negative regulation of reactive oxygen species metabolic process and extracellular matrix/structure organization ( Supplementary file 1C ) . Enrichment by top toxic pathologies revealed proteins associated with lung fibrosis ( Supplementary file 1D ) , a disease associated with aging ( Thannickal et al . , 2014 ) . Network analysis revealed upregulation of pathways associated with the canonical WNT-signaling pathway and cell cycle regulation ( Supplementary file 1E ) . Together , these findings suggest that aged L-MSCs are characterized by cellular senescence associated with SASP factors , oxidative stress , and alterations in regenerative pathways that may account for impaired alveolosphere formation . Altered cellular metabolism has been linked to both senescence and aging ( López-Otín et al . , 2016; Finkel , 2015; Sun et al . , 2016 ) . To compare the energy state between young and aged lung L-MSCs , rates of mitochondrial respiration and glycolysis were analyzed by Seahorse XF Analyzer . Real-time oxygen consumption rates ( OCR ) and extracellular acidification rates ( ECAR ) were significantly higher in aged L-MSCs as compared to young L-MSCs ( Figure 3A and B ) . Aged L-MSCs showed significantly higher basal , maximal , ATP-linked , proton leak , and non-mitochondrial respiration when compared to young L-MSCs with no change in reserve capacity ( Figure 3C-H ) . Higher non-mitochondrial OCRs in aged L-MSCs suggest potential activation of oxygen-metabolizing NADPH oxidases in aged L-MSCs . An energy map profile of the OCR and ECAR data indicated a higher basal rate of both mitochondrial respiration and glycolysis ( Figure 3I ) , suggesting that aged L-MSCs are under a basal state of higher metabolic demand . This higher energy demand under conditions of equivalent nutrient supply was associated with reduced levels of ATP/ADP under steady-state basal conditions ( Figure 3J ) . Measurement of ECAR indicated that , despite similar basal ECAR , higher rates of maximal and non-glycolysis related ECAR were observed in aged L-MSCs . ( Figure 3—figure supplement 1 ) . Together , these data indicate that senescence of L-MSCs is characterized by higher baseline metabolic demand with reduced bioenergetic efficiency . Hydrogen peroxide ( H2O2 ) has emerged as critical regulator of redox signaling and oxidative stress ( Sies , 2017 ) . Replication-induced senescence of human lung fibroblasts results in higher rates of extracellular H2O2 release in association with increased expression of NADPH oxidase 4 ( Nox4 ) ( Sanders et al . , 2015 ) , a gene that is inducible by the pro-senescent/pro-fibrotic mediator , transforming growth factor-β1 ( TGF-β1 ) ( Hecker et al . , 2009 ) . We explored whether L-MSCs isolated from naturally aged mice are associated with higher Nox4-dependent H2O2 secretion; in comparison to young ( 3 months old ) L-MSCs , aged ( 22–24 months old ) , L-MSCs released higher baseline levels of H2O2 , an effect that was further stimulated by exogenous stimulation with TGF-β1 ( Figure 4A ) . This age-dependent , pro-oxidant phenotype of L-MSCs was markedly reduced in aged mice heterozygous for Nox4 ( Nox4-/+; Figure 4B and C ) , implicating Nox4 as a critical mediator of both basal and TGF-β1-induced H2O2 release . Although it is difficult to replicate the tonic , continuous release of H2O2 from a constitutive enzymatic source such as Nox4 in metabolically active cells , we tested the effect of a bolus addition of H2O2 ( 1 µM ) to organoid culture comprising young AEC2s and young L-MSCs; this resulted in a decrease in the size of alveolospheres , without significantly affecting the numbers of alveolospheres formed ( Figure 4—figure supplement 1A–C ) . Recent studies implicate Nox4 in metabolic reprogramming ( Bernard et al . , 2017 ) , although the effects of Nox4 on age-related metabolic dysfunction are unclear . The reduced levels of ATP/ADP in aged L-MSCs were partially rescued in age-matched Nox4-/+ L-MSCs ( Figure 4D ) , implicating a role for Nox4 in the reduced bioenergetic efficiency associated with aging . The steady-state levels of the tricarboxylic acid ( TCA ) cycle metabolites , malate and citrate , were reduced in wild-type aged L-MSCs; these levels recovered to levels similar to young L-MSCs in Nox4-/+ L-MSCs , although only malate achieved statistical significance ( Figure 4E and F ) . Levels of other TCA metabolites that were not significantly altered with aging in L-MSCs were not affected by a deficiency in Nox4 ( Figure 4—figure supplement 1D-G ) . β-gal activity was found to be significantly lower in Nox4-deficient aged L-MSCs as compared to wild-type aged L-MSCs ( Figure 4—figure supplement 1H ) . In addition to metabolic dysfunction , an important feature of senescence reprogramming is the activation of a SASP program . We determined whether Nox4 modulates the release of cytokines/growth factors by aged L-MSCs . Protein ( antibody-based ) arrays showed significant downregulation of SASP-associated cytokines in aged Nox4-/+ L-MSCs vs . aged L-MSCs ( Figure 4—figure supplement 1I ) . We were unable to identify a group of cytokines ( except for coagulation factor III ) that were specifically identified in the dataset comparing young vs . aged L-MSCs based on the antibody-based cytokine array method . Further studies with an unbiased approach such as that employed by mass spectrometric analyses ( Figure 2I ) will be required for more in-depth analyses of Nox4 regulated SASP cytokines/growth factors . Interestingly , Nox4 haploinsufficiency in aged L-MSCs resulted in significant downregulation of OCR as compared to aged L-MSCs ( Figure 4G ) . Nox4-/+ aged L-MSCs showed significantly lower basal , ATP-linked , proton leak , and maximal respiration compared to wild-type aged L-MSCs . Non-mitochondrial respiration was also lower in aged Nox4-/+ L-MSCs , although this did not achieve statistical significance ( Figure 4H ) . Together , these studies indicate that Nox4 contributes to metabolic dysfunction , bioenergetic inefficiency , and SASP programming in L-MSC aging . Based on the observations that Nox4 alters cellular bioenergetics of L-MSCs , we examined the effects of Nox4 on regulating the self-organizing potential of alveolospheres . The aberrant formation of alveolospheres when young AEC2s were cultured with aged L-MSCs were rescued when replaced with age-matched Nox4-/+ L-MSCs ( Figure 4I; Figure 4—figure supplement 1J ) ; quantitation of both alveolospheres counts ( Figure 4J ) and alveolosphere size ( Figure 4K ) was significantly enhanced when these assays were conducted with Nox4-deficient L-MSCs . We were also unable to detect significant changes in alveolosphere formation when the Nox4 inhibitor , GKT137831 ( 10 µM ) , was added to an organoid culture system of young AEC2s and aged L-MSCs ( Figure 4—figure supplement 1K-M ) ; however , it is difficult to ascertain whether this intervention effectively inhibited Nox4 activity in this organoid model . While the phenotype of impaired alveolosphere formation in young AECs and aged L-MSCs was reversed by haploinsufficiency of Nox4 in L-MSCs ( Figure 4I–K ) , this was not sufficient to completely reverse the phenotype in aged AEC2s ( Figure 4—figure supplement 1N-P ) . This finding , along with the observation that wild-type young L-MSCs are sufficient to reverse the aged AEC2 phenotype ( Figure 1H ) , suggests that in addition to reduced expression of Nox4 in aged L-MSCs , the young mesenchyme may support niche function and recovery of alveolosphere formation of aged AEC2s through Nox4-independent mechanisms . Thus , an aging alveolar stem cell niche may involve both a gain of ‘pro-senescent factors’ ( Nox4 ) and loss of ‘rejuvenation factors’ by the mesenchyme; the identification of the latter will require further study . Together , our studies support a critical role for Nox4-dependent oxidative stress , senescence , and bioenergetic insufficiency in restricting the capacity for cellular self-organization in a 3D organoid model of aging and stem/progenitor cell function ( Figure 4L ) . A fundamental difference between non-living matter and life forms on our planet is the ability to extract energy from the environment , primarily through oxidation of carbon-based fuels . Thus , the second law of thermodynamics that governs both living and non-living matter does not strictly apply to living organisms capable of self-renewal through the continuous and dynamic exchange of energy ( from exogenous nutrients ) and mass ( biosynthetic and degradative processes in respiring tissues/organs ) . It follows then that , when this capacity for energy-dependent self-renewal is diminished , the law of entropic degeneration will be operative in living organisms . Aging is associated with derangements in metabolism that influences , and may in fact control , many of the well-recognized hallmarks of aging , including cellular senescence ( López-Otín et al . , 2016; López-Otín et al . , 2013 ) . The inter-relatedness of metabolism with these traditional aging hallmarks is difficult to untangle due to complexities of studies in living organisms and the relative simplicity of 2D cell culture models . Organoids offer the opportunity to study complex living phenomena such as emergent properties and self-organization in biological systems while at the same time allowing for reductionist approaches that provide mechanistic understanding of these complex processes . Using a combination of methods that included a 3D organoid model , biochemical approaches , proteomics , and gene targeting , we demonstrate a critical role for the oxygen metabolizing enzyme , Nox4 , in regulating cellular bioenergetics and senescence that limits stem cell function . Consistent with contemporary theories on aging , Nox4 may function as an antagonistically pleiotropic gene and contribute a number of age-related degenerative disorders , including fibrotic diseases ( Thannickal , 2010; Lambeth , 2007 ) . Interestingly , proteomic analyses of proteins secreted by aged L-MSCs in the current study suggested alterations in cellular redox/oxidative stress ( from gene ontology analysis ) and lung fibrosis ( as a ‘toxic pathology’ ) . These findings , as well as candidate SASP factors , highlight the emerging importance of intercellular communication and stem cell exhaustion as important hallmarks of aging ( López-Otín et al . , 2013 ) . Regenerative mechanisms in adult , mammalian organisms primarily rely on the activation and differentiation of tissue/organ-resident stem cells ( Hogan et al . , 2014 ) . Homeostatic maintenance of these stem cells requires a tightly regulated niche that includes other supporting cell types , in particular stromal cells ( Basil et al . , 2020 ) . An important finding from our studies is the observation that L-MSC aging , relative to AEC aging , largely accounts for the age-related inability to form alveolospheres . Thus , aging of the stem cell niche may be as important , and perhaps more decisive , in promoting certain age-related pathologies . Therapeutic targeting of metabolic aging within the stem cell niche , specifically that of lung-resident fibroblasts/myofibroblasts , may offer a more effective and viable strategy for age-related degenerative disorders such as tissue/organ fibrosis . All reagents , antibodies , assay kits , and mice used in this study are listed in Appendix 1—key resources table . All animal protocols were approved by the Institutional Animal Care and Use Committees ( IACUC ) at the University of Alabama at Birmingham . The mice were acclimatized in the animal facility at least for a week before experiments . Male mice were used in this study for their greater susceptibility to age-related diseases . Mouse lung L-MSCs were isolated and propagated following protocol developed in our laboratory ( Vittal et al . , 2005 ) . L-MSCs were isolated by collagenase digestion of the lungs and anchorage-dependent ex vivo growth on plastic culture dishes in Dulbecco's Modified Eagle Medium ( DMEM ) supplemented with 10% fetal bovine serum , 4 mM L-glutamine , 4 . 5 g/L glucose , 100 U/ml penicillin , 100 μg/ml streptomycin , and 1 . 25 μg/ml amphotericin B ( Fungizone ) , in a humified chamber at 37°C in 5% CO2 , 95% air . AEC2s were isolated and purified from the young and aged mice lungs via enzymatic digestion ( Dispase II ) , cell-specific antibody labeling ( biotinylated Ter-119 , CD104 , CD16/32 , CD45 , CD31 ) , and magnetic separation ( Anti-Biotin MicroBeads ) following published protocols ( Sinha and Lowell , 2016 ) and used immediately . AEC2 purity was determined by fluorescent detection of EpCam+/SFTPC+/CD45-/CD31- cell population using flow cytometry ( Sinha and Lowell , 2016; Bertoncello and Mcqualter , 2011 ) . L-MSCs and AEC2s were seeded in Matrigel mixed with MTEC/plus media ( You et al . , 2002 ) ( 1:1 ) in cell culture inserts ( 24-well format , 0 . 4 µm pore size ) , and co-cultured in MTEC/Plus media as shown in Figure 1A . Matrigel containing alveolospheres were fixed in 4% paraformaldehyde , embedded in HistoGel , dehydrated , and paraffin embedded using standard protocol . 5-µm-thick sections were cut and mounted on glass slides and immunostained for mouse SFTPC , T1-α , PDGFRα , α-SMA , Ki-67 , and histone 2A . X ( H2A . X ) . Briefly , alveolosphere sections were deparaffinized in xylene and hydrated through ethanol series and water . Antigen retrieval was performed using citrate buffer at pH 6 . 0 in a 95°C water bath . Tissue sections were blocked using 5% normal goat or donkey serum and were then incubated in primary antibodies overnight at 4°C . Appropriate IgG isotype controls were also used to determine specificity of staining . Secondary detection was performed using anti-mouse Alexa Fluor 594/488-tagged secondary antibodies . Nuclei were counterstained with Hoechst 33342 dye for immunofluorescence detection . The stained alveolosphere sections were mounted in Vectashield and viewed and imaged in a Keyence BZ-X710 inverted microscope with brightfield as well as fluorescent imaging capability . β-Galactosidase staining was performed on young and aged L-MSCs in culture , according to instructions provided in the Senescence Detection Kit ( Abcam ) . Lipofuscin staining was performed on paraffin sections of alveolospheres following published protocol ( Georgakopoulou et al . , 2013 ) . Cytokine array was carried out in the young and aged L-MSC culture media using Proteome Profiler Mouse XL Cytokine Array kit ( R&D Systems ) as per the manufacturer’s instructions . Proteomics analysis was carried out following established protocols ( Ludwig et al . , 2016 ) with minor changes . Briefly , 5 ml of each cell culture media specimen were concentrated using Amicon Ultra 4 ml , 3 kDa molecular weight cutoff filters ( Millipore ) and protein concentrations were determined using Pierce BCA Protein Assay Kit . Proteins ( 10 µg ) per sample were reduced with DTT and denatured at 70°C for 10 min prior to loading onto Novex NuPAGE 10% Bis-Tris Protein gels ( Invitrogen , Cat . # NP0315BOX ) . The gels were stained overnight with Novex Colloidal Blue Staining kit ( Invitrogen , Cat . # LC6025 ) . Following destaining , each lane was cut into 3 MW fractions and equilibrated in 100 mM ammonium bicarbonate ( AmBc ) , each gel plug was then digested overnight with Trypsin Gold , Mass Spectrometry Grade ( Promega , Cat . # V5280 ) following the manufacturer’s instruction . Peptide digests ( 8 µl each ) were injected onto a 1260 Infinity nHPLC stack ( Agilent Technologies ) and separated using a 75 µm I . D . × 15 cm pulled tip C-18 column ( Jupiter C-18 300 Å , 5 µm , Phenomenex ) . This system runs in-line with a Thermo Orbitrap Velos Pro hybrid mass spectrometer , equipped with a nano-electrospray source ( Thermo Fisher Scientific ) , and all data were collected in CID mode . Searches were performed with a species-specific subset of the UniprotKB database . The list of peptide IDs generated based on SEQUEST ( Thermo Fisher Scientific ) search results was filtered using Scaffold ( Protein Sciences , Portland , OR ) . Gene ontology assignments and pathway analysis were carried out using MetaCore ( GeneGO Inc , St . Joseph , MI ) . The original dataset has been submitted to Dryad repository ( doi:10 . 5061/dryad . x0k6djhj1 ) . Analyses of cellular bioenergetics were performed using the Seahorse XFe96 Extracellular Flux Analyzer ( Agilent , Santa Clara , CA ) . L-MSCs from young and aged mice were cultured for 24 hr in low serum ( 1% ) containing media . Cells were then detached and replated at a density of 25 , 000 cells/well into a XF96 microplate 5 hr prior to assay in the same media . After the cells were attached , XF-DMEM media ( DMEM supplemented with 5 . 5 mM glucose , 1 mM pyruvate , and 4 mM L-glutamine , pH 7 . 4 at 37°C ) was added to the wells , and cells were incubated for 1 hr prior to assay in a non-CO2 incubator at 37°C . Mitochondrial stress test , a parallel measure of basal OCR and ECAR , was carried out following sequential injections of oligomycin ( 1 µg/ml ) , carbonyl cyanide 4- ( trifluoromethoxy ) phenylhydrazone ( FCCP [0 . 6 µM] ) , antimycin-A ( 10 µM ) , and 2-deoxyglucose ( 50 µM ) . This assay was performed using an ADP/ATP ratio assay kit ( Abcam ) following the manufacturer’s instructions . This assay was performed based on protocol developed in our laboratory for quantitative measurement of extracellular H2O2 release from adherent fibroblasts ( Thannickal and Fanburg , 1995 ) . This fluorimetric method relies on the oxidative conversion of homovanillic acid ( HVA ) , a substituted phenolic compound , to its fluorescent dimer in the presence of H2O2 and horseradish peroxidase ( HRP ) . For this experiment , 150 , 000 young , aged , and aged Nox4-/+ mouse lung L-MSCs were seeded in 6-well culture dishes . Next day , cells were serum-starved ( 1% ) overnight and treated with either vehicle or 2 . 5 ng/ml TGF-β1 . Media was aspirated after 16 hr post-treatment; cells were washed with 1 ml of HBSS ( without calcium and magnesium ) and incubated for 2 hr in 1 ml of assay media containing 100 µM HVA and 5 U/ml HRP in HBSS ( containing calcium and magnesium ) . Reactions were stopped adding stop solution and fluorescence was measured in a BioTek microplate reader at excitation and emission maximum of 321 nm and 421 nm , respectively . Extracellular H2O2 release was measured from standard curve generated from a known concentration of H2O2 . The data are normalized to cell number , and H2O2 concentrations are presented as nanomoles/min/106 L-MSCs . Total RNA was isolated from aged and aged Nox4-/+ L-MSCs using RNeasy Mini Kit ( Qiagen ) and reverse transcribed using iScript Reverse Transcription SuperMix for RT- qPCR ( Bio-Rad ) . Real-timePCR reactions were performed using SYBR Green PCR Master Mix ( Life Technologies ) and gene-specific primer pairs for Nox4 and β-actin ( Integrated DNA Technologies; for primer sequences , see Appendix 1—key resources table ) . Reactions were carried out for 40 cycles ( 95°C for 15 s , 60°C for 1 min ) in a StepOnePlus Real Time PCR System ( Life Technologies ) . Real-time PCR data ( 2-ΔΔCt ) is presented as Nox4 mRNA expression normalized to β-actin . Concentrations of TCA metabolites in the young and aged L-MSCs were determined by liquid chromatography tandem mass spectrometry ( LCMS ) following published protocols ( Tan et al . , 2014 ) with minor modifications . Dry stocks of analytically pure standards were weighed out and solubilized in MilliQ water and further diluted with 50% methanol to a 10 µg/ml stock . This stock was diluted to generate concentrations for standards ranging from 0 . 5 to 1000 ng/ml . Methanolic sample extracts were dried under nitrogen ( N2 ) gas and reconstituted in 100 µl of 50% methanol . Methanolic samples and standards were derivatized by addition of 10 µl 0 . 1 M O-benzylhydroxylamine hydrochloride ( O-BHA ) and 10 µl 0 . 25 M N- ( 3-dimethylaminopropyl ) -N′-ethylcarbodiimide hydrochloride ( EDC ) at room temperature ( RT ) for 1 hr . Samples were then transferred to a 13 × 100 mm borosilicate tube for liquid-liquid extraction . Samples were then dried under N2 gas at RT and were reconstituted in 100 µl of 0 . 1% formic acid ( FA ) and transferred to HPLC vials for LCMS analysis . LCMS was carried out with a Prominence HPLC and API 4000 MS . 10 µl of sample was injected onto an Accucore 2 . 6 µm C18 100 × 2 . 1 mm column at 40°C for gradient separation . Column eluent was directed to the API 4000 MS operating system , controlled by Analyst 1 . 6 . 2 software . Post-acquisition data analysis was carried out using MultiQuant v3 . 0 . 3 software . All standard curve regressions were linear with 1/x2 weighting . Sample size calculation was deemed unnecessary as our study contains no clinical component or in vivo intervention in mice and was therefore not computed . In this study , three cohorts of mice ( young , aged , and aged-Nox4-/+ ) , consisting of 3–8 mice per cohort , were used to obtain sufficient quantities of AEC2s and L-MSCs to perform the necessary assays in biological and technical replicates . In this instance , each mouse in a given cohort was considered a biological replicate , while repeated measurements in cells obtained from each animal were considered technical replicates . Student’s t-test was used to compare between two experimental groups . When more than two experimental groups were present , data were analyzed by multifactor analysis of variance ( ANOVA ) . Pairwise comparisons were performed to identify which paired groups presented significant differences and were adjusted by Tukey’s correction . Data were tested for normality distribution by Shapiro–Wilk test before comparative analysis . Non-normal data ( alveolosphere size; Figures 1J and 4K ) were natural log transformed and compared by ANOVA and pairwise comparisons . Differences between the experimental groups were considered significant when p<0 . 05 . All experiments were repeated 2–4 times for reproducibility . Measurement of TCA metabolites was performed only once due to unavailability of age-matched aged Nox4-/+mice in our colony .	Many tissues in the body are capable of regenerating by replacing defective or worn-out cells with new ones . This process relies heavily on stem cells , which are precursor cells that lack a set role in the body and can develop into different types of cells under the right conditions . Tissues often have their own pool of stem cells that they use to replenish damaged cells . But as we age , this regeneration process becomes less effective . Many of our organs , such as the lungs , are lined with epithelial cells . These cells form a protective barrier , controlling what substances get in and out of the tissue . Alveoli are parts of the lungs that allow oxygen and carbon dioxide to move between the blood and the air in the lungs . And alveoli rely on an effective epithelial cell lining to work properly . To replenish these epithelial cells , alveoli have pockets , in which a type of epithelial cell , known as AEC2 , lives . These cells can serve as stem cells , developing into a different type of cell under the right conditions . To work properly , AEC2 cells require close interactions with another type of cell called L-MSC , which supports the maintenance of other cells and also has the ability to differentiate into several other cell types . Both cell types can be found close together in these stem cell pockets . So far , it has been unclear how aging affects how these cells work together to replenish the epithelial lining of the alveoli . To investigate , Chanda et al . probed AEC2s and L-MSCs in the alveoli of young and old mice . The researchers collected both cell types from young ( 2-3 months ) and aged ( 22-24 months ) mice . Various combinations of these cells were grown to form 3D structures , mimicking how the cells grow in the lungs . Young L-MSCs formed normal 3D structures with both young and aged AEC2 cells . But aged L-MSCs developed abnormal , loose structures with AEC2 cells ( both young and old cells ) . Aged L-MSCs were found to have higher levels of an enzyme ( called Nox4 ) that produces oxidants and other ‘pro-aging’ factors , compared to young L-MSCs . However , reducing Nox4 levels in aged L-MSCs allowed these cells to form normal 3D structures with young AEC2 cells , but not aged AEC2 cells . These findings highlight the varying effects specific stem cells have , and how their behaviour is affected by pro-aging factors . Moreover , the pro-aging enzyme Nox4 shows potential as a therapeutic target – downregulating its activity may reverse critical effects of aging in cells .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "short", "report" ]	2021	Mesenchymal stromal cell aging impairs the self-organizing capacity of lung alveolar epithelial stem cells
Previous studies had shown that the integration of genome wide expression profiles , in metabolic tissues , with genetic and phenotypic variance , provided valuable insight into the underlying molecular mechanisms . We used RNA-Seq to characterize hypothalamic transcriptome in 99 inbred strains of mice from the Hybrid Mouse Diversity Panel ( HMDP ) , a reference resource population for cardiovascular and metabolic traits . We report numerous novel transcripts supported by proteomic analyses , as well as novel non coding RNAs . High resolution genetic mapping of transcript levels in HMDP , reveals both local and trans expression Quantitative Trait Loci ( eQTLs ) demonstrating 2 trans eQTL 'hotspots' associated with expression of hundreds of genes . We also report thousands of alternative splicing events regulated by genetic variants . Finally , comparison with about 150 metabolic and cardiovascular traits revealed many highly significant associations . Our data provide a rich resource for understanding the many physiologic functions mediated by the hypothalamus and their genetic regulation . The regulation of body weight and appetite are complex processes , in which hypothalamic nuclei play a pivotal role . Genome wide association studies have shown that DNA sequence variants significantly contribute to variation in metabolic traits both in humans and mice . However , in most cases the connection between genetic variant and final phenotype remains unknown ( Suhre et al . , 2011; Teslovich et al . , 2010; Lappalainen et al . , 2013 ) . In an effort to better understand how genetic variation results in phenotypic differences , many projects in the last decade have focused on genome wide characterization of sequence variants regulating gene expression in different tissues and organisms ( Lappalainen et al . , 2013; Majewski and Pastinen , 2011; Grundberg et al . , 2011 , , 2012 ) . These studies showed that up to 80% of genetic variants associated with complex traits likely act through the regulation of gene expression rather than changing protein sequence and function . Such genes , termed expression quantitative trait loci ( eQTLs ) , offer useful insights into the mechanistic links between genotype and phenotype , providing the eQTLs are characterized with sufficient power and resolution in phenotypically relevant tissues and states ( Pickrell et al . , 2010; Min et al . , 2012; Gaulton et al . , 2015 ) . Mouse is the primary model organism for many cardiovascular traits , including atherosclerosis , metabolic syndrome , obesity , the neural control of metabolism , and diabetes ( Lusis , 2000; Billon et al . , 2010; Allayee et al . , 2003 ) . Dozens of loci contribute genetically to metabolic traits have been identified in mouse models ( Parks et al . , 2013; Lusis , 2012 ) . Indeed , we and others have extensively characterized eQTLs in metabolically relevant tissues of mice , suggesting potential new genes related to obesity ( van Nas et al . , 2010; Keane et al . , 2011; Huang et al . , 2009; Aylor et al . , 2011 ) . Integration of transcriptomic data from liver and adipose with genetic mapping and phenotypes led to mechanistic insights into the complexity of metabolic phenotypes . Yet , hypothalamus , which is not a readily accessible tissue , lacks such high resolution expression data . In fact , only one previous study examined eQTLs in mouse hypothalamus , using mice from 39 inbred strains fed chow diet , using microarray data and 5000 SNPs ( McClurg et al . , 2007 ) . The lack of extensive transcriptome data , which allows mapping of eQTLs and the linking of traits and transcripts , is a major impediment in integrating the hypothalamus with a systems biology of metabolism . In this study , we used RNA-Seq to characterize the hypothalamic transcriptome in 99 inbred strains of mice from the Hybrid Mouse Diversity Panel ( HMDP , [Bennett et al . , 2010; Ghazalpour et al . , 2012] ) fed a high fat , high sugar diet ( Parks et al . , 2013 ) . The advantage of the HMDP is that these strains have all been densely genotyped and carefully phenotyped for about 150 metabolic traits , allowing high resolution genetic mapping of QTLs and eQTLs . We identified thousands of novel isoforms and hundreds of new genes that were not previously annotated , and that may represent variants or transcripts specific to the hypothalamus . The HMDP also allowed us to map QTLs with high resolution and power , identifying both local and trans acting variants . The RNA sequencing data permitted examination of a much broader spectrum of transcriptional features and facilitated analysis not only at the gene level , but also of genetic variants affecting specific isoforms , coding sequences or transcription start sites . Similar to other large-scale RNAseq studies ( Mutz et al . , 2013 ) we identified thousands of novel transcripts , with the vast majority of them only expressed at low levels in a small subset of samples ( Table 1 ) . Nevertheless , in the robust set of transcripts that are expressed at appreciable levels ( FPKM >1 in 50+ samples and FPKM>0 in 100+ samples ) , we still identified 21 , 234 novel isoforms and 485 transcripts originating from 407 novel expressed genes ( Supplementary file 1 ) . Interestingly , the number of novel isoforms in our study is comparable to the number of previously annotated transcripts passing the same filtering criteria ( n = 29 , 305 , Table 1 ) . 10 . 7554/eLife . 15614 . 003Table 1 . Transcriptome assembly and filtering . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 003NoneFilter #1Filter #2†NR‡Loci ( genes ) 40472375911407914079Transcripts3834203570665102450347 known ( = ) coding9965820721 known ( = ) non coding8584 novel isoform ( j ) 25957021234 novel locus ( u ) 11753485 other status12439TSS100073944173253720013CDS4624246242186877643Total features57020753531611632792082Filter #1: Expression values of <1e–6 were rounded to zero , and novel transcripts with all zero values were removed both from expression table and from merged file . Also , all transcripts with class code not "=" , "j" or "u" were removed . †Filter #2: Implementation of detection and expression thresholds ( detected in more than 100 samples and expressed ( fpkm>1 ) in more than 50 ) on each feature separately . ‡Filter_NR: Non redundant features count ( those that do not have 1_2_1 correlation to a gene ) . There are several possible interpretations to as why the 407 genes could be missing from the genomic annotation . First , the 407 novel genes expressed in our data potentially constitute transcripts that are unique to the hypothalamus . Second , the GENCODE M2 annotation used in this study was the most recent available when we started to analyze the data . Since then , however , the annotation has been augmented based on more recently published datasets and improved prediction pipelines . Thus , while the 407 genes are novel relative to the M2 version , they may have been added later . To explore that possibility , we compared our assembly to the latest version of annotation released by GENCODE – M10 ( released January 2016 , http://www . gencodegenes . org/ ) and find that 193 out of 407 loci are still novel . Some of the genes may also represent genomic DNA contamination . However , we consider this possibility less likely since we used stringent filtering criteria based on the number of samples that the genes were expressed in . In terms of genomic properties , such as putative open reading frame length , transcript length or splicing complexity , the novel genes seem to resemble known non-coding genes , suggesting that the majority of the novel genes likely belong to this class . On the other hand , the properties of novel isoforms are consistent with known coding transcripts ( Figure 1 ) . To explore the translational potential of newly identified isoforms and genes , we compared our RNA-Seq data to proteomic data generated from additional hypothalamus samples from the HMDP . Specifically , we translated all known and newly identified transcripts in 6 potential open reading frames , and compared this dataset to the hypothalamic peptide sequences . As expected , >95% of the identified peptides ( Supplementary file 1 ) matched at least one known transcript with the vast majority of these ( >99% ) corresponding to known protein coding transcripts , suggesting high accuracy of the peptide data . In addition , 430 peptides matched either novel isoforms ( n = 401 ) , or novel genes ( n = 29 ) exclusively ( Table 2 ) . Since the genomic properties of the novel genes hint that the majority of them are likely non coding , we do not find their low representation in peptide data surprising . Moreover , manual in-depth characterization of the 29 peptides matching novel genes , with UCSC and NCBI database revealed that almost all of them represent processed pseudogenes , rather than proteins with novel functions . This finding both supports the previously published reports of wide spread transcription of pseudogenes , and their translation ( Kim et al . , 2014; Tay et al . , 2015 ) , while strengthening the suggestion that the majority of novel genes identified in this work are not protein coding . 10 . 7554/eLife . 15614 . 004Figure 1 . Genomic properties of novel genes are similar to known non coding genes . Novel genes and isoforms are defined by Cufflinks class code 'u' and 'j' respectively . Distributions of transcript length ( A ) , and maximal hypothetical peptide length ( B ) of novel genes ( yellow ) , new isoforms ( purple ) , known non coding transcripts ( dashed line ) and known coding transcripts ( solid line ) . Transcriptional complexity ( number of transcripts per locus , ( C ) and splicing complexity ( number of exons per transcript , ( D ) of novel genes , novel isoforms , known coding and know non coding transcripts . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 00410 . 7554/eLife . 15614 . 005Table 2 . Summary of peptide support for transcripts . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 005Peptide matchingAllUniquely mappedKnown transcripts99131016 Protein_coding98391002 ncRNAs7414Novel isoforms40194Novel genes2924Total number103431134Peptides are counted towards supporting novel isoforms only if they do not match any known transcript , and towards supporting novel genes if they do not match either known or novel isoform transcripts . All peptides matching known transcripts were also assigned the most likely transcript . Non coding status was assigned only to peptides that do not match any coding transcripts ( for full details please see Materials and methods ) . We further examined whether multi-mapping reads are a substantial contributor to the measured expression of the novel identified genes . For that purpose , we re-quantified 3 samples from C57BL/6J strain , using uniquely mapping reads only ( mapping quality = 255 ) , and compared the quantification of the novel genes between the two approaches . Not surprisingly , the proportion of uniquely mapped reads contributing to the expression of the genes matched very well between the 3 samples , suggesting that it is an intrinsic gene property and not unique to the individual samples . Importantly , while the low uniqueness of mapping reads may indicate false results , we also noted that among the 29 peptides matching to the novel genes exclusively and uniquely , 18 match genes that do not pass our threshold . In fact , many of gene families share sequence motifs and are homologues , and as such some of the reads would be multi-mapping between them . Thus we chose not to remove these genes from the annotation . We examined expression QTLs in terms of type of the affected features and identified SNPs affecting expression of all transcripts at a locus ( eQTL ) , specific transcript isoforms ( isoQTLs ) , transcription start sites ( tssQTLs ) or open reading frame ( cdsQTLs ) ( Hasin-Brumshtein et al . , 2016 ) . Since the linkage disequilibrium ( LD ) is extensive in the mouse genome , we distinguish between three types of QTLs: local ( within 2 Mb of the gene ) , distant ( on the same chromosome as the gene , distance >2 Mb ) and trans ( SNP and gene reside on different chromosomes ) . The number of identified interactions , and genes affected by such regulation depends on the statistical cutoff of p-value for the interaction one chooses . We examined a set of thresholds ranging from very liberal to stringent . The liberal threshold was previously established by permutation from microarray expression data in the HMDP , i . e . p<1 . 4E–3 for local variants and p<6E–6 for distant and trans . The most stringent threshold used a Bonferroni corrected threshold ( i . e 0 . 01 divided by the number of tests , p<1E-12 ) . As expected , different thresholds resulted in different numbers of QTLs , with local and distant QTLs being more robust to threshold stringency than trans QTLs . However , 85–95% of the genes regulated by distant variants , were also regulated by local variants , suggesting that most of the distant QTLs reflect local signals emulated at a distance as a result of wide-ranging LD , rather than independent regulatory elements . In contrast , 70% of the trans interactions involved genes lacking local signal over the entire set of thresholds . Since distant regulation was mostly redundant to local , and it would be very difficult to determine whether that signal is a result of independent regulation versus extended LD , we chose to focus our analysis on trans and local interactions only , defining trans as trans-chromosomal interactions . If we look at regulation types over a wide range of thresholds , the local signal predominates at the more stringent thresholds reflecting the larger typical effect size in this group ( Figure 2A ) . We identified local and trans QTLs for all expressed feature types , with the most common being eQTLs and isoQTLs ( Figure 2B ) . Notably , while we used kinship matrix specific for our strains , still several genes in our dataset show extensive trans regulation ( horizontal lines in Figure 2B ) suggesting a residual influence of population structure on our mapping results . We also note that regulation of gene expression often occurs at the gene level , than at transcript specific levels ( Figure 2C ) . 10 . 7554/eLife . 15614 . 006Figure 2 . Genetic regulation of expression in the mouse hypothalamus . ( A ) Number of genes affected by trans ( blue ) , local ( yellow ) or both ( striped ) variants as a function of statistical threshold . ( B ) Gene level expression quantitative trait loci ( eQTL , top ) but not transcript specific ( isoQTL , bottom ) show trans eQTL hotspots . Density shows the number of interactions at lower statistical thresholds ( 1e–6 ) , red shows interactions passing 1E-12 threshold . Yellow indicates cis acting variants . ( C ) . Genetic regulation occurs on every level , but gene level regulation is more prevalent than transcript specific cases . Supplementary figure shows correlations between allele expression in F1 and local eQTLs identified in HMDP hypothalamus . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 00610 . 7554/eLife . 15614 . 007Figure 2—figure supplement 1 . Allele specific expression in whole brain correlation to local eQTL . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 007 Classical genome-wide eQTL studies used association analysis of total gene expression levels to map local eQTL , assuming that variation linked to proximal genetic variant indicates cis regulation . Recently , several studies have exploited the single base resolution of RNA-Seq to examine this assumption . RNA-Seq permits the identification allele expression ( ASE ) , a hallmark of functional cis acting regulation , in heterozygotes , such as humans or in mouse F1s . Notably , these studies generally show poor concordance between ASE ratios and previously identified local eQTLs , which had been attributed mostly to different technical aspects of RNAseq and ASE analysis ( Hasin-Brumshtein et al . , 2014; Lappalainen et al . , 2013; Grundberg et al . , 2012 ) . To examine the concordance between local eQTLs obtained from ASE and genome-wide association , we performed RNA-Seq on brain tissue from 20 mice representing two F1 crosses between three of the classical inbred strains used to construct the HMDP . The parental strains were C57Bl/6 ( B ) , A/J ( A ) and C3H ( H ) , and the F1 offspring were BxA and HxB . We compared the effect size of ~2600 local eQTLs , to the average ASE effect in both crosses . While local eQTLs were significantly positively correlated with ASE ( p<2E-16 ) correlation estimates were modest ( R2 = 0 . 2 ) , and lower than between the ASEs in the two sets of F1 mice ( R2 = 0 . 4 , Figure 2—figure supplement 1 ) . Using a previously established pipeline for analysis of alternative splicing events , SUPPA ( see Materials and methods ) , we identified 7564 alternative splicing events affecting 3599 genes ( Table 3 ) . An alternative first exon was most common , accounting for the majority of alternative splicing events with several genes exhibiting multiple alternative first exons ( Figure 3A ) . For other types of events , each most often affected one gene , with some of the genes exhibiting a combination of alternative splicing events . The extent of alternative splicing in each sample was quantified as a percent spliced in ( PSI ) of every event ( Figure 3B ) . This quantification can be regarded as a quantitative trait; however , the distribution of PSI for every type of event suggests an excess of 0 or 1 values ( never or always spliced in , respectively Figure 3E ) . This observation is consistent with alternative splicing being often bimodal rather than a normally distributed continuous measure ( e . g . present or absent splice site ) , however , it also shows that quantitative regulation plays a significant role in ratio of isoforms that arise from the inclusion or exclusion of a splicing event . 10 . 7554/eLife . 15614 . 008Figure 3 . Alternative splicing in the mouse hypothalamus . Alternative splicing events were classified ( see Materials and methods ) into 7 types: alternative 3’ splice ( A3 , blue ) , alternative 5’ splice ( A5 , purple ) , alternative first exon ( AF , orange ) , alternative last exon ( AL , brown ) , mutually exclusive exons ( MX , black ) , retained intron ( RI , green ) , and skipped exon ( SE , dark red ) . All events were quantified in each sample for percent spliced in ( PSI ) . ( A ) Number of alternative splicing events of each type ( solid color ) , and number of genes affected by these events ( light color ) . ( B ) Example of partial exon skipping in Colq gene . DBA shows the complete inclusion of the exon ( therefore PSI = 1 ) , while in C57BL/6 there is partial exon skipping ( PSI = 0 . 78 ) . ( C ) Number of alternative splicing events with and without local QTL signal ( solid and light color respectively ) . ( D ) Alternative splice QTLs are mapping to the same chromosome , for all types of events , indicating that most of genetic regulation is by local ( and likely cis acting ) variants . ( F ) Distance between most significant SNP for each event and gene start . The largest effect is typically within 1 Mb of the gene . ( G ) An example of mapping of mutually exclusive exon event in Nnat gene mapping to SNP rs32019082 . ( E ) Distribution of all PSI values of each event type in all samples . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 00810 . 7554/eLife . 15614 . 009Table 3 . Alternative splicing events . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 009All eventsCis QTL eventsAlternative 3' splice site ( A3 ) 316 ( 266 ) 155 ( 134 ) Alternative 5' splice site ( A5 ) 365 ( 304 ) 220 ( 189 ) Alternative first exon ( AF ) 5288 ( 1874 ) 2776 ( 1278 ) Alternative last exon ( AL ) 507 ( 320 ) 283 ( 208 ) Mutually exclusive exons ( MX ) 44 ( 36 ) 29 ( 27 ) Retained Intron ( RI ) 645 ( 476 ) 356 ( 285 ) Skipped exon ( SE ) 399 ( 323 ) 221 ( 189 ) Total7564 ( 3599 ) 4040 ( 2310 ) Table 2: Number in parenthesis indicates number of distinct genes affected by the events . Contrary to gene expression , all forms of alternative splicing were predominantly regulated by local variants ( Figure 3C , D ) . This is not unexpected , since variation in alternative splicing is most likely to arise from particular sequence variants in the RNA itself . Across all examined categories of alternative splicing , 50–60% of the events were significantly associated with local variants , the strongest signal often residing within 1 Mb of the event ( Figure 3F ) . We observe that for many of the genetically regulated splicing events the data show an excess of 0 or 1 PSI values that correlate with the genetic variant ( e . g Figure 3G ) , rather than a shift in a continuous quantitative distribution . This observation suggests that many of the genetically regulated events in alternative splicing sequence play a deterministic role . Trans eQTLs are not uniformly distributed across the genome , clustering into potential hotspots of genome wide regulation . Such hotspots have been observed in several datasets ( Orozco et al . , 2015; Orozco et al . , 2012; Tian et al . , 2015 ) , and they are thought to represent cases where a cis acting variant affects a regulatory gene , e . g . transcription factor , subsequently affecting expression of multiple targets . To identify trans hotspots we divided the genome into bins of 100 kb , and counted the number of distinct genes which exhibit a trans interaction associated with any of the SNPs in the bin ( Figure 4A ) . Importantly , the bins do not necessarily contain the same number of SNPs , and are an order of magnitude shorter than the typical LD blocks in the inbred mouse genome . Just counting the number of SNPs in a bin that are associated in trans would be confounded by SNP density . However , since LD essentially recapitulates the same interactions over and over again , counting the number of genes per bin , rather than the number of SNPs , is not significantly affected by LD . 10 . 7554/eLife . 15614 . 010Figure 4 . eQTL mapping suggests trans eQTL hotspots in the hypothalamus regulate expression of hundreds of genes . ( A ) Mouse genome was broken into 100 kb bins . The plot presents genome wide counts of genes which expression is associated with SNPs in that region , in trans . ( B ) Zoom of trans eQTL locus on chromosome 15 . Peak SNP ( associated with most genes in trans , rs31703733 ) is shown in red , color of other SNPs indicates r2 to rs31703733 . Lower track shows the 10 genes which expression is associated with rs31703733 locally . C , D , E pertain to the 10 genes associated locally with rs31703733 and therefore potentially mediate the trans effects . ( C ) Summary table about each gene . ( D ) Heatmap showing correlation of expression between genes associated with rs31703733 locally , and genes associated with rs31703733 in trans . Color indicates Pearson correlation coefficient . ( E ) Example correlation between potential regulator ( RApgef3 ) and trait ( HDL levels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 010 There are clearly two very strong trans acting hotspots on chromosome 1 and 15 , which are observable for eQTLs , but not for isoQTLs , tssQTLs or cdsQTLs . Each of the two hotspots span 5–6 Mb and include >500 SNPs regulating >400 genes in trans ( Figure 4B , Supplementary file 2 ) . Functional enrichment analysis of gene targets of these hotspots suggests that the hotspot on chr1 regulates multiple genes involved in nucleotide binding , while genes regulated by chr15 locus are clearly associated with ion transport in synapse activity ( Figure 4C , Supplementary file 2 ) . Moreover , consistent with the hypothesis that trans regulation is mediated by local effects , the trans acting SNPs in these hotspots , are also associated with expression of several local genes ( Figure 4B , C , Supplementary file 2 ) . To identify potential mediators of trans regulation , we examined the trans eQTL hotspot on chr15 in more detail . In this region a cluster of SNPs , which share high linkage disequilibrium , is associated with the expression of the majority of the genes mapping to this locus in trans ( Figure 4B ) . We focused on the SNP with most trans associations ( rs31703733 ) , and found that while the hotspot itself contains dozens of genes , rs31703733 constitutes a local eQTL for only 10 of them ( Figure 4B ) . Moreover , the expression levels of genes regulated in trans by the hotspot as well as 6 genes that were local eQTLs were closely correlated ( Figure 4C ) , forming a coexpression module . Further , this module was significantly associated with triglycerides and cholesterol measurements in the mice . Interestingly , the two strongest local eQTL genes for this hotspot , Endou ( an endonuclease ) and Rapgef3 ( also known as EPAC , cAMP activated guanine exchange factor involved in Ras signaling ) were significantly associated with cholesterol and triglyceride measures in this study as well . Genetic association does not necessarily imply that the associated allele is causative for the change in expression . In fact , the majority of the eQTL SNPs fall into regions with poor or no annotation , making it unlikely to be the actual causative variant . However , surprisingly , peak SNPs in the chr15 hotspot ( SNPs rs31703733 and rs31780670 ) are located within a H3K4me1 histone methylation mark associated with enhancers , and which was detected in all neuronal tissues probed by ENCODE in adult mice ( cortex , cerebellum , olfactory bulb ) , but not in other metabolic tissues such as liver , heart , intestine or lung ( UCSC genome browser ) . Based on this preliminary and indirect evidence one may hypothesize that rs31703733 and rs31780670 potentially affect the enhancer activity and the expression of nearby genes . However , to reach any functional conclusion based on our data would require experimental validation in model systems . Long non coding RNA ( lncRNAs ) are a generally poorly characterized class of RNA molecules , with sometimes unclear classification ( St Laurent et al . , 2015 ) . Commonly used criteria for identification of lncRNAs are a transcript >200 bp long without an obvious potential open reading frame . In contrast to the relative paucity of data regarding the functionality of most lncRNAs , several specific lncRNAs ( e . g . Xist , HOTAIR or H19 ) had been studied extensively and shown to play an essential role in cellular function ( Quek et al . , 2015 ) . Recent large RNA-Seq studies and integrative projects such as ENCODE suggest that lncRNAs likely constitute up to 60% of the transcribed RNAs ( Djebali et al . , 2012 ) . Moreover , in recent years an increasing number of functional studies have shown that lncRNAs play an important role in the regulation of transcription and translation , as well as in cell differentiation , signaling and other processes ( Sun et al . , 2013; Guttman et al . , 2011; Ramos et al . , 2015 ) . Further , lncRNAs are enriched for genetic association signals in genome wide association studies ( Iyer et al . , 2015; Kumar et al . , 2013 ) . GENCODE M2 annotation classifies 4540 lncRNA transcripts into at least 5 biotypes , based on their overlap with protein coding genes ( http://www . gencodegenes . org/mouse_releases/2 . html ) . The two most common biotypes - long intergenic non coding RNAs ( lincRNAs ) and antisense RNAs , account for 97% of all the annotated lncRNAs . Since our data were generated without retaining strand specificity , we focused on lincRNAs only . LincRNAs have been shown to be highly tissue specific ( Cabili et al . , 2011 ) , and therefore it is not surprising that although there are 2417 annotated lincRNA transcripts , we found only 381 expressed in the mouse hypothalamus at our filtering criteria . The 381 transcripts represent 237 known and 144 novel isoforms of 181 lincRNA genes , with both novel and known isoforms showing similar expression levels ( Figure 5A , B ) . We also used length and open reading frame criteria to examine the novel loci . Notably , the RNA length criteria of >200 bp is a commonly accepted parameter , while the length of minimal potential open reading frame varies between 30–100 aa in different studies . Based on these criteria we can identify up to 129 potentially novel lincRNAs expressed in the mouse hypothalamus , with 49 of them being spliced . Since lincRNAs have been shown to be often highly tissue specific , we consider the novel lincRNAs to be good candidates for hypothalamus specific transcripts . 10 . 7554/eLife . 15614 . 011Figure 5 . Expression of long non coding RNAs in the hypothalamus is phenotypically relevant . ( A ) Expression of heatmap known non coding RNAs and novel isoforms of these genes n the HMDP . Six lncRNAs ( top cluster , Meg3 , Gm26924 , Snhg4 , Miat , 6330403K07Rik , and Malat1 ) are highly expressed in almost all samples . ( B ) Novel isoforms of lncRNAs are expressed at a similar level of known ones . ( C ) Long non coding RNAs are associated with multiple phenotypes in the HMDP . ( D ) An example of association between a non coding RNA C330006A16Rik and average food intake . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 011 We used gene expression levels to explore the correlation of 310 lincRNAs ( 181 known and 129 novel genes ) with metabolic traits in HMDP . Up to 35% of lincRNA transcripts significantly ( p<1e–3 ) correlated to at least one phenotype in the HMDP , with a few lincRNAs associated with a multitude of related phenotypes ( Figure 5C , D ) . The statistical threshold was determined by permutations in previous studies of gene expression in HMDP , is less stringent than a Bonferroni correction , and reflects interdependencies among phenotypes . Expression of >30% of lincRNAs is associated with a significant local eQTL , suggesting that a considerable number of lincRNAs in the hypothalamus are playing an important role in translation of genetic regulatory variance into physiologic phenotypes . Notably , six lincRNAs – Meg3 , Gm26924 , Snhg4 , Miat , 6330403K07Rik , and Malat1– were highly expressed in most strains ( Figure 5A upper cluster ) . Meg3 is a known imprinted tumor suppressor gene ( Zhang et al . , 2010 ) and was recently implicated in hepatic insulin resistance ( Zhu et al . , 2016 ) . Miat and Malat1 are both part of nuclear bodies . Knockout models of Malat1 are all viable and normal ( Zhang et al . , 2012; Eißmann et al . , 2012; Nakagawa et al . , 2012 ) , however their response to metabolic challenges , such as high fat diet , had not been assessed . Myocardial infarction associated transcript ( Miat ) was initially linked to myocardial infarction through a genetic association study ( Ishii et al . , 2006 ) . Subsequently , Miat was shown to regulate development of neuronal progenitors , involved in schizophrenia pathogenesis and fear response ( Aprea et al . , 2013; Liao et al . , 2016 ) . Although all of the six highly expressed lincRNAs are not novel , none of them were previously reported to be expressed in the hypothalamus , or to play an established role in the metabolic or reproductive system . Previous investigations have documented several examples of lincRNAs that code for short translated open reading frames ( Anderson et al . , 2015 ) . In addition , recent work has shown that lincRNAs can be associated with ribosomes , and their occupancy is similar to that of protein coding transcripts ( Ingolia et al . , 2014; Ruiz-Orera et al . , 2014 ) . We detected 6 peptides which mapped uniquely to 3 different lincRNAs ( Gm26825 , Gm16295 , and Gm26593 ) , suggesting that translation of lincRNAs occurs rarely , and that their association with ribosomes is more likely to be in the context of translational regulation of other protein coding transcripts . RNAseq provides the opportunity to look at RNA sequence modifications in a quantitative manner . In this study , we examined RNA editing patterns of all possible substitution types over the genome . A recent study showed that genetic variation affects C to U editing in the intestine , both in site specific and genome wide manners in the mouse Diversity Outbred cross ( Gu et al . , 2015 ) . We hypothesized that genetic variation may contribute to the extent of RNA editing either in a site specific ( e . g . by altering editing sites in cis ) or genome wide manner ( e . g . by altering the expression or specificity of editing enzymes ) . Altogether we detected 8462 potential editing sites in 3319 genes . As expected , the majority of edits ( >70% ) were A-to-G modifications , and the total number of detected sites in each strain varied between several hundred to over a thousand ( Figure 6A ) . A comparison of our findings to previously reported editing sites in two databases - DARNED and RADAR - suggests that 75% of these are novel . We then used the proportion of edited reads per site , or per substitution type , across strains as a quantitative trait for mapping genetic variants that contribute to editing . We did not detect any significant genetic association for genome-wide levels of editing , which is consistent with both lack of genetic variation in ADAR in our panel and its very stable expression level across samples ( 49 ± 0 . 27 FPKM ) . 10 . 7554/eLife . 15614 . 012Figure 6 . RNA editing is prevalent at the mouse hypothalamus at low levels . ( A ) A total of 8462 editing sites were identified in the HMDP , with A to G accounting for >70% of the modifications . ( B ) Number of sites identified in each strain ( color coding as in A ) . ( C ) Editing level at 90 sites , that were detected in at least 70% of the samples , were mapped . Heatmap shows variation in editing in these sites among the strains . ( D ) An example of an edited site in Ociad1 gene , and its genome wide mapping result . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 012 Most editing sites were detected only in a small number of strains , which precluded meaningful mapping . For the site specific mapping we therefore focused on 90 sites that were detected in >70% of the strains ( Figure 6C ) . We detected editing QTLs for 3 specific sites . For example , one of the editing sites in Ociad1 ( on chr5 at 73312444 ) had a strong association ( p<1e–8 ) with genetic variation residing on the same chromosome ( Figure 6D ) . Altogether our data suggest that RNA editing occurs at low level in thousands of genes , however the impact of genetic variation on the editing level in mouse brain is limited . There are 150 phenotypes available for the HMDP , reflecting many clinically relevant traits as well as various metabolic measures ( Parks et al . , 2013 ) . We used several approaches to examine the potential role of hypothalamic expression in the various phenotypes . We began by identifying the top 500 transcripts that were significantly correlated with each of the phenotypes ( correlation p values ranged from 4e–3 to <2e–16 , Supplementary file 3 ) . We then analyzed each of these expression sets for enrichment of potential pathways and functional annotation , using DAVID ( Huang et al . , 2009a , 2009b ) . Surprisingly , only a few of those expression sets showed moderate to strong enrichment ( Figure 7B ) . For example , fat mass accumulation between 4 and 8 weeks , was distinctly associated with pathways related to oxidative phosphorylation and energy metabolism , while genes associated with food intake were enriched for ribosome related pathways . Clustering of phenotypes based on the proportion of shared genes associated with each trait ( Figure 7A ) , clearly showed that related phenotypes also shared expression dependencies . For example , up to 70% of the genes most correlated with 'esterified cholesterol' , were also correlated with 'total cholesterol' , and more than 30% were shared with 'HDL' . Together these data validate the notion that related phenotypes share underlying molecular mechanisms , yet these shared genetic correlations may not necessarily correspond to specific readily identifiable pathways . 10 . 7554/eLife . 15614 . 013Figure 7 . Groups of genes are associated with multiple related phenotypes in HMDP , although not necessarily enriched for GO ontology or specific pathways . ( A ) Fraction of co-shared genes among the 500 genes most associated with a phenotype . ( B ) Enrichment analysis of the top 500 genes associated with each of the 150 phenotypes results in a small number of significant associations . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 013 We then analyzed whether the novel genes and isoforms we uncovered potentially contribute to phenotypic diversity . We found 232 novel transcripts associated with at least one phenotype ( Supplementary file 3 ) . For example TCONS_00279690 , was strongly associated with several metabolic traits , such as total mass in response to a high fat , high carbohydrate dietary challenge , initial mouse insulin levels , as well as weight and fat mass ( Supplementary file 3 ) . Similar to known isoforms , expression of ~30% of the novel isoforms was significantly correlated with phenotypes . Correlation coefficients and p values were also similar between known and novel isoforms , suggesting that novel isoforms are as likely to contribute to mouse diversity as previously identified transcripts . The hypothalamus is a heterogeneous brain region containing multiple nuclei that affect different aspects of metabolism and endocrine physiology . One of the drawbacks of our approach , dictated by practical considerations , is that we performed RNA-Seq on the entire hypothalamus rather than particular cell populations or nuclei . This approach likely results in dilution of signals from any one population of cells . Nevertheless , the sensitivity of RNA-seq allows examination of expression of particular cellular markers that are specific to certain cell types . To assess which cell populations are well represented in our data , we examined the expression , genetic regulation and association with phenotypes of some of the known markers of hypothalamic cell populations ( Table 4 ) . Our data show high expression of well-known hypothalamic neuronal markers , such as Agouti-related protein ( Agrp ) , pro-opiomelanocortin ( Pomc ) , hypocretin ( orexin , Hcrt ) and steroidogenic factor-1 ( SF1 ) . In addition , we detect high expression of oligodendrocyte markers and microglia , but only modest expression of epithelial markers . 10 . 7554/eLife . 15614 . 014Table 4 . Expression of hypothalamic markers . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 014GeneMarker forMean expressionlocal eQTLAgRPARC neurons119 . 040NDNPYARC neurons18 . 162NDFoxo1ARC neurons4 . 736NDPOMCARC neurons271 . 9542 . 618E-04Hcrt ( orexin ) LHA neurons128 . 8982 . 053E-05Sf1VMHvl neurons65 . 333NDNkx2-1VMHvl neurons6 . 0671 . 100E-05Tac-1VMHvl neurons36 . 3731 . 624E-09BDNFVMHvl neurons4 . 464NDEsr1Multiple2 . 658NDLEPRMultiple1 . 6442 . 041E-04INSRMultiple9 . 496NDCX3CR1Microglia9 . 7901 . 400E-05AIF-1Microglia98 . 598NDCD68Microglia4 . 162NDItgamMicroglia3 . 599NDMyD88Microglia2 . 3691 . 942E-17Aqp4Astrocytes30 . 041NDSlc1a3Astrocytes120 . 203NDAldh1l1Astrocytes16 . 787NDGfapAstrocytes72 . 5022 . 140E-06VEGF-ATanycytes39 . 559NDCX3CL1Neurons99 . 4119 . 836E-09MogOligodendrocytes34 . 5691 . 807E-04MbpOligodendrocytes1266 . 574NDPlp1Oligodendrocytes569 . 620NDCar4Endothelial10 . 694NDEsamEndothelial11 . 6233 . 184E-04Flt1Endothelial9 . 538NDCldn5Endothelial15 . 594ND Interestingly , we identified genetic variation affecting the expression of some key functional molecules for metabolic regulation and response to high fat diet . For example , myeloid differentiation primary response 88 factor ( Myd88 ) , is a Toll-like receptor ( TLR ) adaptor molecule . This protein mediates fatty acid induced inflammation and leads to leptin and insulin resistance in the central nervous system . Mice with CNS specific deletion of Myd88 , are protected from high fat diet induced weight gain , and development of leptin resistance induced by acute central application of palmitate ( Kleinridders et al . , 2009 ) . Our data show that there is a strong eQTL ( p value 1 . 9e–17 , Table 4 ) modulating expression of Myd88 . In addition , we detect genetic variations that affect expression of key molecules such as leptin receptor ( LepR ) and Pomc . Together , these data suggest that the genetic background of inbred mice is an important factor in functional studies , and that the results of molecular perturbations of hypothalamic metabolic pathways can be modulated by genetics . In addition to exploring the links between genetic variation and expression traits , we also looked into the association of transcript levels with phenotypes . We confirmed known relationships – for example the expression of orexigenic neuropeptide Y ( NPY ) was associated with total body mass ( pvalue = 2 . 4e–6 ) . Interestingly , one of the strongest correlations we observed were between fractalkine receptor Cx3cr1 and fat mass response ( p = 9 . 62E–10 ) . Fractalkine ( Cx3cl1 ) is a chemokine , that was recently implicated in diet induced obesity , insulin regulation and promotion of hypothalamic inflammatory response to fatty acids ( Shah et al . , 2015 ) . However , different models of Cx3cr1 knockout mice resulted in variable results on diet induced obesity . The correlation we found is consistent with previous reports that identify a central role for fractalkine receptor Cx3cr1 as a regulator of diet induced obesity and hypothlamic inflammation ( Lee et al . , 2013 , Morari et al . , 2014 ) . Our results also indicate that the expression of Cx3cr1 is affected by genetic background , and suggest that one possible explanation for the heterogeneity in Cx3cr1 knockout results is the different genetic backgrounds used in those studies . In this work we present a comprehensive picture of the transcriptome of the mouse hypothalamus and its genetic variation and regulation . We identify thousands of new isoforms , and >400 new genes , and show independent support for these being translated into protein , which further validates our data . Notably , transcription of pseudogenes had been noticed previously , and likely plays a role in gene regulation . Recent shotgun proteomic studies of the human proteome strongly suggest that a sizable fraction of pseudogenes and lncRNAs is translated ( Ji et al . , 2015; Kim et al . , 2014 ) . The peptide data from our study supports low level translation of processed pseudogenes and is in line with these results . The hypothalamus is a highly heterogeneous tissue with multiple nuclei and cell types acting in concert . A recent RNA-seq study of fed and food-deprived mice showed that cell type specific transcripts in hypothalamic Agrp and Pomc neurons exhibited specific co-expression networks associated with feeding ( Henry et al . , 2015 ) . In contrast , due to the practical constrains of the study , our data were collected on the entire hypothalamus , and as such would be less sensitive to cell type specific signals . Nevertheless , expression of cell-specific markers and functional molecules showed that our approach recapitulates known correlations between genes and metabolic phenotypes , and identifies new ones . In addition , our data capture the various non neuronal cell types , such as microglia or astrocytes , which are often overlooked in the mostly neuron focused studies of the hypothalamus . These cells are important mediators of hypothalamic inflammation and other processes induced by a high fat diet . Regulation of gene expression in this population impacts every aspect of metabolism . While cell type specific transcriptomics is valuable for understanding cellular processes in hypothalamic neurons , our data provide a robust framework recapitulating transcriptional processes affecting multiple cell populations . Our approach is thus complementary to the cell type-specific transcriptomic efforts . In this study , we showed extensive genetic regulation of transcription and alternative splicing in the hypothalamus and identified two loci which influence transcription of hundreds of genes in trans . In addition , our data indicate that a considerable proportion of the new isoforms and transcripts are significantly correlated to physiological phenotypes . While human studies generally lack the power to address trans regulation on a genome-wide scale , the HMDP provides a powerful resource for such analysis . Indeed , we identified two very strong trans acting hotspots that seem to harbor major regulators of gene expression in hypothalamus . We suggest that the trans effects of genetic variation in these regions are likely mediated by local interactions , which is consistent with previously observed cases ( Small et al . , 2011; Heinig et al . , 2010 ) . Enrichment analyses clearly suggest that each trans eQTLs hotspot regulates a set of functionally enriched genes ( e . g . the hotspot on chromosome 15 is strongly associated with ion transport in synaptic areas ) suggesting a new link between genetic variants in these loci and specific cellular function . We further showed that the genes associated with these hotspots correlate to physiological phenotypes , such as HDL and triglyceride levels providing insight into the mechanism behind correlation of these genotypes with complex traits . The connection between the associated genes and traits does not imply direct causality . For example , ion transport is regulated by circadian rhythm ( Ko et al . , 2009 ) , which in turn is associated with many other aspects of metabolism ( Tsuneki et al . , 2016 ) . Notably , although several examples of trans eQTL hotspots were published and analyzed ( Small et al . , 2011 , Heinig et al . , 2010; Tian et al . , 2016 ) , authenticity of such hotspots remains somewhat controversial , and trans eQTL hotspots may arise due to uncontrolled batch effects ( Breitling et al . , 2008; Kang et al . , 2008; Joo et al . , 2014 ) , which are difficult to distinguish from real interactions . To minimize technical batch effects arising from sequencing , we used a round robin approach where each sample was sequenced as part of several pools ( see details in Materials and methods ) . Batch effects may also arise from random environmental exposures , rather than technical sample preparation . In such case , it is unlikely that those effects would be tissue specific or affect only gene level analysis . We found no indication of these trans eQTL hotspots in the adipose or liver data from the same cohort of mice ( Parks et al . , 2013 ) , nor were the hotspots detected for isoQTL . Thus , while we cannot exclude the possibility that trans eQTL hotspots described in this study may had arisen from unaccounted environmental or technical effects , we believe that this is unlikely , but further molecular studies are required to validate our results . RNA-Seq allowed us to quantify transcripts at different levels of analysis- from the total expression of all isoforms in a locus , to transcript specific estimates and their combinations . Our analysis showed that regulation of the total number of transcripts from a gene is far more common than isoform specific regulation . Nevertheless , we were able to identify specific interactions at every level of detail we explored , namely eQTLs , isoQTLs , tssQTLs and cdsQTLs . In addition , we identified and quantified >7000 alternative splicing events affecting >3500 genes in the hypothalamus , and showed that these events are mostly affected by extensive local genetic regulation . In many cases of alternative splicing QTLs , the associated variant resulted in an excess of either 0 or 1 splice PSI values , suggesting that variants affecting splicing act as strong determinants rather than weak contributors to a complex trait . RNA editing is a result of post transcriptional deamination processes whereby an adenosine is converted to inosine ( A-to-I ) , or cytosine to uracil ( C-to-U ) . Both types of RNA editing are mediated by specific enzymes members of the ADAR family facilitate the A-to-I editing which is most commonly observed in neuronal tissues , while Apobec1 mediates the C-to-U editing , and is primarily expressed in the intestine and liver . Editing is a tissue specific process which usually results in modification of only fraction of the transcripts , and therefore can be regarded as a quantitative trait . Previous studies showed that polymorphisms either in the editing enzymes or in the sequence proximal to editing site affect the extent of editing . Specifically , Gu et al recently reported a sequence variant in Apobec1 which affects in trans multiple C-to-U editing sites in the mouse liver ( Gu et al . , 2015 ) . As expected from previous studies , A-to-I was by far the most common RNA editing event in our study . Further , our data show that RNA editing occurs at low levels in thousands of sites , but is highly variable . Less than <1% of the sites were consistently detected in the adult mouse brain across >75% HMDP strains . Consistent with the lack of coding genetic variation in the ADAR enzymes in our mouse panel , and with their invariable expression levels , we did not observe any trans editing QTL that would affect editing levels of multiple sites . However , we found a few local associations that seem to affect editing of specific sites . For example a strong local editing QTL ( p<e–7 ) was observed for one of the editing sites in the Ociad1 gene ( also known as Asrij ) . This editing QTL is likely due to a sequence variant in or near the Ociad1 gene . Notably , Ociad1 is expressed in stem cells , where it regulates pleuropotency via the JAK/STAT pathway ( Sinha et al . , 2013 ) . Editing of Ociad1 or its expression in neuronal tissue was not reported before . Another significant aspect of our data relates to long non-coding RNAs . LincRNAs have been shown to play an important role in various cellar functions . Recent genome annotations include thousands of known lincRNAs , yet most of them remain functionally uncharacterized , and only a few studies have examined the genetic control of their expression in detail . Moreover , lincRNAs are often expressed in a tissue specific manner , and therefore are not readily identifiable from general expression datasets . In this study , we detected expression of 381 known and novel lincRNA isoforms , and also identified 129 novel , potentially hypothalamus- specific lincRNAs . Our data indicate that lincRNAs are subject to similar variation in expression , and exhibit similar overall genetic control as the coding genes . Specific lincRNAs have been implicated in a variety of phenotypes ( Guttman et al . , 2011; Huarte et al . , 2010; Kumar et al . , 2013 ) , and our data indicate strong correlations between some of the hypothalamic lincRNAs and metabolic phenotypes , such as body weight . The hypothalamus is a very heterogeneous tissue , and one of the major drawbacks of our analysis is that we used whole hypothalamus , rather than dissecting specific nuclei . This limitation stems from practical considerations – meaningful expression QTL analysis requires sacrifice of hundreds of mice , while the dissection of specific hypothalamic nuclei is delicate and time consuming and thus was not feasible within the constraints of this study . Still , this shortcoming is likely to limit our power to detect meaningful associations , rather than introducing spurious ones . Moreover , since our study mostly focuses on genetic regulation of transcription , which was shown to be largely shared among tissues and cell types , we believe that the data presented in this work are not substantially confounded by heterogeneity . To summarize , our data fill a substantial gap and will be useful to the research community . The hypothalamus is composed of multiple nuclei , which are distinct in morphology and function , and many laboratories focus on disentangling the complex interactions that ultimately affect metabolism and behavior . However , genome wide transcriptome analysis of this tissue has not been published , and genetic regulation of transcription as well as tissue specific transcripts remains largely obscure . We believe that our study is complimentary to physiological studies and will facilitate research into the crosstalk between the brain and other metabolic tissues . All of the expression data described in this paper are publicly available from NCBI archives GEO ( GSE79551 ) and SRA ( project number PRJNA314533 ) . Altogether we sequenced 285 samples , from 99 strains of the HMDP . A total of 87 strains had 3 samples per strain , 11 strains had 2 samples per strain , and 2 strains had 1 sample per strain ( a detailed list is in Supplementary file 1 ) . RNA was extracted using Qiazol followed by miRNAeasy kit from Qiagen ( RRID:SCR_008539 ) . Unstranded mRNA libraries were prepared by the UCLA Neuroscience Genomics Core with Illumina standard kits ( TruSeq v3 ) according to standard protocols . All samples were barcoded , and sequenced with ~18 samples per lane , with HiSeq2000 using 50 bp paired-end sequencing protocol . A round robin design was implemented such that biological replicates were sequenced on different lanes , and each sample was part of more than one sequencing pool . Samples were demultiplexed by sequencing facility , forward and reverse read fastq files were supplied for each sample . Read QC was done using FastQC ( RRID:SCR_005539 ) in batch mode . The samples had excellent quality , with all bases exceeding median quality score of 28 , and >98% of the sequences with a mean quality score >28 . The average number of reads per sample was 26 . 3 M . All reads were passed to mapping as is , without trimming or filtering . Samples were mapped to the mm10 genome using STAR aligner version 2 . 3 . 1 ( https://github . com/alexdobin/STAR/releases/tag/STAR_2 . 3 . 1z9 ) . The mm10 sequence was downloaded from http://cufflinks . cbcb . umd . edu/igenomes . html ( UCSC annotation files ) . Reference sequences were built using known splice junctions ( –sjdbGTFfile option ) from known genes annotation file . Mapping was performed allowing up to 3 mismatches per read ( --outFilterMismatchNmax 3 ) , removing non canonical un-annotated junctions ( --outFilterIntronMotifs RemoveNoncanonicalUnannotated ) and allowing up to 10 multiple mappings per read ( --outFilterMultimapNmax 10 ) . Alignment files from all samples of the same strain were merged into one alignment file per strain using the 'merge' function of the samtools package . On average 94 . 1% of the reads mapped , and of the mapped reads 96 . 4% mapped uniquely . STAR also detected on average 2 . 8 M splices per sample , with 98% of them previously annotated . Detailed mapping counts for each sample are in Supplementary file 1 . Our pipeline is summarized in Figure 8 . For the purpose of transcript assembly , sample specific alignment files were pooled into unified BAM alignment files by strain which were then sorted and indexed using samtools . Transcript assembly was done on each strain specific alignment file with Cufflinks v2 . 2 . 0 . , with mouse genome version mm10 , and gtf file of known transcripts as a reference guide ( -g option , reference file downloaded from UCSC genome browser ) , together with bias and multimap option corrections ( -b and –u respectively ) . This step resulted in 99 strain specific transcript assemblies . We then used Cuffmerge to compare those assemblies to GENCODE M2 annotation ( http://www . gencodegenes . org/ ) , and to consolidate them into one unified assembly file representing all transcripts from GENCODE M2 and from the strain specific assemblies ( merged . gtf ) . To obtain FPKM expression values , we run Cuffquant and Cuffnorm v2 . 2 . 1 with default parameters , using the sample specific alignment files and unified assembly ( merged . gtf ) file . 10 . 7554/eLife . 15614 . 015Figure 8 . RNA-Seq analysis framework . General workflow used for analysis of RNA-Seq data in this study . Initial demultiplexed samples ( fastq files ) were aligned to the mouse genome with STAR , merged in one file per strain , and transcripts assembeled with cufflings . The resulting assembly files ( one from each strain ) were merged with GENECODE M2 annotation into unified assembly . The abundance of each transcript in the unified assembly was estimated in sample specific alignment files . DOI: http://dx . doi . org/10 . 7554/eLife . 15614 . 015 Cufflinks assembly and abundance estimation results in assessment of gene expression at multiple levels of genomic resolution - transcripts , transcription start sites , coding sequence and loci ( genes ) . We implemented 3 filtering steps: First , we removed transcripts that are likely to be an assembly or sequencing artifacts , based on following criteria: Comparison to GENECODE M2 indicated that this is not a known transcript ( class code '=' ) , new isoform of a known gene ( class code 'j' ) or novel intergenic transcript ( class code 'u' ) . b . Novel isoforms and intergenic transcripts which expression was <1e-6 FPKM in all samples . Loci and transcription start sites associated exclusively with transcripts removed in steps 1 and 2 . This filtering step removed 7% of the transcripts , with the other 93% ( 535 , 316 features ) treated as potentially true . II . While we consider that all these features may potentially represent true splicing and transcription events , only features that show sufficient variation and expressed at appreciable level are useful in eQTL analysis . Further , many of the potentially new features were expressed either at very low levels or in a small number of samples . Therefore , at the second filtering step we implemented detection and expression thresholds ( FPKM>0 in >=100 samples , and FPKM>1 in >=50 samples ) to focus on ~100 , 000 expressed features that are more likely to produce meaningful mapping results in the HMDP panel . Third , expression values ( FPKM ) of transcription start sites , coding sequences and genes are a sum of expression values of transcripts associated with these features . Consequently , if a locus , transcription start site or coding sequence is associated with only one transcript , the expression information of that feature is identical and thus redundant to the respective transcript . Therefore , such features were removed from eQTL mapping . Altogether , this filtering reduced the number of explored features to ~ 93 , 000 . All features that passed the described above filtering criteria , were analyzed for eQTLs using expression estimates in 282 individual mice representing 100 strains , 193 , 400 SNPs and the fastLMM algorithm using an appropriate kinship matrix that accounts for the HMDP population structure . Cis acting eQTLs were identified by allele expression ( ASE ) as described previously ( Hasin-Brumshtein et al . , 2014 ) . Briefly , allele-specific counts at each exonic SNP , filtered for quality control criteria , were summed in a genespecific manner . The resulting counts were normalized and analyzed for differential expression between the alleles using the DEseq package . To identify RNA editing sites , we re-aligned the reads to the mouse genome ( mm10 ) and transcriptome using the single nucleotide variant-sensitive aligner RASER ( Ahn and Xiao , 2015 ) with the parameters –m 0 . 06 and –b 0 . 03 . We then detected likely mismatch positions considering the read sequence quality score at that position and its position within in a read . We further required potential editing sites to be covered by >=5 total and >=3 edited reads and excluded positions that overlap SNPs or are located in homopolymers , simple repeats , and within 4nts of splice junctions ( Lee et al . , 2013 ) . Proteomic data were collected on 110 samples representing the 99 strains used in this study , plus 10 C57Bl/6 samples . All samples were independent biological replicates of the strains used in this study fed on the high fat , high sugar diet , and not the same hypothalami used for RNA extraction .	Metabolism is a term that describes all the chemical reactions that are involved in keeping a living organism alive . Diseases related to metabolism – such as obesity , heart disease and diabetes – are a major health problem in the Western world . The causes of these diseases are complex and include both environmental factors , such as diet and exercise , and genetics . Indeed , many genetic variants that contribute to obesity have been uncovered in both humans and mice . However , it is only dimly understood how these genetic variants affect the underlying networks of interacting genes that cause metabolic disorders . Measuring gene activity or expression , and tracing how genetic instructions are carried from DNA into RNA and proteins , can reliably identify groups of genes that correlate with metabolic traits in specific organs . This strategy was successfully used in previous studies to reveal new information about abnormalities linked to obesity in specific tissues such as the liver and fat tissues . It was also shown that this approach might suggest new molecules that could be targeted to treat metabolic disorders . A brain region called the hypothalamus is key to the control of metabolism , including feeding behavior and obesity . Hasin-Brumshtein et al . set out to explore gene expression in the hypothalamus of 99 different strains of mice , in the hope that the data will help identify new connections between gene expression and metabolism . This approach showed that thousands of new and known genes are expressed in the mouse hypothalamus , some of which coded for proteins , and some of which did not . Hasin-Brumshtein et al . uncovered two genetic variants that controlled the expression of hundreds of other genes . Further analysis then revealed thousands of genetic variants that regulated the expression of , and type of RNA ( so-called "spliceforms" ) produced from neighboring genes . Also , the expression of many individual genes showed significant similarities with about 150 metabolic measurements that had been evaluated previously in the mice . This new dataset is a unique resource that can be coupled with different approaches to test existing ideas and develop new ones about the role of particular genes or genetic mechanisms in obesity . Future studies will likely focus on new genes that show strong associations with attributes that are relevant to metabolic disorders , such as insulin levels , weight and fat mass .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience", "tools", "and", "resources", "genetics", "and", "genomics" ]	2016	Hypothalamic transcriptomes of 99 mouse strains reveal trans eQTL hotspots, splicing QTLs and novel non-coding genes
The ‘pitchers’ of carnivorous pitcher plants are exquisite examples of convergent evolution . An open question is whether the living communities housed in pitchers also converge in structure or function . Using samples from more than 330 field-collected pitchers of eight species of Southeast Asian Nepenthes and six species of North American Sarracenia , we demonstrate that the pitcher microcosms , or miniature ecosystems with complex communities , are strikingly similar . Compared to communities from surrounding habitats , pitcher communities house fewer species . While communities associated with the two genera contain different microbial organisms and arthropods , the species are predominantly from the same phylogenetic clades . Microbiomes from both genera are enriched in degradation pathways and have high abundances of key degradation enzymes . Moreover , in a manipulative field experiment , Nepenthes pitchers placed in a North American bog assembled Sarracenia-like communities . An understanding of the convergent interactions in pitcher microcosms facilitates identification of selective pressures shaping the communities . Similar selective pressures in geographically distant habitats can cause unrelated organisms to converge in both morphological and functional traits . Pitchers of carnivorous plants have evolved repeatedly and independently to have the same shapes and insect-trapping functions in Southeast Asia , North America , and Australia ( Albert et al . , 1992 ) . Similar selective pressures can also cause the independent emergence of multispecies interactions with parallel physiological or ecological functions , defined as ‘convergent interactions’ ( Bittleston et al . , 2016b ) . The concept of convergent interactions was developed in detail in Bittleston et al . ( 2016b ) , and can be used as a tool to better understand forces influencing interspecific relationships . Here , we investigate whether convergent interactions can be identified between different , independently evolved pitcher plant genera and the arthropods and microbes housed within their pitchers . We hypothesize that the microbial communities formed within the fluids of distantly-related pitcher plant species possess similar community structures and functions , and we test this hypothesis by comparing the bacterial and eukaryotic communities living within the plant-held waters ( phytotelmata ) of pitcher plants from two genera in different plant orders , Nepenthes ( family Nepenthaceae , order Caryophyllales ) native to Southeast Asia and Sarracenia ( family Sarraceniaceae , order Ericales ) native to North America . Microbial communities are complex , and even apparently simple habitats house orders of magnitude more microbial species than plant or animal species ( Horner-Devine et al . , 2004 ) . Parsing the principles shaping microbial community structure and function remains a challenge . But virtually all plants and animals interact with microbes ( van der Heijden et al . , 2008; McFall-Ngai et al . , 2013 ) , and host morphology and ecology are likely to shape associated microbial communities . Convergent plant or animal forms , chemistry , or habitats may control the communities within them ( Bittleston et al . , 2016b ) , analogous to a kind of ‘extended phenotype . ’ Convergently evolved organisms are ideal systems for understanding how different aspects of form or ecology influence microbial communities because they enable distinctions among the effects of evolutionary history , geography , and host morphology or physiology . A nascent understanding of how microbial communities assemble in association with highly specialized , convergently evolved hosts is emerging . Interactions between animals and their gut microbiomes are frequently mediated by diet . For example , animals with fermenting foreguts as distantly related as the hoatzin bird and cow are colonized by microbial communities with similar structures ( Godoy-Vitorino et al . , 2012 ) , and the gut bacteria of disparate animals whose diets consist of ants also possess a convergent community composition ( Delsuc et al . , 2014 ) . In turn , the herbivorous , arboreal ants of different taxonomic groups are associated with specific bacteria likely to supplement their low-nitrogen diets ( Sanders et al . , 2017; Hu et al . , 2018 ) . And convergent shifts in bacterial communities are observed in different cichlid fishes as diets change from herbivory to carnivory ( Baldo et al . , 2017 ) . Similar convergent dynamics are observed in sea urchin larvae ( Carrier and Reitzel , 2018 ) . Animals' microbial associates are likely to impact fitness . The pitchers of carnivorous pitcher plants are the plant analog of an animal gut , and may function in digestion of animal prey . Convergently evolved genera of pitcher plants provide a tractable model and opportunity to explore the assembly and functional potentials of host-associated microbial communities . In fact , pitcher microcosms have long served as elegant models for investigating metacommunities and community assembly ( Kneitel and Miller , 2003; Buckley et al . , 2003; Srivastava et al . , 2004; Armitage , 2017; Bittleston , 2018 ) ; each pitcher pool is a discrete but similar habitat with a unique history . In our study , we take advantage of the pitcher system to focus primarily on comparisons of the communities of convergently evolved plants , testing how unrelated hosts with similar morphologies and functions shape their associated communities . Implicit within our comparisons are myriad community assembly mechanisms and processes , including dispersal and environmental filtering caused by characteristics of the host . Two genera of carnivorous pitcher plants , Sarracenia and Nepenthes , evolved independently in North America and Southeast Asia , respectively ( Albert et al . , 1992 ) . Pitchers are highly modified leaves , and both genera grow pitchers to attract , trap , and digest insects—primarily ants . Pitcher microcosms also house communities of bacteria , fungi , protozoa , and arthropods ( Beaver , 1983; Bradshaw and Creelman , 1984; Kitching , 2000 ) . Pitchers of both genera have characteristic shapes , although forms can vary greatly in color and size among different species . Pitchers of both genera also offer extra-floral nectar to attract prey , possess slippery interiors to trap prey , and secrete digestive enzymes to break down prey tissues ( Juniper et al . , 1989; Adlassnig et al . , 2011; Kurup et al . , 2013 ) . Nepenthes species produce more different kinds , and a greater abundance , of enzymes compared to Sarracenia species , and Sarracenia species may rely more on their bacterial communities for prey degradation ( Butler et al . , 2008; Moran and Clarke , 2010; Baiser et al . , 2011 ) . Pitchers actively absorb nitrogen , phosphorus , and other elements from prey; these nutrients are otherwise scarce in the soils where the plants grow ( Chapin and Pastor , 1995; Schulze et al . , 1997; Ellison , 2006 ) . While the plants are perennial , individual pitchers can last from a few weeks to two years , depending on the species , and are generally most active for the first few weeks to months after opening ( Heard , 1998; Osunkoya et al . , 2008 ) . Pitcher interiors appear to be sterile before opening ( Peterson et al . , 2008; Buch et al . , 2013 ) ( but see [Chou et al . , 2014] ) , and once open , a complex community forms within ( Beaver , 1983; Kitching , 2000; Koopman et al . , 2010; Krieger and Kourtev , 2012; Chan et al . , 2016 ) . Many pitcher-associated organisms are specialists , and are restricted to the pitcher habitat for at least part of their lives ( Fish and Hall , 1978; Beaver , 1983 ) . Various arthropods have co-diversified with their pitcher plant host , suggesting ecological dependence and a shared evolutionary history ( Satler and Carstens , 2016 ) . In fact , even though the species S . purpurea was introduced to Europe over 100 years ago , it houses very few insect inquilines as compared with pitchers in native habitats ( Zander et al . , 2016 ) ; the close associations of pitchers and arthropods may be slow to evolve . To test for convergence between the microcosms of North American Sarracenia and Southeast Asian Nepenthes , we collected fluids from over 330 pitchers of six species of Sarracenia and eight species of Nepenthes from native habitats in the United States , Singapore , and Borneo ( Supplementary file 1 Table S1 ) , and used next generation sequencing to characterize the biodiversity housed in each pitcher . To the best of our knowledge , ours are the first comparisons of the entire communities , encompassing bacteria , microbial eukaryotes , and arthropods , associated with convergently evolved organisms; our sampling is also more intensive than any sampling previously published for pitcher plants . We tested for convergence between communities by comparing species richness , community composition , phylogenetic structure , and functional potential . We hypothesized the living communities housed in unrelated pitchers would converge , both structurally and functionally: tests of the hypothesis would result in similar species richness , phylogenetic structure , and functional potential between the Sarracenia and Nepenthes , as compared to the same parameters measured from surrounding bog water and soil communities . In addition to describing the bacterial and eukaryotic communities of each of the 14 different species of pitcher plants , we explored which features of host species appear to drive patterns of biodiversity . Finally , in a field manipulation , we experimentally tested whether North American insects and microbes would colonize Southeast Asian Nepenthes pitchers when Nepenthes plants were placed in a North American Sarracenia habitat; the experiment tests whether the pitchers of different genera function as similar selective environments when exposed to the same microbial pool . In the aggregate , data provide evidence for the convergence of pitcher microcosms between independently evolved host genera , and identify aspects of pitcher form and physiology underpinning the similarities . To compare the microbial communities within Sarracenia and Nepenthes pitchers , we analyzed DNA samples from pitchers and their surrounding environments using an amplicon sequencing approach , separately characterizing bacteria and eukaryotes ( Figure 1A , Supplementary file 1 Table S2 , Supplementary file 2 Dataset S1 ) . Communities from Southeast Asian and North American pitchers were defined as converging if the communities were more similar to each other than to the communities of the environments immediately surrounding the plants , even despite the vast geographic distance between them . In fact , the Nepenthes and Sarracenia pitcher communities were distinct from and had fewer Operational Taxonomic Units ( OTUs , clustered at 97% sequence similarity; a proxy for species ) than communities in surrounding bog water or soil ( Figure 1B and C , and Supplementary file 1 Table S3 ) . The pattern held for both bacteria and eukaryotes , and was unaffected by sample volume ( no correlation of observed OTUs with sample volume; for bacteria: R = −0 . 003 , p=0 . 984 , and for eukaryotes: R = −0 . 003 , p=0 . 812 ) . Pitcher samples also had significantly lower Shannon diversities ( Mann-Whitney U Test , p<0 . 001 in all comparisons ) than surrounding environments ( Figure 1B ) , and this pattern also held when we controlled for extraction volume ( by analyzing a subset of 155 samples , each extracted from the same volume; Supplementary file 1 Table S3 ) . Overall , pitcher communities were characterized by both decreased richness and evenness as compared to communities from their immediate environments . The composition of pitcher communities was also significantly different from the community composition of surrounding bog water or soil ( Figure 1C , Supplementary file 1 Table S4 . Bacteria: envfit: R2 = 0 . 31 , p<0 . 001 , adonis: R2 = 0 . 08 , p<0 . 001; Eukaryota: envfit: R2 = 0 . 38 , p<0 . 001 , adonis: R2 = 0 . 08 , p<0 . 001 ) . To understand differences in community composition across just one region , we separately tested and analyzed Southeast Asian samples from pitchers , bog water , soil , plastic tubes , or cupped , dead leaves filled with water and sitting on the ground . The communities in water from leaves or from plastic tubes were more similar to pitcher fluid communities than bog water or soil communities ( Figure 1—figure supplement 1 ) . To compare the phylogenetic structures among pitcher communities , we mapped OTUs present in at least 10% of our field ( not experimental ) Nepenthes or Sarracenia samples onto bacterial and eukaryotic phylogenetic trees , together with all OTUs found in bog water and soil ( Figure 2 and associated Figure 2—figure supplement 1 ) . Organisms repeatedly colonizing Nepenthes or Sarracenia pitchers in North America and Southeast Asia tended to be from similar clades of bacteria or eukaryotes ( Figure 2 ) . The pattern was most pronounced in bacteria , and shared families included Microbacteriaceae , Gordoniaceae , Chitinophagaceae , Sphingobacteriaceae , Bradyrhizobiaceae , Rhizobiaceae , Sphingomonadaceae , Burkholderiaceae , Comamonadaceae , Oxalobacteriaceae , Neisseriaceae , Enterobacteriaceae , Moraxellaceae , and Xanthomonadaceae . Across eukaryotes shared clades included dipteran insects , mites , and rotifers . To investigate drivers of community composition among pitchers of each genus , we had recorded species identity and measured the pH and total volume of pitcher fluid associated with each sample . As we extracted DNA from each sample , we recorded DNA concentrations; a proxy for the living biomass within a pitcher ( Marstorp et al . , 2000 ) . In both the Sarracenia and Nepenthes systems , pitcher communities differed significantly among host species ( Figure 3 , Supplementary file 1 Table S4 ) . For bacteria , the effect of host species was similar in Sarracenia ( envfit R2 = 0 . 40 , p<0 . 001; adonis R2 = 0 . 14 , p<0 . 001 ) and Nepenthes ( envfit R2 = 0 . 38 , p<0 . 001; adonis R2 = 0 . 18 , p<0 . 001 ) species; however , for eukaryotes , pitcher host species explained more of the observed variation in Sarracenia species ( envfit R2 = 0 . 42 , p<0 . 001; adonis R2 = 0 . 20 , p<0 . 001 ) as compared to Nepenthes species ( envfit R2 = 0 . 22 , p<0 . 001; adonis R2 = 0 . 15 , p<0 . 001 ) . Pitchers of different species maintain different levels of acidity , although these differences are more pronounced in the genus Nepenthes than in the genus Sarracenia . Certain Nepenthes species can actively raise or lower the acidity of individual pitchers by pumping protons into or out of pitcher fluid ( An et al . , 2001; Moran et al . , 2010 ) . In our sampling of natural populations , we measured values below pH 4 in N . rafflesiana , N . gracilis , and N . stenophylla . But low-pH pitcher fluid does not seem to correlate with the Nepenthes phylogeny: low-pH species are in different clades , separated by species with higher average pH levels ( Meimberg and Heubl , 2006 ) . Furthermore , each species with low pH pitchers also had pitchers with higher pH levels . The large pH gradient across the Nepenthes fluids in our samples was strongly correlated with bacterial community composition , and explained most of the observed variation ( Figure 3A . ordisurf R2 = 0 . 74 , p<0 . 001; mantel r = 0 . 63 , p<0 . 001 ) . This result supports a recent study that also noted a correlation between pitcher fluid pH and Nepenthes bacteria ( Kanokratana et al . , 2016 ) . But the strong effect of pH on bacterial community composition is not driven by Nepenthes species differences per se; significant , high correlations between pH and bacterial community composition are also found within each of the three species with very low pH values when the data of each species are analyzed alone ( Mantel tests: N . gracilis r = 0 . 46 , p<0 . 001; N . rafflesiana r = 0 . 68 , p<0 . 001; N . stenophylla r = 0 . 85 , p<0 . 001 ) . Eukaryotic community composition in Nepenthes was more weakly correlated with pH , and pH explained a smaller portion of the variation ( ordisurf R2 = 0 . 20 , p<0 . 001; mantel r = 0 . 14 , p<0 . 001 ) . In the genus Sarracenia , bacterial ( but not eukaryotic ) community composition correlated with pH ( bacteria: ordisurf R2 = 0 . 15 , p<0 . 001 , mantel r = 0 . 11 , p=0 . 010 ) . For both the Nepenthes and Sarracenia bacterial communities , the relationship of pH with Shannon alpha diversity appeared to be quadratic: Shannon diversity peaked around pH 5 . 5 and was lower at both lower and higher pH levels ( Figure 3—figure supplement 1 ) . The correlation was much stronger for Nepenthes samples ( R2 = 0 . 67 , p<0 . 001 ) , but still significant for Sarracenia samples ( R2 = 0 . 07 , p=0 . 002 ) . Shape emerges as a potential strong influence among the Sarracenia species , but is confounded with species identity: S . purpurea and S . rosea pitchers grow with a shorter , more cylindrical shape , while pitchers of S . alata , S . flava , S . leucophylla and S . rubra grow to a taller , more tapered shape . Although our samples of S . purpurea and S . rosea were collected in Massachusetts and Florida , respectively , the two species are very closely related ( Ellison et al . , 2012 ) . The taller , tapered Sarracenia pitchers have an aspect ratio of width to height below 0 . 2; while the shorter , more cylindrical Sarracenia have an aspect ratio above 0 . 2 , as do the Nepenthes pitchers from this study . Because growth form is confounded with Sarracenia host species identity and phylogeny , we could not analyze it as a separate variable . But pitchers from species of Sarracenia with shorter , wider pitchers tended to have a larger volume of fluid than the taller pitchers , and volume was strongly correlated with Sarracenia bacterial community composition ( ordisurf R2 = 0 . 31 , p<0 . 001 , mantel r = 0 . 15 , p=0 . 006 ) and eukaryotic community composition ( ordisurf R2 = 0 . 18 , p<0 . 001 , mantel r = 0 . 17 , p<0 . 001 ) . Collection site also significantly influenced Sarracenia communities; however , the effect was weaker when we controlled for the fact that not all species grow at all sites ( Supplementary file 1 Table S4 ) . DNA concentration was significantly correlated with Sarracenia bacterial community composition ( ordisurf R2 = 0 . 22 , p<0 . 001 , mantel r = 0 . 11 , p=0 . 035 ) . There was also a weak , but marginally significant correlation between DNA concentration and the Shannon diversity of Sarracenia bacterial communities , which was driven by a few Sarracenia samples with very cloudy fluid and high relative abundances of Enterobacteriaceae OTUs ( Figure 3—figure supplement 2 ) . In a manipulative experiment , we relocated Nepenthes pitcher plants ( propagated in Southeast Asia and purchased through a commercial U . S . nursery ) to a Sarracenia bog in North America to test whether relocated Nepenthes pitchers would acquire communities similar in community structure and phylogenetic composition to those of local Sarracenia . All Nepenthes placed into the Sarracenia habitat were maintained in pots with soil material purchased in the U . S . , and the plants were removed after experiments concluded . We included potted S . purpurea as a control to explore whether growth in a pot influenced community assembly . Target pitchers approaching maturity were manually opened in the bog on Day 1 of each experiment . The experiment also included cylindrical , round-bottomed 50 mL sterile glass tubes , either with or without sterilized insect material ( ‘prey’ ) added as a nutrient control ( Figure 4A ) . During the experiment , we also recorded whether pitchers contained larvae of the pitcher plant mosquito , Wyeomyia smithii , a specialized insect that completes its lifecycle only within Sarracenia purpurea pitchers ( Figure 4B ) . W . smithii larvae regularly colonized their native S . purpurea pitchers ( whether they were growing in the ground or in a pot ) . Surprisingly , they also colonized pitchers of Nepenthes bicalcarata and N . ampullaria , albeit in lower proportions ( Figure 4B ) . The mosquitoes never colonized the more acidic N . gracilis and N . rafflesiana species , nor the experimental glass-tube pitchers . To compare the biodiversity of entire communities , we first re-plotted our beta-diversity results from natural Nepenthes versus natural Sarracenia pitchers ( Figure 4C ) , and found that community composition was significantly different for the two genera . Bacterial assemblages were more similar between the two genera than eukaryotic assemblages , and correspondingly , host genus explained less variation in bacterial than in eukaryotic community composition ( Bacteria: envfit R2 = 0 . 33 , p<0 . 001 , adonis R2 = 0 . 09 , p<0 . 001; Eukaryota: envfit R2 = 0 . 55 , p<0 . 001 , adonis R2 = 0 . 14 , p<0 . 001 ) . We next compared the beta-diversity results of wild Nepenthes to our experimental data of relocated Nepenthes , native Sarracenia , and experimental glass tubes ( Figure 4D ) . When Nepenthes microcosms assembled in a Sarracenia habitat , the assemblages of both bacteria and eukaryotes converged on compositions similar to those of the local Sarracenia and not wild Nepenthes ( Figure 4D ) . The exception to convergence was in Nepenthes pitchers with pH below 4 . The bacterial assemblages in highly acidic pitcher fluids ( generally N . gracilis and N . rafflesiana ) separated from other Nepenthes and Sarracenia pitchers in the same manner as acidic pitcher bacterial assemblages shifted in natural Nepenthes populations , and were instead more similar in structure and phylogenetic composition to wild , acidic Nepenthes ( Figure 4D and Figure 3 ) . Acidity explained most of the variation in bacterial community composition from the experiments ( pH: ordisurf R2 = 0 . 67 , p<0 . 001 , mantel r = 0 . 50 , p<0 . 001 ) and was also a significant predictor of eukaryotic community composition ( pH: ordisurf R2 = 0 . 21 , p<0 . 001 , mantel r = 0 . 11 , p<0 . 001 ) . Region—whether the pitchers were in Harvard Forest , Singapore or Malaysia—explained only a small portion of bacterial variation ( envfit R2 = 0 . 19 , p<0 . 001 , adonis R2 = 0 . 07 , p<0 . 001 ) , but a larger portion of the eukaryotic variation ( envfit R2 = 0 . 41 , p<0 . 001 , adonis R2 = 0 . 10 , p<0 . 001 ) . Communities of bacteria and eukaryotes in our experimental pitchers were different from bog water communities ( not shown ) , but partially clustered with the organisms colonizing the glass tube pitchers ( Figure 4D ) . NMDS plots indicate glass tubes with added prey did not assemble communities more similar to the experimental pitcher communities than glass tubes without added prey . Only a very small portion of the variation in community composition was explained in analyses of pitchers vs . glass tubes; however , the differences were highly significant ( bacteria: envfit R2 = 0 . 05 , p<0 . 001 , adonis R2 = 0 . 02 , p<0 . 001; eukaryotes: envfit R2 = 0 . 09 , p<0 . 001 , adonis R2 = 0 . 03 , p<0 . 001 ) . The analysis suggests a sterile , pitcher-shaped form is almost , but not quite entirely , sufficient for acquiring a pitcher plant-like microcosm ( Figure 4D ) . To investigate the functional potential of pitcher microbiomes , we generated metagenomes from 24 field-collected ( not experimentally relocated ) pitcher samples ( 16 Nepenthes and 8 Sarracenia ) . When compared with other published metagenomes for soil , lake , and phyllosphere samples ( Supplementary file 1 Table S5 ) , pitcher plant community metagenomes were more enriched in gene pathways for fatty acid degradation , fermentation , and the biosynthesis of cell wall materials and non-proteinogenic amino acids , while non-pitcher metagenomes were more enriched in gene pathways for metabolic precursors ( the biosynthesis of proteinogenic amino acids , tRNA charging , glycolysis , Calvin cycle and folate transformations ) ( Figure 5A ) . In terms of overall functional potential as measured by Kegg Orthology ( KO ) groups using the Bray-Curtis dissimilarity metric in an NMDS plot , pitcher metagenomes were highly variable and were most similar to other phyllosphere communities ( Figure 5B ) . Among Nepenthes and Sarracenia metagenomes , KO gene families in Sarracenia purpurea clustered close to Nepenthes ampullaria , N . gracilis and N . reintwardiana . Other Sarracenia ( those with a tapered shape ) and other Nepenthes ( those with more acidic fluid ) appeared to be more dissimilar in terms of functional potential . To probe the functional similarities of Nepenthes and Sarracenia communities more deeply , we chose to compare abundances of enzymes involved in the degradation of complex polysaccharides and proteins . We specifically chose to focus on chitinases ( K01183 , GH families 18 and 19 ) because chitin is the main component of insect exoskeletons and can be used as both a carbon and nitrogen source . Because pitchers evolved to trap insect prey , we hypothesized pitcher plant microbiomes would have the genes to digest chitin . We also chose to focus on key enzymes involved in proteins and amino acid degradation ( aminopeptidase N , lysine decarboxylase , ornithine decarboxylase , and glutamate dehydrogenase ) , because nitrogen mineralization via the microbiome may assist pitcher plants in nitrogen acquisition from prey . Finally , we chose to compare cellulases across the metagenomes , as we hypothesized microbiomes of pitcher plants would be less involved in breaking down plant material , compared to microbiomes of other habitats . Although individual pitcher samples showed considerable variability , Nepenthes and Sarracenia microbiome metagenomes did in fact have high relative abundances of chitinase , lysine decaboxylase , and ornithine decarboxylase genes compared to metagenomes of bacterial communities collected from other habitats ( Figure 5C ) . But aminopeptidase N levels were not significantly higher in pitcher plants than in other habitats , and surprisingly , glutamate dehydrogenases were , in fact , significantly lower in pitcher plants . The glutamate dehydrogenase pathway can be either catabolic or anabolic , and this dual activity might explain our result . Cellulases were significantly lower in pitcher plant microbiomes , as hypothesized . Our results are consistent with recent proteomics data documenting high levels of amino acid and carbohydrate metabolism pathways in Sarracenia pupurea ( Northrop et al . , 2017 ) . As the specific functions of the microbial communities within pitchers become better understood , we anticipate our data will be used to more explicitly measure and compare the functions of these different microbiomes . Evidence for the convergence of communities within the carnivorous pitchers of Nepenthes and Sarracenia is strong and pitcher characteristics appear to regulate fundamental aspects of community biodiversity . First , in nature , the bacterial and eukaryotic communities inside both Nepenthes and Sarracenia pitchers are less species rich and less even than communities in surrounding soil or bog water; pitcher habitats favor a subset of available species ( Figures 1 and 2 ) . Second , although the communities within the two genera of pitchers are made up of different species , organisms tend to be closely related and from the same phylogenetic clades ( Figure 2 ) . This pattern was especially pronounced among bacteria . Third , both Nepenthes and Sarracenia pitcher microbiomes are depleted in pathways for metabolic precursors and enriched in pathways involved in fatty acid degradation and cell wall biosynthesis; pitcher microbiomes also possess a high relative abundance of both chitinases and genes involved in amino acid degradation ( Figure 5 ) . Results suggest the pitcher microbiomes of both genera function as decomposers of insect prey . Finally , Nepenthes pitchers experimentally placed into a Sarracenia habitat assembled Sarracenia-like communities ( Figure 4 ) . Convergently evolved pitchers appear to cause convergent interactions ( Bittleston et al . , 2016b ) between the two genera and their associated pitcher microcosms . Among species within a genus , fluid acidity was the strongest driver of beta-diversity , specifically for Nepenthes bacterial communities; this same characteristic explains aspects of our manipulative experiment . When microcosms of Nepenthes and Sarracenia pitchers assembled in parallel in a common environment , pitchers with similar fluid acidity held communities more alike in composition ( Figure 4 ) . Furthermore , while the communities formed in sterile glass tubes with or without sterilized insect ‘prey’ appeared somewhat similar to plant pitcher communities , communities were still statistically different—suggesting that a general tube-like form that drowns insects in rainwater is almost , but not completely , sufficient for generating a pitcher-like microcosm . Other unmeasured characteristics of real Nepenthes and Sarracenia pitchers , including for example the production or abundance of plant-produced digestive enzymes or nectar , oxygen levels , and temperature , are also likely to cause these plant-formed pitcher microcosms to be more similar to each other than they are to glass tube microcosms . The pitcher environment appears to be the dominant selective force shaping community composition in pitchers . Fewer eukaryotes consistently colonize Nepenthes or Sarracenia pitchers , as compared to bacteria ( Figure 2 ) ; the pattern is consistent with stronger habitat filtering with increasing body size , as observed in bromeliad phytotelmata ( Farjalla et al . , 2012 ) . Within each pitcher plant genus , collection site explained less of the observed variation than characteristics of the pitcher itself , suggesting that environmental filtering has a larger influence than dispersal . At a broad scale , comparisons between Southeast Asia and North America reveal that the regional pools of microbial organisms are distinct , and dispersal likely plays a stronger role in differentiating the microbial communities of these two continents . However , despite the differences in regional pools , organisms from the same phylogenetic groups colonize the pitcher microcosms found on opposite sides of the globe . Convergent interactions , albeit on a much smaller scale , mirror the biome concept ( Odum , 1971 ) : the same functional groups of plants and animals are found in different regions of the world when those regions possess similar climate and soil conditions . For example , clumped grasses , crustose lichens , and jumping rodents are found in xeric shrublands globally , including in Australia and Arizona , just as Chitinophagaceae bacteria , Sphingomonadaceae bacteria , and Histiostomatidae mites are found in the unrelated pitcher microcosms of Southeast Asia and North America . However , beyond scale , there is a second fundamental difference between the two concepts: convergent interactions are , by definition , interactions among different living organisms , and thus there is a potential for reciprocal feedback , and potentially coevolution . The concept of convergent interactions can be used to better understand the selective pressures structuring microbial biodiversity elsewhere , outside of the pitcher plant system . In addition to diet ( Godoy-Vitorino et al . , 2012; Delsuc et al . , 2014; Baldo et al . , 2017; Sanders et al . , 2017; Hu et al . , 2018 ) , many other features of convergently evolved organisms appear to cause associations with similar microbial communities . For example , bacterial communities associated with fungus-growing ants , beetles , and termites in different regions of the world have dominant community members from the same genera , with convergent functional potential ( Aylward et al . , 2014 ) , and microbial symbionts of sponges are functionally equivalent across phylogenetically divergent hosts ( Fan et al . , 2012 ) . Convergent interactions can provide predictions about community structure and function , which can be tested across systems . The framework provides a tool to explore compositional and functional similarities of whole ecosystems , potentially enabling an understanding of the fundamental evolutionary and environmental drivers structuring microbial communities . Nepenthes pitchers from three co-occurring species ( N . ampullaria , N . gracilis and N . rafflesiana ) were sampled from three sites in Singapore ( Kent Ridge Park , Bukit Timah Nature Preserve , and between Lower and Upper Peirce Reservoir Park ) in January 2012 . Additional pitchers from the same species and sites were sampled in March 2013 and March 2014 . Pitchers from an additional five co-occurring species ( N . veitchii , N . tentaculata , N . stenophylla , N . reinwardtiana , and N . hirsuta ) were sampled from the Maliau Basin , Borneo in March 2014 . Sarracenia pitchers from five species ( S . alata , S . flava , S . leucophylla , S . rosea and S . rubra ) were sampled from thirteen sites from Mississippi to Florida along the U . S . Gulf Coast in June 2014 and a sixth species ( S . purpurea ) was sampled from Harvard Forest in Massachusetts in July 2014 . For details of which species were sampled from which sites see Supplementary file 1 Table S1 . Sites were considered different if separated by more than 0 . 1 degree of latitude or longitude . Contents of each pitcher were collected with sterile , single-use plastic transfer pipettes and placed into empty , sterile plastic tubes . Fluids were mixed within each pitcher using the pipette before collecting to homogenize any differences by depth . Volumes and pH levels of all pitcher fluids were recorded , except for our first collection in Singapore in 2012 ( for more detail on Singapore sampling see [Bittleston et al . , 2016a] ) . The pitchers of some species can have large volumes ( e . g . 100–500 mLs ) ; for higher volume samples , we estimated total volume and collected a well-mixed subsample from the pitcher . We measured pH with colorpHast strips ( EMD Millipore ) by removing small amounts of fluid with additional sterile pipettes . To preserve DNA , we added cetyltrimethylammonium bromide and salt solution ( hereafter ‘CTAB’; final concentrations: 2% CTAB , 1 . 4 M NaCl , 20 mM EDTA , 100 mM Tris pH 8 ) to each sample in the same volume as the collected fluid . All samples were processed the same day as collection , except for Maliau Basin samples that were refrigerated overnight and processed the next morning , due to time constraints . After CTAB addition , samples were transported at room temperature to Harvard University , and subsequently frozen . Protocols reflected pitcher plant habitats ( Figure 1A ) . When wet , we collected bog samples from the surrounding environment , and when dry , we collected soil samples and water either from fallen leaves or from sterile tubes placed in the environment , as follows: Singapore , March 2013—soil , Gulf Coast , June 2014—soil and bog water; Massachusetts , July 2014—bog water; Singapore , February 2014—sampling from plastic tubes left out for one month to collect rainwater and acquire microbial communities; Maliau Basin , January 2014—sampling from soil and water held in fallen leaves . All soil samples were collected from the surface organic layer in approximately 7 mL volumes . See Dataset S1 for sample details . In summer 2013 we set up experiments to manipulate Nepenthes within a Sarracenia habitat , the Tom Swamp bog at Harvard Forest in Petersham , MA ( USA ) . Four different species of Nepenthes ( N . ampullaria , N . bicalcarata , N . gracilis and N . rafflesiana ) were purchased from Borneo Exotics via the ExoticPlantsPlus nursery in New York . Nepenthes plants were maintained in a greenhouse for two months after arriving from Southeast Asia , and then a growth chamber for a few weeks while pitchers were maturing for use in the field . Sarracenia purpurea plants were purchased from Meadowview Biological Research Station in Virginia , potted using purchased sphagnum peat and perlite , maintained in a greenhouse for three months , and used in the experiments as a control for whether growth in a pot influenced community assembly . Experiment I had six treatments: S . purpurea growing naturally in the bog , S . purpurea in pots , N . ampullaria , N . gracilis and N . rafflesiana in pots , and empty , 50 mL sterile glass tubes used as a rough mimic of the pitcher shape . There were eight stations , and the experiment ran from June 26 – July 17 . Pitchers were sampled for subsequent sequencing on days 14 and 21 . Experiment II had five treatments: S . purpurea in the bog , S . purpurea in pots , N . ampullaria in pots , sterile glass tubes , and sterile glass tubes each filled with 30 mg of autoclaved , ground wasps as a nutrient and prey control . There were five stations , and the experiment ran from July 17 – September 4 . Pitchers were sampled for subsequent sequencing on days 14 , 35 , and 49 . Experiment III had six treatments: S . purpurea in the bog , N . ampullaria , N . rafflesiana and N . bicalcarata in pots , sterile glass tubes , and sterile glass tubes with prey . There were five stations , and the experiment ran from July 24 – September 10 . Pitchers were sampled for subsequent sequencing on days 15 , 35 , and 48 . To sample , we collected 750 uL of fluid from experimental pitchers and tubes using sterile transfer pipettes , as described above . At sampling , we also noted the presence or absence of pitcher plant mosquito larvae ( Wyeomyia smithii ) in pitchers . On the last day of each experiment , we collected entire pitcher contents . On the last days of Experiments II and III , we collected samples of bog water . All samples were stored in small tubes in a cooler with ice , and brought back to the laboratory , where they were frozen the same day . Subsequent analyses target only the last day’s sample from each pitcher or tube , so that no pitcher or tube is included more than once in the dataset . Once we had collected samples , we turned our attention to DNA extraction and sequencing . When removing fluid for DNA extraction , we took care to avoid macroscopic organisms . We concentrated sample fluid by filtration or isopropanol precipitation and centrifugation . Concentration protocol did not affect community composition ( samples processed by different techniques clustered together ) . We extracted DNA by bead-beating concentrated materials with buffer and phenol-chloroform , and then proceeding with a standard phenol-chloroform extraction ( Sambrook and Russell , 2001 ) . We included a negative control with each set of extractions , and discarded the samples from one extraction set found to have measurable amounts of DNA in the negative control . We measured DNA quantity , and then re-extracted DNA from a few samples with very low DNA amounts , using a larger initial volume . DNA extracts with dark coloration ( suggesting high levels of polyphenols ) were cleaned using a MoBio Powerclean kit . DNA concentrations of successful extractions were fluorometrically quantified a final time using a Quant-iT High-Sensitivity dsDNA Assay Kit ( Invitrogen ) . Samples were sent to Argonne National Laboratories for Illumina MiSeq next-generation amplicon sequencing . The Earth Microbiome Project’s barcoded 16S and 18S primers ( Amaral-Zettler et al . , 2009; Caporaso et al . , 2012 ) were used to amplify DNA in separate runs , with PCR amplification and sequencing executed according to the Earth Microbiome Project protocols ( http://www . earthmicrobiome . org/emp-standard-protocols ) . The 16S primers target the V4 region of the ribosomal RNA gene and are used to characterize prokaryotic communities , while the 18S primers target the V9 region and are used to characterize eukaryotes . The amplicon sequencing datasets can be accessed from the Sequence Read Archive as NCBI BioProject PRJNA448553 . To explore functional gene diversity in pitcher plant microbiomes , we conducted shotgun metagenomic sequencing with 24 of our pitcher samples . Sixteen samples were sequenced at the High Impact Research Institute in Malaysia ( HIR ) at the University of Malaya , and eight at the Bauer Core Facility ( RRID:SCR:001031 ) at Harvard University . The HIR set targeted two samples from each of four different species of both Nepenthes and Sarracenia . DNA was extracted as described above , but 2–4 extractions were done for each sample and resulting DNA was pooled to increase the amount available for sequencing . DNA was sheared with a Covaris at settings aiming for average lengths of 350 base pairs ( bp ) , and libraries were prepared using a TruSeq DNA PCR Free HT Kit . DNA library concentrations were measured using a KAPA Library Quantification Kit , and the qualities were tested with a Bioanalyzer . DNA libraries were then pooled in equal concentrations and sequenced on the Illumina HiSeq 2500 platform in four Rapid Run , paired-end , 100 bp lanes . The Bauer Core Facility set included eight additional Nepenthes samples ( two samples from each of four different species ) . Here , we used the same DNA extractions as previously used for metabarcoding . DNA samples were sheared with a Covaris at 500 bp , and prepared with a KAPA LTP Library Prep Kit . Due to lower initial DNA quantities , the samples were subject to 2–9 cycles of PCR before final quantification using a PerfeCta NGS Library Quantification Kit , quality testing with a Bioanalyzer , and pooling of samples in equal concentrations . These libraries were sequenced in one-third of an Illumina HiSeq 150 bp paired-end Rapid Run lane . Shotgun metagenomic data can be accessed via the Argonne National Laboratory metagenomics server MG-RAST ( RRID:SCR:004814 ) : http://www . mg-rast . org/linkin . cgi ? project=mgp15454 . To generate Operational Taxonomic Units ( OTUs ) , amplicon data were clustered using QIIME ( Quantitative Insights Into Microbial Ecology , RRID:SCR:001905 ) versions 1 . 8 and 1 . 9 ( Caporaso et al . , 2010 ) on Harvard University’s Odyssey computing cluster . We joined forward and reverse reads using fastq-join , then split libraries with a PHRED quality cut-off of 20 to remove low-quality sequences , and used UCLUST ( version 1 . 2 . 21q ) open-reference clustering to form groups of sequences into OTUs with 97% similarity . Resulting numbers of sequences and OTUs are summarized in Supplementary file 1 Table S2 . Phylogenetic trees were generated using QIIME default settings for 16S; when generating the18S alignment and tree we set the allowed gap fraction to 0 . 8 and the entropy threshold to 0 . 0005 . We assigned taxonomy with the greengenes version 13_8 ( Greengenes Database Consortium ) and SILVA version 111 databases for 16S and 18S , respectively . For 18S , we used the BLAST method to assign taxonomy , as UCLUST assignment was poor . For subsequent analyses of 18S sequences , we used only OTUs assigned to Eukaryota . To calculate alpha diversity of our samples , we first discarded any samples with fewer than 6500 or 4300 sequences for the 16S and 18S datasets , respectively . We then built rarefaction plots of soil , bog , Nepenthes , and Sarracenia samples using the observed species metric in QIIME and standard deviations across each category ( Figure 1B ) . We also calculated the Shannon diversity index and standard deviations across the same categories ( Figure 1B , Supplementary file 1 Table S3 ) . To ensure that sample volume was not driving differences in alpha diversity between environmental and pitcher plant samples , we used a linear model to test for a correlation between sample volume and observed OTUs . We also re-calculated Shannon diversity and standard deviations using a subset of 155 samples where DNA was extracted from the same volume for all samples ( Supplementary file 1 Table S3 ) . To explore beta diversity among samples , we first removed any observation of an OTU with less than 10 sequences per sample in order to minimize the probability of including sequencing errors , including barcode misassignments ( Bokulich et al . , 2013; Nelson et al . , 2014 ) . We accounted for uneven sequencing across the samples by subsampling our OTU tables to 4000 sequences per sample . We calculated dissimilarity matrices with the unweighted Unifrac metric ( Lozupone and Knight , 2005 ) and Bray-Curtis using R packages picante and phyloseq ( Kembel et al . , 2010; McMurdie and Holmes , 2013; RRID:SCR:013080 ) , and ran non-metric multidimensional scaling ( NMDS ) analyses using the vegan R package ( Oksanen et al . , 2013; RRID:SCR:011950 ) ( Figures 1C , 3 , 4 and 5 and Figure S1 ) . We used the functions envfit and ordisurf to fit environmental factors or vectors ( respectively ) to our ordinations and to analyze main effects . We also calculated dissimilarity matrices using the Bray-Curtis metric , to take into account how relative abundances among OTUs might influence beta-diversity analyses . For both the unweighted Unifrac and Bray-Curtis measures , we used mantel tests to test for correlations of the dissimilarity matrices with pH and volume , and permutational multivariate analyses of variance ( PERMANOVAs , function adonis in vegan ) ( Anderson , 2001 ) to test the explanatory power of factors including plant species and collection site ( Supplementary file 1 Table S4 ) . We adjusted P-values within each group to account for multiple comparisons using the Benjamini-Hochberg procedure ( Benjamini and Hochberg , 1995 ) . To examine phylogenetic patterns among Nepenthes pitchers , Sarracenia pitchers , and environmental samples , we chose to focus on relatively common OTUs , removing OTUs containing fewer than 100 sequences across all our samples as well as those not present in at least 10% of either Nepenthes or Sarracenia microbiome samples . We then subsampled the OTU table for each category to 2000 sequences per sample , combined all observations of the OTUs by category ( e . g . Nepenthes , Sarracenia or environment ) , and normalized by the number of samples in each category . We filtered our previously generated 16S and 18S phylogenetic trees using the resulting OTU tables and plotted them with the Interactive Tree of Life ( iToL ) program ( Letunic and Bork , 2011 ) ( Figure 2 ) . The bacterial tree was rooted with Archaea , and the eukaryotic tree was rooted in Streptophyta ( land plants and most green algae ) . We added barcharts along the outer edge of the trees , displaying the natural log of the abundance for each OTU in each category , and gray dots to each branch with bootstrap support of 0 . 7 or higher ( Figure 2 ) . The same trees , with branch lengths and tree scales included , are shown in Figure 2—figure supplement 1 . The branches of the figure supplement trees were colored by either phylum ( for bacteria ) or by broad taxon levels ( for eukaryotes ) . For the shotgun metagenomic data , we combined forward and reverse reads from all lanes for each sample , and used Trimmomatic to remove barcodes and low-quality sequences . We used HUMAnN2 ( HMP Unified Metabolic Analysis Network 2 , [Abubucker et al . , 2012] ) on Harvard University’s Odyssey computing cluster to identify individual reads by comparing and annotating reads to reads of known function , build profiles of identified functional genes for each sample , and normalize numbers of sequences across samples . We next compared our metagenomes to publicly available metagenomes from soil , lake , and phyllosphere habitats using MG-RAST ( Glass et al . , 2010; RRID:SCR:004814 ) and the NCBI’s Sequence Read Archive ( SRA; RRID:SCR:004891; accession numbers are listed in Supplementary file 1 Table S5 ) . We analyzed these metagenomes in the same way as our own data , also using HUMAnN2 . To detect differentially abundant gene pathways in pitcher plants vs . the other metagenomes , we subset gene pathways to those with abundances and variances in the top 50% of the dataset , and used LEfSe ( Linear Discriminant Analysis Effect Size , Segata et al . , 2011 ) to identify statistically significant features . We reported the pathways with the five largest linear discriminant values for each group ( Figure 5 ) . We made NMDS plots of KO functional matrices with Bray-Curtis distances and a beanplot of chitinases using the vegan and beanplot ( Kampstra , 2008 ) packages in R . We tested for differences in gene family abundances between pitcher plant and comparison metagenomes using Mann-Whitney U tests ( function wilcox . test in R ) and we adjusted P-values to control for false discoveries using the Benjamini-Hochberg procedure ( Benjamini and Hochberg , 1995 ) ( Figure 5 ) .	The ecosystems found across the Earth , including in forests , lakes and prairies , consist of communities of plants , animals and microbes . How these organisms interact with each other determines which ones grow and thrive . We still do not understand how communities form: why different species exist where they do , and what enables them to survive in different locations . This knowledge is particularly limited with regard to communities of microbes because they are hard to see and count . Pitcher plants are an ideal system for studying how communities and ecosystems assemble . The pitcher-shaped leaves of these plants each contain small aquatic communities of microbes and arthropods ( including insects and mites ) that can be relatively easily studied . Because unrelated groups of plants have evolved pitchers at different times and on different continents , these communities can also be used to explore how evolutionary history and the current environment determine which species thrive in a particular location . Bittleston et al . sampled the DNA of the communities living within 330 pitchers from various North American and Southeast Asian pitcher plant species . This revealed that very distantly related plants on opposite sides of the planet have pitchers that host similar communities , with the organisms found in one pitcher plant often closely related to the organisms found in others . The genes within the community’s DNA also shared many functions , with the majority of shared genes devoted to digesting captured insect prey . Bittleston et al . also relocated pitcher plants from Southeast Asia to grow alongside North American species and found the same microbes and arthropods colonizing both groups , indicating that the different types of pitchers present a similar habitat . Overall , the results of the experiments performed by Bittleston et al . suggest that certain kinds of interactions between species ( such as between the pitcher plants and their microbes ) can evolve independently in different parts of the world . Researchers can use these interactions to learn more about how communities and ecosystems form . With a greater understanding of the Earth’s ecosystems , it will be easier to protect them and predict how they will fare as global conditions change .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "evolutionary", "biology" ]	2018	Convergence between the microcosms of Southeast Asian and North American pitcher plants
The cerebral cortex contains multiple areas with distinctive cytoarchitectonic patterns , but the cellular mechanisms underlying the emergence of this diversity remain unclear . Here , we have investigated the neuronal output of individual progenitor cells in the developing mouse neocortex using a combination of methods that together circumvent the biases and limitations of individual approaches . Our experimental results indicate that progenitor cells generate pyramidal cell lineages with a wide range of sizes and laminar configurations . Mathematical modeling indicates that these outcomes are compatible with a stochastic model of cortical neurogenesis in which progenitor cells undergo a series of probabilistic decisions that lead to the specification of very heterogeneous progenies . Our findings support a mechanism for cortical neurogenesis whose flexibility would make it capable to generate the diverse cytoarchitectures that characterize distinct neocortical areas . The mammalian cerebral cortex contains a wide diversity of neuronal types heterogeneously distributed across layers and regions . The most abundant class of neurons in the cerebral cortex are excitatory projection neurons , also known as pyramidal cells ( PCs ) . In the neocortex , PCs can be further classified into several subclasses with unique laminar distributions , projection patterns and electrophysiological properties ( Greig et al . , 2013; Jabaudon , 2017; Lodato and Arlotta , 2015 ) , and currently available data suggest that several dozen distinct transcriptional signatures can be distinguished among them ( Tasic et al . , 2018 ) . The relative abundance of the different types of PCs largely determines the distinct cytoarchitectonical patterns observed across different regions of the mammalian neocortex ( Brodmann and Gary , 2006 ) . The diversity of excitatory neurons emerges from progenitor cells in the ventricular zone ( VZ ) of the developing neocortex known as radial glial cells ( RGCs ) ( Malatesta et al . , 2000; Miyata et al . , 2001; Noctor et al . , 2001 ) . RGCs divide symmetrically to expand the progenitor pool during early stages of corticogenesis . Subsequently , they undergo asymmetric cell divisions to generate clones of PCs directly or indirectly via intermediate progenitor cells ( IPCs ) ( Lui et al . , 2011; Taverna et al . , 2014 ) . The characteristic vertical organization of migrating PCs in the developing neocortex led to the ‘radial unit hypothesis’ , which postulates that PCs in a given radial column are clonally related ( Rakic , 1988 ) . However , the precise mechanisms through which RGCs generate diverse cytoarchitectonic patterns throughout the neocortex remain to be elucidated . The most commonly accepted view of cortical neurogenesis is based on the notion that RGCs are multipotent and generate all types of excitatory neurons following an exquisite inside-out temporal sequence ( Leone et al . , 2008; Molyneaux et al . , 2007; Rakic et al . , 1994 ) . Consistently , progenitor cells cultured in vitro reproduce the temporal sequence of cortical neurogenesis ( Gaspard et al . , 2008; Shen et al . , 2006 ) , and genetic fate mapping experiments have shown that cortical progenitors identified by the expression of the transcription factors Fezf2 and Sox9 are multipotent in vivo ( Guo et al . , 2013; Kaplan et al . , 2017 ) . In contrast to this view , other studies have suggested the existence of fate-restricted cortical progenitors , which would only generate PCs for certain layers of the neocortex ( Franco et al . , 2012; García-Moreno and Molnár , 2015 ) . However , the interpretation of these results remains a matter of controversy ( Eckler et al . , 2015; Gil-Sanz et al . , 2015 ) . Our current framework for understanding cortical neurogenesis largely relies on studies that consider RGCs as a homogeneous population . Consistent with this view , recent clonal analyses of the developing neocortex led to the conclusion that progenitor cell behavior conforms to a deterministic program through which individual RGCs consistently generate the same neuronal output ( Gao et al . , 2014 ) . This would suggest that variations in the organization of cortical areas would exclusively rely on mechanisms of lineage refinement at postmitotic stages , such as programmed cell death . Alternatively , the absence of detailed quantitated data of individual PC lineages or methodological caveats may have prevented the identification of a certain degree of heterogeneity in the neuronal output of individual RGCs . In this study , we have used three complementary approaches to circumvent some of the intrinsic technical biases associated with each of the previously used methods to systematically investigate the clonal organization of PC lineages in the cerebral cortex . Our results provide a detailed quantitative assessment of the neurogenic fate of individual VZ progenitor cells that reveal a large diversity of PC lineage configurations . These findings support a stochastic model of cortical neurogenesis through which a limited number of progenitor cell identities could generate the diverse of cytoarchitectonical patterns observed in the neocortex . To study the cellular mechanisms underlying the generation of PCs in the neocortex , we analyzed the output and organization of neuronal lineages generated by individual progenitor cells . To this end , we first used replication-deficient retroviral vectors that integrate indiscriminately in mitotic cells but only identify cell lineages with fluorescent proteins following Cre-dependent recombination ( Ciceri et al . , 2013 ) . To specifically label PC lineages , we injected a very low titer cocktail of conditional reporter retroviruses ( rv::dio-Gfp and rv::dio-mCherry ) into the lateral ventricle of Neurod6Cre/+ mouse embryos ( also known as Nex-Cre ) , in which Cre expression is confined to postmitotic PCs ( Goebbels et al . , 2006 ) ( Figure 1a ) . Using this approach , we achieved sparse labeling and avoided biasing the tagging of progenitor cells by the expression of specific genetic markers ( Cepko et al . , 2000 ) . To identify the developmental stage at which progenitor cells become neurogenic in the cortex , we injected retroviruses at different embryonic days ( E9 . 5 to E14 . 5 ) and analyzed the organization of individual PC clusters at postnatal day ( P ) 21 ( Figure 1a ) . Since a single copy of the viral vector is stably integrated into the host genome , retroviral infection leads to the labeling of only one of the two daughter cells resulting from the division of the infected progenitor cell . Consequently , infection of progenitor cells in the ventricular zone ( VZ ) of the pallium labels PC lineages in three main configurations depending of the mode of division of the infected progenitor ( Figure 1b ) : ( 1 ) a large cluster containing more than one lineage , which results from the infection of a self-renewing progenitor cell dividing symmetrically; ( 2 ) a single lineage , which results from the infection of a progenitor cell undergoing its last symmetric division; and ( 3 ) a partial lineage , which results from a neurogenic division of a progenitor cell . In this later case , partial lineages may contain the majority of neurons in the clone , if integration occurs in the progenitor cell , or one or two neurons , if the integration occurs in a neuron or an IPC , respectively . We observed clusters of neurons with the characteristic morphology of PCs at all stages examined . Systematic mapping at P21 revealed very sparse labeling and widespread distribution of clones throughout the entire neocortex ( Figure 1c–c” and Figure 1—figure supplement 1 ) . The spatial segregation of the lineages was confirmed by the virtual absence of green and red clones within 500 µm of each other in all experiments analyzed ( Figure 1—figure supplement 1 ) . We quantified the number of PCs per clone at P21 following viral infection at different embryonic stages and observed that lineages contain progressively smaller progenies ( Figure 1d ) . This is consistent with the notion that VZ progenitors undergo proliferative symmetric cell divisions early during corticogenesis before they become neurogenic and begin self-renewing via asymmetric divisions ( Götz and Huttner , 2005; Kriegstein and Götz , 2003 ) . Since neurogenic divisions label one or two neurons in 50% of the cases ( Figure 1b ) , the fraction of one- and two-cell clones found after retroviral infection is indicative of the proportion of neurogenic VZ progenitor cells at each embryonic stage . We observed that these clones represent ~50% of the lineages at E12 . 5 ( Figure 1e ) . Consistent with previous reports using other methods ( Gao et al . , 2014 ) , these results indicated that the onset of cortical neurogenesis begins immediately before E12 . 5 , and that at this stage most VZ progenitor cells are already neurogenic . Thus , we focused subsequent analyses on this stage . We first examined lineages labeled at E12 . 5 that contained more than two cells , which correspond to the progeny of a VZ progenitor cell ( Figure 2a ) . Consistent with classical models of cortical neurogenesis , we found that most VZ progenitor cells ( 63% ) infected with retroviruses at E12 . 5 produce translaminar lineages containing neurons in both deep ( V and VI ) and superficial ( II-III and IV ) layers of the neocortex ( Figure 2b , e ) . However , we also observed a substantial fraction of lineages in which PCs were confined to either deep ( Figure 2c , e ) or superficial ( Figure 2d , e ) layers ( 15% and 22% , respectively ) . The distribution of single-cell and two-cell clones following infection of VZ progenitor cells at E12 . 5 further support the existence of cortical lineages restricted to superficial layers of the neocortex . As expected from the normal progression of neurogenesis in translaminar lineages , most single-cell and two-cell clones ( which result from the labeling of a neuron or an IPC , respectively ) were located in deep layers of the cortex ( Figure 2f–m ) . However , in these experiments , we also identified a small fraction of single-cell and two-cell clones in superficial layers of the neocortex ( Figure 2f–m ) . This suggested that some VZ progenitor cells generate PCs for superficial layers of the neocortex in their earliest neurogenic divisions . We noticed that the clonal size of laminar-restricted lineages is typically smaller than that of translaminar clones ( Figure 2e ) . One explanation for this difference could be that laminar-restricted lineages represent sub-clones resulting from the labeling of IPCs that undergo more than one round of cell division , generating four to five neurons with a laminar-restricted distribution . To test this hypothesis , we carried out lineage-tracing experiments at single-cell resolution using low-dose tamoxifen administration in Tbr2CreERT2/+;RCL-Gfp pregnant mice at E12 . 5 ( Figure 2—figure supplement 1a ) , which led to the sparse labeling of IPCs and their progenies ( Pimeisl et al . , 2013 ) . We analyzed 73 IPC-derived lineages at P21 and exclusively found one-cell and two-cell clones , with no evidence for larger clones within our sample ( Figure 2—figure supplement 1b–d ) . Although the existence of IPCs that undergo more than one cell division cannot be completely excluded , these results indicated that this is not common at this developmental stage . Consequently , IPCs are unlikely to be the origin of laminar-restricted lineages . Our retroviral tracing experiments suggested that the neuronal output of neocortical VZ progenitor cells is significantly more heterogeneous than previously described , including translaminar , deep- and superficial-layer restricted lineages . However , several technical limitations may contribute to the observation of laminar-restricted lineages , as retroviral tracing may lead to the incomplete labeling of neuronal lineages . For example , the existence of deep layer-restricted lineages might be due to the silencing of the viral cassette after a few rounds of cell division ( Cepko et al . , 2000 ) , which would prevent the expression of GFP or mCherry in superficial layer PCs . There are also alternative explanations for the observation of superficial layer-restricted lineages in the retroviral-tracing experiments . First , infected progenitors might have become neurogenic at slightly earlier stages and have already produced a wave of deep layer PCs before infection , which would therefore not be labeled by the retrovirus . Second , the entire set of deep layer neurons might have been generated during the first neurogenic division of a VZ progenitor cell , which would not be labeled in some cases due to the retroviral integration mechanism . To overcome these technical limitations , we took advantage of the Mosaic Analysis with Double Markers ( MADM ) technique , a genetic method widely used to fate-map cellular lineages at high resolution ( Hippenmeyer et al . , 2010; Zong et al . , 2005 ) . We used the Emx1-CreERT2 mice ( Kessaris et al . , 2006 ) to induce MADM sparse labeling of VZ progenitor cells following tamoxifen administration at E12 . 5 ( Figure 3a ) . We specifically focused our analysis on G2-X MADM segregation events that result in the labeling of an unbalanced number of daughter cells with either green or red fluorescent proteins and report the outcome of asymmetric divisions in VZ progenitor cells ( Zong et al . , 2005 ) . Consistent with the retroviral lineage tracing experiments , we found that the vast majority of MADM lineages adopt a translaminar configuration ( Figure 3b , d ) . In addition , we also identified some lineages in which PCs were confined to layers V and VI , thereby confirming the existence of cortical lineages restricted to deep layers of the neocortex ( Figure 3c , d ) . We observed that the fraction of deep layer-restricted lineages labeled with MADM ( ~7% ) is smaller than that obtained with retroviral tracing ( 15% ) , which suggested that reporter silencing might exist in some clones in the retroviral lineage tracing experiments . In contrast , we did not recover a significant number of superficial layer-restricted lineages in MADM experiments ( Figure 3d ) . The MADM experiments suggested that the observation of superficial layer-restricted lineages in retroviral experiments might be artifactual , a result of the incomplete retroviral labeling of neuronal lineages . We reasoned that if this were the case , the analysis of the MADM sub-clones ( i . e . only one of the two colors in the lineage ) containing more than two cells should lead to a similar fraction of ‘artifactual’ lineages , since these would essentially correspond to those labeled by retroviral infection missing the first division of VZ progenitor cells ( Figure 3—figure supplement 1 ) . This analysis indeed identified a small fraction of MADM sub-clones as ‘apparent’ superficial layer-restricted lineages ( ~12% ) , which was nevertheless significantly smaller than those identified in the retroviral experiments ( 22% ) . This indicated that although some of the superficial layer-restricted lineages observed in retroviral experiments were artifactual , others might not be . One important difference between both approaches is that MADM G2-X recombination events occur exclusively in mitotic cells ( Zong et al . , 2005 ) , while retroviral labeling does not strictly depend on cell division . Retroviruses require cell division for their integration into the genome , but the infection is independent of cell cycle stage ( Cepko et al . , 2000 ) . Thus , we hypothesized that MADM may not consistently label a fraction of quiescent or slowly dividing progenitors , which could otherwise be targeted by retroviral infection . To test this idea , we carried out a new set of lineage-tracing experiments using a third , complementary method . In brief , we traced cortical lineages at single cell resolution using low-dose tamoxifen administration in Emx1-CreERT2;RCL-Gfp ( RCL-Gfp also known as RCE ) pregnant mice at E12 . 5 ( Figure 4a ) , in which labeling of VZ progenitor cells should be independent of cell cycle dynamics . Since this method does not distinguish between lineages derived from symmetric or asymmetric cell divisions , we limited our analysis to lineages with a maximum of 12 cells , the larger clonal size of neurogenic lineages in the Emx1-CreERT2;MADMTG/GT experiments ( clones with more than 12 cells account for less than 5% of the neurogenic lineages and largely include [87%] the outcome of symmetrically-dividing progenitor cells ) . Consistent with the other approaches , the majority of lineages ( ~75% ) labeled by injection of Emx1-CreERT2;RCL-Gfp mice with low tamoxifen doses at E12 . 5 were translaminar ( Figure 4b , e ) . We also confirmed that ~13% of the lineages were restricted to deep cortical layers ( Figure 4c , e ) . In addition , we found that ~11% of the lineages consist of PCs confined to superficial layers of the neocortex ( Figure 4d , e and Figure 4—figure supplement 1 ) . In sum , the combined results of three different sets of lineage-tracing experiments suggested that translaminar ( ~80% ) , deep layer-restricted ( ~10% ) and superficial layer-restricted ( ~10% ) lineages are generated at the onset of neurogenesis in the developing neocortex . We next explored the precise organization of cortical lineages derived from VZ progenitor cells at E12 . 5 . In lineage-tracing experiments using Emx1-CreERT2;RCL-Gfp mice ( Figure 5a ) , we observed that only about a quarter of traced lineages contains neurons in every cortical layer from II to VI , and every other clone lacks neurons in one or multiple cortical layers ( Figure 5b–d , f ) . For instance , a significant proportion of translaminar lineages lack PCs in layer V ( Figure 5b , f ) or layer IV ( Figure 5c , f ) but , considered collectively , PC lineages adopt every possible configuration of laminar distributions in the neocortex ( Figure 5f ) . The heterogeneous organization of cortical lineages was not exclusively observed in the experiments performed in Emx1-CreERT2;RCL-Gfp mice; similar results were obtained in retroviral lineage tracing ( Figure 5—figure supplement 1a–e ) and MADM experiments ( Figure 5—figure supplement 1f–j ) . Although the clonal size of these lineages also exhibits great heterogeneity , it seems to follow a bimodal distribution , with maximums at approximately four and eight cells ( Figure 5e ) . Intriguingly , these two maximums largely correspond to restricted and translaminar lineages respectively , reinforcing the previously described link between clonal size and laminar configuration . We characterized the organization of translaminar lineages with PCs in every layer by quantifying the relative proportion of neurons in deep and superficial layers . This analysis revealed that these cortical lineages typically showed a bias toward the production of PCs for superficial layers , although a minority of lineages displayed a preference toward deep layers or a balanced distribution across deep- and superficial layers ( Figure 5g ) . In general , the total amount of cells in superficial and deep cortical layers was slightly anti-correlated . To further explore the molecular diversity of PCs in these lineages , we stained P21 brain sections from Emx1-CreERT2;RCL-Gfp mice induced at E12 . 5 with antibodies against Ctip2 and Satb2 , two transcription factors whose relative expression defines different types of PCs with unique patterns of axonal projections ( Greig et al . , 2013; Lodato and Arlotta , 2015 ) . We identified four PC subclasses based on the expression of these markers and their laminar distribution ( Figure 5—figure supplement 2 ) : Cortico-cortical projection neurons ( CCPN ) , subcerebral projection neurons ( SCPN ) , cortico-thalamic projection neurons ( CThPN ) and heterogeneous projection neurons ( HPN ) ( Harb et al . , 2016 ) . Using this classification , we found that nearly a quarter of all-layer translaminar lineages were composed exclusively by CCPNs , while multiple different combinations of PC identities comprise the remaining lineages ( Figure 5h ) . Of note , only a minor fraction of all cortical lineages contains the entire complement of subtypes identified . Altogether , our experiments revealed that PC lineages exhibit a great degree of heterogeneity in the number and identities they comprise . The observed heterogeneity in cortical lineages likely emerges during neurogenesis . However , it is possible that selective cell death of specific PCs might contribute to the heterogeneous organization of cortical lineages . Recent studies have shown PCs undergo apoptosis during early postnatal stages ( Blanquie et al . , 2017; Wong et al . , 2018 ) . To explore the contribution of cell death to the heterogenous configuration of cortical lineages , we labeled clones by injecting a low dose of tamoxifen in Emx1-CreERT2;RCL-Gfp mice at E12 . 5 and analyzed their laminar organization at P2 , prior to the period of PC death ( Wong et al . , 2018 ) . We detected no significant differences in the average clonal size or in the relative frequency of P2 translaminar and laminar-restricted lineages compared to P21 ( Figure 5—figure supplement 3a–d ) . In addition , we observed that the diversity of lineage patterns was remarkably similar between P2 and P21 ( Figure 5—figure supplement 3e ) . We also noticed a tendency ( χ2 test , p=0 . 099 ) for the fraction of lineages with PCs in every layer to be larger and the frequency of lineages lacking PCs in layer five to be smaller at P2 compared to P21 ( Figure 5—figure supplement 3e ) . These experiments suggested that although cell death may have a subtle impact in refining the final diversity of lineages and their relative proportion , such heterogeneity should arise directly during the process of cortical neurogenesis . The variability in size and composition of PC lineages raises questions about the developmental mechanisms underlying their genesis . We first asked whether lineage structure is relevant for cortical cytoarchitectural development . It is formally possible that the diversity in laminar composition of cortical lineages is simply the consequence of a random process of PC generation in which the only boundary condition is the relative number of PCs that populate each layer of the cortex . To test this , we used the lineages mapped in the primary somatosensory cortex ( S1 ) of Emx1-CreERT2;RCL-Gfp mice , which are meant to collectively generate a common pattern of laminar densities . We randomly permuted the PCs obtained from the different lineages while maintaining each neuron’s laminar identity and the total number of cells in each lineage . If lineage structure were to exclusively affect the control of PC laminar fractions at population level , permuted lineages should be expected to match experimental data . As expected , the permutation process left unaltered the clonal size distribution and number of cells per layer observed in the experimental data ( Figure 6—figure supplement 1a , b ) . However , it failed to replicate the observed anti-correlation in neuron numbers between superficial and deep layers ( Figure 6—figure supplement 1c ) . In addition , we observed that the laminar configuration of the permuted lineages differed from the experimental data ( data not shown ) . These results indicated that the laminar distribution of neurons within each lineage arises specifically from an organized pattern of neurogenesis . One possibility to explain lineage diversity is the existence of multiple different VZ progenitor cell types with restricted potential to generate specific classes of PCs ( Franco and Müller , 2013 ) . However , the observed heterogeneity in lineage configurations may also arise from equipotent VZ progenitor cells that are subject to stochastic factors controlling their output , as proposed for the retina ( He et al . , 2012 ) . To establish the feasibility of the latter scheme , we used a Bayesian approach to model the outcome of cortical progenitor cells following stochastic developmental programs ( Diana , 2019; copy archived at https://github . com/elifesciences-publications/SampLin ) . This method involved using a set of probabilistic rules for generating lineages , and subsequently inferring the number of rules required for the assignment of all lineages observed in the experimental data ( see Materials and methods for details ) . To avoid having to account for variability potentially attributable to differences in the distribution of lineages across areas of the neocortex , we only considered experimental data obtained from lineages mapped in S1 of Emx1-CreERT2;RCL-Gfp mice . We reasoned that some lineage configurations observed in our experiments ( such those containing cells exclusively in deep cortical layers ) could derive from an early interruption of the developing lineage , reflecting an early terminal division , or a progenitor cell undergoing cell death after a few rounds of division . Since the genesis of such lineages might therefore not arise from specific developmental programs , these configurations were also not taken into account for the inference of the stochastic rules governing this process . The Bayesian inference approach revealed that models using one or two progenitor types are sufficient to produce a diversity of lineage compositions as found in our experimental data ( Figure 6a , e , f ) . However , the approach failed to reproduce the anti-correlation in cell numbers between superficial and deep layers found experimentally ( Figure 6d ) and tended to underestimate the fractions of superficial and small lineages in our data set ( Figure 6c , e ) . This suggests that while simple stochastic processes acting mostly on a single homogeneous population of VZ progenitor cells can originate a vast diversity of outcomes as observed in our experiments , some experimental observations may arise from additional developmental programs and from features characteristic of the sequential process of neurogenesis . Having established that the stochastic behavior of a small number of progenitor types could in principle account for lineage diversity , we next explored specifically how diversity can result from the sequential dynamics of stochastic neuron generation . To this end , we simulated cortical progenitor behavior using models based on four basic rules derived from experimental knowledge ( Figure 7a and Figure 7—figure supplement 1a ) . First , in silico progenitors would generate neurons for different layers sequentially , following the observed inside-out pattern . Second , each in silico progenitor would have a set , randomly selected number of opportunities to generate neurons in each layer . Third , for any progenitor , the decision to generate a neuron would be probabilistic , with cell generation probabilities varying by cortical layer but equal for all opportunities within the same layer . Thus , a progenitor type was defined by its specific combination of cell generation probabilities across layers . Fourth , to simulate the chances of premature terminal division and/or progenitor death , we introduced a probabilistic chance of lineage interruption at each opportunity for cell generation . In silico lineages generated using this model were then compared with the experimental lineages mapped in the primary somatosensory cortex ( S1 ) of Emx1-CreERT2;RCL-Gfp mice . We set cell generation probabilities in each layer to match the total laminar fractions of PCs ( Figure 7b and Figure 7—figure supplement 1b ) , as well as the clonal size distribution ( Figure 7c and Figure 7—figure supplement 1c ) observed in those lineages . In agreement with the Bayesian approach , we found that a stochastic model based on a single , equipotent VZ progenitor cell ( Model 1 ) , that is a single set of cell generation probabilities , was able to reproduce the majority of experimentally observed lineage features . Modeled lineages recapitulated the existence of restricted lineages as well as all observed laminar configurations of translaminar lineages ( Figure 7—figure supplement 1e–f ) . However , this model generated lineages with an exaggerated anti-correlation in the number of cells in superficial versus deep layers ( Figure 7—figure supplement 1d ) and failed to reproduce the bimodal distribution of clonal sizes , underestimating the fraction of small lineages ( those containing 3–4 cells ) . In addition , the fraction of lineages restricted to superficial layers , which largely contribute to the small lineage sizes , was also underestimated ( Figure 7—figure supplement 1c , e ) . These results suggested that the fraction of small superficial lineages is unlikely to arise from a single stochastic program common to all cortical progenitors . We then generated a second model with two different sets of cell generation probabilities , defining two different progenitor populations ( Model 2 ) . In this model , the majority of progenitors belonged to a population generating the larger lineages , while a second , smaller population generated small progenies biased towards superficial layer PC fates ( Figure 7a ) . We found that this model faithfully reproduced all the experimental features in our data: total laminar fractions ( Figure 7b ) , bimodal distribution of clonal sizes ( Figure 7c ) and negative correlation in superficial versus deep layers ( Figure 7d ) . In addition , the relative proportions of translaminar and laminar-restricted lineages were identical to those measured experimentally ( Figure 7e ) . Finally , translaminar modeled lineages exhibited similar laminar configurations to the experimental lineages ( Figure 7f ) . In sum , mathematical modeling suggests that a stochastic mechanism of cortical neurogenesis based on two independent progenitor cell populations best approximates the experimental data . Finally , we explored whether the proposed stochastic model with two progenitor populations ( Model 2 ) would be able to generate different ratios of layer-specific neurons under different circumstances ( i . e . different cell generation probabilities ) , which would robustly account for the emergence of cytoarchitectural differences across neocortical areas . To this end , we quantified the fraction of PCs in each layer of the primary somatosensory ( S1 ) and visual ( V1 ) cortices in Neurod6Cre/+;Fucci2 mice , in which all PCs in the neocortex are labeled with a nuclear fluorescent marker . As expected , we found important differences in laminar cytoarchitecture between both regions ( Figure 7g ) . Remarkably , we found that subtle tuning of generation probabilities for both areas was sufficient to replicate the different laminar ratios in silico ( Figure 7h ) . This suggests the stochastic mechanisms of neurogenesis described here would suffice to generate the diverse cytoarchitectonic patterns observed across neocortical areas . Understanding how individual lineages contribute to the production and organization of PCs is essential to articulate a coherent framework of cortical development . The analysis of the output of progenitor cells in the developing rodent cortex expands over three decades and has relied on four approaches: retroviral labeling ( Luskin et al . , 1988; Noctor et al . , 2001; Noctor et al . , 2004; Price and Thurlow , 1988; Reid et al . , 1995; Walsh and Cepko , 1988; Walsh and Cepko , 1992 ) , mouse chimeras ( Tan et al . , 1998 ) , MADM ( Beattie et al . , 2017; Gao et al . , 2014 ) and genetic fate-mapping ( Eckler et al . , 2015; Franco et al . , 2012; García-Moreno and Molnár , 2015; Gil-Sanz et al . , 2015; Guo et al . , 2013; Kaplan et al . , 2017 ) . These studies often led to contradictory results , which has prevented the emergence of a consistent model . The prevalent view is that each progenitor cell in the developing pallium is multipotent and generates a cohort of PCs that populate all layers of the neocortex except layer I ( Eckler et al . , 2015; Gao et al . , 2014; Guo et al . , 2013; Kaplan et al . , 2017 ) , as originally conceived in the radial unit hypothesis ( Rakic , 1988 ) . In contrast , some authors have suggested that many cortical progenitor cells are fate-restricted to generate PCs that exclusively occupy deep or superficial layers of the neocortex ( Franco et al . , 2012; Franco and Müller , 2013; García-Moreno and Molnár , 2015; Gil-Sanz et al . , 2015 ) . Here , we have used three different methods ( retroviral labeling , MADM and genetic fate-mapping ) to investigate the clonal production of cortical neurons by capitalizing on the synergy that emerges from the advantages of each individual approach . Our results indicate that this multi-modal approach is required to comprehensively capture the complex behavior of progenitor cells in the developing cortex . Retroviral labeling has two important limitations: it only labels hemi-lineages and is prone to silencing , which may prevent the identification of the entire progeny of a progenitor cell ( Cepko et al . , 2000 ) . Conversely , retroviral labeling targets progenitor cells indiscriminately and , consequently , is not biased toward a particular genetic fate ( Cepko et al . , 2000 ) , as is the case for genetic strategies . MADM , on the other hand , has the enormous advantage of identifying both sister cells resulting from a cell division . However , G2-X MADM events require progenitor cells to undergo cell division at the time of induction because it directly relies on Cre-dependent inter-chromosomal mitotic recombination ( Zong et al . , 2005 ) . Our results revealed that MADM does not reliably label a small fraction of progenitor cells present in the pallial VZ at E12 . 5 that gives rise to cohorts of PCs exclusively located in superficial layers of the neocortex . These lineages were however observed in both retroviral labeling experiments and in genetic tracing experiments using the same genetic driver ( Emx1-CreERT2 ) as in the MADM experiments , which strongly suggests that some Emx1+ progenitor cells producing exclusively superficial layer PCs in the developing cortex are not targeted by the MADM approach . We hypothesize that these progenitors might be quiescent or slow-dividing progenitors at this stage and become more active at later stages of development . Finally , although the use of genetic fate-mapping strategies ( e . g . Emx1-CreERT2;RCL-Gfp ) is a powerful method to investigate cortical lineages , it has the important constraint of not being able to distinguish between symmetric proliferating and asymmetric neurogenic divisions . This hampers the analysis of clonal sizes , which can be otherwise accurately assessed with MADM except for the lineages that are not detected with this method . Previous clonal analyses based on MADM lineage tracing experiments led to the suggestion that individual progenitor cells in the pallial VZ produce a unitary output of approximately eight excitatory neocortical neurons distributed throughout superficial and deep layers of the neocortex ( Gao et al . , 2014 ) . However , those studies failed to identify lineages with restricted laminar patterns ( either deep or superficial layer restricted clones ) . Consequently , they also underestimated the fraction of lineages with relatively small clonal size . In contrast , our analysis of neurogenic lineages revealed a bimodal distribution of clonal sizes with defined peaks centered at approximately four and eight cells , which largely correspond to the contribution of laminar-restricted and translaminar lineages , respectively . Previous studies have suggested that some neocortical progenitor cells generate laminar-restricted lineages of PCs ( Franco et al . , 2012; Gil-Sanz et al . , 2015 ) . In our experiments , approximately one in six cortical progenitor cells generate laminar-restricted lineages . The existence of lineages restricted to deep layers of the neocortex was observed with all three methods used in this study . Although some variation exists in the relative fraction of deep layer-restricted lineages observed with the different approaches , these differences lie within the expected experimental noise considering the relatively small number of lineages that belong to this category . In addition , both retroviral labeling and genetic fate-mapping experiments identified a fraction of cortical progenitor cells that generate PCs that exclusively populate the superficial layers of the neocortex . It is conceivable that these lineages reflect the output of progenitor cells that had already produced an earlier cohort of deep layer neurons prior labeling . If this where the case , one should expect to observe similar results in MADM experiments . In such experiments , however , we did not recover a significant fraction of superficial lineages . Therefore , the discrepancy between the results of genetic fate-mapping and MADM experiments , in which the same mouse strain is used as the driver for recombination ( Emx1-CreERT2 ) , suggests that these fate-restricted lineages arise from progenitor cells that are not actively dividing at E12 . 5 . This hypothesis is consistent with the identification of a population of self-renewing progenitors with limited neurogenic potential during the earliest phases of corticogenesis ( García-Moreno and Molnár , 2015 ) . The existence of superficial layer-restricted cortical lineages is further supported by the identification of IPCs as early as E12 . 5 that generate superficial layer PCs ( this study and Mihalas et al . , 2016 ) , when the majority of deep layer PCs are being generated . Since IPCs derive from VZ progenitor cells ( Haubensak et al . , 2004; Miyata et al . , 2004; Noctor et al . , 2004 ) and cortical neurogenesis begins at these stages in the mouse , this observation reinforces the idea that some progenitors are tuned to generate superficial layer PCs from early stages of corticogenesis . Although it is formally possible that some translaminar lineages in the retroviral and genetic fate-mapping experiments could arise from symmetric divisions generating two laminar-restricted lineages ( one superficial and one deep ) , this possibility is unlikely considering the results of the MADM experiments . The results of those experiments indicate that symmetric divisions at E12 . 5 generate at least one translaminar sub-lineage in virtually all cases ( 58/59 ) , suggesting that most translaminar lineages arise from progressive divisions of individual progenitor cells . Our study also revealed that , independently of the laminar distribution , individual cortical progenitor cells generate lineages with very diverse combinations of PC types . Cortical progenitors are thought to undergo progressive changes in their competency to generate different layer-specific types of PCs ( Desai and McConnell , 2000; Oberst et al . , 2019; Rakic , 1974 ) . Consistent with this idea , our results reveal that most cortical progenitors generate diverse types of excitatory neurons . However , since many cortical progenitor cells fail to generate neurons for at least one layer of the neocortex , the majority of cortical lineages does not include the entire diversity of excitatory neurons . In other words , the fraction of individual cortical lineages that would be considered as ‘canonical’ – that is containing all three main classes of excitatory projection neurons ( CCPN , SCPN and CThPN ) – is significantly smaller than previously anticipated . Considering the variance in clonal size and lineage composition of neocortical lineages , our results indicate that cortical progenitor cells exhibit very heterogeneous patterns of neuronal generation and specification . This interpretation challenges the view that the neuronal output of RGCs is deterministic ( Gao et al . , 2014 ) . Our results indicate that stochastic developmental programs , in which cortical progenitors undergo a series of probability-based decisions for the generation of the different PC fates , are capable of generating the wide diversity of lineage configurations observed in our experiments . Therefore , in spite of the great diversity of configurations that exist among individual neocortical lineages , our results suggest that their genesis does not require a corresponding heterogeneity in VZ progenitors . The model proposed here is somewhat reminiscent of that described for the developing rodent and zebrafish retina ( Gomes et al . , 2011; He et al . , 2012 ) . In line with our findings , stochastic mechanisms based on a single set of probability rules explain the genesis of most , but not all neuronal types in the mammalian retina ( Gomes et al . , 2011 ) . It is presently unclear whether laminar-restricted lineages arise from a pool of progenitor cells separate from those generating translaminar lineages or should simply be considered as extreme examples of the enormous diversity of lineage configurations uncovered by our study . The generation of lineages restricted to deep layers might be due to premature terminal division or death of the progenitor cell ( Blaschke et al . , 1996; Mihalas and Hevner , 2018 ) , as considered in our models , but the existence of superficial layer-restricted lineages is more difficult to explain . Moreover , our mathematical model best reproduces the complex cytoarchitecture of the neocortex when two distinct progenitor cell identities are considered . Previous studies have identified morphological heterogeneity among pallial VZ progenitor cells ( Gal et al . , 2006 ) . However , there is limited evidence for important molecular differences among these cells ( Mizutani et al . , 2007; Pollen et al . , 2014; Telley et al . , 2016 ) . In the absence of a definitive molecular signature , our results suggest that while a homogeneous population of progenitor cells following a common developmental program explains most of the observed outcomes , it fails to generate the fraction of small superficial lineages observed in the experiments . The introduction of a second population of progenitor cells is required to reproduce these lineages . Although these findings might suggest the existence of a small fraction of cortical progenitors tuned to preferentially generate superficial lineages , it should not be taken as a definitive proof of fate-restriction in cortical progenitors . In our model , replicating the experimental data does not require such progenitors to be restricted , but simply biased toward generating superficial fates . Our study suggests that progenitor cells in different cortical areas are likely constrained by different probabilistic rules , which would contribute to the generation of the diverse cytoarchitectonic patterns found across the neocortex . Although the number of lineages recovered from each cortical region is insufficient to provide conclusive evidence for major regional differences , lineages located in different cortical areas seem to exhibit features that reflect the local cytoarchitecture . For instance , lineages lacking layer IV neurons were abundantly found in the retrosplenial cingulate and motor cortices , where this layer is remarkably small . How and when stochastic neurogenic decisions are made remains to be elucidated , but they likely depend on the influence of extrinsic and intrinsic signals on parameters such as cell cycle length , the asymmetric inheritance of cell components , the generation of dividing ( IPCs ) versus postmitotic progeny , and the membrane potential of progenitor cells ( Haydar et al . , 2003; Lange et al . , 2009; Pilaz et al . , 2009; Roccio et al . , 2013; Vitali et al . , 2018; Wang et al . , 2009 ) . Local signals in different neocortical areas would contribute to the tuning of progenitor cell behaviors to output different cytoarchitectures without the requirement of regional-specific progenitor populations . Consequently , this model allows great flexibility in the generation of heterogeneous cortical cytoarchitectures without the requirement of a large number of progenitor identities . The specification of a very small number of progenitor cells with competence to adapt their neurogenic behavior to different probabilistic rules based on their location within the neocortical neuroepithelium represents the most parsimonious and robust mechanism for the generation of cortical circuitry . The following transgenic mouse lines were used in this study: Neurod6Cre ( Goebbels et al . , 2006 ) RRID: MGI:4429427 , Emx1-CreERT2 ( Kessaris et al . , 2006 ) RRID: IMSR_JAX:027784 , Tbr2CreERT2 ( Pimeisl et al . , 2013 ) RRID: MGI:5499789 , RCL-Gfp ( Sousa et al . , 2009 ) RRID: MGI:4420759 , MADMTG ( JAX 013751 ) RRID: IMSR_JAX:013751 , MADMGT ( JAX 013749 ) RRID: IMSR_JAX:013749 and RCL-Fucci2 RRID: IMSR_HAR:6899 . The MADMTG and MADMGT alleles were generated by inserting dTN-term–GfpC-term and GfpN-term–tdTC-term sequences in the Hipp11 locus ( Hippenmeyer et al . , 2010 ) . RCL-Fucci2 are reporter mice in which a fluorescent ubiquitination-based cell cycle indicator ( Fucci ) consisting of mVenus-hGem and mCherry-hCdt1 sequences linked by a T2A linker were flanked by loxP sited and inserted into the ROSA26 locus ( Mort et al . , 2014 ) . All adult mice were housed in groups and kept on reverse light/dark cycle ( 12/12 hr ) regardless of genotypes . Only time-mated pregnant female mice that have undergone in utero surgeries were house individually . Both male and female mice were used in all experiments . In utero experiments where performed at different developmental stages that range from E9 . 5 to E14 . 5 . For histological analyses , mice ages range from P2 to P30 . All procedures were approved by King's College London and IST Austria , and were performed under UK Home Office project licenses , and in accordance with Austrian Federal Ministry of Science and Research license , and European regulations ( EU directive 86/609 , EU decree 2001–486 ) . The day of vaginal plug was considered as embryonic day ( E ) 0 . 5 and the day of birth as postnatal day ( P ) 0 . Cre-dependent conditional retroviral stocks encoding EGFP and membrane-bound mCherry reporters ( rv::dio-eGfp and rv::dio-mCherry ) ( Ciceri et al . , 2013 ) were produced as previously described ( Tashiro et al . , 2006 ) . In brief , Moloney murine leukemia viruses ( MoMLV ) were produced by transfecting HEK293T cells ( RRID: CVCL_6911 ) with retroviral plasmids ( dio-eGfp or dio-mCherry , pCMV-Vsvg , and pCMV-GAG-pol ) using lipofectamine 2000 . Forty-eight hours post-transfection , the supernatant was collected , concentrated and purified by two sequential rounds of ultracentrifugation . The viral pellet was re-suspended in sterile PBS and stored in aliquots at −80°C . Viral stocks for dio-eGfp and dio-mCherry were produced in the same plates and mixed before concentration by ultracentrifugation . For in utero injections , pregnant females were deeply anesthetized with isoflurane and the abdominal cavity was incised to expose uterus . Conditional retroviruses were injected at low-titer into the telencephalic ventricles of E9 . 5 , E10 . 5 , E11 . 5 , E12 . 5 and E14 . 5 mouse embryos using an ultrasound-guided imaging system ( Visualsonic ) coupled with a nanoliter injector as previously described ( Ciceri et al . , 2013; Pla et al . , 2006 ) . Some experiments were performed using the rv::dio-eGfp exclusively . After the procedure , the uterine horns were place back in the abdominal cavity and the wound was surgically sutured . The female was then placed in a 32°C recovering chamber for 30 mins post-surgery before returning to standard housing conditions . Emx1-CreERT2;RCL-Gfp and Tbr2CreERT2/+;RCL-Gfp pregnant females received a single intraperitoneal injection of low-dose ( 1 ng/kg ) tamoxifen dissolved in corn oil at E12 . 5 . MADM clones were generated as described previously ( Beattie et al . , 2017; Hippenmeyer et al . , 2010 ) . In brief , timed pregnant females were injected intraperitoneally with tamoxifen dissolved in corn oil at E12 . 5 at a dose of 2–3 mg/pregnant dam . Live embryos were recovered at E18–E19 through caesarean section , fostered , and raised for further analysis at P21 . Postnatal mice were perfused transcardially with 4% paraformaldehyde ( PFA ) in PBS and the dissected brains were fixed for 2 hr at 4°C in the same solution . Brains were serially sectioned at 100 µm on a vibratome ( VT1000S , Leica ) or on a freezing microtome ( SM 2010R , Leica ) and free-floating coronal sections were then subsequently processed for immunohistochemistry as previously described ( Pla et al . , 2006 ) . The following primary and secondary antibodies where used: chicken anti-GFP ( 1:2000 Aves lab cat . no . GFP-1020 , RRID:AB_10000240 ) , rabbit anti-DsRed ( 1:500 Clonetech cat . no . 632496 , RRID:AB_10013483 ) , goat anti-mCherry ( 1:500 Antibodies-Online cat . no . ABIN1440057 , RRID:AB_11208222 ) , rat anti-Ctip2 ( 1:500 Abcam cat . no . Ab18465 , RRID:AB_2064130 ) , mouse anti-Sabt2 ( 1:500Abcam cat . no . Ab51502 , RRID:AB_882455 ) , rabbit anti-Sabt2 ( 1:1000 Abcam cat . no . Ab34735 , RRID:AB_2301417 ) , goat anti-Tle4 ( 1:200 gift from Stefano Stifani ) , anti-chicken IgY ( H+L ) 488 ( 1:400 Molecular Probes cat . no . A-11039 , RRID:AB_2534096 ) , anti-mouse IgG1 647 ( 1:400 Molecular Probes cat . no . A-21240 , RRID:AB_2535809 ) , anti-mouse IgG ( H+L ) biotinylated ( 1:400 Vector laboratories cat . no . BA-2000 , RRID:AB_2313581 ) , anti-rat IgG ( H+L ) 555 ( 1:400 Molecular Probes A-21434 , RRID:AB_2535855 ) , anti-goat IgG ( H+L ) 555 ( 1:400 Molecular Probes cat . no . A-21432 , RRID:AB_2535853 ) , and anti-rabbit 488 ( 1:400 Molecular Probes cat . no . A-21206 , RRID:AB_2535853 ) . Images were acquired using fluorescence microscopes ( DM5000B , CTR5000 and DMIRB from Leica or Apotome . 2 from Zeiss ) coupled to digital cameras ( DC500 or DFC350FX , Leica; OrcaR2 , Hamamatsu ) with the appropriate emission filter sets or in inverted confocal microscopes ( Leica TCS SP8 and Zeiss LSM800 Airyscan ) . All modeling of progenitor behavior was performed using MATLAB ( MathWorks; RRID:SCR_001622 ) . To avoid overfitting variability that could correspond to differences in progenitor behavior across cortical areas , simulations were compared to the lineages observed in primary somatosensory cortex ( S1 ) , using the Emx1-CreERT2;RCL-Gfp experiments . The structural similarity of the results of a model with the experimental data was assessed based on three parameters: proportion of cells per layer , clonal size distribution and Spearman correlation ( r ) values for number of cells in upper versus lower layers . For each parameter , we computed a normalized z-score measure by taking the difference between the experimental value and the average value across simulation repeats , and then dividing by the standard deviation across simulation repeats . Thus , z-score values over one would reflect a distance between experimental and modeled data larger than the standard deviation between simulation repeats . To generate randomly permuted cortical lineages , neurons observed in the Emx1-CreERT2;RCL-Gfp experiments were permuted among lineages while maintaining their laminar identities . This operation was repeated 1000 times , providing average and standard deviation values that were then used to compare with the experimental results . To perform statistical inference on the number of categories required to explain the distribution of lineages throughout cortical layers , we employed a statistical model where N observed lineages are grouped in K progenitor types . Each type t = 1 , … , K is associated to a vector of four probabilities pt = {pt ( II/III ) , pt ( IV ) , pt ( V ) , pt ( VI ) } representing the probabilities of any progenitor in the class to generate neurons in each of the four layers . We assume that each observed lineage can be assigned to a unique progenitor type based on its occupancy distribution . Progenitor types are associated with frequencies ft , reflecting how likely a lineage is to belong to type t . The occupancy probabilities and the relative frequencies for each type as well as the number of types K required can be obtained using Bayesian inference according to the Bayes’ theoremp ( t1:N , p1:K , f1:K\|S ) =P ( t1:N , S\|p1:K , f1:K ) ⏞likelihood⋅P ( p1:K;f1:K ) ⏞priorP ( S ) ⏟marginallikelihoodwhere S is the count matrix whose elements Sij indicate how many neurons in lineage i belong to layer j . The Bayes’ theorem provides the posterior distribution of the model parameters p and f as well as the type assigned to each lineage conditional to the observations . Our Bayesian model can be viewed as the following two-step generative process: The likelihood of a given lineage assignment and count matrix can be written asP ( t1 , . . . , K , S\|p1 , . . . , K , f1:K ) = ( ∏t=1Kftnt ) ∏i=1N∏j∈layers[ptiSij ( 1−ptiNmax−Sij ) ]σij , where we introduced the binary variable σij to denote whether the matrix element Sij is included in the likelihood ( in which case σij=1 ) or not ( σij=0 ) . In particular , the selection variable σij for each lineage was set in such a way to exclude for each lineage the most superficial empty layers not followed by an occupied layer . The corresponding zero counts in the matrix S might be spurious due to external processes stopping the lineage at early stages . To perform Bayesian inference , we used Dirichlet priors on the relative frequencies and Beta distributions as priors on the occupancy probabilities p1 , … , K . To draw samples of model parameters and progenitor types from the posterior distribution we implemented a Gibbs sampler ( Diana , 2019 ) which combines data likelihood and prior distributions to explore the parameter space efficiently . In order to draw statistical samples of the number of classes K , we employed the Dirichlet process prior technique which allows us to remove existing classes or introduce new ones when assigning lineages to classes within the Gibbs sampler ( Diana , 2019 ) . Probabilistic models 1 and 2 simulated 100 progenitors undergoing cell generation sequentially , following the in vivo inside-out pattern ( Figure 7—source code 1 ) . In each layer , in silico progenitors took a number of stochastic decisions for neuron generation; at each decision , a new neuron could be generated , or alternatively , the chance could be skipped without neuron generation . Sequential generation of neurons thus used the following parameters . Number of opportunities per layer was set randomly and could vary between a minimum of one and a maximum equal to the maximum number of cells found for that layer in any single experimental lineage across our three types of experiments . This parameter establishes the number of stochastic decisions available to the progenitor and reflects the size of the temporal ‘window’ within which a progenitor can generate neurons for a given layer . Probability of cell generation , also layer-specific , gave the likelihood that a neuron is actually generated at each decision point . Simulations were repeated 100 times . Lineages smaller than three cells or larger than 12 cells were discarded from analysis . For each model , the set of laminar division probabilities was adjusted to fit the experimental data regarding clonal size and laminar fractions of cells . Model 1 used a unique progenitor , that is a single set of laminar division probabilities . Model 2 incorporated an additional population and was fit by varying both the relative size of the two populations and the values of their division probabilities , including how probabilities varied across layers . Cell distributions and clonal spatial configuration . In all the experiments , brain sections were sequentially analyzed in rostral-to-caudal order and PC clones throughout the entire neocortex were identified as sparse , spatially separated cell clusters . The boundaries between cortical layers were traced based on nuclear ( DAPI ) staining and the laminar position of each cell was recorded accordingly . PC clones were classified as translaminar , infragranular and supragranular clones according to the laminar position of the neurons belonging to each clone . Cortical areas were identified based on the reference atlas of adult mouse brain ( Allen Brain Atlas; http://www . brain-map . org ) . In Emx1CreER;MADMTG/GT experiments , lineages derived from symmetric divisions ( defined as lineages with three or more cells expressing each reporter ) were excluded . In the Emx1CreER;RCL-Gfp experiments , lineages derived from symmetric divisions ( defined as lineages containing more than 12 neurons ) were excluded . Lineages containing one or two cells were also excluded in Emx1CreER;MADMTG/GT and Emx1CreER;RCL-Gfp experiments . In retroviral experiments , one-cell and two-cell clones were considered separately in E9 . 5 , E10 . 5 , E11 . 5 and E12 . 5 experiments . In E14 . 5 experiments , two-cell lineages were considered , while one-cell clones were not quantified . This is due to the fact that , unlike at earlier time points , RGCs may be undergoing their last neurogenic division at this stage , and thus lineages with a minimum of two cells may be derived from targeted apical progenitor cells . Pyramidal cell types . Brain sections were stained for markers of cortical projection neuron identity and classified based on the relative expression of the transcription factors Ctip2 and Satb2 in four main subtypes: Cortico-cortical ( CCPN ) , sub-cerebral ( SCPN ) , cortico-thalamic ( CthPN ) and heterogeneous ( HPN ) projection neurons . This last type was defined as layer V cells expressing both Ctip2 and Satb2 markers , which have been recently described as a distinct identity33 . Images were captured using a confocal microscope and analyzed using a custom algorithm written in MATLAB ( Mathworks ) . In brief , cell nuclei were segmented using the disk morphological function based on size and thresholds of fluorescence intensity over background . Cells were categorized as expressing high or low levels of the transcription factors Ctip2 and Satb2 and further subclassified as CCPN ( Ctip2Low/Satb2High ) , SCPN ( Ctip2High/Satb2Low ) or HPN ( Ctip2High/Satb2High ) , based on the combination of marker expression . To distinguish between CThPN from CCPN in layer VI we used the following criteria: CCPN ( Ctip2Low/Satb2High OR Ctip2Low/Satb2Low ) or CThPN ( Ctip2High/Satb2Low ) . This allowed for the subclassification of layer V and layer VI cells based on the same set of markers . We verified these criteria by staining brain sections for the transcription factor Tle4 ( Figure 5—figure supplement 2 ) , a well-established specific marker of cortical CThPN identity ( Molyneaux et al . , 2015 ) . Layer VI cells expressing high levels of both transcription factors were not classified , and lineages containing those cells were excluded from the quantification . Laminar ratios . To quantify the actual densities of PCs in different cortical layers , Neurod6Cre mice were crossed with Fucci2a reporter mice . The density of labeled red nuclei in each cortical layer was quantified from five representative serial sections of the somatosensory and visual cortex . Z-stacks were then 3d reconstructed and quantified using Imaris 8 . 1 . 2 ( Bitplane; RRID:SCR_007370 ) . Statistical tests . Error bars in all graphs indicate standard deviation ( std ) unless otherwise stated in the legends . Comparisons of distributions over fractions of a total ( e . g . Figure 6e , f ) were analyzed using Fisher’s exact test or Chi-square test . Average clonal size between lineages analyzed at P2 and P21 were analyzed using Mann-Whitney U-test . All statistical tests are specified in the figure legends .	Recognizable by its deep outer folds in humans , the cerebral cortex is a region of the mammalian brain which handles complex processes such as conscious perception or decision-making . It is organized in several layers that contain different types of ‘excitatory’ neurons which can activate other cells . The various areas of the cortex have different characteristics as they contain various proportions of each kind of neurons . Stem cells are cells capable to divide and create various types of specialized cells . The excitatory neurons in the cortex are created during development by stem cells known as radial glial cells . These cells divide several times , giving rise to different types of neurons in sucessive divisions , presumably thanks to internal molecular clocks . In the cortex , it is generally assumed that an individual radial glial cell produces all the different types of excitatory neurons . However , studies have suggested that certain cells could be specialized in creating specific types of neurons . To explore this question , Llorca et al . used three complementary approaches to follow individual radial glial cells and track the neurons they created in mouse embryos . This helped to understand how groups of stem cells work together to build the cortex . The experiments revealed that radial glial cells differ more than anticipated in the number and the types of neurons they generate , and rarely produce all types of excitatory neurons . In other words , the output of individual radial glial cells is not always the same . The results by Llorca et al . suggest that as radial glial cells divide , they undergo a series of probabilistic decisions – that is , in each division the cells have a certain probability to generate a specific type of neuron . Consequently , the resulting lineages are rarely identical or contain all types of excitatory neurons , but collectively they generate the full diversity of excitatory neurons in the cortex . Ultimately , new insights into how excitatory neurons form and connect in the brain may be used to help understand psychiatric conditions where circuits in the cortex might be impaired , such as in autism spectrum disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2019	A stochastic framework of neurogenesis underlies the assembly of neocortical cytoarchitecture
Ryanodine receptor type I-related myopathies ( RYR1-RMs ) are a common group of childhood muscle diseases associated with severe disabilities and early mortality for which there are no available treatments . The goal of this study is to identify new therapeutic targets for RYR1-RMs . To accomplish this , we developed a discovery pipeline using nematode , zebrafish , and mammalian cell models . We first performed large-scale drug screens in C . elegans which uncovered 74 hits . Targeted testing in zebrafish yielded positive results for two p38 inhibitors . Using mouse myotubes , we found that either pharmacological inhibition or siRNA silencing of p38 impaired caffeine-induced Ca2+ release from wild type cells while promoting intracellular Ca2+ release in Ryr1 knockout cells . Lastly , we demonstrated that p38 inhibition blunts the aberrant temperature-dependent increase in resting Ca2+ in myotubes from an RYR1-RM mouse model . This unique platform for RYR1-RM therapy development is potentially applicable to a broad range of neuromuscular disorders . The ryanodine receptor type I ( RyR1 ) is a calcium release channel located in the terminal cisternae of the sarcoplasmic reticulum ( SR ) in skeletal muscle . During excitation-contraction coupling ( ECC ) , RyR1 is activated by the voltage sensing L-type calcium channel dihydropyridine receptor ( DHPR ) , located in the transverse tubule ( T-tubule ) membrane . Together , the T-tubule and two adjacent SR terminal cisternae form a junctional membrane unit referred to as the triad ( Jungbluth , 2007; Dowling et al . , 2014; Jungbluth et al . , 2018 ) . Mutations in the RYR1 gene are the most common cause of non-dystrophic muscle disease in humans ( Colombo et al . , 2015; Gonorazky et al . , 2018; Jungbluth et al . , 2018 ) . RYR1 mutations are associated with a wide range of clinical phenotypes , collectively referred to as RYR1-related myopathies ( RYR1-RM ) , that can include wheelchair and ventilator dependence , and dynamic symptoms such as exercise induced myalgias , heat stroke , and malignant hyperthermia ( Klein et al . , 2012; Amburgey et al . , 2013; Snoeck et al . , 2015; Jungbluth et al . , 2016; Matthews et al . , 2018 ) . Despite their relatively high prevalence and associated morbidities , there are currently no approved pharmacological therapies for patients with RYR1-RM . Much of what is known about the function of RyR1 and the impact of its mutations on skeletal muscle comes from animal models . Well described recessive models of RYR1-RM include the C . elegans unc-68 mutant ( null mutant with impaired motility [Maryon et al . , 1996; Maryon et al . , 1998] ) , the relatively relaxed zebrafish ( loss of function ryr1b mutant with impaired motility and early death ( Hirata et al . , 2007 ) , and the ‘dyspedic’ Ryr1 null mouse ( perinatal lethal [Buck et al . , 1997; Avila and Dirksen , 2000] ) . In addition , two compound heterozygous mouse models of recessive RYR1-RM were with recently generated and characterized ( Brennan et al . , 2019; Elbaz et al . , 2019 ) . These models are complimented by ‘knock-in’ mutants in mice that mirror specific dominant human mutations , including the I4895T mutant ( associated with central core disease and referred to as the IT model ) ( Zvaritch et al . , 2007; Zvaritch et al . , 2009; Lee et al . , 2017 ) , the R163C mutant ( associated with malignant hyperthermia ) ( Yang et al . , 2006 ) , and the Y522S mutant ( associated with malignant hyperthermia and referred to as the YS mouse ) ( Chelu et al . , 2006; Durham et al . , 2008; Lanner et al . , 2012; Yarotskyy et al . , 2013 ) . Previous work using these models identified potential therapeutic targets for RYR1-RM ( for a comprehensive review , see Lawal et al . , 2018 ) including anti-oxidants ( Durham et al . , 2008; Dowling et al . , 2012; Michelucci et al . , 2017 ) , ER stress modulators ( Lee et al . , 2017 ) , and chemicals that influence the binding of RyR1 to modifying partners ( e . g . S107 , which promotes RyR1/calstabin1 interaction ) ( Lehnart et al . , 2008; Bellinger et al . , 2008; Andersson et al . , 2011 ) . However , as of yet none of these targets has successfully translated to patients , though N-acetylcysteine was tested in a recently completed clinical trial ( ClinicalTrials . gov identifier: NCT02362425 ) , where it failed to achieve its primary endpoint ( Todd et al . , 2020 ) . There is thus a critical need to identify and develop new treatment strategies . With the goal of identifying new therapies for RYR1-RM , we set out to establish a novel multi-species translational pipeline ( Figure 1 ) . This pipeline is based on the functional conservation of RyR1 across many species , and takes advantage of specific attributes of C . elegans ( ability to rapidly screen thousands of compounds ) , zebrafish ( large-scale testing in a vertebrate model ) , and mammalian cell lines ( translatability to humans ) . We screened several thousand compounds , and discovered that p38 inhibition modifies RyR1 phenotypes in all three systems . Our study identifies a new potential therapeutic strategy for RYR1-RM , outlines the utility of multi-species drug discovery , and lays the groundwork for future similar screens for other neuromuscular disorders . We performed a drug screen using the unc-68 ( r1162 ) C . elegans model of RYR1-RM ( Figure 2 ) . This model has a deletion in the worm ryanodine receptor , lacks RyR protein expression by western blot , and manifests an unc-68 null phenotype characterized by an ‘uncoordinated’ ( unc ) movement phenotype , defective pharyngeal pumping , impaired calcium regulation , and reduced fitness ( Maryon et al . , 1996; Maryon et al . , 1998 ) . We first considered using a liquid-based movement assay ( i . e . , the C . elegans ‘thrashing assay’ ( Maryon et al . , 1996; Maryon et al . , 1998 ) as the basis for our drug screen because unc-68 mutants thrash at lower rates than WT ( Figure 2—figure supplement 1A ) , but found that automatable methods for this assay were sufficiently variable to prevent use in a screen of thousands of compounds . Instead , we developed a sensitized screen based on our observation that nemadipine-A , an inhibitor of the dihydropyridine receptor ( Kwok et al . , 2006 ) , induces developmental growth arrest in unc-68 mutants ( Figure 2—figure supplement 1B ) . Specifically , unc-68 worms exposed to 25 μM nemadipine-A arrest at the L1-L3 larval stage , while the majority of wild type N2 strain treated with nemadipine-A have either normal development or , in a small percentage , arrest at the L4 stage ( Figure 2—figure supplement 1B ) . Based on this , we screened for chemicals that could overcome this growth arrest by counting the number of L4 and adult stage worms after six days of exposure to both nemadipine-A and chemical ( Figure 2A ) . We evaluated 3700 chemicals in duplicate from a combination of libraries . We screened 770 worm-bioactive ( a . k . a . ‘wactives’ ) and non-wactives at 7 . 5 μM and 60 μM , with doses based on bioactivity established in previous screens with C . elegans ( Burns et al . , 2015 ) . We also screened 880 kinase inhibitors ( GlaxoSmithKline Published Kinase Inhibitor Set ) at 60 μM . Lastly , we screened 1280 drugs from the US Drug Collection of clinical trial stage compounds at 60 μM ( MicroSource Discovery Systems ) . The concentration used for the latter two libraries was based on previous C . elegans screens performed in the 25–60 μM range ( Kwok et al . , 2006; Burns et al . , 2010; Otten et al . , 2018 ) . This concentration range is used to overcome the poor bioaccumulation of exogenous compounds in C . elegans ( Burns et al . , 2010 ) . The primary screen found 278 chemical conditions out of 3700 ( ~7 . 5% ) that contained at least one L4 or adult worm in duplicate wells ( Figure 2B ) . However , 62 single DMSO vehicle control wells out of 760 ( ~8 . 2% ) and 360 single experimental wells out of 7400 ( ~4 . 9% ) also contained at least one unc-68 mutant that reached the L4 or adult stage , which we collectively refer to as ‘random escapees’ . Based on this , we concluded that many of the 278 chemicals may potentially be false positives . We prioritized 145 of the 278 chemicals for re-testing because they demonstrated in duplicate wells the most complete ( i . e . highest number of L4 and adult ) suppression the phenotype . Of note , we only counted the number of ‘rescued’ worms ( L4 and adult ) per well , and did not count the exact number of L1-L3 worms in each well . Precise counts would have allowed us to calculate the true proportion of worms which escaped growth suppression and thus more accurately determine which chemicals promoted statistically meaningful growth arrest rescue values as compared to random escapee wells . We instead addressed this by performing a post-hoc analysis using an assumption of a standard number of worms plated in each well ( n = 20 , where realistically this number ranged from 15 to 25 ) . We estimated the % arrest based on this assumption , using a formula where the number of L1-L3 larvae = 20 – ( true count of L4-adult ) and found that 142 of the original 278 chemical wells were statistically different from random escapee wells ( Figure 2—figure supplements 2–3 ) . Given that 88% of those identified as statistically significant were in our prioritization group of 145 , this provided us with confidence that they merited additional examination . We re-tested the 145 chemicals in duplicate and found that 74 reproducibly suppressed the nemadipine-A-induced growth suppression of unc-68 mutants ( Figure 2B ) . Among these 74 ‘hits’ , we identified five wactives and seven non-wactives , 44 kinase inhibitors , and 18 compounds from the MicroSource library ( Supplementary file 1 ) . For several of the strongest suppressors , we performed dose-response analyses to confirm the effect of suppression ( Figure 2—figure supplement 4A ) . Many of the chemicals in our libraries have overlapping targets and/or functions ( e . g . there are several EGFR inhibitors in the kinase library and several steroid hormones in the US Drug Collection ) . Therefore , we reasoned that hits that are overrepresented among groups of structurally and/or functionally related chemicals should be prioritized for further testing in zebrafish . To determine enrichment based on structural similarity , we compared the chemical fingerprints of the hits with those of the chemicals in each library by hierarchical clustering of their Tanimoto scores ( Figure 2C and Figure 2—figure supplement 5; Burns et al . , 2015 ) . This method shows how structurally similar chemicals cluster with one another . Using this methodology , we found that p38 inhibitors were significantly overrepresented ( p=0 . 0022 , Fisher’s exact test ) when considering the total number of p38 inhibitors in the kinase inhibitor library . Interestingly , structurally dissimilar p38 inhibitors suppressed the phenotype ( clusters 5 , 6 , 16 , and 19 in Figure 2C ) , suggesting that inhibition of their common target was responsible for the activity . We then applied the same post-hoc analysis to the re-tested molecules as we did with the primary screen to visualize changes in developmental stage distribution after chemical treatment ( Figure 2—figure supplement 4B–G ) . As shown , p38 inhibitors were among the strongest suppressors identified from the inhibitor set . Finally , we validated p38 as a target by knocking down the p38 MAPK orthologs pmk-1 , pmk-2 , pmk-3 alone and in combination , and we observed a modest increase in the proportion of unc-68 mutants that escaped nemadipine-A induced growth arrest with each of the four conditions ( Figure 2D ) . Additionally , from the 44 hits in the kinase library we found that EGFR , PERK , and PLK1 inhibitors were overrepresented ( p=0 . 0060 , p=0 . 0109 , p=0 . 0326 , respectively ) . Multiple GSK3 ( n = 9 ) and PI3K ( n = 3 ) inhibitors also strongly suppressed the phenotype , but these were not statistically overrepresented ( p=0 . 4093 and p=0 . 1027 , respectively ) . Among the 18 positive hits from the MicroSource library , several chemical classes were overrepresented ( Figure 2—figure supplement 5 ) : riboflavins ( *p=0 . 0002 ) , surfactants ( p=0 . 0002 ) , benzophenones ( p=0 . 0018 ) , anthracenes ( p=0 . 0279 ) , salicylates ( p=0 . 0416 ) , and tricyclic antihistamines ( p=0 . 0416 ) . Interestingly , two DHPR inhibitors out of nine present in the library ( *p=0 . 0063 ) were identified as suppressors of the growth arrest induced by nemadipine-A , itself a DHPR inhibitor . This may reflect competition for receptor binding which diminishes the effect of nemadipine-A , or perhaps a higher effective concentration which becomes agonistic to the receptor . Of note , riboflavin and riboflavin 5-phosphate sodium both strongly suppressed growth arrest ( Figure 2—figure supplement 4A , E ) and they appear to be structurally distinct from every other chemical in the US Drug Collection ( cluster 11 , Figure 2—figure supplement 5 ) . Similarly , thiostrepton suppressed growth arrest strongly ( Figure 2—figure supplement 4E ) , is a structurally unique molecule and overrepresented among the 18 hits ( p=0 . 0141 ) . Altogether , the overrepresented groups of chemicals and the structurally unique molecules were prioritized for follow-up testing in ryr1b mutant zebrafish . In parallel , we performed a screen in a zebrafish model of RYR1-RM . Zebrafish have two ryr1 paralogs . Recessive mutations in ryr1a cause no overt phenotype , while recessive mutations in ryr1b result in abnormal swim behavior and early lethality after 11–13 days of life ( Hirata et al . , 2007 ) . ryr1a; ryr1b double mutants exhibit no movement and have a median survival of 5 days of life ( Figure 3—figure supplement 1A–B; Chagovetz et al . , 2019 ) . We used the double mutants for our screen because of their obvious motor phenotype and because variability in the ryr1b single mutant motor phenotype precluded large-scale screening ( Figure 3—figure supplement 1C ) . We screened 436 kinase inhibitors at 10 μM from the DiscoveryProbe Kinase Inhibitor Library ( ApexBio ) and 1360 drugs at 10 μM from the US Drug Collection ( MicroSource Discovery Systems ) , using improvement in motility of the ryr1a; ryr1b double mutants ( as measured by touch-evoked escape response ) as the primary outcome measure . We did not identify a single compound that was able to promote movement in these fish . Using our prioritized list based on enrichment modeling , we next sought to determine if any positive hits from our C . elegans screen could improve phenotypes in either the ryr1a; ryr1b double mutants or the ryr1b single mutant zebrafish . We used motility as our outcome measure , examining both touch-evoked escape response and optogenetically induced swimming as per our previously established testing methodology ( Sabha et al . , 2016 ) . Among the overall group of 74 hits , we tested 4/5 wactives , 7/7 non-wactives , 16/18 MicroSource drugs , and several inhibitors from the six classes of kinase inhibitors over-represented among the hits . First , we tested all of these at a single concentration of 10 μM for the ability to promote movement in the ryr1a; ryr1b double mutants , which in the untreated state lack any movement . Consistent with our large-scale screen done with the ryr1a; ryr1b mutants , none of these chemicals improved the double mutant phenotype ( Figure 3 ) . We then tested all of these chemicals for the ability to suppress the abnormal touch-evoke escape response of ryr1b mutants , but these chemicals did not modulate this phenotype either . However , the utility of this phenotype as a readout of improvement may be unreliable given that it only transiently exists between 3–4 days of life , after which time the ryr1b mutants are indistinguishable by eye from their WT siblings ( Hirata et al . , 2007 ) . Hence , we assayed many of these chemicals on the ryr1b single mutants using our quantitative optogenetic motility assay . We tested for chemical-genetic interactions , which we defined as unexpected larval movement not resulting from the predicted combined effects of genotype and chemical treatment ( see Methods ) . In other words , we assessed whether the movement speed of ryr1b mutants treated with a chemical was different from that which could be predicted from the combination of the ryr1b mutant effect plus the effect of the chemical on wild type . The p38 inhibitors SB239063 and PH-797804 decreased motility in wildtype controls , but had no effect on ryr1 mutants ( Figure 4A–B ) . The absence of effects in ryr1 mutant models stands in contrast to predicted additive and multiplicative effects for these compounds ( Figure 4A'–B' ) , suggesting that RyR1-related pathways mediate the effects of SB239063 and PH-797804 . Additionally , we found significant positive interactions for three wactives , a PI3K inhibitor , and the anti-psoriatic drug anthralin ( Figure 4—figure supplement 1A–E ) . We did not observe chemical-genetic interactions with two additional p38 inhibitors , SB203580 and SB202190 ( Figure 4C–D ) or with 18 other ‘hits’ ( Figure 4—figure supplements 1–3F–W ) . Notably , we did not identify any chemicals that significantly improved ryr1b mutant movement speed relative to untreated mutant controls . We sought to examine the potential translatability of our findings to mammalian models of RYR1-RM . To accomplish this , we tested the effect of two p38 inhibitors on RyR1-dependent Ca2+ release in C2C12 mouse myotubes . We examined this in wild type C2C12 cells and in a C2C12 Ryr1 knockout line that we created using CRISPR/Cas9 gene editing . This new line contains a bi-allelic frameshift deletion mutation in Ryr1 ( which we refer to as ‘KO’ ) . Successful targeting of the Ryr1 locus was demonstrated by Sanger sequencing , lack of off-target mutations verified by whole genome sequencing , and absence of RyR1 protein expression confirmed by western blot analysis ( Figure 5—figure supplement 1 ) . We measured intracellular calcium release from RyR1 in response to acute application of 10 mM caffeine ( Tong et al . , 1997; Meissner , 2017 ) in wild type control and Ryr1 KO C2C12 myotubes after 24 hr incubation with either SB203580 or SB202190 . As expected , Ryr1 KO myotubes treated with DMSO vehicle control lacked caffeine-induced Ca2+ release ( Figure 5A–B ) . KO myotubes treated with SB203580 or SB202190 , however , exhibited a dose dependent increase in caffeine-induced calcium release ( Figure 5A–B ) . Conversely , these chemicals impaired caffeine-induced calcium release in control WT C2C12 myotubes ( Figure 5A–B ) . To interrogate the pathway specificity of SB203580 or SB202190 , and to corroborate their positive effect , we examined caffeine-induced calcium release in the setting of siRNA knockdown of p38 isoforms α , β , and γ . Using commercially available siRNA we achieved roughly 50–80% knockdown of p38 MAPK targets as compared to non-targeting negative control siRNA ( Figure 5—figure supplement 2 ) . In Ryr1 KO C2C12 cells , siRNA knockdown of Mapk11 ( p38β ) promoted increased caffeine-induced calcium release in KO cells versus negative control siRNA ( Figure 5C ) . Knockdown of Mapk14 ( p38α ) and Mapk12 ( p38γ ) also increased calcium release in Ryr1 KO myotubes but to a lesser degree . These data thus suggest that in KO C2C12 cells , p38 inhibition ( either via chemical or genetic inhibition ) is able to promote intracellular calcium release independent of RyR1 . This is consistent with the ability of p38 inhibitors and RNAi to suppress the unc-68 ( i . e . RyR1 null ) phenotype in C . elegans . Unlike with SB203580 or SB202190 treatment , siRNA knockdown of p38 isoforms did not impair Ca2+ release in WT cells ( Figure 5C ) , perhaps suggesting that inhibition may be partially caused by off-target effects of the chemicals . Alternatively , this may instead be a reflection of the incomplete p38 knockdown we achieved with siRNA , or could reflect siRNA toxicity , as caffeine-induced calcium release from WT cells was lower with control siRNA treatment when compared to DMSO-treated conditions . The fact that p38 chemical inhibitors negatively modulate swim behavior in WT zebrafish and caffeine-induced calcium release in C2C12 cells opens the possibility that they may serve as modifiers of phenotypes related to RyR1 hyperexcitability . To test this , we examined calcium dynamics in myotubes from the YS mouse model of RYR1-RM . The YS model contains a point mutation analogous to the Y522S mutation found in patients with malignant hyperthermia and central core pathology ( Quane et al . , 1994 ) . The YS mutation enhances both the sensitivity of RyR1 to activators ( e . g . DHPR , caffeine , 4-chloro-m-cresol ) and the temperature-dependence of RyR1 Ca2+ leak , alterations that underlie the MH susceptibility and exertional heat stroke phenotypes of these mice ( Chelu et al . , 2006; Durham et al . , 2008; Lanner et al . , 2012 ) . We examined if the temperature dependent increase in resting myoplasmic Ca2+ concentration in YS myotubes was abrogated by the p38 inhibitor SB203580 . Overnight incubation of YS myotubes with 10 μM SB203580 significantly reduced the temperature-dependent increase in resting Ca2+ observed in YS myotubes ( Figure 5D ) . We established a unique ‘multi species’ pipeline for drug discovery for RYR1-RM . This platform is rapid and robust , and provides the ability to examine multiple different types of in vivo models . Our study lays the ground work for its future use in RYR1-RM drug development , and for establishment of similarly platforms for other rare diseases . All zebrafish experiments were performed in accordance with all relevant ethical regulations , specifically following the policies and guidelines of the Canadian Council on Animal Care and an institutionally reviewed and approved animal use protocol ( #41617 ) . No additional ethical approval was required for our experiments with the invertebrate nematode worm C . elegans . Chemical libraries used in this study include the US Drug Collection ( 1280 compounds; MicroSource Discovery Systems Inc ) , the DiscoveryProbe Kinase Inhibitor Library ( 436 compounds; APExBIO ) , the GlaxoSmithKline Published Kinase Inhibitor ( PKI ) Set ( 880 compounds; William Zuercher ) , and a collection of 770 worm-bioactives ( ‘wactives’ ) that were identified in a screen for bioactive small molecules in C . elegans ( Burns et al . , 2015 ) . Nemadipine-A ( #5619779 ) , optovin analog 6b8 ( #5707191 ) , and wactives/non-wactives ( ID numbers in Supplementary file 1 ) were purchased from ChemBridge . Chemicals from the US Drug Collection identified as ‘hits’ in the C . elegans screen were purchased from Sigma-Aldrich for testing in zebrafish . Kinase inhibitors representative of those identified as ‘hits’ in the C . elegans screen , including p38 inhibitors , were purchased individually from APExBIO or selected from the DiscoveryProbe Kinase Inhibitor Library for testing in zebrafish and cell lines . All animals were cultured under standard methods at 20°C ( Burns et al . , 2015 ) . The wild type ( N2 ) , and unc-68 ( r1161 ) and unc-68 ( r1162 ) ( TR2170 and TR2171 , respectively ) strains of Caenorhabditis elegans were obtained from the C . elegans Genetics Center ( University of Minnesota ) . The ‘thrashing assay’ is performed by counting the waveforms propagated by individual C . elegans in one minute as previously described ( Maryon et al . , 1996 ) . The protocol for the 96-well liquid-based chemical screens was described previously ( Burns et al . , 2015 ) . Briefly , nematode growth media ( NGM; for recipe see Burns et al . , 2015 ) was used to concentrate saturated E . coli HB101 bacteria two-fold ( NGM-HB101 ) . Nemadipine-A ( NEM ) was added to NGM+HB101 to a final concentration of 31 . 25 μM/0 . 5% DMSO ( NEM+NGM+HB101 ) . A total of 40 μL of NEM+NGM+HB101 was dispensed into each well of a 96-well plate , and 300 nL of chemical dissolved in DMSO was pinned into the wells using a 96-well pinning tool ( V and P Scientific ) . At Day 0 , approximately 20 synchronized first larval-stage ( L1 ) , unc-68 ( r1162 ) ( TR2171 ) worms obtained from an embryo preparation were added to each well in 10 μL of M9 buffer ( Burns et al . , 2015 ) . The unc-68 ( r1162 ) strain TR2171 was selected for screening because its growth rate was comparable to N2 wild type whereas strain TR2170 grew noticeably slower than N2 . The final concentration of dimethyl sulfoxide ( DMSO ) in the wells was 1% v/v . Plates were sealed with parafilm , wrapped in damp paper towels to reduce evaporation in wells , and incubated for 6 days at 20°C while shaking at 200 rpm ( New Brunswick I26/I26R shaker , Eppendorf ) . A stereomicroscope was used to assess developmental stage after 6 days incubation . All preliminary screens and re-tests were performed in duplicate . Post-hoc statistical analysis was performed by assuming 20 worms in each well and assigning each individual worm in L1-L3 , L4 , or adult stage a rank score of 1 , 2 , and 3 , respectively . Next , the developmental stage distributions were compared to the random escapee wells using a nonparametric Kruskal-Wallis test with Dunn’s multiple comparisons post-test in GraphPad Prism 8 . RNAi knock-down was carried out in 96-well plate liquid culture as previously described ( Lehner et al . , 2006 ) . Synchronized L1-stage unc-68 ( r1162 ) mutant worms were fed E . coli HT115 bacteria expressing double-stranded RNA targeting pmk-1 , pmk-2 , pmk-3 or a combination of the three from the Source BioScience RNAi library . The E . coli HT115 strain carrying the L4440 empty vector was used as a control for RNAi machinery induction , and pop‐1 RNAi which produces a severe embryonic lethality phenotype was used as a control for RNAi induction efficiency . A total of 40 μL of bacterial suspension containing nemadipine-A or DMSO was dispensed into each well of a flat-bottomed 96-well plate . Approximately 25 unc-68 ( r1162 ) L1 worms were dispensed into the wells in 10 μL of M9 buffer . The final concentration of nemadipine-A in the wells was 25 μM in 0 . 4% DMSO v/v . Plates were sealed with parafilm and incubated at 20°C with shaking at 200 rpm . On Day six the plates were observed under a dissection microscope and the distribution of developmental stages in each condition was assessed . Each experiment was performed in quadruplicate and repeated at least three times . For statistical analysis , each individual worm in L1-L3 , L4 , or adult stage was assigned a rank score of 1 , 2 , and 3 , respectively , and the developmental stage distributions were compared to the negative empty vector control using a nonparametric Kruskal-Wallis test with Dunn’s multiple comparisons post-test in GraphPad Prism 8 . Pairwise Tanimoto coefficient scores were calculated for each chemical ‘hit’ and compound in the screening library using OpenBabel ( http://openbabel . org ) as previously described ( Burns et al . , 2015 ) . Tanimoto coefficient scores were hierarchically clustered in Cluster 3 . 0 using an unweighted Euclidean distance similarity metric with complete linkage clustering and visualized in TreeView as previously described ( Folts et al . , 2016 ) . For enrichment analysis , chemicals were counted based on the number of structurally similar members in each cluster ( Tanimoto scores were >0 . 55 for the majority of members in a given cluster ) and Fisher’s exact test ( GraphPad Prism 8 ) was used to calculate enrichment . In this study , we used ryr1a and ryr1b mutant alleles that have been previously characterized ( Hirata et al . , 2007; Chagovetz et al . , 2019 ) . Both mutants result in loss of RYR1 protein expression from the mutant allele . For follow-up screens , we generated single ryr1b-/- mutants via incross of ryr1b+/- carriers . Phenotypic analysis of all ryr1 mutants was performed on a stereomicroscope . All chemical stocks were prepared in DMSO and added to egg water at 0 . 1–0 . 5% of the final volume to prepare working concentrations ( depending on chemical solubility ) . Equal volumes of vehicle solvent were used in all conditions for a single assay . Note that methylene blue was not added to the egg water . Dishes or 96-well plates were sealed with parafilm , wrapped in aluminum foil , and incubated at 28 . 5°C until the assay date . Different volumes and culture dish formats were used depending on the endpoint assay . The US Drug Collection ( 1280 compounds ) and the DiscoveryProbe Kinase Inhibitor Library ( 436 compounds ) were screened for chemicals that could promote motility in ryr1a; ryr1b double mutants . Library stocks of 10 mM in DMSO were added to 0 . 1% of the final volume to egg water to prepare a 10 μM screening concentration . Specifically , 150 µL of egg water was added to every well of two separate 96-well plates . Double mutants were generated by in-cross of ryr1a-/-;ryr1b+/- mutants . Embryos were manually dechorionated at one dpf . Then at two dpf two double mutant embryos were added to each well of one plate while two phenotypically wild type siblings were added to the second plate as a control . We reasoned that testing two larvae would be sufficient to detect if a chemical rescued the striking immotile phenotype . Care was taken to lower embryos to the bottom of a Pasteur pipette tip and deposit embryos into the well by surface tension to minimize changes to well volume . Next , 250 µL of the drug library was prepared at 40 µM working concentration in a separate 96-well plate by adding 1 µL of the 10 mM stock to 249 µL of egg water . Next , 50 µL of the 40 µM working concentration was added to the 150 µL water containing embryos to give a final concentration of 10 µM drug . After 24 hr incubation in chemical , motility of 3 dpf wild type and double mutant larvae was assessed by touch-evoked response ( Hirata et al . , 2007 ) under a stereomicroscope . At three dpf , ryr1b mutants were segregated from wild type ( ryr1b+/+ or ryr1b+/- ) siblings based on their phenotype and distributed into sterile 6 cm tissue culture dishes containing 10 mL of egg water plus chemical . Please note that care was taken to evenly distribute larvae from one pair of parents across all conditions , that is sibling-matching , to minimize potential effects arising from clutch-to-clutch variability . Additionally , dysmorphic or underdeveloped larvae were not used . All assays were performed at 4 dpf after 24 hr incubation using the ZebraBox platform ( ViewPoint ) and 10 μM optovin analog 6b8 as previously described ( Sabha et al . , 2016 ) . Using GPower Version 3 . 1 ( Faul et al . , 2009 ) , we calculated that a sample size n = 9 larvae per group would allow us a 99% probability of detecting a difference between WT and ryr1b mutants given their group means and standard deviations in movement speed ( mm/s ) . For the majority of chemicals , two independent experiments with n = 9 or 10 larvae per group were performed . To compare effects from independent experiments , movement speed of individual larvae in each group were normalized to the average movement speed of DMSO treated WT siblings from the same assay . This would allow us to estimate chemical-genetic interactions based on an unexpected change in the difference between WT and ryr1b following chemical treatment . We adapted the formulas for calculating additive and multiplicative models of genetic interactions Figure 1 in Baryshnikova et al . ( 2013 ) to calculate and visualize expected chemical-genetic interactions based on larval movement speed . Statistical significance for the interaction was calculated by two-way ANOVA with Tukey’s multiple comparison post-test ( GraphPad Prism 8 ) . The original C2C12 ( ATCC CRL1772 ) was purchased from American Type Culture Collection ( Manassas , VA , USA ) . The sgRNA sequence ( 5’-AGGAGAGAAGGTTCGAGTTG-3’ ) against Ryr1 in Exon six was designed by the online CRISPR Design Tool ( http://tools . genome-engineering . org ) and cloned at the BbsI site into pSpCas9 ( BB ) −2A-Puro ( PX459 ) V2 . 0 ( Addgene plasmid ID: 62988 ) . The CRISPR plasmids are transfected into C2C12 cells by electroporation ( Amaxa Nucleofector Lonza ) . Seventy-two hours later , the transfected cells were selected in medium containing 4 µg/mL puromycin for three days and then subcloned into 96-well plates . Once at sufficient cell density , the genomic DNA of subclones was analyzed by Sanger sequencing . Primers used were ( forward ) 5’-GTGTGACGGGAGTCCCAAAT-3’ and ( reverse ) 5’-ACTGGGCATGCCAATGATGA-3’ . Cells were tested and found negative for mycoplasma . Protein was isolated in RIPA buffer from C2C12 myotubes at 5 days after starting differentiation . Cells are incubated with 100 nM SB202190 , 10 μM SB203580 , DMSO and without treatment for 24 hr from day4 . A total of 30 μg of total protein was run on either 4 . 5% SDS-acrylyamide gel for RyR1 or 15% SDS-acrylamide gel for β-actin . Blots were run for 3–3 . 5 hr for Ryr1 or for 2–2 . 5 hr for βactin at 100 V and transferred overnight at 20 V . The membrane was blocked in 3% bovine serum albumin ( BSA ) in Tris Buffered Saline with Tween 20 ( TBST ) for 1 hr at room temperature before incubating with primary antibodies overnight at 4°C . Antibodies used were anti-Ryanodine receptor antibody 34C ( Developmental Studies Hybridoma Bank ) at 1:100 dilution and anti-beta actin antibody ( Abcam ) at 1:5000 dilution . After three washes in TBST , blots were incubated with Anti-Mouse IgG-HRP conjugate ( Bio-Rad ) at 1:10000 dilution . Blots were imaged by chemiluminescence ( Western Lightning Plus-ECL , PerkinElmer ) using the Gel Doc XR + Gel Documentation System ( BioRad ) , and band signal intensities determined using ImageLab software ( BioRad ) . C2C12 myoblasts were seeded at 2 . 5 × 104 cells·cm−2 density in 24-well plates ( Falcon ) containing glass coverslips coated with 5 μg/cm2 collagen and grown in DMEM with 20% ( v/v ) fetal bovine serum ( FBS ) . Once cells reached 80–90% confluency , media was changed to differentiation media ( DMEM with 2% ( v/v ) horse serum , 1 μg/mL insulin and 50 μg/mL gentamicin ) representing Day 0 of differentiation . At Day 4 , myotubes were transfected for 6 hr with 50 pmol ON-TARGETplus siRNA ( Horizon Discovery ) against p38α/Mapk14 ( 5’-GCAAGAAACTACATTCAGT-3’ ) , p38β/Mapk11 ( 5’-ATGAGGAGATGACCGGATA-3’ ) , p38γ/Mapk12 ( 5’-AATGGAAGCGTGTGACTTA-3’ ) or a non-targeting negative control ( 5’-TGGTTTACATGTCGACTAA-3’ ) in complex with 5 μL Lipofectamine RNAi/MAX ( Invitrogen ) in 500 μL Optimem solution ( Gibco ) . Fresh differentiation medium was added until Day five when myotubes were transfected again following the same protocol . Finally , myotubes were maintained in differentiation media until Day 6 . At Day 6 , myotubes were either used for calcium measurements or qPCR analysis . It is important to note that siRNA knockdown was started after differentiation into myotubes because siRNA knockdown of either the α , β , or γ isoforms of p38 prevents C2C12 myoblast differentiation into myotubes ( Wang et al . , 2008 ) . RNA was extracted with the RNeasy Mini kit ( Qiagen ) from n = 3 independent myotube cultures for each siRNA condition . Each sample was run in triplicate using SYBR Green master mix Applied Biosystems StepOne system ( ThermoFisher ) . Comparative ΔΔCT method was used to determine relative expression of target genes after siRNA knockdown . Briefly , target gene expression was normalized to endogenous control gene Tbp as an endogenous control to obtain ΔCT values . Next , relative fold change in gene expression for each siRNA condition was calculated by normalizing to the mean ΔCT value for negative control siRNA . The following primer pairs were used: Tbp ( Forward: 5’-TGCTGCAGTCATCATGAG-3’; Reverse: 5’-CTTGCTGCTAGTCTGGATTG-3’ ) , p38α/Mapk14 ( Forward: 5’-CAGCAGATAATGCGTCTGACGGG-3’; Reverse: 5’-GCGAAGTTCATCTTCGGCATCTGG-3’ ) , p38β/Mapk11 ( Forward: 5’-CCAGCAATGTAGCGGTGAACGAG-3’; Reverse: 5’-GCATGATCTCTGGCGCCCGGTAC-3’ ) , p38γ/Mapk12 ( Forward: 5’-CACTGAGGATGAACCCAAGGCC-3’; Reverse: 5’-CTCCTAGCTGCCTAGGAGGCTTG-3’ ) . Intracellular Ca2+ measurements were obtained from Fura-2 ( Invitrogen ) AM-loaded myotubes as described previously ( Goonasekera et al . , 2007 ) . Briefly , myotubes were differentiated for 5–6 days on glass bottom dishes and loaded with 5 μM Fura-2 AM for 45 min at 37°C in a normal rodent Ringer’s solution consisting of 145 mM NaCl , 5 mM KCl , 2 mM CaCl2 , 1 mM MgCl2 , 10 mM HEPES , pH 7 . 4 . Coverslips of Fura-2–loaded cells were then mounted in a tissue chamber on the stage of an epifluorescence-equipped inverted microscope ( Zeiss ) . Cells were sequentially excited at 340- and 380 nm wavelength and fluorescence emission at 510 nm was collected using a high-speed CCD camera ( Hamamatsu ) . The results are presented as the ratio of 340/380 nm . Maximal increase or peak change in intracellular Ca2+ by induction of 10 mM caffeine was defined as the difference between peak and 10 s of baseline fluorescence ratios prior to addition of caffeine . To better visualize differences in peak change across multiple treatment groups , 340/380 ratios for each individual myofiber were normalized to their own 10 s of baseline 340/380 ratios prior to addition of caffeine . Data are also presented without normalization in Figure 5—figure supplement 1D–F . Resting [Ca2+]i was measured in primary myotubes derived from Ryr1Y524S/+ ( YS ) mice cultured on glass bottom dishes using Fura-2 ( Invitrogen ) and a temperature controlled chamber ( Lanner et al . , 2012 ) . The day before measurements , myotube cultures were incubated overnight in either vehicle or 10 μM SB203580 . Fura-2 was excited at 340 nm and 380 nm and images of myotubes were collected using a CCD camera connected to a TILL monochromator ( TILL Photonics Inc ) . The myotubes were first imaged at temperature just above room temperature ( 25°C ) . The temperature of the bathing solution was then raised to 37°C and the same cells were re-imaged again at 37°C for paired statistical comparison . Ratio images of 340/380 nm for Fura-2 were created using TILLvisION software and analyzed offline using ImageJ software . Resting free calcium concentration ( [Ca2+]i ) was calculated using a calibration curve of Fura-2 as described previously ( Lanner et al . , 2012 ) . Tests of statistical significance were performed using Microsoft Office Excel 2008 ( Microsoft ) and GraphPad Prism eight for Mac OSX ( GraphPad Software ) . Differences were considered to be statistically significant at p<0 . 05 ( ) , p<0 . 01 ( ) , or p<0 . 001 ( * ) . All data unless otherwise specified are presented as mean ± SEM .	Muscle cells have storage compartments stuffed full of calcium , which they release to trigger a contraction . This process depends on a channel-shaped protein called the ryanodine receptor , or RYR1 for short . When RYR1 is activated , it releases calcium from storage , which floods the muscle cell . Mutations in the gene that codes for RYR1 in humans cause a group of rare diseases called RYR1-related myopathies . The mutations change calcium release in muscle cells , which can make movement difficult , and make it hard for people to breathe . At the moment , RYR1 myopathies have no treatment . It is possible that repurposing existing drugs could benefit people with RYR1-related myopathies , but trialing treatments takes time . The fastest and cheapest way to test whether compounds might be effective is to try them on very simple animals , like nematode worms . But even though worms and humans share certain genes , treatments that work for worms do not always work for humans . Luckily , it is sometimes possible to test whether compounds might be effective by trying them out on complex mammals , like mice . Unfortunately , these experiments are slow and expensive . A compromise involves testing on animals such as zebrafish . So far , none of these methods has been successful in discovering treatments for RYR1-related myopathies . To maximize the strengths of each animal model , Volpatti et al . combined them , developing a fast and powerful way to test new drugs . The first step is an automated screening process that trials thousands of chemicals on nematode worms . This takes just two weeks . The second step is to group the best treatments according to their chemical similarities and test them again in zebrafish . This takes a month . The third and final stage is to test promising chemicals from the zebrafish in mouse muscle cells . Of the thousands of compounds tested here , one group of chemicals stood out – treatments that block the activity of a protein called p38 . Volpatti et al . found that blocking the p38 protein , either with drugs or by inactivating the gene that codes for it , changed muscle calcium release . This suggests p38 blockers may have potential as a treatment for RYR1-related myopathies in mammals . Using three types of animal to test new drugs maximizes the benefits of each model . This type of pipeline could identify new treatments , not just for RYR1-related myopathies , but for other diseases that involve genes or proteins that are similar across species . For RYR1-related myopathies specifically , the next step is to test p38 blocking treatments in mice . This could reveal whether the treatments have the potential to improve symptoms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "medicine" ]	2020	Identification of drug modifiers for RYR1-related myopathy using a multi-species discovery pipeline
Pigment Epithelium Derived Factor ( PEDF ) is a secreted factor that has broad biological activities . It was first identified as a neurotrophic factor and later as the most potent natural antiangiogenic factor , a stem cell niche factor , and an inhibitor of cancer cell growth . Numerous animal models demonstrated its therapeutic value in treating blinding diseases and diverse cancer types . A long-standing challenge is to reveal how PEDF acts on its target cells and the identities of the cell-surface receptors responsible for its activities . Here we report the identification of transmembrane proteins PLXDC1 and PLXDC2 as cell-surface receptors for PEDF . Using distinct cellular models , we demonstrate their cell type-specific receptor activities through loss of function and gain of function studies . Our experiments suggest that PEDF receptors form homooligomers under basal conditions , and PEDF dissociates the homooligomer to activate the receptors . Mutations in the intracellular domain can have profound effects on receptor activities . Employing physiological pathways to impede pathological processes has been a fruitful approach in developing effective therapeutics for human disease . There exists a natural factor that can inhibit pathogenesis of several major diseases and has surprisingly diverse therapeutic value . This factor is called Pigment Epithelium-Derived Factor ( PEDF ) ( Dawson et al . , 1999; Tombran-Tink and Barnstable , 2003 ) and was originally identified as a strong protective factor for neurons ( Tombran-Tink and Barnstable , 2003 ) . It was also initially known as EPC-1 , a factor that is downregulated by more than 100-fold in aged compared to young human fibroblasts ( Pignolo et al . , 1993 ) . In an unbiased search for new antiangiogenic factors , PEDF was identified as the most potent endogenous inhibitor of angiogenesis ( Dawson et al . , 1999 ) . PEDF inhibits endothelial cell migration and angiogenesis even in the presence of strong proangiogenic factors ( Dawson et al . , 1999 ) . It specifically targets new vessel growth without affecting pre-existing vessels . In numerous animal models , PEDF has been shown to have potent therapeutic effects in treating several major human diseases through its neurotrophic , anti-angiogenic , antitumorigenic and antimetastatic activities . In addition to treating major blinding diseases such as ischemia-induced retinopathy , diabetic retinopathy , glaucoma and age-related macular degeneration ( Stellmach et al . , 2001; Semkova et al . , 2002; Miyazaki et al . , 2011 ) , PEDF has been shown to inhibit the growth of a wide variety of cancer types including melanoma , neuroblastoma , osteosarcoma , hepatoblastoma , Lewis lung carcinoma , chondrosarcoma , gastric carcinoma , glioma , Wilm's tumor , prostate cancer , and pancreatic cancer ( Doll et al . , 2003; Ek et al . , 2006; Fernandez-Garcia et al . , 2007; Broadhead et al . , 2009 ) . In addition , PEDF has also been identified as a stem cell niche factor ( Pumiglia and Temple , 2006; Ramirez-Castillejo et al . , 2006; Andreu-Agullo et al . , 2009; Elahy et al . , 2012 ) and an anti-inflammatory factor ( Zamiri et al . , 2006; Zhang et al . , 2006b ) . PEDF is widely expressed in many tissues such as the eye , brain , spinal cord , bone , liver , heart and lung . PEDF is also naturally present in the blood ( Petersen et al . , 2003 ) . PEDF level was found to decrease during cellular senescence and aging ( Pignolo et al . , 1993; Tombran-Tink et al . , 1995; Francis et al . , 2004 ) and in many pathological conditions . A significant decrease in PEDF level in the eyes has been observed in patients with age-related macular degeneration and diabetic retinopathy , two major blinding diseases characterized by neovascularization ( Ogata et al . , 2001; Spranger et al . , 2001; Holekamp et al . , 2002; Ogata et al . , 2002; Boehm et al . , 2003 ) . PEDF levels were also found to decrease with age in human eyes ( Ogata et al . , 2004; Smith and Steinle , 2007; Steinle et al . , 2008 ) . PEDF expression has been inversely related to metastasis in a variety of cancer types such as gliomas ( Guan et al . , 2003 ) , lymphangiomas ( Sidle et al . , 2005 ) , hepatoma ( Matsumoto et al . , 2004 ) , melanoma ( Orgaz et al . , 2009 ) , lung cancer ( Zhang et al . , 2006a ) , pancreatic cancer ( Uehara et al . , 2004 ) , and prostate cancer ( Halin et al . , 2004 ) . A long-standing challenge has been to understand how PEDF acts on different cell types and its fundamental transmembrane mechanisms . Uncovering the transmembrane pathways of PEDF would lead to a better understanding of its fundamental mechanisms and the development of new therapeutic strategies . After many years of effort in trying both existing and new strategies on a variety of native tissues and cell types , we were unable to identify the PEDF receptor , likely due to its low abundance and transient nature of expression . Since PEDF's actions do not match any well-characterized receptors , we reasoned that its receptor is likely new and uncharacterized . We tested human orphan receptors and transmembrane domain proteins of unknown function for their ability to bind native PEDF on the cell surface and found two transmembrane domain proteins that confer cell-surface PEDF binding and have other properties expected of PEDF receptors . These two membrane proteins both have a large extracellular domain , a transmembrane domain , and an intracellular domain , and share about 50% homology . These two membrane proteins are called plexin domain containing 1 ( PLXDC1 ) and plexin domain containing 2 ( PLXDC2 ) . Gene and protein expression studies have revealed overlapping but distinct tissue expression patterns of PLXDC1 ( St Croix et al . , 2000; Gaultier et al . , 2010 ) and PLXDC2 ( Leighton et al . , 2001; McMurray et al . , 2008; Miller-Delaney et al . , 2011; Boheler et al . , 2014 ) . A prerequisite for a cell-surface receptor is the ability to confer cell-surface binding to the extracellular ligand . We found that expression of PLXDC1 or PLXDC2 confers extracellular PEDF binding to live cells ( Figure 1A–C ) . PLXDC1 or PLXDC2 with a deletion of the extracellular domain no longer binds PEDF , while deletion of the intracellular domain has no effect on PEDF binding ( Figure 1D–F ) . The domain structures of PLXDC1 and PLXDC2 are depicted in Figure 1—figure supplement 1 . Staining with an extracellular epitope on live cells showed that all these proteins are expressed on the cell surface ( Figure 1G–L ) . These experiments demonstrated that PEDF binds to cell-surface transmembrane domain proteins PLXDC1 and PLXDC2 through their extracellular domains . 10 . 7554/eLife . 05401 . 003Figure 1 . The binding of PEDF to PLXDC1 and PLXDC2 on cell surface . Upper panel: Binding of biotinylated PEDF to control HEK293 cells ( A ) or HEK293 cells transfected with PLXDC1 ( B ) , PLXDC2 ( C ) , intracellular domain deleted PLXDC2 ( D , PLXDC2-dC ) , extracellular domain ( ECD ) deleted PLXDC1 ( E ) , or ECD deleted PLXDC2 ( F ) . Binding was detected by stretavidin-alkaline phosphatase , shown as deep purple color . Lower panel: Live cell staining of an epitope tag of control HEK293 cells ( G ) or HEK293 cells transfected with PLXDC1 ( H ) , PLXDC2 ( I ) , intracellular domain deleted PLXDC2 ( J , PLXDC2-dC ) , ECD-deleted PLXDC1 ( K ) , or ECD deleted PLXDC2 ( L ) . All constructs have the epitope tag engineered after the secretion signal at the N-terminus . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 00310 . 7554/eLife . 05401 . 004Figure 1—figure supplement 1 . PLXDC1 and PLXDC2 schematic diagrams and alignment showing the definitions of domains . ( A ) Schematic diagrams of domains in PLXDC1 ( red ) and PLXDC2 ( orange ) . ( B ) Alignment of human ( h ) and mouse ( m ) PLXDC1 and PLXDC2 . Junctions of domains are indicated . Identical regions are highlighted in yellow . The junction between the secretion signal and the rest of the protein is indicated by a red arrowhead ( PLXDC1 ) or an orange arrow head ( PLXDC2 ) . The junction between domain A and B is indicated by the letter A in a circle . The junction between domain B and C is indicated by the letter B in a circle . The junction between domain C and D is indicated by the letter C in a circle . The end of domain D is indicated by the letter D in a circle . Domain B contains a region that is homologous to nidogen . Domain D includes the plexin homology domain . Putative location of the transmembrane domain is indicated by a black square . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 004 Studying PEDF receptors requires robust and accessible cellular assays for gain and loss of function studies . We used three cell types that respond robustly and reproducibly to PEDF as cellular models to study PEDF receptors: macrophage cell RAW267 . 4 , endothelial cell SVEC4-10 , and neuronal cell 661W . These three cell types represent three distinct cellular targets of PEDF . We found that both PLXDC1 and PLXDC2 are expressed in these three cellular models , consistent with previous knowledge that both RAW267 . 4 and SVEC4-10 cells express PLXDC1 ( Wang et al . , 2005; Gaultier et al . , 2010 ) . In macrophage RAW267 . 4 , PEDF is known to stimulate secretion of IL-10 , an anti-inflammatory cytokine ( Zamiri et al . , 2006 ) . To perform loss of function studies , we screened for siRNAs that can effectively knockdown PLXDC1 or PLXDC2 expression in RAW26 . 4 cells ( Figure 2—figure supplement 1 ) . The most effective siRNA was used in subsequent functional assays . We found that knocking down of either PLXDC1 or PLXDC2 led to a substantial decrease in PEDF response ( Figure 2A ) . Conversely , transfection of either receptor into macrophages further augments PEDF-induced secretion of IL-10 without increasing basal activity ( basal activity is defined as receptor activity without PEDF treatment ) ( Figure 2B ) . Either receptor lacking the cytoplasmic domain no longer has this activity , consistent with role of the cytoplasmic domain in cellular signaling ( Figure 2B ) . Tyrosine 481 in human PLXDC1 is a highly conserved residue in the cytoplasmic domain and is a potential phosphorylation site . PLXDC1 with a mutation of this single residue in the cytoplasmic domain ( Y481F ) has highly enhanced PEDF-mediated response without increasing the basal activity of the receptor ( Figure 2D , E ) . One potential mechanism is that phosphorylation of this residue dampens receptor signaling and the mutation prevents this inhibition . PLXDC1 transfected cell shows about 100% increased activity in response to 2 nM PEDF as compared to control cells , while PLXDC1-Y481F cells showed about 400% increased activity in response to PEDF ( Figure 2D ) . This profound stimulatory effect on PEDF signaling by mutating a single intracellular conserved residue in PLXDC1 supports its role in PEDF signaling . 10 . 7554/eLife . 05401 . 005Figure 2 . The roles of PLXDC1 and PLXDC2 in PEDF-induced IL-10 secretion by macrophage cell RAW267 . 4 . ( A ) siRNA-mediated knockdown of PLXDC1 or PLXDC2 substantially suppresses PEDF-stimulated IL-10 secretion . Activity of control transfected cells with PEDF treatment is defined as 1 . = p < 0 . 01 . ( B ) Transfection of either PLXDC1 or PLXDC2 cDNA enhances RAW cell's PEDF-stimulated IL-10 secretion , while PLXDC1 or PLXDC2 lacking the cytoplasmic domain ( PLXDC1-dC or PLXDC2-dC ) do not show significantly different secretion from control EGFP transfection . Activity of control cells with PEDF treatment is defined as 1 . = p < 0 . 01; NS = not significant . ( C ) Alignment of human , bovine , mouse , and rat PLXDC1 cytoplasmic tail and the location of the putative phosphorylated residue ( residue number according to human PLXDC1 ) . ( D ) Comparing PEDF-induced IL-10 secretion by RAW267 . 4 transfected with PLXDC1 and PLXDC1-Y481F . Mutation Y481F on the cytoplasmic tail of PLXDC1 greatly enhances its response to PEDF . Activity of control transfected cells without PEDF treatment is defined as 1 . ( E ) PEDF concentration-dependent stimulation of IL-10 secretion from control , PLXDC1 , and PLXDC1-Y481F transfected cells ( from D ) . Activity of PLXDC1-Y481F cells at 16 nM PEDF is defined as 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 00510 . 7554/eLife . 05401 . 006Figure 2—figure supplement 1 . Unbiased screening for effective siRNAs that knock down PLXDC1 or PLXDC2 expression in macrophage cell RAW267 . 4 . The definition and source of each siRNA is described in ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 006 Using endothelial cell SVEC4-10 , we established a highly effective and reproducible assay to study PEDF-mediated endothelial cell death ( Figure 3 ) . We found that PEDF—mediated cell death was completely suppressed by siRNA-mediated knockdown of PLXDC2 , but there was no suppression by siRNA knockdown of PLXDC1 ( Figure 3A and Figure 3—figure supplement 1 ) . Since the cytoplasmic domain of each receptor is expected to be involved in downstream signaling , we tested the effect of expression of the cytoplasmic domain fused to the transmembrane domain of another membrane protein ( DCC ) ( Stein and Tessier-Lavigne , 2001 ) without the extracellular domain of each receptor . Interestingly , transfection of PLXDC2 cytoplasmic domain linked to the DCC transmembrane domain is sufficient to cause cell death independently of PEDF . In contrast , the cytoplasmic domain of PLXDC1 does not have this activity ( Figure 3B ) . We also found that mutations in two cytoplasmic residues that are potential phosphorylation sites in the PLXDC2 tail enhance the effect of PLXDC2 in causing cell death ( Figure 3B ) . These experiments demonstrated that PLXDC2 is responsible for mediating PEDF-mediated cell death in this cell type , while there is no detectable role of PLXDC1 . In addition , the cytoplasmic domain of the PLXDC2 is sufficient to trigger downstream activity . 10 . 7554/eLife . 05401 . 007Figure 3 . The PLXDC2 dependence of PEDF's effect on endothelial cell SVEC4-10 . ( A ) PEDF promoted-cell death of SVEC4-10 cells is suppressed by siRNA-mediated knockdown of PLXDC2 , but not PLXDC1 . Survival of control siRNA tranfected cells without PEDF treatment is defined as 1 . Statistical significance is shown on the top . ** = p < 0 . 01 , and NS = not significant . ( B ) Left panel: Schematic diagrams of full length receptors and the fusion proteins for the receptor cytoplasmic tails . The cytoplasmic tail of the receptor is fused to the TM domain of DCC , which is fused to the secretion signal of alkaline phosphase at the N-terminus . Alignment of human , mouse , rat and bovine PLXDC2 cytoplasmic tails shows complete conservation ( bottom ) . Locations of potential phosphorylation sites are indicated . Right panel: Expression of cytoplasmic tail of PLXDC2 , but not the cytoplasmic tail of PLXDC1 promotes SVEC4-10 cell death . PLXDC2 double mutant S506A/Y511F has greater activity . Survival of control EGFP transfected cells without PEDF treatment is defined as 1 . Statistical significance of the comparison of cells without PEDF treatment ( with the control cells without PEDF treatment ) is shown in blue . Statistical significance of the comparison of cells with PEDF treatment ( with the control cells with PEDF treatment ) is shown in red . * = p < 0 . 05 , = p < 0 . 01 , and NS = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 00710 . 7554/eLife . 05401 . 008Figure 3—figure supplement 1 . Unbiased screening for effective siRNAs that knock down PLXDC1 or PLXDC2 expression in endothelial cell SVEC4-10 . The definition and source of each siRNA is described in ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 008 To assay PEDF's neurotrophic activity , we used 661W cells , a neuronal cell line derived from cone photoreceptors ( Tan et al . , 2004; Kanan et al . , 2008 ) . We found that PEDF treatment effectively protects 661W cells against oxidative damage . Knocking down PLXDC1 , but not PLXDC2 , in 661W cells abolishes the protective effect of PEDF ( Figure 4A and Figure 4—figure supplement 1 ) . Conversely , using gain of function analysis , we showed that transfection of PLXDC1 further enhances the protective effect of PEDF and that transfection of the cytoplasmic domain of PLXDC1 protects 661W independent of PEDF ( Figure 4B ) . These experiments suggest that the cytoplasmic domain of the receptor is responsible for triggering the intracellular events during receptor activation . The dependence of 661W cell on PLXDC1 for the survival promoting effect of PEDF is in contrast to SVEC4-10's dependence on PLXDC2 to mediate the cell death effect of PEDF . 10 . 7554/eLife . 05401 . 009Figure 4 . PEDF's neurotrophic effect on 661W cells depends on PLXDC1 . ( A ) PEDF treatment protects 661W cells against hydrogen peroxide-mediated oxidative damage . siRNA-mediated knockdown of PLXDC1 , but not PLXDC2 abolishes the protective effect of PEDF . The survival of control siRNA transfected cell without treatment is defined as 1 . ( B ) Transfection of PLXDC1 or the cytoplasmic domain of PLXDC1 fused to DCC's TM domain enhances protection of 661W cells against damage caused by hydrogen peroxide . The effect of DCC-PLXDC1 C-tail is independent of PEDF . Cell survival of transfected control is defined as 1 . Statistical significance is shown on the top . = p < 0 . 01; NS = not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 00910 . 7554/eLife . 05401 . 010Figure 4—figure supplement 1 . Unbiased screening for effective siRNAs that knock down PLXDC1 or PLXDC2 expression in neuronal cell 661W . The definition and source of each siRNA is described in ‘Materials and methods’ . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 010 Although PLXDC1 and PLXDC2 have an architecture reminiscent of membrane receptors with extracellular , transmembrane and intracellular domains , they do not belong to any well-characterized receptor families and represent a family of their own . What is the consequence of PEDF's binding to these receptors ? To answer this question , we performed further mechanistic studies on PEDF/receptor interaction . We hypothesized that PEDF might affect the oligomerization states of the receptors . We found that PLXDC1 and PLXDC2 form homooligomers in the absence of PEDF . We designed experiments to compare homooligomerization and heterooligomerization and found that PLXDC1 and PLXDC2 preferentially form homooligomers ( Figure 5—figure supplement 1 ) . Using deletion series of the extracellular domains , we identified domain D as an important domain for oligomerization ( Figure 5A ) . We noticed that the last residue in the cytoplasmic domain of both PLXDC1 and PLXDC2 is a conserved cysteine . Since intracellular cysteines are in the reduced state , we tested whether copper phenanthroline-induced thiol oxidation ( Zhou et al . , 2009 ) could crosslink these cysteines . Indeed , we found that copper phenanthroline promoted the formation of covalent dimers , consistent with the close proximity of the two cysteines on the cytoplasmic domain ( Figure 5B ) . This covalent dimer can be cleaved under reducing conditions ( Figure 5B ) , consistent with its linkage through a disulfide bond catalyzed by copper phenanthroline . Interestingly , incubation with PEDF before copper phenanthroline treatment inhibited the formation of the dimer and promoted the formation of the monomer , indicating that the cytoplasmic tail is no longer in close contact with another cytoplasmic tail ( Figure 5B ) . 10 . 7554/eLife . 05401 . 011Figure 5 . Receptor oligomerization . ( A ) PLXDC1 deletion series with a Rim tag following the N-terminal secretion signal were purified together with HA-tagged full length PLXDC1 by purifying the Rim tag ( diagrams on the left ) . Anti-Rim Western is shown on the top and anti-HA Western is shown on the bottom for the elutions . HA-PLXDC1 no longer copurifies if domain D is deleted . ( B ) Copper phenanthroline [Cu ( II ) Phe] treatment creates covalent receptor dimer through oxidation of the free cysteine residue on the cytoplasmic tail ( schematic diagram on the left ) . Cu ( II ) Phe oxidation creates disulfide bond-linked covalent PLXDC1 dimer , as indicated in the Western blot for the receptor . PEDF inhibits dimer formation as shown by increased monomer band on a non-reducing gel after Cu ( II ) Phe oxidation ( red asterisk ) . The disulfide bond-linked dimers are sensitive to DTT treatment as shown in the reducing gel on the right . Molecular weight markers are in kD . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 01110 . 7554/eLife . 05401 . 012Figure 5—figure supplement 1 . PLXDC1 and PLXDC2 preferentially form homooligomers . ( A ) Schematic diagram of experimental design to compare homooligomer and heterooligomer formation of PLXDC1 is shown on the left . Purification of the Rim tag is indicated by an antibody symbol . Anti-Rim purification of Rim-PLXDC1 from cells coexpressing Rim-PLXDC1 , HA-PLXDC1 and PLXDC2 leads to HA-PLXDC1 purification , but not PLXDC2 purification ( Experiment 1 ) . This experiment suggests that PLXDC1 preferentially forms homooligomers . As a control , HA-PLXDC1 does not get purified in the absence of Rim-PLXDC1 ( Experiment 2 ) . Input is the starting material before anti-Rim purification . ( B ) Schematic diagram of experimental design to compare homooligomer and heterooligomer formation of PLXDC2 is shown on the left . Anti-Rim purification of Rim-PLXDC2 from cells coexpressing Rim-PLXDC2 , HA-PLXDC2 and PLXDC1 leads to HA-PLXDC2 purification , but not PLXDC1 purification ( Experiment 1 ) . This experiment suggests that PLXDC2 preferentially forms homooligomers . As a control , HA-PLXDC2 does not get purified in the absence of Rim-PLXDC2 ( Experiment 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 012 To pinpoint the receptor domain that interacts with PEDF , we used a deletion series of PLXDC1 to perform copurification analysis with PEDF and found that extracellular domain B plays an important role in binding to PEDF ( Figure 6A ) . To demonstrate PEDF's effect on receptor oligomerization in live cells , we developed an assay to visualize PEDF-induced receptor dissociation on the cell surface . Epitope-tagged extracellular domain of PLXDC1 is associated with the cell surface through its binding to the coexpressed full length untagged PLXDC1 ( Figure 6B ) . Incubation of the cells with PEDF causes the dissociation of the extracellular domain and the loss of the epitope tag from the cell surface ( Figure 6B ) . This live cell-based assay again demonstrated the ability of PEDF to dissociate receptor oligomer . 10 . 7554/eLife . 05401 . 013Figure 6 . Structure/function and real-time analysis of PEDF/receptor interaction . ( A ) PEDF with HA tag after the N-terminal secretion signal is copurified with different deletion mutants of PLXDC1 , which are all tagged with a Rim tag following the N-terminal secretion signal ( diagrams on the left ) . Purification of the Rim-tagged proteins with HA-PEDF is shown on the right . The upper Western is the anti-Rim Western and the lower Western is anti-HA Western to detect PEDF . Deletion of domain B largely abolishes the interaction between PLXDC1 and HA-PEDF . ( B ) An assay to study PEDF-mediated disruption of receptor dimers on the cell surface . Schematic diagram is on the left . PEDF displaces Rim-tagged PLXDC1 extracellular domain ( green ) bound to cell surface PLXDC1 ( gray ) to cause detachment of the extracellular domain . Immunostaining of Rim-tagged PLXDC1 extracellular domain , Rim-PLXDC1N , ( cotransfected with PLXDC1 ) on live cell surface is shown in the middle two pictures ( green signal ) . Blue color is nucleic acid stain DAPI . Upper picture: control ( no PEDF ) . Lower picture: PEDF treated . Quantitation of bound Rim-tagged PLXDC1 extracellular domain on the cell surface with or without PEDF treatment is shown on the right . ( C ) An assay to study PEDF-mediated disruption of receptor dimerization in real time . Schematic diagram of the experimental design is shown on the left . The cytoplasmic tail of PLXDC1 is linked to CFP or YFP , which are in close proximity due to receptor dimerization . PEDF suppresses the FRET signal between CFP and YFP if it disrupts the association of the receptor dimer . Right panel: PEDF , but not a control protein ( nidogen ) causes a time-dependent decrease in FRET signal between PLXDC1-CFP and PLXDC1-YFP . Both PEDF and nidogen were added at time 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 013 To further observe PEDF's effect on receptor oligomerization in real time , we developed a fluorescence resonance energy transfer ( FRET ) -based assay . We coexpressed PLXDC1 fused to a Cyan Fluorescent Protein ( CFP ) and PLXDC1 fused to a yellow fluorescent protein ( YFP ) at the C-terminus and observed a time-dependent decrease in FRET signals after addition of PEDF , but not a control extracellular protein nidogen ( Figure 6C ) . To show that this decrease in FRET signal was due to receptor dissociation , we crosslinked the cysteine residues on the cytoplasmic domain of PLXDC1 using sulfhydryl-specific crosslinker bismaleimidoethane ( BMOE ) before PEDF addition and found that this crosslinker prevents PEDF-dependent suppression of the FRET signal ( Figure 7 ) . However , mutation of the cysteine to serine ( C500S ) prevents the blocking effect of BMOE on PEDF ( Figure 7 ) . 10 . 7554/eLife . 05401 . 014Figure 7 . Crosslinking of the terminal cysteine of PLXDC1 prevents PEDF's effect on receptor dimerization . ( A ) Schematic diagrams of the experimental design . ‘C’ indicates the cysteine residue located at the C-terminus of PLXDC1 . ‘S’ indicates mutation of this residue to serine . ( B and C ) Crosslinking of the cysteine on the cytoplasmic tail of wild-type PLXDC1 ( PLXDC1-WT ) by sulfhydryl-specific crosslinker BMOE prevents the PEDF-dependent decrease of FRET signal between PLXDC1-CFP and PLXDC1-YFP . ( D and E ) BMOE has no effect on PEDF-dependent decrease of FRET signal between PLXDC1-CFP and PLXDC1-YFP if the C-terminal cysteine is mutated to serine ( PLXDC1-C500S ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05401 . 014 In summary , using three different techniques ( visualization of receptor oligomers in SDS-PAGE , visualization of receptor oligomers on the cell surface , and tracking receptor interaction in real-time ) , we found that PEDF receptors self-associate to form homooligomers and that PEDF has the ability to dissociate receptor homooligomers so that the cytoplasmic tails are no longer in contact with each other . This effect of PEDF is consistent with the ability of membrane-tethered cytoplasmic tail to activate downstream pathways . Both PLXDC1 and PLXDC2 are cell-surface transmembrane domain proteins and confer cell-surface binding to PEDF , as expected of cell-surface receptors . PLXDC1 and PLXDC2 represent a receptor family of their own that has two members in human . Through both loss of function and gain of function studies using three distinct cellular models that respond robustly to PEDF , we showed that both receptors have the expected properties of PEDF receptors in transducing PEDF signal and have cell type-specific roles . How does PEDF activate its cell-surface receptors ? Our experiments suggest that these receptors form homodimers under basal conditions , and the dimerization functions to inhibit self-activation . PEDF activates its receptors through dissociation of the dimer . PEDF's ability to dissociate the receptor dimer was demonstrated using three independent techniques . Without oligomerization provided by the extracellular domains , the cytoplasmic domain of the PEDF receptor is sufficient to activate downstream signaling . We also showed that modulating the cytoplasmic domain strongly affects receptor signaling . For example , a mutation in a conserved tyrosine residue in the cytoplasmic domain of PLXDC1 greatly enhanced PEDF-induced and PLXDC1-dependent IL-10 secretion from macrophages . Structure and function analysis revealed an important PEDF-interacting domain and a dimerization domain in the receptors . PLXDC1 and PLXDC2 are homologous membrane proteins with overlapping but distinct expression patterns in physiological and pathological conditions , as revealed by gene and protein expression studies . PLXDC1 ( also called tumor endothelial marker 7 ) was discovered as one of the genes enriched in many types of human tumor endothelial cells ( St Croix et al . , 2000; Schwarze et al . , 2005; Beaty et al . , 2007; Lu et al . , 2007; van Beijnum et al . , 2009 ) . PLXDC1 was also found to be highly expressed in the endothelial cells of another human disease- diabetic retinopathy- and is highly specific to diseased blood vessels ( Yamaji et al . , 2008 ) . PLXDC1 is also expressed on the macrophage cell surface ( Gaultier et al . , 2010 ) and is downregulated by LRP-1 , a large membrane protein involved in endocytosis of a variety of cell surface receptors ( Herz et al . , 1990; Herz and Strickland , 2001 ) . In contrast , PLXDC2 , but not PLXDC1 , was found as one of the genes that increase in expression during cellular senescence ( Schwarze et al . , 2005 ) , as one of the E2F1 target genes repressed by serum ( Hallstrom et al . , 2008 ) , as a gene negatively correlated with malignant cell transformation in tumors and its disturbance increases tumor volume ( McMurray et al . , 2008 ) , and as a candidate axon guidance molecule ( Leighton et al . , 2001 ) . Microarray analysis also revealed PLXDC2 , but not PLXDC1 , as one of the markers of adult stem cells ( Noh , 2006 ) . Proteomic analysis also revealed PLXDC2 on the cell surface of human pluripotent stem cells ( Boheler et al . , 2014 ) . PLXDC2 is also known as a mitogen for neuroprogenitors ( Miller-Delaney et al . , 2011 ) . Interestingly , expression patterns of both PEDF and PLXDC2 have been previously linked to cellular growth states . PEDF expression correlates with G0 growth arrest in fibroblasts ( Pignolo et al . , 1993 , 2003 ) and has been demonstrated to induce cell cycle arrest of glioma cells ( Zhang et al . , 2007 ) . The correlation between PLXDC2 and cell proliferation ( Miller-Delaney et al . , 2011 ) , senescence ( Schwarze et al . , 2005 ) , transformation ( McMurray et al . , 2008 ) and cell death ( Hallstrom et al . , 2008 ) has been noted in several studies . Consistent with these earlier reports , our study identified PLXDC2 as playing the dominant role in a PEDF-induced endothelial cell death model . From a different perspective , this study identified the extracellular ligand for two membrane proteins and showed that they function as cell-surface receptors that can transduce an extracellular signal . These receptors represent a new type of cell-surface receptor . They oligomerize in the basal state and are activated by ligand-induced dissociation . Mechanistically , they behave oppositely from many known single transmembrane domain signaling receptors that are activated by ligand-induced dimerization . However , detailed mechanisms can be complex , as exemplified by the human grown hormone receptor . For a long time after its original discovery more than two decades ago ( Leung et al . , 1987; Cunningham and Wells , 1989; Cunningham et al . , 1989 , 1991a , 1991b ) , the growth hormone receptor was hypothesized to be activated by ligand-induced receptor dimerization , as revealed by classic studies . A recent study suggested that its conformational change is similar to that of scissors ( Brooks et al . , 2014 ) . PEDF receptors have at least two mechanisms to diversify their function and regulation . First , both human PLXDC1 and PLXDC2 have several isoforms . Second , receptors can couple to different immediate downstream molecules . The immediate downstream molecules that transduce the PEDF receptor signal ( the equivalent of G-proteins for G-protein coupled receptors ) are still unknown . As shown by G-protein coupled receptors , distinct cellular responses can be governed not only by distinct cell-surface receptors , but also by distinct downstream molecules . Because PEDF is a multifunctional factor , understanding the fundamental mechanisms of its transmembrane receptors in different cellular contexts will help to develop potent and specific small molecule-based therapeutics in treating diseases . To purify PEDF , we performed large-scale transfection of human PEDF cDNA into HEK293T cells using Jetprime reagent ( Polyplus-transfection SA , Illkirch , France ) . Six hours after transfection , the cells were washed twice with Hank's Balanced Salt Solution ( HBSS ) and changed to serum free medium ( SFM ) . PEDF is naturally present in several isoforms as revealed by isoelectric focusing ( Tombran-Tink et al . , 1995 ) . Not all PEDF isoforms are equally active ( Duh et al . , 2002; Subramanian et al . , 2012 ) . Consistent with previous reports ( Duh et al . , 2002; Subramanian et al . , 2012 ) , we found that more negatively charged PEDF isoforms are more biologically active . Therefore , we purified PEDF using sequential anion exchange chromatography . Briefly , PEDF was purified from conditioned SFM 24 , 48 , or 72 hours after transfection using a combination of Q sepharose ( Amersham/GE Healthcare , Little Chalfont , United Kingdom ) and polyethyleneimine column ( AX-300 , Eprogen , Downers Grove , IL ) . The conditioned SFM was dialyzed against binding buffer ( 20 mM Tris , pH 7 . 5 and 50 mM NaCl ) overnight at 4°C . Dialyzed medium was then applied to Q sepharose equilibrated with the binding buffer . The column was washed with 10 bed volumes of binding buffer before elution using buffer containing 20 mM Tris , pH 7 . 5 with 100 , 200 , 300 , or 400 mM NaCl . High performance liquid chromatography ( HPLC ) using polyethyleneimine column was performed to further purify PEDF . Briefly , buffer exchange for the Q fraction containing PEDF was performed for three times by diluting with 10 vol of column buffer ( 25 mM Tris , pH 8 . 4 ) and concentrating in an Amicon Ultra-4 filter ( Millipore , Billerica , MA ) before being loaded onto the HPLC system . HPLC was performed using the Agilent 1100 series liquid chromatography system with a diode-array detector . Proteins were separated on polyethyleneimine column using column buffer with increasing NaCl concentrations as the mobile phase ( from 0 to 1 . 8 M in 8 min and the 1 . 8 M sodium concentration was maintained for another 4 min ) . The flow speed of mobile phase was 0 . 5 ml/min , and four fractions were collected every minute . Elution fractions with significant A280 value were saved . Buffer exchange was performed for three times by diluting with 10 vol of phosphate buffered saline ( PBS ) and concentrating in an Amicon Ultra spin filter . The final volume of each concentrated elution was 0 . 5 ml . The presence of PEDF was confirmed by both total gel staining and Western blot analysis . Protein sterilization was achieved using Ultrafree Durapore 0 . 22 μm filter ( Millipore ) . We also produced PEDF with a 6XHis tag followed by the HA tag at the N-terminus after the secretion signal and PEDF with an 8XHis tag at the C-terminus of PEDF . We found that tagging significantly diminishes the biological activity of PEDF . Therefore , all biological assays were performed using untagged PEDF . The negative effect of epitope tagging on PEDF is likely one of the reasons that PEDF receptors are difficult to identify . The domains for human PLXDC1 ( numbered according to the full length receptor with the secretion signal ) were: domain A ( 19–127 ) , domain B ( 128–242 ) , domain C ( 243–292 ) , and domain D ( 293–359 ) . We also created chimeras containing the transmembrane domain ( TM ) of another single transmembrane protein DCC ( Stein and Tessier-Lavigne , 2001 ) and the C-terminus of the receptors . The DCC TM domain was fused to the secretion signal of alkaline phosphatase and Rim tag at the N-terminus and the cytoplasmic tail of PLXDC1 or PLXDC2 at the C-terminus . A monoclonal antibody has been produced against the Rim tag , which has 14 residues ( NETYDLPLHPRTAG ) ( Illing et al . , 1997 ) . The most likely positions of phosphorylation sites in the cytoplasmic domains of human PLXDC1 and PLXDC2 were identified through PhosphoSite Plus , a bioinformatics resource to identify potential protein phosphorylation sites ( http://www . phosphosite . org/staticUsingPhosphosite . do ) . Residue numbers are according to human isoform 1 . The secretion signal for alkaline phosphatase ( AP ) followed by the Rim tag was engineered at the N-terminus of each extracellular domain . PEDF was biotinylated using sulfo-NHS-SS-biotin ( Pierce , Rockford , IL ) after overnight dialysis in PBS at 4°C . After biotinylation , free biotin was removed by further overnight dialysis in PBS at 4°C and the degree of biotinylation was assessed by visualizing the shifting of molecular weight in SDS-PAGE gels after incubation with streptavidin . The advantage of biotinylation is that biotin is a tag much smaller than peptide tags or fusion protein tags and is less likely to interfere with biological activities . In addition , biotin is added after protein production and folding and allows sensitive detection . However , excessive biotinylation can inactivate proteins due to the modification of key lysine residues . To prevent excessive biotinylation , the ideal degree of biotinylation is about 90% , as judged by shifting in molecular weight after binding to streptavidin . Biotinylated PEDF ( 20 nM ) was added to transfected or control cells grown on a fibronectin-coated dish in HBSS with 10 mM HEPES , pH 7 . 5 and 2 mg/ml BSA at room temperature for 1 hour . After two continuous washes with HBSS , 10 mM HEPES , pH 7 . 5 , the cells were fixed using freshly made 4% paraformaldehyde in HBSS , pH 7 . 5 for 20 min . The cells were heated in HBSS at 65°C for 1 hr to inactive endogenous AP activity . After blocking in 5 mg/ml BSA in PBS for 1 hour , the cells were incubated with streptavidin-AP diluted in 5 mg/ml BSA in PBS . After four washes using PBS , AP activity was visualized using NBT/BCIP ( Thermo Scientific , Waltham , MA ) . Membranes were prepared from HEK293 cells transfected with the receptors using PBS and 5 mM EDTA , which helps to keep free cysteine residues in the reduced state . After one wash using PBS , the membrane was resuspended in PBS and incubated with or without PEDF for 3 hr at room temperature . Oxidation-induced disulfide bond formation was catalyzed by 0 . 5 mM Cu ( II ) Phe . After 5 min , EDTA was added to each reaction to 50 mM to stop the oxidation reaction . The membranes were spun down , resuspended and boiled in SDS loading buffer with or without DTT for loading onto a SDS-PAGE gel . We developed a live cell-based assay to study the ability of PEDF to dissociate receptor complexes . PLXDC1 extracellular domain with a Rim tag following the N-terminal secretion signal of alkaline phosphatase is cotransfected into COS-1 cells with wild-type PLXDC1 with no tag . The Rim-tagged extracellular domain of PLXDC1 associates with the cell surface through its interaction with the extracellular domain of the full length PLXDC1 . 1 day after transfection , the cells were washed once with HBSS and incubated overnight at 37°C in SFM with 5 mg/ml BSA with or without 50 nM PEDF . The next day cell surface associated Rim tagged protein is assessed through live cell staining by anti-Rim antibody by incubating with antibody diluted in SFM with 5 mg/ml BSA for 60 min at 37°C . After antibody binding , the cells were washed with HBSS and fixed in freshly made 4% paraformaldehyde in HBSS , pH 7 . 5 for 20 min . Rim antibody was detected through immunostaining using anti-mouse secondary antibody . Fluorescent signals were quantified using Nikon NIS Elements AR Analysis software . The survival of SVEC4-10 endothelial cells ( ATCC , Manassas , VA ) upon PEDF treatment was analyzed by the MTT assay . Briefly , SVEC4-10 cells were grown in 10% FBS in DMEM containing penicillin and streptomycin until confluency . Cell death was initiated by splitting confluent SVEC 1:10 in serum free media ( SFM ) and 1 mg/ml BSA with or without 20 nM PEDF . PEDF was added before cell addition . Trypsin used in cell splitting was neutralized by defined trypsin inhibitor ( Gibco ) and removed by spinning down the cells . Cell viability was assessed 24 hr later after cell plating . MTT assay was done by incubating cells with 100 μg/ml MTT reagent ( 3- ( 4 , 5-dimethylthiazol-2-yl ) -2 , 5-diphenyltetrazolium bromide ) in SFM for 3 hr at 37°C . Dimethyl sulfoxide ( DMSO ) was added to each well after MTT was removed . The absorbance of the purple color from the formazan formed was measured and quantified using POLARstar Omega ( BMG Labtech , Ortenberg , Germany ) at 534 nm . For cell death assay on siRNA transfected cells , cell plating in SFM ( with or without PEDF addition ) was done after two rounds of 48-hour transfection . For cell death assay on DNA transfected cells , the cells were split at 1:15 ratio 16 hours before transfection and were transfected using Jetprime reagent ( about 40–50% confluency during transfection ) . Cell plating in SFM and 1 mg/ml BSA ( with or without PEDF addition ) was done 16 hours after transfection . All assays were performed in 96-well plates in triplicate . We found that PEDF protects cone-derived 661W cells from hydrogen peroxide-mediated oxidative damage . 661W cells were grown in 10% FBS , 40 μg/l of hydrocortisone 21-hemisuccinate , 40 μg/l of progesterone , 32 mg/l of putrescine , 40 μl/l of β-mercaptoethanol in DMEM containing penicillin and streptomycin and treated with 10 nM PEDF for 20 hours . After addition of 2 . 5 mM H2O2 for 1 hr ( to achieve about 90% cell death in control cells the next day ) , the media was replaced with fresh media and the cells were continuously grown for 24 hours . Cell survival was quantified using the MTT assay as described above . For siRNA transfected cells , 10 nM PEDF was added after two rounds of 48-hour transfection . For DNA transfection , the cells were split at 1:5 ratio 16 hours before transfection and were transfected using Jetprime reagent . Four hours after transfection , the media was changed to fresh media and PEDF was added . Macrophage cell line RAW267 . 4 ( ATCC ) was grown in RPMI1640 media with 10% FBS ( Gibco Certified Performance Plus FBS ) , penicillin and streptomycin . For DNA transfection , the cells were split at 1:6 ratio 16 hours before transfection and were transfected using Jetprime reagent . Eight hours after transfection , the media was changed to fresh media and PEDF was added . PEDF-induced IL-10 secretion from macrophages was assayed using mouse IL-10 ELISA kit from Southern Biotech 18 hours after PEDF addition . For siRNA transfection , media was changed to fresh media and PEDF was added after two rounds of 48-hour transfection . All final assays were performed in 96-well plates in triplicate . We found that longer culture of RAW267 . 4 cells without cell replating leads to more responsiveness to PEDF . Since siRNA experiments need longer cell culture time , cells are consistently more responsive to PEDF than cells in DNA transfection experiments , which are done only 1 day after transfection . After screening many siRNA transfection reagents including X-tremeGENE ( Roche , Basel , Switzerland ) , siTran ( Origene , Rockville , MD ) , Jetprime ( Polyplus-transfection SA ) , RNAiMAX ( Life Technologies , Carlsbad , CA ) , and GenMute ( SignaGen , Rockville , MD ) , we found that the most effective siRNA transfection reagent is RNAiMAX . Since the three cell types ( RAW267 . 4 , SVEC4-10 and 661W ) that were used as cellular models to study PEDF receptors are all mouse cells , siRNAs targeting mouse genes were tested . For mouse PLXDC1 , siRNAs tested included Dharmacon D-060224-01 ( siRNA-1 ) , Dharmacon D-060224-02 ( siRNA-2 ) , Dharmacon D-060224-03 ( siRNA-3 ) , Dharmacon D-060224-04 ( siRNA-4 ) , Invitrogen 4390771-s90877 ( siRNA-5 ) , Invitrogen 4390771-s90878 ( siRNA-6 ) , Dharmacon smart pool L-060224-01 ( siRNA-7 ) and Origene 866091-SR46066C ( siRNA-8 ) . For mouse PLXDC2 , siRNAs tested included Dharmacon D-059538-01 ( siRNA-1 ) , Dharmacon D-059538-02 ( siRNA-2 ) , Dharmacon D-059538-03 ( siRNA-3 ) , Dharmacon D-059538-04 ( siRNA-4 ) , Invitrogen 4390771-n380220 ( siRNA-5 ) , Invitrogen 4390771-n380229 ( siRNA-6 ) , Dharmacon smart pool L-059538-01 ( siRNA-7 ) , and Origene 866094-SR416812C ( siRNA-8 ) . For mouse LRP-1 , siRNAs tested included Origene SR423695A-866095 ( siRNA-1 ) and SR423695B-866096 ( siRNA-2 ) . Control siRNA was from Invitrogen . The most effective siRNA was transfected through reverse transfection using RNAiMAX in antibiotic free culture medium at 50 nM concentration with a cell splitting ratio of 1:5 . To achieve a high transfection rate and knockdown effect , reverse transfection was performed twice consecutively following the manufacturer's protocol . At 48 hr after transfection , the cells were reverse transfected again using the same siRNA for the second round of knockdown . Functional assays were performed 48 hours after the second round of reverse transfection . Gene expression levels were analyzed by RT-PCR . Briefly , total RNA was extracted from cells with a kit ( Qiagen , Hilden , Germany ) . Total RNA concentration was measured by Nanodrop ( Thermo Scientific ) . Total RNA was then used to generate cDNA by using ThemoScript Reverse transcriptase ( Life Technologies ) . Mouse PLXDC1 was amplified by 5′-GGAGGCAGAAGGCAAGACATGCG-3′and 5′-CGTGGAGGCCGAGCAGTGCTGA-3′ . Mouse PLXDC2 was amplified by 5′-CTGCCAGCCGGGATCTGTGGGTTAACATAGACC-3′ and 5′-GGGAAGTGGAGTCATCTCCACAGCTGAGATGTTGG-3′ . Rim-tagged proteins were purified using the anti-Rim antibody- conjugated to CNBr-activated Sepharose 4 Fast Flow beads ( Amersham/GE Healthcare ) . Briefly , cells were washed once with HBSS and lysed in well with 1% Triton X-100 in HBSS and protease inhibitors for 30 min on ice . Cell lysate was spun at 16 , 000×g , 4°C for 10 min to remove insoluble materials . Cell lysate was applied to anti-Rim antibody conjugated beads , and rotated for 4 hr at 4°C . The beads were washed three times using 0 . 1% Triton X-100 in HBSS by spinning down at 1000×g for 30 s and eluted in 0 . 1% Triton X-100 in 0 . 1 M Glycine , pH = 2 . 3 for 15 min at room temperature . Tris ( pH 9 . 5 ) was added to 0 . 1 M to neutralize the elution before the samples were analyzed . HA-tagged proteins were detected using a monoclonal anti-HA antibody . To compare homooligomerization and heteroligomerization , anti-Rim purification was performed 24 hr after cells were transfected with Rim-tagged PLXDC1 ( 20% ) , HA-tagged PLXDC1 ( 40% ) and untagged PLXDC2 ( 40% ) in one experiment and Rim-tagged PLXDC2 ( 20% ) , HA-tagged PLXDC2 ( 40% ) and untagged PLXDC1 ( 40% ) in another experiment . Copurified receptors were detected either by anti-HA antibody or antibody specific to PLXDC1 or PLXDC2 . Polyclonal antibodies against the N-terminal peptide of human PLXDC1 ( SPQPGAGHDEGPGSGWAAKGTVRG ) and the N-terminal peptide of human PLXDC2 ( KPGDQILDWQYGVTQAFPHTE ) were produced by conjugating the peptides to KLH before immunization of rabbits ( Genemed Synthesis , San Antonio , TX ) . Antibodies were purified from rabbit crude sera using the corresponding peptide conjugated to Affigel ( Bio-Rad , Hercules , CA ) . CFP and YFP proteins were fused to the C-terminus of PLXDC1 and PLXDC2 to detect oligomerization of PEDF receptors . Three glycine linkers were added between YFP/CFP and the C-terminal tail of PLXDC1 or PLXDC2 . FRET analysis was performed similarly as described ( Kawaguchi et al . , 2011 ) . Briefly , membranes were prepared from HEK293 cells that coexpress PLXDC1-CFP and PLXDC2-YFP . CFP-YFP FRET was measured in black flat bottom 96-well plates ( Microfluor 2 , Thermo Scientific ) using simultaneous dual emission optics in POLARstar Omega with excitation filter 422-20 and emission filters 470-12 and 530-10 . The background signal of each reaction was measured before PEDF was added to the membrane suspension to initiate the reactions . The signal from each time point was the average of 20 measurements . After all the measurements were done , the signals were calculated as the ratio of emissions at 530 nm over emissions at 470 nm to observe the dynamic change in FRET . To crosslink the C-terminal free cysteine using BMOE ( Pierce ) , membrane preparations were made in PBS and 5 mM EDTA . BMOE was added to the membrane suspension at a concentration of 2 mM . The reaction was carried out at room temperature for 1 hour . Concentrated DTT solution was added to 5 mM to quench the reaction . After incubation at room temperature for 10 min , 1 ml of HBSS/HEPES ( HBSS with 10 mM HEPES , pH 7 . 5 ) was added to the membrane suspension . After the membranes were pelleted down , the resulting membrane pellets were washed once and resuspended in HBSS/HEPES for FRET measurement .	Many cells in our body release signals that trigger responses in other cells . A protein called PEDF is a signal released from a variety of cells that can prevent the formation of new blood vessels , protect cells in the retina and brain from damage and stop cancer cells from growing . Experiments using model animals have also demonstrated that PEDF could be used to treat a variety of eye diseases that lead to blindness and many types of cancer . PEDF is found in tissues including the brain , eye , liver , heart and lung , but it was not known how cells sense this signal . Cells are expected to have specific proteins called receptors on the cell surface membrane to detect PEDF and transmit the signal into the cell; however , the identity of these receptors has remained a long-standing unsolved puzzle . Cheng , Zhong , Kawaguchi et al . have now identified two human proteins that act as receptors for PEDF . These proteins—known as PLXDC1 and PLXDC2—span the cell surface membrane , and bind to PEDF on the outside of the cell . PLXDC1 and PLXDC2 are expressed on different types of cells that respond to PEDF . Furthermore , PEDF was unable to act upon cells that had been engineered to make less of these two receptors . This study also revealed that each receptor can play different roles in different cell types . For example , exposing one type of cell from blood vessels to PEDF would normally kill them , but cells without PLXDC2 ( but not those without PLXDC1 ) could survive PEDF treatment . Furthermore , PEDF treatment protects a type of neuron against environmental damage , and this activity depends on PLXDC1 , but not PLXDC2 . How do the receptors transmit the PEDF signal from the outside of the cell to the inside of the cell ? Cheng , Zhong , Kawaguchi et al . found that when PEDF is not present , both PLXDC1 and PLXDC2 form complexes containing more than one copy of either receptor . When PEDF binds to the receptors , it causes these complexes to disassemble and this activates further downstream signaling events inside the cell . Understanding PEDF receptors and their mechanisms will open the way to developing new drugs that target these receptors to treat human diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2014	Identification of PLXDC1 and PLXDC2 as the transmembrane receptors for the multifunctional factor PEDF
We asked how a new , complex trait evolves by selecting for diurnal oscillations in the budding yeast , Saccharomyces cerevisiae . We expressed yellow fluorescent protein ( YFP ) from a yeast promoter and selected for a regular alternation between low and high fluorescence over a 24-hr period . This selection produced changes in cell adhesion rather than YFP expression: clonal populations oscillated between single cells and multicellular clumps . The oscillations are not a response to environmental cues and continue for at least three cycles in a constant environment . We identified eight putative causative mutations in one clone and recreated the evolved phenotype in the ancestral strain . The mutated genes lack obvious relationships to each other , but multiple lineages change from the haploid to the diploid pattern of gene expression . We show that a novel , complex phenotype can evolve by small sets of mutations in genes whose molecular functions appear to be unrelated to each other . Biological oscillators demand dynamical interactions between multiple components and have evolved many times to drive various behaviors , from heartbeats , on the timescale of seconds , to circadian clocks and seasonal clocks , on the time scale of days and years . The budding yeast , Saccharomyces cerevisiae , lacks homologs to genes that make up the core oscillator or light-sensing pathways of fungal circadian rhythms ( Brunner and Kaldi , 2008; Idnurm and Heitman , 2010; Salichos and Rokas , 2010 ) and does not display any 24-hr behavior capable of sustained autonomous oscillations in a constant environment ( Eelderink-Chen et al . , 2010; Robertson et al . , 2013 ) . As an example of a novel trait , we evolved a 24-hr oscillator in the budding yeast to determine how biologically complex machinery could evolve from proteins and signaling networks that lack any known connection to oscillatory behavior . Because circadian clocks can be entrained by environmental signals and have a temperature-independent period , whereas the behavior we have evolved represents an autonomous oscillator with a period of roughly 24 hr that controls a biological output , we refer to the circuit we have evolved as a diurnal oscillator , referring to its period , rather than the distinction between day and night . Our initial goal was to use fluorescence activated cell sorting ( FACS ) and fluorescent reporters to select for diurnal regulation of gene expression . We expressed yellow fluorescent protein ( YFP ) from the promoter of the FLO1 gene ( Verstrepen et al . , 2005; Rando and Verstrepen , 2007; Verstrepen and Fink , 2009 ) and elevated the mutation rate about 100-fold by mutating the proof-reading activity of DNA polymerase ∂ ( POL3 ) ( Jin et al . , 2005 ) . We used FACS to select for variation in YFP fluorescence over a 24-hr period: each cycle began with selection of the dimmest events , the selected cells were allowed to proliferate , and the population was then selected , 10 hr later; for the brightest events , the selected cells were allowed to proliferate , and the dimmest events were selected , 14 hr later , completing the cycle ( Figure 1A ) . We evolved two parallel populations using two different strengths of selection: the top and bottom 6% ( E6 ) or 15% ( E15 ) of the events detected by FACS . The selected cells proliferated for roughly six divisions before the next selection . Our scheme selects for synchrony rather than entrainment: twice a day , we select the cells with high or low fluorescence , rather than varying an environmental cue and selecting for a response to it . The human analogy would be to go to a large , international airport and select only those travelers who had arrived from the same time zone , rather than taking all newly arrived passengers and using alternating light and dark cycles to entrain their clocks over a period of several days . 10 . 7554/eLife . 04875 . 003Figure 1 . Cell association clock evolves from selecting for high and low YFP expression . ( A ) Selection scheme for a 24-hr oscillator: cells were collected from the dimmest 6% ( E6 ) of the distribution of YFP fluorescing events , grown exponentially at <1 × 106 cells/ml for ∼10 hr and then the brightest 6% of events were collected and the cells in this population were grown for ∼14 hr at <1 × 106 cells/ml . The selection cycle was repeated for 30 days . The E15 population was evolved in the same manner , but the dimmest and brightest 15% of events were collected in the two phases of the selection . ( B ) Scheme for synchronizing YFP oscillations from an unsynchronized population: the dimmest events from an unsynchronized population ( blue box ) are collected , thus selecting for cells at the single cell phase of the oscillation . After 10 hr of growth , distributions are first recorded and the brightest YFP events ( clump of cells ) are collected , grown for 14 hr , distributions are recorded again and the dimmest events are selected to complete the cycle . This cycle is repeated to keep the population maximally synchronized . The dashed red lines represent the fluorescence distribution of the ancestral population . ( C ) Representative density plots showing oscillations in YFP fluorescence of a FACS-synchronized sample of the E6 and E15 populations after terminating the selection , and a clone isolated from E6 ( E6-1 ) . The red outline overlay shows the average ancestral distribution . ( D ) Representative DIC images of synchronized populations immediately before FACS selection for low ( single-cell phase ) and high ( multi-cell phase ) YFP fluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 00310 . 7554/eLife . 04875 . 004Figure 1—figure supplement 1 . Evolution of 24-hr cell aggregation clock . ( A ) Representative density plots showing RFP oscillations of a FACS-synchronized sample of the final E6/E15 populations and a clone isolated from E6 ( E6-1 ) . ( B ) Ancestor cells expressing cerulean fluorescent protein ( CFP+ ) were mixed with the populations E6 , E15 and the clone E6-1 ( none of which expressed CFP , ( CFP− ) ) , sonicated to single cells , and imaged . There are no detectable changes in YFP or RFP levels , per cell , between ancestral and evolved populations . YFP cannot be detected by microscopy in any strain . ( C ) Representative YFP density plots of E6-1 cells either synchronized by FACS ( synch ) or passaged through the FACS by collecting cells from the entire YFP distribution ( unsynch ) at identical 10 and 14-hr intervals for 3 days . ( D ) Oscillations were quantified by calculating the log ( F75/F25 ) ratio of synchronized and unsynchronized cultures of E6-1 . See Figure 3A for a definition of the F75:F25 ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 004 After 30 days of selection , the intensity distributions resulting from the selections for high and low fluorescence differed . To analyze the behavior of the evolved populations in more detail , we cultured them from stocks frozen that had been prepared at the end of the original selection . When they are first thawed , these populations are asynchronous , presumably because the manipulations associated with freezing and thawing disrupt the synchrony between the oscillations of different individuals . We therefore subjected these cultures to the same , twice daily , selection used for the evolution: the dimmest events are collected , cells are allowed to proliferate for 10 hr , the population distribution of fluorescence is recorded by FACS , and the brightest events are collected . After 14 hr of proliferation , distributions are recorded , and the dimmest events are collected to complete the cycle ( Figure 1B ) . This protocol is a form of selection synchrony: it synchronizes a population by selecting those members that are at the peaks and troughs of a cycle of fluorescence intensity , rather than by applying a stimulus that alters the phase of the oscillations until all members of the population have the same phase ( induction synchrony or entrainment ) . After synchronization , both evolved populations showed a cyclical change in the fluorescence distribution , with the E6 population giving larger shifts than the less strongly selected E15 ( Figure 1C ) . It takes more cycles of synchronization of E15 to achieve the full amplitude of daily oscillation than it does for E6 , which is largely synchronized within 24 hr of selection . Our starting strain also contained red fluorescent protein ( RFP ) under the control of the ACT1 ( actin ) promoter , allowing us to ask if RFP fluorescence , which had not been selected on , also cycled . To our surprise , the distribution of RFP also cycled ( Figure 1—figure supplement 1A ) , prompting us to examine the cells at different phases of the cycle . Just before their scheduled selection for low fluorescence , the population was predominantly single-celled , and just before their selection for high fluorescence , most cells were in multicellular clumps ( Figure 1D ) . The fluorescence per cell is constant over 24 hr , for both YFP and RFP , and thus the cyclical variation in intensity seen by FACS is entirely due to changes in cell association ( Figure 1—figure supplement 1B ) . Because we see similar fluctuations in the fluorescence from PACT1-RFP as we do from PFLO1-YFP , we believe that we would have selected for oscillations in cell aggregation with any promoter that drove the expression of a fluorescent protein to levels that were sufficient for our FACS-based selection . The cyclical behavior of our evolved populations could be due to interactions between two or more different genotypes . To eliminate this possibility , we isolated and analyzed fifteen clones from each population after subjecting them to FACS-based synchronization: all 30 clones produced strong oscillations in YFP and RFP distributions , and we analyzed the clone that showed the strongest oscillations ( E6-1 ) in detail ( Figure 1C , D ) . Except for the free-run experiments , described later in the paper , all other experiments examined populations that were subject to FACS-based selection twice a day with a 10-hr interval between the selection for the dimmest and brightest events and a 14-hr interval between the selection for the brightest and dimmest events . The cycle is not a response to daily variations in environmental factors such as temperature . If populations were responding to environmental fluctuations , they should eventually synchronize in response to these fluctuations . To look for this behavior , we subdivided a single population of clone E6-1 and subjected it to two different treatments: synchronizing it by FACS selection , as described above , or subjecting it to identical manipulations , except that we collected all the events that passed through the FACS , without regard to their fluorescence intensity . The first population oscillated , showing primarily single cells just before the selection for low YFP fluorescence , and primarily clumps just before the selection for high YFP fluorescence , whereas the unsynchronized population showed very similar distributions at these two times ( Figure 1—figure supplement 1C , D ) . We conclude that we have evolved a cyclical change in cell association with a period of about 24 hr , and cells that show this oscillation can be synchronized by selecting on their level of fluorescence as a proxy for the number of cells in a clump . To analyze the oscillations in cell association , we followed the behavior of E6-1 cells grown at low population density using two different labels . The first , cerulean fluorescent protein ( CFP ) was expressed from a constitutive promoter ( PACT1 ) to distinguish two genetically identical subclones: one expressed CFP ( CFP+ ) and the other did not ( CFP− ) . The second was introduced by labeling cells , just after FACS selection , by covalently linking Oregon Green 488-X , a yellow fluorescent molecule , to their cell walls ( Hoch et al . , 2005 ) . Because a daughter’s cell wall is entirely new , the original cells retain the yellow label and their daughters are unlabeled ( Barral et al . , 2000 ) . We followed these cultures as they proliferated and eventually switched to the other phase of the morphological oscillation ( Figure 2A , B and Figure 2—figure supplement 1A ) . Single cells that were selected for by sorting for low YFP fluorescence and covalently labeled ( T = 0 hr ) , proliferated to become clumps composed of a single yellow cell surrounded with non-yellow cells that had the same CFP expression state as the original single cell . The absence of clumps containing more than one yellow cell or both CFP+ and CFP− cells in the same clump shows that cells emerging from the single-cell phase build lineage-based clumps . Consistent with this interpretation , we saw synchronous oscillations in cell association even at very high dilution ( ∼103 cells/ml ) . Multicellular clumps , which were covalently labeled at the time of peak aggregation gave rise to single cells by a combination of two mechanisms: fragmenting to produce smaller clumps and proliferating to produce single-celled offspring . Smaller clumps consisting of yellow cells and newly born non-yellow cells were observed 3 hr after ( T = 13 hr ) the peak aggregation time ( 10 hr ) . By 18 hr , the population was a mixture of medium-sized and smaller clumps and single cells , and by 24 hr non-yellow , single cells , and small clumps were the bulk of the population , representing the completion of one cycle . When unsynchronized CFP+ and CFP− subclones were sonicated to break up all the clumps and cultured together for 90 min , clumps that re-associated were composed of a mixture of CFP+ and CFP− cells showing that clumps form by cells sticking together rather than failing to separate after division ( Figure 2—figure supplement 1B ) . These results show that cells oscillate between two states over a 24-hr period: single cells produce offspring that immediately stick to their mothers to assemble a lineage-based clump , which later produces a population of single cells and small clumps as a result of clumps fragmenting and producing single cells which divide to produce a mixture of small clumps and single cells ( Figure 2C ) . 10 . 7554/eLife . 04875 . 005Figure 2 . 24-hour autonomous oscillations in dynamic aggregate assembly . ( A ) Representative images of a synchronized E6-1 population at different points through the 24-hr cycle . The experiment contained two genetically similar subclones , one expressing cerulean fluorescent protein ( CFP+ ) and the other not ( CFP− ) . After FACS selection ( 0 and 10 hr ) cell walls were covalently labeled with Oregon Green 488-x N-hydroxy succinimidyl ester , making them fluoresce brightly . Yellow-labeled single cells give rise to the multicellular phase by sticking to their daughters and clumps of different lineages do not mix ( i . e . , there are no CFP clumps with both CFP+ and CFP− cells ) . Labeled clumps give rise to single cells and smaller clumps by clump fragmentation and producing single-celled daughters . Time scale is in reference to the initial collection of single cells ( Low YFP selection , T = 0 hr ) over a 24-hr period . ( B ) The population from the experiment described in ( A ) was quantified by calculating the frequency of clumps containing 1–4 , 5–10 , and >10 cells . There is a synchronous rise in clump size during the transition from single cells to clumps and a gradual decay of clump sizes during the transition from clumps to single cells . ( C ) Schematic representation of oscillator dynamics: single cells give rise to the multicellular phase by sticking to their progeny to form a lineage-based clump . At the peak multicellular phase , clumps give rise to the single cell phase by fragmentation and producing single-celled daughters some of which adhere to their progeny to produce small clumps . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 00510 . 7554/eLife . 04875 . 006Figure 2—figure supplement 1 . Dynamics of oscillator aggregate assembly . ( A ) Low magnification images of synchronized E6-1 cultures that contained two subclones , one labeled with CFP ( CFP+ ) and one without CFP ( CFP− ) . Single cells that were selected as dim events by FACS and whose cell walls were labeled with Oregon green 488-X assemble a lineage-based clump and do not adhere to other cell lineages ( i . e . , no CFP ( +/− ) clumps ) ( left panels ) . At the peak multicellular phase , clumps were labeled with Oregon green 488-X and followed . These cells give rise to the single cell stage by fragmenting and producing non-adherent daughters so that the majority of the population is single celled by the end of the complete , 24-hr cycle ( right panels ) . ( B ) Schematic of the mixing experiment performed by mixing , sonicating , and incubating asynchronous CFP+ and CFP− populations: cell separation defects will result in single colored clumps but adhesive cells will stick non-specifically to each other and produce clumps containing cells of both colors . Representative image of a single clump ( white outline ) showing E6-1 cells form clumps by adhering to each other . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 006 Because we typically sort populations twice every 24 hr , their behavior could reflect a damped oscillation , driven by selection , rather than a true oscillator , which should continue to cycle , even in the absence of selection . We therefore asked if our synchronized populations could free run: continue to oscillate in the absence of both entraining signals and periodic selection for cells in particular phases of the oscillation . We performed two experiments: one on populations and another to follow lineages initiating from single cells . In the population experiment , E6-1 was subjected to three different conditions . ( 1 ) Unsynchronized: cells that had been passaged without earlier synchronization were passed through the FACS machine and diluted twice every 24 hr , with intervals of 10 and 14 hr between the two dilutions , but all cells that passed through the FACS machine were collected , regardless of their fluorescence level and then diluted . ( 2 ) Synchronized: a population was synchronized by selection for 3 days with selection at alternating 10 and 14 hr intervals . The culture was synchronized by FACS for another 3 days , by FACS selection and dilution twice every 24 hr , with intervals of 10 and 14 hr between the two dilutions . ( 3 ) Free run: a population was synchronized by selection for 3 days at 10:14-hr intervals . After 3 days of synchronization ( marked by a dotted line ) , this culture was not subject to FACs but was simply diluted twice every 24 hr , with intervals of 10 and 14 hr between the two dilutions for an additional 3 days . The initial , synchronized population used to start the free run was the same one that was used to start the population with continuing synchronization . Thus , all three conditions were diluted with intervals of 10 and 14 hr between the two daily dilutions and maintained at a density that allowed continual exponential proliferation . The free run populations continued to show detectable oscillations , which were undetectable in a control population that had not been synchronized ( Unsynch ) ( Figure 3A ) . We next monitored how individual lineages oscillate without any prior FACS-based synchronization . A single cell was grown in a well of a 96-well plate and imaged every 3 . 5 hr . After 24 hr , the resulting population was collected and diluted back to single objects , regardless of clump size , into three separate new wells and followed every 3 . 5 hr for another 24 hr . This procedure was repeated one more time for each lineage ( i . e . , 3 cells from each of 3 sub-lineages ) for a third day , as shown by the cartoon in Figure 3B . This protocol dilutes cells once every 24 hr to keep cells in exponential growth throughout the experiment . Cell lineages produced sustained oscillations for 3 days ( Figure 3C ) with many lineages remaining in phase with each other . In some lineages , their phase shifted within a 24-hr period , but the overall period , over 3 cycles was approximately 24 hr . These results suggest that we have evolved an autonomous oscillator capable of sustained oscillations in the absence of an environmental stimulus . 10 . 7554/eLife . 04875 . 007Figure 3 . The evolved oscillator free runs for at least three cycles . ( A ) Three populations were compared . ( 1 ) Unsynchronized: cells that had been passaged without earlier synchronization were passed through the FACS machine and diluted twice every 24 hr , with intervals of 10 and 14 hr between the two dilutions , but all cells that passed through the FACS machine were collected , regardless of their fluorescence level and then diluted . ( 2 ) Synchronized: a population was synchronized by selection for 3 days with selection at alternating 10 and 14 hr intervals . The culture was synchronized by FACS for another 3 days , by FACS selection and dilution twice every 24 hr , with intervals of 10 and 14 hr between the two dilutions . ( 3 ) Free run: a population was synchronized by selection for 3 days at 10:14 hr intervals . After 3 days of synchronization ( marked by a dotted line ) , this culture was not subject to FACs but was simply diluted twice every 24 hr , with intervals of 10 and 14 hr between the two dilutions for an additional 3 days . All populations were diluted to ensure the population always reached the same final , pre-dilution density of ∼ 3 × 105 cells/ml , well below the density that corresponds to the end of mid-log phase growth for budding yeast ( 3 × 107 cells/ml ) . The initial , synchronized population used to start the free run was the same one that was used to start the population with continuing synchronization ( Synch ) . The measured variable , F75/F25 is based on the distribution of fluorescence and pulse widths at each time point . For both measurements , the lowest value is set to 0% and the highest to 100% , with values scaled linearly in between and the number of events that lie between 0 and 25% for both fluorescence intensity and pulse width is counted as F25 , and the number of events that lie between 75 and 100% for both fluorescence intensity and pulse width is counted as F75 , and we plot log ( F75/F25 ) . ( B ) Schematic of the protocol for analyzing individual lineages . A single cell is deposited in a microtiter well , and its progeny are observed every 3 . 5 hr for 24 hr , before taking 3 objects randomly ( single cells or cell clumps ) , and depositing them in fresh wells , observing for 24 hr , and finally taking 3 objects randomly ( cells or cell clumps ) , from each of the three wells and depositing them in fresh wells , and observing for 24 hr . ( C ) Individual traces of lineages originating from three separate cells ( Lineage 1 , 2 , 3 ) . An unsynchronized population was used to establish lineages that arose from single cells , which were diluted once every 24 hr and never experienced any FACS synchronization . A single cell was placed in 100 µl of medium in a microtiter well and allowed to proliferate for 24 hr that corresponds to a maximum of 16 divisions , a 64 , 000-fold increase in cell density from 10 cells/ml to 640 , 000 cells/ml , more than 40-fold below the density at which the exponential rate of cell proliferation starts to fall . Individual cells and clumps were imaged every 3 . 5 hr for 72 hr . At each time point , at least 50% of the cells and clumps in the wells were imaged and the average two-dimensional area was calculated . All lineages produce approximately 24-hr oscillations without any FACS-based synchronization . Some lineages alter their phase but continue to produce ∼24-hr oscillations . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 007 Because there could be multiple oscillating lineages , each with a different genetic basis , in our evolved population , we analyzed the behavior of four individual clones from the E6 population , focusing on the clone E6-1 . We used bulk segregant analysis ( Yvert et al . , 2003; Segre et al . , 2006; Birkeland et al . , 2010; Koschwanez et al . , 2013 ) to find candidates for the mutations that caused E6-1 to oscillate and verified them by engineering them into the ancestral , un-evolved strain . We began by crossing the evolved clone to its ancestor , putting the resulting diploid through meiosis , selecting for 24-hr oscillations on the resulting spores to isolate a pool of haploid progeny that showed robust cycling , and sequencing this pool's genomic DNA at high coverage . The full set of mutations needed to cause cycling should be present in the vast majority of the selected progeny , whereas the allele frequency of the neutral mutations that have accumulated during evolution should be ∼50% ( Birkeland et al . , 2010; Koschwanez et al . , 2013 ) . We classified eight mutations that were present in ≥95% of the selected cells as putative causal mutations ( Figure 4A ) . We recreated the evolved oscillator by engineering these eight mutations into a wild type laboratory strain . After synchronization , the oscillations of the recreated strain are indistinguishable from those of E6-1 ( Figure 4B ) and produce a similar free run behavior as E6-1 both as populations ( Figure 4C ) and as single lineages ( Figure 4D ) . 10 . 7554/eLife . 04875 . 008Figure 4 . Reconstruction of an evolved oscillator ( E6-1 ) . ( A ) Table of putative causal mutations . The mutations lie in the coding regions of the genes , and the allele number indicates the amino acid substitution that results . An asterisk denotes creation of a stop codon before the midpoint of the open reading frame . ( B ) The eight putative causal mutations were engineered into a wild type laboratory strain ( Recreated ) , which was synchronized and its oscillations were compared to the evolved E6-1 clone by plotting the log ( F75/F25 ) , as explained in the legend to Figure 3A . ( C ) The recreated strain shows a similar ability to oscillate without selection as the evolved clone E6-1 . Both cultures were synchronized by FACS for 3 days ( only the last cycle is shown ) and then allowed to free run by growing them exponentially at a constant temperature in the dark . ( D ) Individual lineages produce 24-hr oscillations in the absence of FACS-synchronization when subjected to the protocol described in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 008 We performed three tests to investigate the role of the different mutations in producing the evolved phenotype: deleting the mutated gene , replacing it with a wild-type allele , and investigating selection for or against the mutant allele when we selected for different levels of cell association . We investigated the role of removing genes because two of the causative mutations were stop codons , suggesting that other causative mutations might also inactivate the genes in which they occurred . Each causative mutation could be individually replaced by a gene deletion without destroying the oscillation ( Figure 5A ) . For each of these single gene deletions , the corresponding wild-type version of the same gene was added back and tested for oscillations . Surprisingly , only two genes substantially altered the oscillations: SIR4 made E6-1 single-celled and YJL070C produced constitutive clumps ( Figure 5B ) . To determine how the mutations whose wild-type expression had no effect were contributing to the oscillator , pools of spores from a diploid heterozygous for the eight mutations were selected for four ‘sub-traits’ of the oscillator: batch culture was used to select for exponential growth and FACS to select for constitutive clumps , single cells , and stochastic clumping . After 5 days of selection , the eight mutations were PCR amplified and Sanger sequenced from genomic DNA of the surviving pool of spores to estimate the frequency of the ancestral and evolved allele ( Koschwanez et al . , 2013 ) ( Figure 5C ) . In three independent replicate experiments , sir4-100 was selected for in all conditions , scw11-E261* was selected for in constitutive clumps , and three mutations , vps5-S520L , pet127-D650N , and whi2-R127* were positively selected for in single cells and selected against in clumping cells . Different combinations of HBN1 , IDS2 , and YJL070C were selected for between replicate experiments in both clumping and single cell sub-traits , making it hard to assess their significance . Because of this variability , and the absence of a phenotype when HBN1 or IDS2 were either deleted or replaced with wild-type genes , it is unclear if these genes are required for the evolved oscillations , and if they are , how they contribute to them . 10 . 7554/eLife . 04875 . 009Figure 5 . Analysis of putative causal mutations . ( A ) Genes containing causal mutations were individually deleted in E6-1 and tested for oscillations after applying the standard synchronization protocol to an unsynchronized culture . The metric for oscillation , F75/F25 is described in the legend to Figure 3A . ( B ) E6-1 strains containing a wild type copy ( WT ) of the indicated gene , at the LEU2 locus , and deletion of the mutant allele from its endogenous locus were tested for oscillations after applying the standard synchronization protocol to an unsynchronized culture . ( C ) Effect of different mutant alleles on oscillator subtraits . The recreated strain was crossed to the ancestor to generate a pool of spores carrying all possible combinations of the wild type and evolved alleles at the genes that harbored putative causal mutations in clone E6-1 . After selection , genomic DNA from the surviving spores was purified and the eight loci were PCR amplified and Sanger sequenced to estimate the relative frequency of the wild type and mutated alleles . ( D ) Gene expression changes in clone E6-1 , relative to its ancestor . Six asynchronous E6-1 samples were analyzed as replicates against six aggregated ancestor samples . Haploid-specific genes ( hsg ) , shown in red , are amongst the most repressed genes of E6-1 . Overexpressed genes in the E6-1 are distributed among a variety of cellular processes . Data reported as the log2 ratio of E6-1 and ancestor fragments per kilobase per million fragments mapped ( FPKM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 00910 . 7554/eLife . 04875 . 010Figure 5—source data 1 . Gene expression changes in clone E6-1 , relative to its ancestor . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 01010 . 7554/eLife . 04875 . 011Figure 5—figure supplement 1 . Restoration of haploid gene expression abolishes E6-1 aggregates . ( A ) Mating competency was used as a proxy for expression of haploid specific genes . Strains were crossed to MATa or MATα tester strains and replica-plated to media that selects for successful mating . E6-1 cannot mate because it has lost haploid-gene expression but deletion of HMLα ( hmlαΔ ) restores haploid gene expression as assayed by restored mating ability . ( B ) Representative DIC images of unsynchronized populations . Deletion of HMLα in E6-1 reduces the frequency of large multicellular aggregates to a level indistinguishable from the ancestor . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 011 The nature of the mutated genes does not produce simple hypotheses for the mechanism of the oscillator: they have a wide range of molecular functions and no pair of them has a previously described functional relationship . In search of other clues , we measured gene expression to look for changes in E6-1 . We performed two comparisons . The first was between synchronous and asynchronous populations . We isolated RNA at the same times from three cultures: the peak periods of single cells and multicellular clumps for a synchronized E6-1 culture , an unsynchronized E6-1 culture passaged through the FACS machine , but without sorting , and an ancestral population subjected to the synchronization protocol . We compared each population to the other two and failed to find a significant excess of genes that appeared to cycle in the evolved , synchronized population . The second comparison was to examine the mean expression of genes in the evolved and the ancestral populations . There were 155 genes whose expression differed between the ancestral and evolved populations ( Figure 5D and Figure 5—source data 1 ) . The most striking of these were genes involved in sexual behavior . Haploid-specific genes ( hsg ) , which are normally expressed in haploids and repressed in diploids , were repressed in the evolved clone; 8 of the 10 most strongly repressed genes were haploid-specific . This phenotype is due to the causal mutation that inactivates SIR4 , a gene required for transcriptional silencing of sub-telomeric genes , and the mating genes at the two silenced loci , HMLα and HMRa ( Rine and Herskowitz , 1987; Herskowitz , 1988; Pillus and Rine , 1989 ) . In sir4 cells , the expression of a and α information from these two loci mimics a diploid cell and represses haploid-specific genes . Consistent with the effect of expressing SIR4 in E6-1 , deleting HMLα from the evolved clone , which removes the only source of α information restores the haploid pattern of gene expression and eliminates cell clumping ( Figure 5—figure supplement 1A , B ) . We investigated the evolutionary trajectory of the E6 population by asking when the causative mutations found in the E6-1 clone appeared and how widespread they were by the end of our evolution . Each mutation was PCR amplified and Sanger sequenced at different time points over the course of the evolution to determine its frequency to an accuracy of roughly 10% . None of the mutations in E6-1 reached a frequency above 0 . 4 , two were undetectable in the overall population , and the remaining six were first detected between 12 and 30 days of evolution ( Figure 6A ) . The low allele frequencies at the final time point of our evolution have two possible interpretations: the evolved population contains a single oscillating lineage , which has yet to eliminate a substantial fraction of non-oscillating lineages , or it is a mixture of genetically distinct oscillator lineages . All the clones from the evolved populations cycled suggesting that the populations have multiple lineages . We therefore performed bulk segregant analysis on three additional clones from the strongly selected population ( E6-2 , 3 , 4 ) ( Figure 6B and Figure 6—source data 1A ) . Our results show that two cycling populations have evolved entirely independent , and each of these has split into two subpopulations . E6-2 contained 6 of the 8 causative mutations found in E6-1 and had three additional putative causative mutations . E6-3 and E6-4 share three mutations , but have no mutations in common with the lineage that produced E6-1 and E6-2 . Analyzing the entire open reading frame of WHI2 , the one gene that was mutated in all 4 sequenced clones , suggests that there are at least five independently evolved oscillators in the E6 and E15 populations , since one population ( E6 ) contains 2 mutant and one wild-type allele , and the other contains one mutant and one wild-type allele ( Figure 6—source data 1B ) . 10 . 7554/eLife . 04875 . 012Figure 6 . Genetically distinct clones evolve same oscillator through similar molecular changes . ( A ) The frequencies of the causal mutations found in E6-1 were measured by Sanger sequencing a region surrounding each mutation from frozen stocks of the E6 population taken at the indicated times during the experimental evolution . The frequencies of yjl070c-S188G and pet127-D650N were too low to measure by Sanger sequencing , whose detection threshold is an allele frequency of 10% . ( B ) Bulk segregant analysis and whole genome sequencing on three other isolates from the E6 population . Mutations that segregated at frequencies above 0 . 95 in at least one lineage are listed . All four clones contain a mutation in WHI2 , but the mutation in E6-1 and E6-2 is a different allele from that in E6-3 and E6-4 . Mutations with asterisks did not segregate above 95% in both lineages: WHI2-101 segregated at 80% in E6-2 and TPS1 segregated at 88% in E6-3 . Given the read depth , the probability of producing these deviations from the expected frequency of a non-causal mutation ( 0 . 5 ) by chance is 6 × 10−5 and 3 × 10−8 , respectively . ( C ) Quantitative PCR ( qPCR ) was used to compare the mRNA abundance of the two genes at HMLα , which encodes two regulators of mating type- and haploid-specific gene ( hsg ) expression , and three representative haploid-specific genes from eight clones ( E6-1 , 2 , 3 , 4 , 5 and E15-1 , 2 , 3 ) to the ancestor . All eight clones have increased gene expression from HMLα and reduced haploid-specific gene expression . ( D ) The importance of repressing haploid-specific gene expression for the evolved phenotype was assessed by comparing cells that expressed both alleles of the mating type information ( HMLα ) to cells that expressed only MATa information ( hmlαΔ ) . The frequency of clumps larger than 3 to 4 cells was compared in asynchronous cultures of HMLα and hmlαΔ derivatives of each clone . Four clones from E15 and three from E6 cannot form clumps when HMLα is deleted . E6-5 was unaffected by deletion of HMLα . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 01210 . 7554/eLife . 04875 . 013Figure 6—source data 1 . Amino acid change of candidate causal mutations . ( A ) Table of the candidate causal mutations identified in clones E6-1 , 2 , 3 , 4 and the resulting amino acid change . ( B ) The entire WHI2 gene was PCR-amplified , Sanger sequenced , and aligned to the wild type sequence of WHI2 to identify mutations . DOI: http://dx . doi . org/10 . 7554/eLife . 04875 . 013 We asked whether the oscillations in other lineages also depend on the silencing of haploid specific genes . We measured mRNA levels of the two genes in HMLα , one of which , HMLα2 , inhibits haploid-specific gene expression , and three representative haploid-specific genes in eight other clones derived from our two evolved populations ( E6 and E15 ) . All nine clones have increased expression of HMLα and reduced expression of haploid-specific genes when compared to the ancestor ( Figure 6C ) . Furthermore , deleting HMLα in these clones resulted in a loss of multicellular aggregates in all but one clone ( Figure 6D , E6-5 ) , even though only one of the clones ( E6-2 ) has the sir4 allele found in E6-1 . Do the oscillations that we detect have any relationship to other oscillations that have been detected in budding yeast ? Fluctuations in metabolic activity can be driven by temperature oscillations with a 24-hr period but rapidly die out at constant temperature ( Eelderink-Chen et al . , 2010 ) , and a particular regime of growth to high density can induce prolonged metabolic oscillations whose period is 4–5 hr ( Tu et al . , 2005 ) . We think it is unlikely that either of these cycles are connected to the behavior we observe: the 24-hr cycles do not free-run and the shorter cycles have the wrong period and can only be seen at cell densities higher than those that we used . We cannot , however , exclude the possibility that the behavior we see is the result of coupling changes in cell–cell adhesion to a previously unknown circadian clock in budding yeast . How similar is the diurnal oscillator we evolved to naturally occurring circadian clocks ? All circadian clocks share four features: ( 1 ) a limit-cycle , cell autonomous cellular oscillator that continues to oscillate in the absence of environmental cues ( free run ) ; ( 2 ) the oscillator maintains a roughly 24-hr period over a range of temperatures ( temperature compensation ) ; ( 3 ) the period of the internal oscillator can be synchronized , or entrained , to the phase of a daily environmental cue ( e . g . , light , temperature ) ; and ( 4 ) the ability to control cellular and organismal behavior ( Young and Kay , 2001; Tauber et al . , 2004; Bell-Pedersen et al . , 2005; Rosbash , 2009 ) . Our results show the yeast oscillator we evolved has two of these four behaviors: its roughly 24-hr period is maintained in the absence of environmental cues or selection on the output of the oscillator ( Figures 2B , 3A , 3C , 4C , and 4D ) , and it regulates a cellular behavior , the ability of cells to form clumps . Further investigations are required to determine if the yeast oscillator uses the combination of positive and negative feedback loops that form the core of circadian clocks and could be evolved to be temperature-compensated and entrained by environmental signals . Our experiment shows that mutating genes in multiple , unrelated biochemical networks can produce evolutionary novelty and reveals that identifying adaptive mutations is often insufficient to explain the mechanism of these new phenotypes . The ability of multiple combinations of mutations to produce the same complex phenotype leads to two inferences . First , we know relatively little about the circuitry of many existing molecular pathways and evolving novel traits may help illuminate previously unknown connections between pathways . Second , that multiple mutations are capable of producing unexpected changes in cellular physiology by connecting previously unrelated pathways to produce new circuits . The few mutations that are shared between different oscillating lineages suggest that there are multiple , functionally different ways that such rearrangements can lead to the same phenotype . The common loss of silencing despite little commonality in other mutated genes between evolved clones has two implications . First , it suggests that wild yeast may regulate their clumpiness during the diploid phase , which accounts for most of their life cycle , and that our evolved populations have exploited this regulation to respond to our selection for oscillatory behavior . Second , the independent evolution of the same trait appears to have a mixture of common and diverse features . All nine of the populations we examined have switched from the haploid- to the diploid-specific pattern of gene expression and all four lineages analyzed in detail have mutated WHI2 . The repeated occurrence of these genetic changes suggests that they greatly increase the number of subsequent mutations that could lead to the appearance of an oscillator . Whether the later mutations act on a single pathway or multiple , different pathways will require the detailed dissection of the oscillator's mechanism . It would be interesting to determine whether a selection for oscillation in diploids , where the inactivation of genes requires two independent genetic changes , would take a similar trajectory and produce the same mechanism that we have uncovered in haploids . The inconsistency among bulk segregant analysis , gene deletions and wild type reversions illustrates the difficulty in identifying the contribution of different mutations to a complex , evolved phenotype . We can nevertheless make inferences about the molecular underpinnings of evolutionary novelties . Four of the six mutations that are strongly identified as causal mutations are loss of function mutations . The mutations in SCW11 and WHI2 are nonsense mutations , whereas those in SIR4 and YJL070C can be replaced by gene deletions , but not by restoration of the wild-type gene . The preponderance of loss of function mutants suggests that sophisticated functions can be produced primarily by inactivating genes , rather than modifying or increasing their activities . Whether similar paths are followed in nature is likely to depend on the strength and unidirectionality of selection and the extent of antagonistic pleiotropy: the contrast between the benefits of a mutation in one environment and its costs in others . We speculate that short bursts of intense selection , such as those that occur when a species occupies a new and unoccupied niche , could lead to adaptation that is primarily driven by gene inactivation , whose short-term benefits exceeds its costs . These costs could later be reduced either by suppressor mutations that would minimize the harmful effects of loss of function mutations , while maintaining the evolved phenotype , or even by reversion of the original mutation , once the evolved phenotype had been stabilized by further mutations . Further investigation will be required to determine how often loss of function mutations participate in a variety of evolutionary processes , ranging from cancer to the formation of new species and forms of biological organization . The W303 S . cerevisiae strain background was used for all experiments . Supplementary file 1 provides a detailed list of each strain used . Standard rich media ( YPD ) was used for all experiments: 2% Peptone , 2% D-Glucose , and 1% Yeast-Extract supplemented with Penicillin/Streptomycin ( Sigma P0781 ) . Analysis and sorting was performed on a MoFlo Legacy ( Dako Cytomation/Beckman Coulter ) with two excitation lasers and respective filter settings: 488 nm-550/30 , and 594 nm-630/40 . FACS FCS files were exported as tab delimited files using FlowJo software and plotted using R . Ancestral cells were inoculated into liquid media from a single colony and first grown for >12 hr in exponential phase at 30°C . At a fixed time of the day , cultures were centrifuged at 3000 rpm for 5–10 min and placed on ice . Cells were immediately brought to the FACS where ∼5 × 105 cells were collected from the bottom 6% or 15% of the YFP distribution . Data were collected from the first several thousand cells to set the gates for the selection that immediately followed . Cells were grown at 30°C in 40 ml YPD in beveled flasks for ∼10 hr rotating at 110 rpm , centrifuged , and placed on ice for FACS selection . ∼5 × 105 cells were collected from the top 6% or 15% of the YFP distribution , grown at 30°C in beveled flasks containing 100 ml YPD for ∼14 hr . This cycle was repeated for 30 continuous days . Cell densities were occasionally measured before FACS selection to ensure the population stayed below 1 × 106 cells/ml . A small portion of the evolving population was occasionally frozen in 15% glycerol to establish a historical record . After 30 days of selection , clonal lines were isolated by spotting single events with FACS from the top 6% or 15% of the YFP distribution onto YPD plates . Resulting colonies were restreaked again on YPD to ensure the colonies were from a single cell . For evolved populations , E6 and E15 , a portion of the frozen stock was inoculated directly into liquid media and diluted to first grow exponentially overnight for >12 hr at 30°C . For clones , a single colony was inoculated into liquid media and diluted overnight for >12 hr , 30°C . At a fixed time of the day , which differed between different experiments , cells were centrifuged at 3000 rpm for 5–10 min , placed on ice , and immediately brought to the FACS machine . The dimmest 6–10% YFP expressing cells were collected from the unsynchronized population and grown in liquid media at <5 × 10^5 cells/ml for 10 hr at 30°C . Cells were then centrifuged , placed on ice , and brought to the FACS . Data were collected from the first several thousand cells to set the gates for the selection that immediately followed . The brightest 6–10% YFP cells were collected and grown at <5 × 10^5/ml for 14 hr at 30°C . Cells were prepared for FACS , data were collected while gates were set and the dimmest 6–10% YFP cells were collected to begin the next cycle . Unsynchronized populations were prepared for FACS in an identical manner to synchronized populations but cells were collected randomly from the entire YFP distribution . For free run experiments , populations were synchronized by FACS for 3 days and then split into two cultures . One culture was synchronized for another 3 days and the other was kept exponentially growing by transferring ∼1 × 104 cells into fresh media and analyzing the remaining cells by FACS every 10 and 14 hr for 3 days . A time 0 hr , a single cell was deposited into a well of a 96-well plate containing 10 µl of YPD and immediately imaged with an inverted microscope . 190 µl of YPD was slowly added to the well , and the plate was placed at 30°C , unshaken . Every 3 . 5 hr , the plate was removed from the incubator , gently tapped by hand , quickly imaged , and placed back at 30°C . After 24 hr , the cells were collected and three cells regardless of their clump size were randomly chosen from the population by depositing every 200th FACS event from the entire YFP distribution into new individual wells containing 10 µl of YPD . Cells were imaged and 190 µl of YPD was added to each well and placed at 30°C , unshaken and imaged every 3 . 5 hr as described above . After another 24 hr , the three populations were removed from the plate and three cells from each population were randomly deposited into new individual wells ( 9 total ) using FACS as described above and imaged every 3 . 5 hr for 24 hr . For each time point , the average area of cell clumps was determined using ImageJ . A minimum of 250 cells were counted for each time point always making sure to count at least 50% of the population by choosing random fields of view to quantify . qPCR was performed using published methods ( Koschwanez et al . , 2013 ) . See Figure 3A for a definition of the F75:F25 ratio . A small fraction of the glycerol stock from each time point of the evolved population was collected and genomic DNA was immediately purified . A small region ( ∼300 bp ) surrounding each mutation was PCR amplified and Sanger sequenced . Frequencies were calculated as the ratio of the mutant and wild type peak heights on the sequencing electropherogram . Experimental procedures are described in the text . A description of the microscope and software used for acquisition is described elsewhere ( Koschwanez et al . , 2013 ) . All images were modified for publication using imageJ . Cells were centrifuged , washed twice in PBS , and incubated with 100 µg of Oregon Green 488X succinimidyl ester ( Life Technologies ) in 250 µl of 0 . 1 M Sodium Bicarbonate in PBS for 3 to 5 min at room temperature . Dye was quenched with an excess of YPD , cells centrifuged , washed , and resuspended in YPD . For Figure 6D , the ratio of cells above and below the pulse width ( Picot et al . , 2012 ) size of 14 . 5 µM polystyrene beads ( Spherotech ) was reported . To determine the clump size that corresponded to the pulse width of 14 . 5 µM beads , ancestor and E6-1 cells were collected by FACS onto a glass slide from a narrow gate that matched the 14 . 5 µM beads pulse width distribution . ∼500 cells were scored for their clump size three separate times by eye with a microscope . The average clump size was the same for ancestor and E6-1 cells . E6-1 was mated to yGW 556 , a MATα derivative of its ancestor and the resulting diploid was put through meiosis . The resulting spores were released from the ascus by treating the population with 50 µl of 2 mg/ml zymolyase ( Zymo Research ) diluted in water for 1 hr at 30°C . 450 µl of 1% Triton X-100 diluted in water was added and cells were sonicated for 10–20 s . Cells were centrifuged at 6000 rpm for 1 min and resuspended in YPD and grown overnight in exponential phase . 5 × 105 spores were selected in bulk for oscillations in the same manner E6 was originally evolved . Because one of the causal mutations eliminates silencing of HML and HMR , the spores that can cycle are incapable of mating . After 10 days of selection , genomic DNA was purified from the remaining pool of cells and sequenced at high coverage . The frequency of mutations in the mapped reads was scored using published methods ( Ares , 2012; Koschwanez et al . , 2013 ) . Six putative causal mutations were identified and engineered into a wild-type strain but were insufficient to recreate the oscillator phenotype . This strain was mated to E6-1 , sporulated , and individual spores were tested for oscillations . Equal amounts of 30 oscillating spores were pooled and genomic DNA was sequenced to identify the remaining two putative causal mutations in SIR4 and SCW11 . The recreated oscillator strain was mated to yGW 556 , a MATα derivative of its ancestor to produce a diploid heterozygous for the eight causal mutations . To generate a pool of spores that could not mate during bulk selection and contained all possible combinations of E6-1 mutations , PSTE2-URA3 , a marker that selects for MATa spores when plated on media lacking uracil and LEU2-sir4-100 , a marker that selects for the sir4-100 allele ( which is sterile ) when plated on media lacking leucine were introduced into the diploid strain . This strain was sporulated and the resulting spores were released from the ascus as described above and plated on either media lacking uracil or leucine . Equal numbers of cells from each media were pooled together , a fraction of this pool was frozen down to measure the initial frequency of each mutation ( starting population ) and the remaining cells were selected by FACS for the noted traits twice a day for 5 days . For exponential growth , cells were diluted by batch culture twice a day to keep them in exponential phase . After 5 days , genomic DNA was isolated from the pool of surviving spores , and the allele frequency of each mutation was determined by methods described above . Genomic DNA ( gDNA ) was purified by resuspending >1 × 108 cells in 50 µl of 0 . 5 M EDTA ( pH = 7 . 5 or 8 . 0 ) , 200 µl filtered water , and 2 . 5 µl of 20 mg/ml Zymolyase and incubating at 37°C for 1 hr . 50 µl of miniprep mix ( 0 . 20 M EDTA [pH = 8 . 0] , 0 . 40 M Tris [pH = 8 . 0] , 2% SDS ) was added , mixed by inversion and incubated at 65°C for 30 min . 63 µl of 5 M KAc was added , mixed by inversion , and incubated on ice for 30 min . The cell lysate was centrifuged at 15K rpm for 10 min , and the supernatant was transferred to a new tube containing 720 µl of 100% Ethanol . gDNA was pelleted by centrifugation at 15K rpm and resuspended in 100 µl of water containing 1 µl of RNaseA ( 10 mg/ml ) ( Sigma ) and incubated for 1 hr at 37°C . 2 µl of Proteinase K ( 20 mg/ml ) ( Sigma ) was added and incubated for 2 hr at 37°C . 300 µl of isopropanol was added , and gDNA was pelleted by centrifugation at 15K rpm , washed in 80% Ethanol , air dried for 10 min , and resuspended in 10 mM Tris , pH 8 . 0 . Total RNA was isolated following published protocols ( Ares , 2012; Koschwanez et al . , 2013 ) . Genomic DNA and RNA sequencing libraries were made using Illumina Truseq DNA and RNA kits , respectively . Genome and mRNA sequencing was performed on an Illumina Hiseq 2000 , reading 150 base pair paired-end reads .	In living things , many important behaviors—including animal heartbeats and sleep patterns—happen in cycles . Machines called oscillators , which are found inside cells , control these behaviors . There are many different oscillators and they share some common features , despite involving different genes . Each oscillator is formed of a set of genes that interact with each other to drive regular cycles lasting seconds , hours , or even months . The oscillators do not need any cues from the environment to maintain these cycles . However , cues such as light or temperature can keep the oscillator synchronized with the environment . To ask how complex machines like oscillators could evolve , Wildenberg and Murray inserted a gene that makes a fluorescent protein into budding yeast , a single-celled species that does not have an oscillator with a period of 24 hr . These yeast cells were then selectively grown through a few hundred generations to experimentally evolve a yeast strain where the levels of protein fluorescence regularly alternated over 24-hr periods . Wildenberg and Murray then carried out further experiments to discover the cause of the pattern of protein fluorescence . These revealed that the pattern was due to the yeast cells alternating between forming clumps of multiple cells and living separately . The genes that mutated to create the cycles of cell clumping in the yeast all appear to have unrelated roles . The 24-hr oscillator that evolved in the yeast has some of the features of the biological oscillators found in nature . It maintains regular cycles even without any cues from the environment and it can control a cell behavior , but the oscillator appears to be unable to accept cues from the environment , a universal property of naturally evolved circadian clocks . Further work to understand how the genes work together in the oscillator will help to better understand how 24-hr oscillators in nature can evolve from genes that lack any 24-hr behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2014	Evolving a 24-hr oscillator in budding yeast
Social interactions have a major impact on well-being . While many individuals actively seek social situations , others avoid them , at great cost to their private and professional life . The neural mechanisms underlying individual differences in social approach or avoidance tendencies are poorly understood . Here we estimated people’s subjective value of engaging in a social situation . In each trial , more or less socially anxious participants chose between an interaction with a human partner providing social feedback and a monetary amount . With increasing social anxiety , the subjective value of social engagement decreased; amygdala BOLD response during decision-making and when experiencing social feedback increased; ventral striatum BOLD response to positive social feedback decreased; and connectivity between these regions during decision-making increased . Amygdala response was negatively related to the subjective value of social engagement . These findings suggest a relation between trait social anxiety/social avoidance and activity in a subcortical network during social decision-making . Pursuing interpersonal relationships is a powerful human drive . Social relationships contribute to the feeling that life has meaning ( Baumeister and Leary , 1995 ) and social isolation is a major health risk factor ( House et al . , 1988; Cacioppo et al . , 2015 ) with an influence on mortality risk comparable with smoking or alcohol consumption ( Holt-Lunstad et al . , 2010 ) . Several social stimuli and situations can act as a reward ( Krach et al . , 2010 ) , including beautiful faces ( Aharon et al . , 2001; O'Doherty et al . , 2003 ) , praise and attention ( Izuma et al . , 2008 ) , anticipation of positive social feedback ( Spreckelmeyer et al . , 2009 ) and interactions in the game Cyberball ( Kawamichi et al . , 2016 ) . However , the value of engaging in social interactions varies across individuals ( Cheek and Buss , 1981 ) . Social situations are a source of pleasure for sociable people for which they invest time and money , such as when inviting friends for celebrating one’s birthday ( Dunn et al . , 2008 ) . In contrast , socially anxious people tend to avoid social situations in order to avoid receiving negative social feedback and experiencing feelings such as fear and shame ( Clark and Wells , 1995 ) . The neural mechanisms underlying the decision to engage in a social situation and their variations across individuals are unclear . To better understand these mechanisms , we first aimed to quantify how much individuals value engaging in a simple social situation . We measured the choice frequency of engaging in the social situation as opposed to obtaining a monetary amount , from which we inferred the value that engagement has for a given individual . A reduced frequency of choosing the social situation would indicate a reduced value of the situation , and a tendency to avoid the social situation . We hypothesised that considering whether to engage in a social situation and experiencing its outcome would evoke avoidance-related neural signals proportional to the individual level of social anxiety , specifically increased BOLD response in the amygdala , a neural structure strongly associated with fear and avoidance ( Shackman and Fox , 2016; Terburg et al . , 2018 ) , decision-making ( Jenison et al . , 2011; Grabenhorst et al . , 2012 ) and stimulus valuation ( Paton et al . , 2006 ) . Further , we hypothesised that the reduced value of the social situation associated with higher levels of social anxiety would be reflected in reduced reward-related activations in the ventral striatum ( Schultz et al . , 1997; O'Doherty et al . , 2004 ) in response to positive outcomes of the situation . Finally , we hypothesised that social anxiety may affect functional connectivity between amygdala and ventral striatum , a pathway implicated in active avoidance behaviour ( Ramirez et al . , 2015 ) . As expected , participants’ likelihood of choosing the safe option and thus avoid the gamble increased as a function of the amount offered as an alternative to the gamble , with both human and computer partners ( Figure 2A ) . Based on the choice frequencies , we calculated the certainty equivalent ( CE50 ) of the gamble against each partner ( CE50 Human; CE50 Computer ) , which represents a measure of the subjective value of engaging in the situation . We expected that the value of engaging in the gamble against the human , but not against the computer , would be lower in participants with more pronounced social anxiety traits ( =higher LSAS score ) . We thus computed the difference between the estimated values of the two gambles ( CE50 Human minus CE50 Computer ) , which yields a measure of the value of social engagement controlled for variations in risk preference . A multiple regression model with LSAS , AQ and gender as predictor variables could explain individual differences in the value of social engagement ( global F ( 3 , 64 ) = 3 . 2 , p=0 . 028 , R²=0 . 13 ) . Only LSAS contributed significantly to the prediction of the interindividual variations in the value of social engagement ( value decreased with LSAS: t ( 64 ) = −2 . 99 , p=0 . 004 , Pcorr = 0 . 012 Bonferroni-corrected for the use of three predictor variables , unstandardized coefficient = −0 . 01 Euro/LSAS point , adjusted R2 = 0 . 09 , Figure 2C , Supplementary file 1C ) , while variations in AQ and gender did not ( respectively: t ( 64 ) = 1 . 75 , p=0 . 084; t ( 64 ) = 0 . 24 , p=0 . 81 ) . To test if differences in avoidance as a function of social anxiety were due to differences in the perception of the social feedback , we analysed valence ratings of the feedback stimuli . As expected , ratings of positive and negative feedback differed ( F ( 1 , 66 ) = 355 . 8 , p<0 . 001 , ηp²=0 . 84 , higher ratings for positive feedback ) , ratings for human and computer partners did not ( F ( 1 , 66 ) = 0 . 4 , p=0 . 43 , ηp²=0 . 005 ) , and ratings differed more between negative and positive human feedback than computer feedback ( interaction between partner and positive vs . negative feedback: F ( 1 , 66 ) = 234 . 8 , p<0 . 001 , ηp²=0 . 78 ) . Importantly , none of these ratings varied with social anxiety ( all F ( 1 , 66 ) < 0 . 24 , all p>0 . 6 , all R²<0 . 01 ) . We replicated our findings in a separate group of 47 participants and verified the specificity of the effects of social anxiety by including more personality trait measures . A multiple linear regression with LSAS , AQ , gender , BDI and STAI - T as predictor variables could explain individual differences in the value of social engagement ( global F ( 5 , 39 ) = 4 . 3 , p=0 . 004 , R²=0 . 35 ) . Again , only LSAS contributed to predicting interindividual variations in the value of social engagement ( value decreased with LSAS: t ( 39 ) = −3 . 4 , p=0 . 002 , Pcorr = 0 . 01 Bonferroni-corrected for the use of five predictor variables , unstandardized coefficient = −0 . 007 Euro/LSAS point , adjusted R2 = 0 . 27 , Figure 2D , Supplementary file 1D ) , while variations in AQ , gender , BDI and STAI-T did not ( all \|t ( 39 ) \|<1 . 4 , p>0 . 17 ) . Individual values of engaging in the gamble against human and computer partners ( CE20 , CE50 and CE80 ) obtained in this behavioural experiment were used during scanning of these participants in a slightly modified version of the task ( Materials and methods ) . Participants underwent fMRI scans after the behavioural experiment; the data of 42 participants were analysed ( data from five participants with excessive head motion were excluded; see Materials and methods ) . To increase comparability of the BOLD data across participants , we aimed to equalise the ratio of safe vs . risky choices taken by each participant during scanning by using individually calculated CE20 , CE50 and CE80 for the gamble against human and computer partners ( see above and Materials and methods ) . Data acquired during scanning revealed that , as expected , the proportion of safe choices increased from CE20 through CE50 to CE80 ( F ( 2 , 80 ) = 53 . 9 , p<0 . 001 , ηp²=0 . 57 ) , that the proportion of safe choices did not significantly vary between human and computer partners ( F ( 1 , 40 ) = 2 . 35 , p=0 . 13 , ηp²=0 . 06 ) , that there was no significant interaction between these factors ( F ( 2 , 80 ) = 1 . 96 , p=0 . 15 , ηp²=0 . 05 ) , and crucially that there were no effects of social anxiety on these decisions ( all F ( 1 , 40 ) < 2 . 3 , all p>0 . 1 , all ηp²<0 . 06 ) . Two participants who never chose the risky option against the human partner during the scan were excluded from further analysis because their neural response during that crucial experimental condition could not be assessed . Based on the regression results linking social anxiety to variations in the subjective value of engaging in the social situation , we estimated a cost of social avoidance associated with social anxiety . The average LSAS across participants of both studies was 29 . 02 , their average earning 1 . 93 Euros , and the slope −0 . 008 Euros/LSAS point; therefore a person with an LSAS of 0 would earn 1 . 93 + 0 . 00829 = 2 . 162 Euros or 12% more than the average participant , while a person with an LSAS of 60 , likely to suffer from generalised social phobia ( Rytwinski et al . , 2009 ) , would earn 1 . 93–0 . 00831 = 1 . 682 Euros or 12 . 8% less than the average participant and 22 . 2% less than a person with an LSAS of 0 . This study shows that trait social anxiety is associated with reduced subjective valuation of engaging in a social situation , and amygdala and ventral striatum activation and functional connectivity differences related to social anxiety during social decision-making , both at the decision stage and when experiencing the outcome of a social situation . Interestingly , both relatively more anxious and relatively more sociable participants deviated from economically optimal decisions in order to either seek ( more sociable participants ) or avoid ( more anxious participants ) the social situation . This mirrors real-life behaviour , where sociable people spend money to interact with other people whereas socially anxious participants may instead spend money in order to avoid social interactions . Crucially , the observed behavioural effects were ( i ) specific to interactions with a human partner and thus not due to general risk aversion , which is increased in anxiety ( Hartley and Phelps , 2012 ) , ( ii ) not related to differences in the subjective perception of the human feedback stimuli as a function of anxiety , and ( iii ) specific for social anxiety ( measured with the LSAS ) and not with measures of related but non-specific traits such as general anxiety ( STAI-T ) , depression ( BDI-II ) or autistic traits ( AQ ) . BOLD signal data acquired during this task revealed that amygdala activation may be related to the decision-making process: participants’ safe or risky choices could be decoded from activation at the time of decision-making in the right amygdala . These findings are compatible with the previously reported involvement of the primate amygdala in social decision-making ( Chang et al . , 2015; Grabenhorst et al . , 2013 ) . While decoding accuracy did not vary with social anxiety , amygdala activation , both at the time of social decision-making and during the outcome phase of the social situation ( i . e . when participants experienced social feedback ) , varied with social anxiety and with the value of social engagement estimated prior to the scan . These findings are compatible with findings of increased amygdala responses to threat stimuli in people with higher trait anxiety ( e . g . Etkin et al . , 2004 ) and extend this association to the domain of social decision-making . Nucleus accumbens response to receiving positive social feedback , that is winning the game of dice against the human compared to winning against a computer partner , decreased with social anxiety . In contrast , no effects of social anxiety were found on the response to ‘no win’ outcomes . These results are compatible with previous findings indicating that social anxiety is associated with reduced striatal response to different kinds of social rewards ( Richey et al . , 2014; Sripada et al . , 2013 ) . Social anxiety influenced the functional connectivity between the left nucleus accumbens and both amygdalae , and between right nucleus accumbens and the perigenual anterior cingulate cortex ( pACC ) . Connectivity between amygdala and nucleus accumbens during the decision to engage in the social interaction increased with social anxiety . These findings are consistent with the importance of this connection for avoidance behaviour: Disruption of the ( basolateral ) amygdala - nucleus accumbens ( shell ) projection in rats has been shown to impair avoidance behaviour ( O'Doherty et al . , 2004 ) . Our findings suggest that in humans , functional connectivity between these regions when engaging in a social interaction is greater in people with more pronounced social anxiety . In contrast , nucleus accumbens – pACC connectivity decreased proportionally with increasing social anxiety . Several previous studies have reported a reduced functional connectivity in social anxiety between parietal , limbic and executive network regions during resting state or in response to presentation of emotional faces ( Brühl et al . , 2014 ) . The pACC is functionally dissociable from subgenual anterior cingulate ( Pezawas et al . , 2005; Apps et al . , 2016 ) , connected to ventral striatum ( Kunishio and Haber , 1994; Margulies et al . , 2007 ) and reduced in volume in people with a genotype associated with increased anxiety-related temperamental traits and risks for depression ( Pezawas et al . , 2005 ) . pACC has also been associated with the response to social stress and negative social feedback ( Dedovic et al . , 2009; Lederbogen et al . , 2011 ) as well as with specific social cognitive functions such as the tracking of other people’s motivation ( Apps et al . , 2016 ) . A dysfunction of pACC and its regulatory influence on amygdala and ventral striatum has even been associated with poor response to chronic social defeat , a possible pathophysiological mechanism of schizophrenia ( Selten et al . , 2017 ) . Our findings are compatible with the idea that these differences in regulatory influence of pACC on subcortical circuits are also observable during social decision-making . Our neuroimaging findings described above reveal a multifaceted variation in the subcortical neural correlates of both decision-making and processing of social feedback associated with social anxiety and social avoidance . These response variations are compatible with increases of threat-related responses , reduced social reward signals , increased avoidance-related network activity and reduced top-down feedback on these networks . While causal relationships between these findings cannot be assessed , all may play a role in social avoidance and social anxiety . Interestingly , the combination of higher threat-related amygdala and lower reward-related ventral striatum activity have recently been associated with increased risk of developing alcohol use disorder in response to stress ( Nikolova et al . , 2016 ) . As engaging in social decisions can be stressful for many people , paradigms such as ours may allow to investigate the link between social decisions , social stress and mental disorders . Further studies specifically designed to investigate connectivity between these structures will be required to elucidate the details of the neural mechanism underlying social avoidance . Avoidance behaviour was not reflected in explicit valence judgments of the feedback stimuli . Our effects thus seem not to have been driven by differences in the conscious interpretation of the feedback but rather by differences in the desire to expose oneself to such feedback . While reduced explicit approachability ratings for positive facial expressions have been reported in social anxiety ( Campbell et al . , 2009 ) , socially anxious individuals in another study have shown avoidance behaviour without reduced valence ratings ( Heuer et al . , 2007 ) . It thus appears that while implicit measures often show avoidance , effects on explicit measures may be task-dependent ( Staugaard , 2010 ) . If our task can be taken as an example of behaviour in real-life social interactions , it may be used to quantify one aspect of the costs resulting from the avoidance behaviour observed in social anxiety . Our linear regression results suggest that persons likely to suffer from generalised social phobia would earn about 22% less in this game than people with an LSAS score of 0 . It may be interesting to combine our experimental approach with estimations based on patient’s costs related to medical care or productivity loss ( Wittchen et al . , 2000; Patel et al . , 2002; Stein et al . , 2005; Acarturk et al . , 2009 ) in future studies . It may also be interesting to run our study on patients with diagnosed social anxiety to evaluate whether our task could be useful as an experimental rather than a self-report-based measure to identify socially anxious individuals , an advance aligned with the NIMH’s RDoC proposal ( Insel , 2014 ) . There are several limitations to the work presented here . First , the number of participants in both studies is relatively low for the assessment of interindividual differences . While the sample size of Study two roughly corresponds to the number of participants recommended by a power analysis based on the results of Study 1 ( see Materials and methods ) , effect size estimates for inter-individual differences based on such a relatively low number of participants are bound to be imprecise ( see size of the confidence bounds in Figures 2 , 4 and 5 ) ( Schönbrodt and Perugini , 2013 ) . One should thus be careful when interpreting our results , including our estimates of the costs of social avoidance . Another issue concerns our study sample: our participants were young German participants , most of them students at the local university , who were willing to subject themselves to behavioural and neuroimaging experiments performed by unknown experimenters; such situations are likely to be quite distressing for individuals with severe levels of social anxiety . This restriction is reflected in the limited range of LSAS scores of our sample: only 11% of participants showed LSAS levels found in people suffering from generalised social phobia ( >60 ) . We must thus exercise caution when drawing inferences about clinical populations based on our results . Next , while we could significantly decode participants’ choice from activation in the right amygdala , the average accuracy was quite low . Therefore , caution must be used in interpreting this finding: the amygdala cluster identified in our analysis is unlikely to be the major contributor to participants’ choices . Further studies specifically designed for a decoding analysis and investigating additional brain regions are required to better understand the neural mechanism underlying the decision-making process in our task . Despite these considerations , the fact that we could replicate the link between social anxiety and our experimental measure of the value of social engagement across two studies , with almost identical slope estimates , suggests that our approach has identified an interesting link between a relevant but subjectively reported , real-world behavioural trait and a controlled experiment . The studies fulfilled all relevant ethical regulations and were approved by the local ethics committee of the Medical Faculty of the University of Bonn , Germany . All subjects gave written informed consent and the studies were conducted in accordance with the latest revision of the Declaration of Helsinki . Subjects were remunerated for their time ( 10 Euros/hour ) and received game earnings ( 0–6 Euros ) . 68 healthy participants ( 24 male , mean age 25 . 2 , range 20 to 37 ) participated in Study 1 ( behaviour only ) and 47 healthy participants ( 19 male , mean age 28 . 5 , range 22 to 40 ) participated in Study 2 ( behaviour and fMRI ) . Participants were recruited from the local population through advertisements on online blackboards at the University of Bonn and on local community websites , and through flyers posted in libraries , university cafeterias and sports facilities ( recruitment period: May 2016 – August 2017 ) . The number of participants recruited in Study two roughly corresponds to the sample size estimated for a point biserial model test based on the results of Study 1 [R2 = 0 . 13 , one-tailed test with alpha error = 0 . 05 and power ( 1-beta ) =0 . 8 , sample size = 43; GPower 3 . 1 ( Faul et al . , 2007 ) ] . Participants were remunerated for their time ( 10 Euros/hour ) and received game earnings ( 0–6 Euros ) . The data from seven participants in Study two were excluded from the fMRI data analysis: five participants moved their head too much for reliable motion correction ( >3 mm or >3° ) and two were excluded because they never choose the risky option in trials with human partners , prohibiting the analysis of neural responses during this condition . Data of individual participants were excluded from the ROI analysis if less than 10 voxels were included in their first-level GLM mask for the ROI considered , indicating insufficient MR signal quality . All subjects gave informed consent and the ethics committee of the Medical Faculty of the University of Bonn , Germany approved all studies . Study 1 consisted of one behavioural experiment , Study 2 ( different participants ) consisted of the same behavioural experiment and a subsequent fMRI experiment . The task was inspired by previous studies quantifying the value of social stimuli ( Deaner et al . , 2005 ) . It was implemented in MATLAB ( Version R2016b; RRID:SCR_001622; The MathWorks , Inc , Natick , MA ) using the Psychtoolbox extensions ( RRID:SCR_002881; http://psychtoolbox . org ) . Participants decided on each trial whether to play a game involving a gamble against a partner ( risky option ) or not ( safe option ) ( Figure 1A and B ) . The game was played in two separate blocks of 126 trials against human partners ( Figure 1C; 21 trials with each of the six partners ) and a computer partner . Each trial proceeded as follows ( Figure 1A ) : The risky option ( gamble against the partner ) was shown as an image on one side of the screen , and the safe option ( amount of money obtainable as an alternative ) was shown on the other side of the screen ( screen sides were chosen randomly on each trial ) . Participants then chose the risky or safe option under no time pressure . If the participant chose the risky option , virtual dice were rolled for the participant and the partner , and the higher number won ( 50% winning chance , explicitly stated to participants; equal numbers led to immediate repetition of the dice roll , not displayed ) . The outcome was revealed after a delay of 1 s: if the participant won , they received 3 Euros , if they did not , they received 0 Euros . Importantly , the participant witnessed the partner’s reaction ( =feedback ) to the outcome of the game at the same time . Feedback from human partners consisted in videos ( duration 1 s ) of facial expressions of admiration in case of winning , and condescension in case of no win . As in previous neuroimaging studies ( Schultz and Pilz , 2009; Schultz et al . , 2013 ) , we used videos of facial expressions instead of pictures as feedback because such stimuli have higher ecological validity and lead to improved recognition of emotions ( Krumhuber et al . , 2013 ) : dynamic displays of emotions are judged as more natural ( Sato and Yoshikawa , 2004 ) , lead to higher judgments of intensity and arousal ( Biele and Grabowska , 2006 ) , improve the recognition of subtle facial expressions ( Ambadar et al . , 2005; Cunningham and Wallraven , 2009 ) and evoke stronger and more widespread neural responses than static faces ( Schultz and Pilz , 2009; Schultz et al . , 2013; Puce et al . , 1998; Sato et al . , 2004; Fox et al . , 2009; Arsalidou et al . , 2011 ) , particularly in brain areas associated with social cognition ( Schultz et al . , 2013; Puce et al . , 1998; Furl et al . , 2007 ) . The facial expressions were selected from a validated database ( Kaulard et al . , 2012 ) . Feedback from the computer partner were abstract symbols ( green ‘check’ mark for win or red ‘X’ for no win ) . Perceived valence ratings about all feedback stimuli were collected after the experiment using a visual analogue scale ( range 0–100 ) . The safe option consisted in an amount of money varied between 0 and 3 Euros across trials ( 0 , 0 . 5 , 1 , 1 . 5 , 2 , 2 . 5 , or 3 Euros , equal probability , random order , 18 trials per amount ) . The game’s certainty equivalent ( termed CE50; equivalent to the game’s subjective value ) was defined as the amount of payoff a participant would have to receive to be indifferent between that payoff and the game , and was estimated by fitting a cumulative Gaussian function to each participant’s choice probabilities observed for the amounts offered in the safe option ( Figure 2A and B ) . This CE50 ( and in Study 2 , the amounts needed to obtain 20% and 80% safe choices , CE20 and CE80 ) was calculated for the computer and for the human partners . One trial was randomly selected and paid out at the end of the experiment . Under the simplest assumption ( i . e . neutral attitude towards risk , choose ‘play’ if alternative amount is less than 1 . 5 Euros , ‘play’ or ‘no play’ at equal rates if alternative amount is 1 . 5 Euros , and ‘no play’ if alternative amount is above 1 . 5 Euros ) , the average earning per trial ( and thus per game , as only one trial was ultimately rewarded ) would be 1 . 93 Euro . In Study 2 , participants first performed the same experiment as described above and were then scanned while performing a modified version of the task immediately after the behavioural experiment . The fMRI task included the following modifications: ( i ) partner ( computer or human ) was determined randomly on each trial; ( ii ) at the decision stage , participants were presented with names ( first names associated with the humans , or the word ‘computer’ ) instead of images; ( iii ) the alternative monetary amounts varied between the three possible values CE50 , CE20 and CE80 , determined individually , in order to equate the number of safe and risky choices across individuals with different anxiety levels; ( iv ) the temporal intervals between trials and between phases of the trail ( decision stage and outcome ) were varied between 2 and 11 s following a Gamma distribution to dissociate the responses in the different parts of the trial . There were 72 trials per scanning run ( 36 per human and computer partner ) , and two runs per participant . Social anxiety trait levels of each participant using the self-report version of the Liebowitz Social Anxiety Scale [LSAS , ( Rytwinski et al . , 2009 ) ] . Additional traits were obtained using Beck’s Depression Index [BDI , ( Beck et al . , 1996 ) ] , Spielberger’s Trait Anxiety Scale [STAI-T , ( Spielberger and Gorsuch , 1970 ) ] and the Autism-Spectrum Quotient [AQ , ( Baron-Cohen et al . , 2001 ) ] . Statistics on the behavioural data and neural activation data from regions of interest ( see below ) were performed using JASP software ( JASP Version 0 . 9 . 2; RRID:SCR_015823; JASP Team 2018; jasp-stats . org ) . As data were normally distributed , tests performed included multiple linear regression , repeated-measures ANOVA , and t-tests , including normality tests ( Q-Q plot; Kolmogorov-Smirnov goodness-of-fit test ) . Whole-brain activation statistics were performed with SPM12 software ( RRID:SCR_007037; Wellcome Trust Centre for Neuroimaging , London , UK; http://www . fil . ion . ucl . ac . uk/spm ) running in MATLAB with corrections for multiple comparisons ( for details , see ‘fMRI data analysis’ below ) . All statistical tests were two-tailed . Bayes factors were calculated using default priors and express the probability of the data given H1 relative to H0 ( BF10 , values larger than one are in favour of H1 ) . Effect sizes were calculated using standard approaches implemented in JASP software . Imaging data were collected on a 3T Siemens TRIO MRI system ( Siemens AG , Erlangen , Germany ) with a Siemens 32-channel head coil . Functional data were acquired using a T2 echo-planar imaging ( EPI ) BOLD sequence , with a repetition time ( TR ) of 2500 ms , an echo time ( TE ) of 30 ms , 37 slices with voxel sizes of 2 × 2×3 mm3 , a flip angle of 90° , a field of view of 192 mm and PAT two acceleration . To exclude subjects with apparent brain pathologies and facilitate normalisation of the functional data , a high-resolution T1-weighted structural image was acquired , with a TR of 1660 ms , a TE of 2540 ms , 208 slices with voxel sizes of 0 . 8 × 0 . 8×0 . 8 mm3 and a field of view of 256 mm . Data were then preprocessed and analysed using standard procedures in SPM12 . The first five volumes of each functional time series were discarded to allow for T1 signal equilibration . The structural image of each participant was coregistered with the mean functional image of that participant . Functional images were corrected for head movement between scans by a 6-parameter affine realignment to the first image of the time-series and then re-realigned to the mean of all images . The structural scan of each participant was spatially normalised to the current Montreal Neurological Institute template ( MNI305 ) by segmentation and non-linear warping to reference tissue probability maps in MNI space , and the resulting normalisation parameters were applied to all functional images which were then resampled at 2 × 2×2 mm3 voxel size , then smoothed using an 8 mm full width at half maximum Gaussian kernel . Time series were de-trended by the application of a high-pass filter ( cut-off period , 128 s ) . For the multivariate decoding analysis , data preprocessing was identical except for omission of the normalisation and smoothing steps . Functional data were analysed using a two-stage approach based on the general linear model ( GLM ) implemented in SPM12: individual participants’ data were modelled with a fixed effects model , and their summary data were entered in a random effects model for group statistics and inferences at the population level . For the main ( i . e . non-decoding ) analysis , the fixed effects model implemented a mass univariate analysis applied to normalised data . For the decoding analysis , the same fixed effects model was applied to non-normalised data , and patterns of parameter estimates were used for a multivariate ( multivoxel ) analysis . The accuracy maps resulting from the decoding analysis were then normalised to MNI space and entered into a random effects model for group statistics . The fixed-effects model included the following event types per session: decision to play or not and win or loss outcomes , all modelled separately for the human or computer partner , resulting in eight event types . Regression coefficients ( parameter estimates ) were estimated for each voxel of each participant’s brain . For the main ( univariate ) analysis , linear contrasts were applied to the individual parameter estimates of the response to the experimental conditions , resulting in contrast images . These were subjected to a group-wise random effects ANOVA in order to identify brain regions sensitive to the human partner , and brain regions sensitive to reward , using the BOLD responses collected during the outcome phase of the trial . The former regions were identified by subtracting responses to trials with computer partners from responses to trials with human partners , using a significance threshold of p<0 . 05 , with family-wise error correction for multiple comparisons ( FWE ) at the voxel level across the whole brain . The latter , reward-sensitive regions were identified by subtracting responses to losses from responses to wins in the region of interest , using a threshold of p<0 . 05 FWE-corrected across voxels in the anatomically-defined ventral striatum , as provided in the WFU Pickatlas toolbox for SPM ( RRID:SCR_007378; http://fmri . wfubmc . edu/software/pickatlas ) ( Maldjian et al . , 2003 ) . We restricted our analyses to these regions and to the regions functionally connected to them . Parameter estimates of the response during the decision phase of the trials were then extracted from these regions of interest using MATLAB , as follows . All parameter estimates used for further analysis were conducted based on a leave-one-out procedure ( Kriegeskorte et al . , 2009 ) to avoid non-independence bias in data analysis . Specifically , we ran n = 38 leave-one-subject-out group-level GLM analyses , and each of these GLMs was used to define the clusters of interest for the subject left out ( Esterman et al . , 2010 ) . To further reduce circularity or ‘double-dipping’ issues , the contrasts used to define the clusters ( i . e . , outcome human >outcome computer for the human-sensitive regions and win >loss for the reward-sensitive regions ) were not calculated on the extracted data . The influence of anxiety traits on these neural responses was then assessed using linear regression models implemented in JASP , with social anxiety level as independent variable and parameter estimates as dependent variable . We used the Decoding Toolbox ( https://sites . google . com/site/tdtdecodingtoolbox/ , version 3 . 991 ) for SPM , applied directly to the single-subject fixed effects model described above ( data were not normalised and not smoothed ) . Independent variables were the decision ( play or not ) and the partner ( human or computer ) . Features were the parameter estimates for the corresponding regressors in the model; no dimensionality reduction was used . We ran a classification searchlight analysis on each participant's data , designed to decode the decision to play or not ( chance accuracy = 50% ) , using a support vector machine classifier with default parameters ( LIBSVM; RRID:SCR_010243; parameters: C = 1 , type C-SVC , linear kernel ) , 9 mm search radius , trained on the data of one run and tested on the data of the other run . The resulting individual decoding accuracy maps minus chance were normalised to MNI space and entered into a second-level random effects t-test analysis against 0 in SPM , designed to select voxels with reliable above-chance accuracy values in the participant group . Results were restricted to the ROIs ( amygdalae and bilateral nucleus accumbens , defined anatomically using the WFU Pickatlas toolbox ) , and thresholded at p<0 . 05 , with family-wise error correction for multiple comparisons at the voxel level within the ROIs . For tests of the effects of social anxiety on decoding accuracy , accuracy scores were averaged within each ROI and compared across participants using linear regression . We performed a PPI analysis following the standard procedure in SPM8 , using the gPPI toolbox ( McLaren et al . , 2012 ) ( version 13 . 1 , http://www . nitrc . org/projects/gppi ) . We extracted BOLD signals ( eigenvariates ) from our regions of interest ( amygdala and nucleus accumbens clusters identified as responding more to the human than to the computer partner; see above ) . We then constructed , for each region of interest , another GLM for the PPI analysis identical to the GLM described above , including 1 ) the same psychological regressors ( n = 8 per session ) ; 2 ) the BOLD signal from the region of interest as a physiological factor ( n = 1 per session ) ; 3 ) a set of psychophysiological interaction ( PPI ) factors , which are an interaction of the deconvolved BOLD signal in the region of interest and the psychological factors of interest ( we limited the analysis to the decision phase of the trial; n = 4 conditions per session ) ; 4 ) and movement parameters estimated during motion correction included as confound regressors . All of the regressors except for the physiological factors and the movement parameters were convolved with a canonical HRF . For each participant , regression coefficients of the PPI factors were estimated for each voxel of each participant’s brain , as in the primary GLM analysis described above . Linear contrasts were applied to the individual parameter estimates of the response to the PPI regressors , resulting in contrast images . These were subjected to a group-wise random effects ANOVA in order to identify brain regions showing a change in connectivity with the region of interest at the time of decision-making as a function of the partner , decision taken and participant anxiety level . The threshold used to identify clusters of interest was p<0 . 05 , with family-wise error correction for multiple comparisons at the cluster level across all the voxels of the brain based on an uncorrected threshold of p<0 . 001 at the voxel level . Parameter estimates of the PPI regressors based on the nucleus accumbens activation were extracted from the amygdala ROIs in order to assess the connectivity between these structures using MATLAB . The data that support the findings of this study are available from the corresponding author upon reasonable request .	Your relationships with the people around you – friends , family , colleagues – have a strong influence on your overall life happiness . Even so , many people struggle to engage with the people around them . Social interactions can be stressful and many people choose to avoid them , even at a cost . Being able to measure these tendencies experimentally is a first useful step for assessing social avoidance without relying on people’s , often biased , recollections of their actions and behaviours . But how can a tendency to avoid social situations be quantified ? And what can an experiment to measure this tendency reveal about the neural underpinnings of social avoidance ? Schultz et al . asked volunteers to play a social game . If they played , the volunteers had the chance to win three euros , but they could choose not to play and receive a fixed amount of money , which varied across trials between zero and three euros . This approach allowed Schultz et al . to quantify how much the volunteers valued playing the game . The game involved playing with other virtual human partners , who gave either positive or negative social feedback depending on the outcome of the game in the form of videos of facial expressions . In a non-social control experiment , a computer gave abstract feedback in the form of symbols . Schultz et al . found that the value people placed on playing the social game varied with their level of social anxiety ( established using a standard questionnaire ) . The more anxious people attributed less value to engaging in the game . Neuroimaging experiments revealed that the activity and connectivity between the amygdala and ventral striatum , two parts of the brain involved in processing emotions and reward-related stimuli , varied according to people’s levels of social anxiety . Social interactions have a major impact on the quality of life of both healthy people and those with mental disorders . Developing new ways to measure and understand the differences in the brain linked to social traits could help to characterise certain conditions and document therapy progress . Methods to quantify social anxiety and avoidance are also in line with efforts to explore the neuroscience behind the full range of human behaviour .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	A human subcortical network underlying social avoidance revealed by risky economic choices
Each of the olfactory sensory neurons ( OSNs ) chooses to express a single G protein-coupled olfactory receptor ( OR ) from a pool of hundreds . Here , we show the receptor transporting protein ( RTP ) family members play a dual role in both normal OR trafficking and determining OR gene choice probabilities . Rtp1 and Rtp2 double knockout mice ( RTP1 , 2DKO ) show OR trafficking defects and decreased OSN activation . Surprisingly , we discovered a small subset of the ORs are expressed in larger numbers of OSNs despite the presence of fewer total OSNs in RTP1 , 2DKO . Unlike typical ORs , some overrepresented ORs show robust cell surface expression in heterologous cells without the co-expression of RTPs . We present a model in which developing OSNs exhibit unstable OR expression until they choose to express an OR that exits the ER or undergo cell death . Our study sheds light on the new link between OR protein trafficking and OR transcriptional regulation . Seven transmembrane G-protein coupled receptors ( GPCRs ) , are diverse and the largest superfamily of receptors . Their roles are well established in sensing various stimuli including odorants , tastants , light , hormones , neurotransmitters and proteins . Some GPCRs require the presence of specific accessory proteins such as chaperones , vesicular targeting molecules and co-receptors for their cell surface expression ( Lu et al . , 2003; Salahpour et al . , 2004; Dey and Matsunami , 2011; Wu et al . , 2003 ) . Mammalian olfactory receptors ( ORs ) , which are GPCRs ( Buck and Axel , 1991 ) , are retained in the ER when expressed in non-olfactory cells . RTP1 ( Receptor Transporting Protein 1 ) and RTP2 ( Receptor Transporting Protein 2 ) , both single transmembrane proteins strongly and exclusively expressed in the peripheral olfactory organs ( Lu et al . , 2003; Saito et al . , 2004; Zhuang and Matsunami , 2008; Gimelbrant et al . , 1999 ) , greatly enhance the trafficking of ORs to the cell surface of heterologous cells . However , the role played by the RTPs in vivo remains unclear . The mouse genome encodes over one thousand intact OR genes ( Niimura et al . , 2014 ) , which are expressed in a singular and monoallelic fashion in each olfactory sensory neuron ( OSN ) ( Shykind et al . , 2004; Chess et al . , 1994; Serizawa et al . , 2000; Malnic et al . , 1999 ) . Each OR is not chosen at random; rather , OSNs express different ORs with dramatically varying probabilities ( Khan et al . , 2011; Ibarra-Soria et al . , 2014 ) . OSNs in the olfactory epithelium ( OE ) are organized in overlapping zones defined by the expression of each OR ( Ibarra-Soria et al . , 2014; Ressler et al . , 1993; Vassar et al . , 1993; Miyamichi et al . , 2005; Kanageswaran et al . , 2015; Saraiva et al . , 2015 ) as well as in a pseudostratified manner with progenitor cells forming the basal layer and mature neurons forming the upper layers . Mature OSN dendrites project into the nasal cavity forming a dendritic knob at the surface of the OE where they express ORs to interact with odorant molecules . Mature OSN axons expressing the same OR project to the olfactory bulb ( OB ) to converge onto specific glomeruli ( Mombaerts et al . , 1996; Ressler et al . , 1994; Vassar et al . , 1994; Hayar et al . , 2004; Aungst et al . , 2003; Gire et al . , 2012 ) . When the β2 Adrenergic Receptor ( β2AR ) is expressed instead of an OR , the β2AR –expressing OSNs target their axons to the OB and form glomeruli ( Feinstein et al . , 2004a , 2004b; Omura et al . , 2014; Nakashima et al . , 2013 ) . Hence , the development of the peripheral olfactory system is dependent on functional GPCRs . The mechanisms by which an OSN makes an OR choice have not been fully elucidated . Locus control-region like enhancers scattered on the genome and relative location of ORs from these elements have important roles in determining the probabilities of OR gene choice ( Khan et al . , 2011; Serizawa et al . , 2003; Markenscoff-Papadimitriou et al . , 2014 ) . Various epigenetic mechanisms , for example histone modification by lysine demethylase ( LSD1 ) , allow the escape of an OR gene from repression ( Magklara et al . , 2011; Lyons et al . , 2013 , 2014; Armelin-Correa et al . , 2014; Kilinc et al . , 2016 ) . OR expression is unstable until one OR is functionally expressed which then represses the expression of other OR alleles via negative feedback signaling through the unfolded protein response ( UPR ) and G proteins ( Serizawa et al . , 2003; Dalton et al . , 2013; Wang et al . , 2012; Li and Matsunami , 2013; Lewcock and Reed , 2004; Ferreira et al . , 2014; Nguyen et al . , 2007; Abdus-Saboor et al . , 2016 ) . Here , we generated Rtp1 and Rtp2 double knockout mice ( RTP1 , 2DKO ) to investigate their role in the functioning and development of the olfactory system in vivo . We show that the RTP1 , 2DKO have OR trafficking defects , a substantial reduction in the number of mature OSNs , and an overall diminished olfactory capacity . Unexpectedly , we found that some ORs are overrepresented ( referred to as oORs ) while others are underrepresented ( referred to as uORs ) in RTP1 , 2DKO . Cells expressing a uOR lack stable gene choice in the mutant compared to wild-types while cells expressing an oOR do not show this instability , a result that links OR protein trafficking and OR transcriptional regulation . In order to study the role played by RTP1 and RTP2 in regulating OR expression and trafficking in vivo , we consecutively knocked out these genes while the intervening ~500 kb genomic region was not disrupted in ES cells ( Figure 1A ) . Following chimeric mice production and germline transmission , we established mouse lines with Rtp1 and Rtp2 double knock out alleles . We found no phenotypic difference between Rtp1 ( +/+ ) ;Rtp2 ( +/+ ) ( wild-type ) and Rtp1 ( +/− ) ;Rtp2 ( +/− ) ( het ) mice . The Rtp1 ( −/− ) ;Rtp2 ( −/− ) homozygous mutants ( RTP1 , 2DKO ) showed no gross defects outside the olfactory system . Heterozygous crosses gave rise to wild-type , heterozygous and homozygous adults in roughly 1:2:1 ratio ( Rtp1 ( +/+ ) ;Rtp2 ( +/+ ) 19 , Rtp1 ( +/− ) ;Rtp2 ( +/− ) 34 , Rtp1 ( −/− ) ;Rtp2 ( −/− ) 15 , n = 10 mating pairs ) , suggesting no embryonic or postnatal lethality . We validated the absence of RTP1 and RTP2 transcripts in RTP1 , 2DKO by performing RNA in situ hybridization ( Figure 1B ) . 10 . 7554/eLife . 21895 . 003Figure 1 . Deletion of RTP1 and RTP2 causes defects in the OE . ( A ) Strategy for knocking out RTP1 and RTP2 in series . ( B ) RNA in situ hybridization with probes specific to RTP1 and RTP2 in both wild-type and RTP1 , 2DKO mice showing that the knock out mice do not express either of these proteins . Scale bar = 50 μm . ( C ) Schematic depiction of M71-IRES-tau GFP . ( D ) Antibody against M71 ( red ) stains the dendrite in the wild-type OE ( top ) but not the RTP1 , 2DKO OE . On the other hand , the antibody against GFP shown in green stains the entire neuron from RTP1 , 2DKO;M71-IRES-tauGFP mice , which shows that M71 positive OSNs have dendrites . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 003 It has been previously shown that RTP1 and RTP2 promote cell surface expression of ORs in the heterologous expression assays ( Saito et al . , 2004; Zhuang and Matsunami , 2007 ) . Therefore , we used M71-IRES-tauGFP mice in which Olfr151 ( also known as M71 and MOR171-2 ) expressing OSNs co-express tauGFP to examine the OSNs for OR trafficking defects ( Feinstein et al . , 2004a ) ( Figure 1C ) . In the RTP1 , 2DKO;M71-IRES-tauGFP OE , GFP staining was observed in the dendrites of Olfr151 positive OSNs ( Figure 1D ) , indicating that the morphology of their OSNs remains unchanged . In contrast , immunostaining against Olfr151 ( Barnea et al . , 2004 ) was restricted to the cell body , indicating these OSNs are unable to traffic the OR to the dendrite ( Figure 1D ) . Altogether , the data suggest that RTP1 and RTP2 are essential for OR trafficking . Upon examination of the OE , we found that its thickness was significantly reduced in RTP1 , 2DKO mice . ( p=0 . 02 paired student t test ) ( Figure 2A ) . We therefore examined the expression of various OSN developmental markers and signaling molecules in the OE to evaluate areas occupied by mature and immature OSNs in RTP1 , 2DKO . We compared OMP and adenylate cyclase 3 ( ACIII ) , markers for mature neurons ( Carter et al . , 2004; Rogers et al . , 1987 ) , in 21 day old RTP1 , 2DKO mice and their littermates ( Figure 2B , Figure 2—figure supplement 1 ) . We measured the area occupied by RNA in situ hybridization signal against OMP and found that mice showed an average of 22% reduction in RTP1 , 2DKO when compared to the wild-type ( p<0 . 0001 , paired student t test , wild-type mean area 71% ± 5 ( SD ) , RTP1 , 2DKO mean area 49% ± 4 ( SD ) ) ( Figure 2C , See methods for details ) . Comparison of the OMP positive layer from wild-type and RTP1 , 2DKO OE collected at 1-day-old , 21-day-old and 6-month-old mice showed a significant reduction in OMP expression at 1 day and 21 days ( p=0 . 0003 , Mann Whitney U test , p=0 . 0003 , Mann Whitney U test ) but not at 6 months ( Figure 2D ) . Immunohistochemical analysis of expression of adenylate cyclase 3 ( ACIII ) , a signaling molecule expressed in mature OSNs ( Wei et al . , 1998; Wong et al . , 2000; Col et al . , 2007 ) showed a 17% decrease in the area occupied by the staining in 21 day old RTP1 , 2DKO OE ( p=0 . 0001 , Mann Whitney U test , wild-type mean area 44% ± 9 ( SD ) , RTP1 , 2DKO mean area 27% , ± 3 ( SD ) ) ( Figure 2B ) . Consistent with OMP expression , we observed a significant difference in ACIII expression at 1 day old ( p=0 . 0079 , Mann Whitney U test ) but not at 6 months ( Figure 2E ) . 10 . 7554/eLife . 21895 . 004Figure 2 . RTP1 , 2DKO mice have fewer mature sensory neurons . ( A ) Paired comparison of the thickness of the OE measured at five matched positions ( see methods ) between RTP1 , 2DKO and their wild-type littermate . ( p=0 . 02 , paired student t test ) . ( B ) RNA in situ hybridization against OMP ( top ) , GAP43 ( bottom ) and IHC against ACIII ( middle ) at 1 day , 21 days and 6 months old . Scale bar = 25 μm . ( C ) Quantification of the percent area occupied by OMP RNA in situ hybridization signal from matched positions in the OE . Pair wise student t test shows a significant reduction in the area occupied by OMP staining in RTP1 , 2DKO . Error bars indicate SEM , p<0 . 0001 , Paired student t test . ( D ) Comparison of percent area occupied by OMP between RTP1 , 2DKO and their het/wild-type littermates at different ages showing that RTP1 , 2DKO has fewer mature OSNs at 1 day ( p=0 . 0003 Mann Whitney U test ) and 21 day ( p=0 . 0003 Mann Whitney U test ) but there is no difference at 6 months ( p=0 . 7 , Mann Whitney U test ) . ( E ) Quantification of the area occupied by ACIII staining between RTP1 , 2DKO and their control genotype ( hetetrozygous or wild-type ) littermates at different ages showing that RTP1 , 2DKO has fewer mature OSNs at 1 day ( p=0 . 0079 Mann Whitney U test ) and 21 day ( p<0 . 0001 Mann Whitney U test ) but there is no significant difference at 6 months ( p=0 . 1143 , Mann Whitney U test ) . ( F ) Quantification of the area occupied by GAP43 staining between RTP1 , 2DKO and their het or wild-type littermates at different ages showing that RTP1 , 2DKO has more immature neurons at 21 day ( p=0 . 0343 Mann Whitney U test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 00410 . 7554/eLife . 21895 . 005Figure 2—source data 1 . OE thickness and percent area occupied by the OMP layer , ACIII layer and GAP43 layer in the wild-type and RTP1 , 2DKO . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 00510 . 7554/eLife . 21895 . 006Figure 2—figure supplement 1 . Low magnification view of OMP in situ hybridization signals in OE sections . ( A ) Matching OE sections from Wild-type ( left ) and RTP1 , 2DKO ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 006 GAP43 is a marker for immature olfactory neurons in the OE ( Meiri et al . , 1988; Verhaagen et al . , 1990; Treloar et al . , 1999 ) and area occupied by it shows a 7% increase in the area of the OE it occupies in 21-day-old RTP1 , 2DKO ( p=0 . 03 Mann Whitney U test , wild-type mean area 20% , ± 5 ( SD ) , RTP1 , 2DKO mean area 27% , ± 6 ( SD ) ) ( Figure 2B ) . No significant difference in the GAP43 positive layer is observed between RTP1 , 2DKO and their littermates at 1 day nor at 6 months ( Figure 2F ) . Upon observation of fewer OSNs in RTP1 , 2DKO mice and lack of OR trafficking to the cilia ( Figure 1D ) , we sought to test the olfactory ability by electroolfactogram ( EOG ) . We tested a diverse set of 7 odorants , in both wild-type and RTP1 , 2DKO littermates . Wild-type mice show robust EOG responses to all odorants at concentrations as low as 0 . 01% ( Figure 3A ) . In contrast , RTP1 , 2DKO mice showed striking deficits in their response . Responses to most odors were identical to the blank stimulus ( air only ) , although some sensitivity was maintained for a subset of odorants ( 2-heptanone , amyl acetate , isomenthone ) compared to the wild-types ( Figure 3B–C ) . Thus , the reduction of mature OSNs and the loss of surface OR expression corresponds to a dramatic loss of odorant sensitivity in RTP1 , 2DKO . 10 . 7554/eLife . 21895 . 007Figure 3 . Diminished activity in response to odorants in RTP1 , 2DKO . ( A ) Electroolfactograms show the response to seven odorants wild-type . The grey line denotes the air only blank averaged over multiple interleaved trials interspersed within the series . ( B ) RTP1 , 2DKO responses to the same odors ( C ) Quantification of the EOG amplitudes for each of the seven odorants showing that only a few of the odors elicit responses from the RTP1 , 2DKO OE and these responses are lower than the wild-type . Each bar represents the difference between the peak of the odor minus the peak of the air only blank . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 007 To obtain a comprehensive view of gene expression changes in RTP1 , 2DKO , we performed an RNA-Seq on isolated whole olfactory mucosa including the OE and surrounding tissues . Differential expression analysis comparing RTP1 , 2DKO to wild-type littermates revealed that 3 . 8% of all genes ( 926/24 , 661 ) were differentially expressed between the two genotypes , among which 805 were downregulated and 121 were upregulated in RTP1 , 2DKO ( FDR corrected p<0 . 05 , see Experimental Procedures for details ) . Canonical signaling molecules known to be expressed in mature OSNs including Gnal ( Gαolf ) , Adcy3 ( ACIII ) , and Cnga2 were less abundant in the RTP1 , 2DKO consistent with a reduced number of mature OSNs in absence of RTP1 and RTP2 . We found no significant difference in the expression levels of housekeeping genes like Gapdh and β actin ( Supplementary file 1 ) ( Kouadjo et al . , 2007 ) , neither did we see any compensatory increase in other RTP family members Rtp3 or Rtp4 . We then asked whether the loss of RTP1 and RTP2 equally affected all ORs . In a comparison between wild-type and RTP1 , 2DKO we found that 62% of intact ORs ( 678/1088 ) were significantly affected by the loss of RTP1 and RTP2 ( Figure 4A ) . Close to half of the annotated intact ORs ( 562/1088 ) were downregulated in RTP1 , 2DKO ( FDR corrected p<0 . 05 ) , consistent with fewer OSNs in the mutant . Unexpectedly however , a small subset of OR transcripts ( 116/1088 ) were upregulated in RTP1 , 2DKO mice ( FDR corrected p<0 . 05 ) ( Figure 4B , Figure 4—figure supplement 1 ( A-B ) ) . 10 . 7554/eLife . 21895 . 008Figure 4 . Representation of ORs in RTP1 , 2DKO . ( A ) Comparison of all transcripts between the wild-type and RTP1 , 2DKO , the green dots represent ORs , higher read counts for ORs are observed in the wild-type compared to RTP1 , 2DKO . ( B ) A comparison of the expression levels of ORs between the wild-type ( x – axis ) and RTP1 , 2DKO ( y-axis ) . Red indicates uORs and oORs with p<0 . 01 , blue indicates p<0 . 05 . ( C ) A volcano plot showing the fold change of the expression levels ( x-axis ) of the ORs between wild-type and RTP1 , 2DKO using read counts normalized by OR genes . Red dots are ORs with p<0 . 01 , blue: p<0 . 05 , yellow dots signify candidate uORs chosen for validation . ( D ) Representative images for an in situ analysis with a probe specific to Olfr522 ( uOR ) where there are fewer positive OSNs in RTP1 , 2DKO when compared to the wild-type . Scale bar = 25 μm . ( E ) Quantification of the OSNs expressing uORs shown in ( C ) ; all the tested ORs showed smaller fractions of positive OSNs in RTP1 , 2DKO compared to the wild-type . p<0 . 05 , Mann-Whitney U Test , n = 3 mice . ( F ) Volcano plot showing oORs with read counts normalized by OR genes . ( G ) Representative images for an in situ hybridization analysis with a probe specific to Olfr414 ( oOR ) where there are more positive OSNs in RTP1 , 2DKO when compared to the wild-type . Scale bar = 25 μm . ( H ) Quantification of the OSNs expressing oORs shown in ( G ) ; all the tested ORs showed greater fractions of positive OSNs in RTP1 , 2DKO compared to wild-type . p<0 . 05 , Mann-Whitney U Test , n = 3 mice . ( I ) Plot of the mean abundance where each dot represents a single olfactory receptor classified as an uOR/ oOR/ NS based on normalization by ORs . The horizontal bars denote mean abundance ( FPKM ) . oORs are significantly more abundant than uORs , NS are less abundant than both oORs and uORs ( p<0 . 0001 , one-way ANOVA , Tukey’s post hoc test ) . ( J ) zoomed in view of the plot showing uOR/oOR and NS abundance , horizontal bars denote mean abundance ( FPKM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 00810 . 7554/eLife . 21895 . 009Figure 4—source data 1 . Percent positive cell counts for the uORs and oORs in Figure 4E and 4 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 00910 . 7554/eLife . 21895 . 010Figure 4—figure supplement 1 . Representation of ORs in RTP1 , 2DKO using all genes . ( A ) Volcano plot showing the fold change of the expression levels ( x-axis ) of the ORs between wild-types and RTP1 , 2DKO using read counts normalized by all genes from our sequencing data . Y axis indicates the FDR . Red dots are ORs with p<0 . 01 , blue: p<0 . 05 , yellow dots signify candidate uORs chosen for validation in Figure 4E . ( B ) Volcano plot showing the fold change of the expression levels ( x-axis ) of the ORs between wild-types and RTP1 , 2DKO using read counts normalized by all genes from our sequencing data . Y axis indicates the FDR . Red dots are ORs with p<0 . 01 , blue: p<0 . 05 , yellow dots signify candidate oORs chosen for validation in Figure 4H . ( C ) Table depicting the number of OR genes that are either underrepresented , overrepresented or not significantly changed based on the data set used to normalize OR read counts . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 010 The disparity in the abundance of transcripts for these ORs raised the possibility of a difference in probabilities of OSNs expressing each OR in RTP1 , 2DKO . To remove any possible confounding variables from non-OSN cells , we normalized our read counts using only reads mapped on intact ORs and found that 531/1088 were underrepresented and 202/1088 were overrepresented ( FDR corrected p<0 . 05 ) ( Figure 4C , F , Figure 4—figure supplement 1C ) . To further validate changes in numbers of OSNs expressing individual ORs in RTP1 , 2DKO mice , we carried out RNA in situ hybridization with probes against either underrepresented ORs ( uORs ) ( Figure 4C ) or overrepresented ORs ( oORs ) ( Figure 4F ) . For all uORs tested , fewer OSNs were positive in RTP1 , 2DKO ( Figure 4D–E ) ( p<0 . 05 Mann-Whitney U test ) . In stark contrast , the frequencies for the tested oORs were greater in RTP1 , 2DKO ( p<0 . 05 , Mann-Whitney U test ) ( Figure 4G–H ) . These results demonstrate that OR gene choice is biased in RTP1 , 2DKO towards a specific subset of receptor types . Curiously , we found that oORs as a group are more abundantly expressed than uORs in the wild-type . The OR genes that were not classified as either underrepresented nor overrepresented ( NS , not significant ) exhibited a wide range of changes in expression levels between the wild-type and RTP1 , 2DKO , but are expressed at significantly lower abundance levels than both oORs and uORs ( Figure 4I ) ( p<0 . 0001 one-way ANOVA , Tukey’s post hoc test ) . We wondered what happens to the proportion of OSNs expressing uORs and oORs in RTP1 , 2DKO mice at different ages . We performed RNA in situ hybridization with a ( 1 ) a probe mix containing 11 uORs and ( 2 ) a probe mix containing 25 oORs , all expressed in the dorsal region of the OE on 1-day , 21-day and 6-month-old OE . In the case of the uOR mix , wild-type showed an increase for OSNs expressing the uORs we tested both at 21 days and 6 months ( Figure 5A–B ) consistent to the increasing proportion of mature OSNs in the OE indicated by larger OMP and ACIII layers ( Figure 2D–E ) . However , RTP1 , 2DKO showed no obvious increase in the fraction of cells expressing these ORs with age , while on the other hand , in RTP1 , 2DKO the number of neurons expressing oORs showed a dramatic increase both from 1 day old to 21 days and from 21 days to 6 months ( p<0 . 0001 one-way ANOVA , Tukey’s post hoc test ) ( Figure 5C–D ) demonstrating that the RTP1 , 2DKO OE is progressively populated by oORs . 10 . 7554/eLife . 21895 . 011Figure 5 . The proportion of OSNs expressing oORs increases in older RTP1 , 2DKO . ( A ) Representative images from 1 day , 21 day and 6 month OE stained with a probe mix against 11 of the uORs expressed in the dorsal OE . ( B ) Quantification of the percent dorsal uOR positive cells at different ages in RTP1 , 2DKO and their het or wild-type littermates . The fraction of cells positive for this probe significantly increases with age only in wild-type ( p<0 . 0001 one-way ANOVA , Tukey’s post hoc test ) . ( C ) Representative images from 1 day , 21 day and 6 month OE stained with a probe mix against 25 of the oORs expressed in the dorsal OE . ( D ) Quantification of the percent dorsal oOR positive cells at different ages between RTP1 , 2DKO and their het or wild-type littermates . The fraction of cells positive for this probe mix significantly increases with age in RTP1 , 2DKO ( p<0 . 0001 one-way ANOVA , Tukey’s post hoc test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 01110 . 7554/eLife . 21895 . 012Figure 5—source data 1 . Percent positive cell counts for the uOR and oOR probe mix at 1 day , 21 day and 6 month old OE . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 012 We wondered whether the OR expression bias arose due to the effect of RTP on the regulatory elements of an OR’s gene locus or the protein . In our initial investigation , we did not find an obvious pattern or clustering for the genomic locations nor did we find obvious conserved residues or motifs amongst uORs or oORs ( Figure 6A , Supplementary file 2 ) . In an attempt to causally identify the basis of the bias , we used a mouse expressing β2AR-IRES-LacZ from the Olfr151 locus ( Feinstein et al . , 2004b ) ( Figure 6B ) and asked whether the numbers of OSNs expressing Olfr151 or β2AR are similarly affected in RTP1 , 2DKO . We chose Olfr151 as it is a uOR and β2AR because it is a non-OR GPCR , capable of reaching the cell surface without the RTPs in heterologous culture and can replace a functional OR in native OSNs ( Omura et al . , 2014 ) . We saw that fewer Olfr151 expressing OSNs were present in RTP1 , 2DKO ( p=0 . 0028 , Mann-Whitney U test ) , as expected for a uOR . Strikingly , more β2AR positive OSNs were present in RTP1 , 2DKO compared with wild-type ( p<0 . 001 , Mann-Whitney U test ) ( Figure 6C–D ) , suggesting that it is the protein sequence and not the locus of the OR that determines whether a given OR is underrepresented or overrepresented . 10 . 7554/eLife . 21895 . 013Figure 6 . OR protein sequences determine representation in RTP1 , 2DKO . ( A ) Phylogenetic tree showing uORs in black and oORs in red . ( B ) Schematic depiction of β2AR-IRES-tau LacZ . ( C ) The percent Olfr151 positive cells is smaller in RTP1 , 2DKO mouse ( left panels ) . β2Adrenergic receptor expressed from the Olfr151 locus shows more β2AR cells in RTP1 , 2DKO mouse ( right panels ) . ( D ) Quantification of the percent positive Olfr151 and β2AR cells in wild-types vs RTP1 , 2DKO p<0 . 05 Mann-Whitney U test , n = 3 mice . ( E ) Representative FACS data graphing the number of cells ( y-axis ) vs the intensity of phycoerythrin staining expressing Rho tagged uORs ( x-axis ) . Each color represents an individual uOR . ( F ) Representative FACS data graphing the number of cells ( y-axis ) vs the intensity of phycoerythrin staining expressing Rho tagged oORs ( x-axis ) . Each color represents an individual oOR . ( G ) Comparison of the normalized geometric mean of the compensated PE intensity for all uORs vs all oORs tested . The geometric means are normalized to Olfr78 ( p=0 . 0483 Mann-Whitney U test , uOR n = 23 genes , oOR n = 24 genes ) . Every geometric mean is calculated by counting the PE intensity across 10 , 000 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 01310 . 7554/eLife . 21895 . 014Figure 6—source data 1 . Normalized geometric mean for PE intensity obtained from our FACS experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 014 Given that β2AR is known to be efficiently trafficked to the cell surface when heterologously expressed in the absence of the RTPs , we asked whether uORs and oORs show differential capabilities in cell surface trafficking in heterologous cells . We carried out live cell surface staining of HEK293T cells transfected with either Rho tagged uORs or oORs in the absence of RTP1 and RTP2 . In order to quantify the surface staining , we carried out FACS to measure the surface OR levels . oORs as a group showed more OR surface expression than uORs ( p<0 . 05 , Mann-Whitney U test ) ( Figure 6E–G ) . Notably , the ORs that show most robust cell surface expression were all oORs . Even though trafficking mechanisms between OSNs and HEK293T cells are likely to be different , our data are consistent with the idea that RTP-independent trafficking of ORs may be related to increased frequencies of OSNs expressing oORs . Our results thus far suggest that OSNs expressing oORs are able to function despite the loss of the RTPs . In order to test this , we examined whether OSNs expressing uORs or oORs co-express OMP ( Figure 7A ) . We found that the number of immature OMP-negative OSNs expressing uORs are similar in RTP1 , 2DKO and het controls , whereas the number of OMP-positive OSNs expressing uORs show a 69% decrease in RTP1 , 2DKO . ( p=0 . 024 Mann Whitney U test ) ( Figure 7B–D ) . 10 . 7554/eLife . 21895 . 015Figure 7 . OSNs expressing oORs from RTP1 , 2DKO can mature and function . ( A ) Representative images showing the colocalization of Olfr923 ( uOR ) and Olfr78 ( oOR ) ( green ) with OMP ( red ) for het ( top ) and RTP1 , 2DKO ( bottom ) . OMP negative OSNs are indicated with arrows . ( B ) Quantification of OMP positive OSNs for uORs and oORs as a group . For uORs there is a significant decrease ( p=0 . 024 Mann Whitney U test , het n = 96 , RTP1 , 2DKO n = 33 ) in the number of OMP positive OSNs in RTP1 , 2DKO whereas no significant difference is observed for oORs ( p=0 . 4427 Fisher’s exact test , het n = 126 , RTP1 , 2DKO n = 213 ) . ( C ) Quantification of the number of OSNs co-expressing OMP for individual uORs . ( D ) Quantification of the number of OSNs co-expressing OMP for individual oORs . ( E ) Representative images for pS6 staining ( green ) along with either a uOR ( Olfr923 ) or an oOR ( Olfr1395 ) in response to their cognate ligands . het ( left ) shows pS6 induction , whereas RTP1 , 2DKO ( right ) shows pS6 induction in response to an odor that stimulates the oOR but not the one that stimulates the uOR . All pS6 positive neurons are indicated by white arrows . ( F ) Quantification of the fold change in pS6 staining ( pixel ) intensity for Olfr923 positive cells in het ( grey ) and RTP1 , 2DKO mice ( red ) in response to 1% acetophenone . There is a significant increase in the pS6 induction in het ( p<0 . 0001 one-way ANOVA , Tukey’s post hoc test ) but not RTP1 , 2DKO when subject to the odor . ( G ) Quantification of the fold change in pS6 staining pixel intensity for Olfr1395 positive cells in het ( grey ) and RTP1 , 2DKO mice ( red ) in response to 1% TMT . There is a significant increase in the pS6 induction in both het and RTP1 , 2DKO when subject to the odor ( p=0 . 0002 , one-way ANOVA , Tukey’s post hoc test ) . ( H ) Wild-type ( left ) and RTP1 , 2DKO OE ( right ) stained with an antibody against the active form of caspase3 ( red ) and OMP ( green ) . The white arrows indicate cells expressing active caspase 3 in the OMP positive layer and the blue arrows show the active caspase 3 positive cells outside the OMP positive layer . Scale bar = 50 μm . ( I ) Quantification of the percentage of active caspase3 positive cells . Error bars indicate SEM , p<0 . 01 , Mann-Whitney U Test ( n = 3 mice ) ( J ) Quantification of OMP and active caspase 3 double staining showing that RTP1 , 2DKO have significantly more OSNs undergoing cell death both in mature and immature OSN layers compared to wild-type . ( p=0 . 029 , Fisher’s exact test ) DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 01510 . 7554/eLife . 21895 . 016Figure 7—source data 1 . Numbers of uOR and oOR neurons found within the OMP layer and outside it . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 01610 . 7554/eLife . 21895 . 017Figure 7—source data 2 . Normalized pS6 staining intensity for Olfr923 and Olfr1395 positive cells from het and RTP1 , 2DKO OE in response to 1%acetophenone and 1%TMT respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 017 To evaluate the function of OSNs expressing uORs or oORs , we chose a uOR and an oORs that have been previously deorphanized . Olfr1395 is an oOR found to respond to 2 , 5-dihydro-2 , 4 , 5-trimethylthiazoline ( TMT ) and Olfr923 , a uOR , to acetophenone in vivo ( Jiang et al . , 2015 ) . Using the induction of phospho ribosomal protein S6 ( pS6 ) , a marker for neuronal activation ( Knight et al . , 2012 ) , we found that OSNs expressing Olfr1395 in both het and RTP1 , 2DKO were activated to its cognate ligand TMT ( p<0 . 0001 one-way ANOVA , Tukey’s post hoc test ) ( Figure 7E , G ) . In contrast OSNs expressing Olfr923 in het , but not RTP1 , 2DKO were activated by their cognate ligand acetophenone ( Figure 7E–F ) . These data show that OSNs expressing oORs mature and function in the RTP1 , 2DKO . Immunostaining against active caspase 3 , a cell death marker , suggested that RTP1 , 2DKO mice have an increased number of OSNs undergoing cell death ( p<0 . 01 , Mann-Whitney U test ) ( Cowan et al . , 2001 ) ( Figure 7H–I ) . Active caspase three staining in conjunction with OMP suggested that more OSNs in immature and mature layers both undergo cell death in RTP1 , 2DKO ( Figure 7H ) . Notably , immature OSNs in the OMP negative layer rarely undergo cell death in wild-type ( Figure 7J ) ( Jia et al . , 2010 ) . Together , these observations are consistent with the idea that OSNs expressing uORs are more likely to undergo cell death in RTP1 , 2DKO . The above findings reveal the expression of a biased OR repertoire in RTP1 , 2DKO , with about a half of the ORs being underrepresented but a small subset of ORs overrepresented . Previous studies have shown that the unfolded protein response ( UPR ) plays an important role in OR gene choice mechanisms . During the initial phase of OR expression , the UPR pathway gives rise to an increased translation of nuclear activating transcription factor ( nATF5 ) over an upstream inhibitory ORF via eIF2α signaling ( Godin et al . , 2016 ) . Once the OSN has matured , OR gene choice is stabilized and the UPR is relieved ( Dalton et al . , 2013 ) . We hypothesized that the lack of RTP1 and RTP2 causes persistent UPR in OSNs expressing uORs , leading to unstable OR gene choice in these OSNs , which in turn contributes to the skewed OR repertoire in mice lacking RTP1 and RTP2 . We first asked whether the expression of nATF5 , a marker for UPR , is different in RTP1 , 2DKO . Indeed , we observed that there were more nATF5 positive OSNs in RTP1 , 2DKO mice ( p<0 . 001 , Mann-Whitney U Test ) and some of these OSNs were located closer to the apical surface , a phenomenon that was not observed in the wild-types , suggesting that nATF5 expression persists during the OSN development in RTP1 , 2DKO animals ( Figure 8A–B ) . Similarly , expanded expression was observed for LSD1 , a histone demethylase whose expression depends on nATF5 ( Figure 8C ) . Next , we asked whether the ectopic expression of nATF5 in RTP1 , 2DKO is due to their protein sequence or the gene locus . We first compared the co-expression of nATF5 and Olfr151 between the wild-type and RTP1 , 2DKO using M71-IRES-tauGFP mice , and then asked whether the same co-localization occurs for β2AR expressed from the Olfr151 gene locus using β2AR-IRES-LacZ mice . We observed that the number of OSNs co-expressing Olfr151 and nATF5 was significantly higher in RTP1 , 2DKO than in the wild-type ( p<0 . 05 , Fisher’s exact test ) . In contrast , the number of OSNs co-expressing β2AR and nATF5 in the β2AR:IRES:tauLacZ mouse was not different from the wild-type ( p=1 , Fisher’s exact test ) ( Figure 8D–F ) . As Olfr151 but not β2ARs require RTPs for surface expression , these data suggest that delivery of ORs to the membrane plays a role in terminating the UPR . 10 . 7554/eLife . 21895 . 018Figure 8 . nATF5 expression persists in OSNs expressing uORs but not oORs in RTP1 , 2DKO . ( A ) Expanded expression pattern of nATF5 is observed for RTP1 , 2DKO Scale bars = 25 μm . S . C . = Sustentacular cells . ( B ) Analysis of individual sections of wild-type and RTP1 , 2DKO OE for ATF5 expression , RTP1 , 2DKO mice have a larger number of nATF5 positive cells and they are more apically situated compared to wild-type p=0 . 0049 , Mann-Whitney U test , n = 3 mice . ( C ) Similar expanded expression pattern is observed for LSD1 . Scale bars = 25 μm . S . C . = Sustentacular cells . ( D ) Representative images of Olfr151 and nATF5 in wild-type vs RTP1 , 2DKO OE . Inset: higher magnification , arrow head: nATF5 and Olfr151 co-localization . ( E ) Representative images from β2Adrenergic receptor IRES tauLacZ ( β2AR ) with antibody staining against LacZ indicating β2AR positive OSNs and nATF5 . ( F ) Quantification of the number of OSNs positive for both Olfr151 and nATF5 ( left ) and β2AR and nATF5 ( right ) . In RTP1 , 2DKO , there was a significant increase in the number of nATF5-Olfr151 double positive OSNs p=0 . 0022 , Fisher’s exact test , n = 3 mice but not β2AR and nATF5 double positive neurons . p=1 , Fisher’s exact test , n = 3 mice . ( G ) Left: Representative images for Olfr1444 ( uOR ) ( red ) co-localization with nATF5 ( green ) signal in wild-type OE ( left ) and RTP1 , 2DKO ( right ) . The inset shows a higher magnification of a single OSN positive for Olfr1444 and nATF5 . Right: Representative images for Olfr1056 ( oOR ) shown in red . ( H ) Top: Percent nATF5 positive OSNs expressing 7 uORs in the wild-type vs the RTP1 , 2DKO . The solid points indicate the overall mean of all uORs and the solid line shows that more RTP1 , 2DKO OSNs expressing uORs co-localize with nATF5 . p=0 . 007 , Mann-Whitney U test , n = 2 mice . Bottom: Percent nATF5 positive OSNs expressing 6 oORs in the wild-type vs the RTP1 , 2DKO . ( p=0 . 937 , Mann-Whitney U test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 01810 . 7554/eLife . 21895 . 019Figure 8—source data 1 . Percent uOR/oOR positive cells that co-localize with nATF5 in wild-types and RTP1 , 2DKO . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 019 We next asked whether the expression of nATF5 in RTP1 , 2DKO mice was different in OSNs expressing uORs or oORs . In order to answer this question , we carried out fluorescent in situ hybridization against 7 uORs and 6 oORs along with immunohistochemistry for nATF5 and quantified colocalization of the OR signal with nATF5 staining . As expected , higher numbers of OSNs expressing uORs colocalized with nATF5 in RTP1 , 2DKO as a group ( p=0 . 007 , Mann-Whitney U test ) , whereas no significant difference was observed when oORs were tested ( p=0 . 937 , Mann-Whitney U test ) ( Figure 8G–H ) . These observations suggest that OSNs expressing uORs contribute to the expanded expression of nATF5 in RTP1 , 2DKO and are further consistent with the idea that surface trafficking of ORs is linked to turning off UPR . Increased nATF5 levels in uOR-expressing OSNs of RTP1 , 2DKO mice suggests a lack of stable OR gene choice in these neurons . This led us to hypothesize that OSNs that initially express a uOR may later turn off the OR and stabilize the expression of another OR . To directly address this , we used a lineage tracing strategy to study the stability of OR gene choice in RTP1 , 2DKO ( Shykind et al . , 2004; Dalton et al . , 2013; Abdus-Saboor et al . , 2016 ) . In our assay ( Figure 9A ) , we crossed a mouse that had one allele expressing Cre recombinase under the Olfr151 promoter , M71-IRES-Cre , to a mouse that had the Cre inducible fluorescent reporter Rosa26-lox-stop-lox-tdTomato ( Madisen et al . , 2010 ) . In this mouse , any OSN expressing the M71-IRES-Cre allele at any point in time , would give rise to the permanent expression of tdTomato even if the OSN went on to express a different OR gene . We counted the number of tdTomato positive neurons that were also Olfr151 positive ( double positive ) , which reflect the OSNs that initially chose as well as stably express Olfr151 . OSNs that were tdTomato positive but not Olfr151 positive are the ones that switched their initial gene choice . We found that only 17% of tdTomato positive cells from RTP1 , 2DKO OE ( n = 24/140 neurons ) also express Olfr151 in comparison to 31% OSNs from the wild-type ( n = 79/258 neurons ) ( p<0 . 05 , Fisher’s exact test ) and 38% ( n = 66/172 neurons ) in the heterozygotes ( p<0 . 05 , Fisher’s exact test ) ( Figure 9B–C ) . This suggests that the loss of RTPs leads to the frequent termination of Olfr151 gene expression . 10 . 7554/eLife . 21895 . 020Figure 9 . Unstable expressing of M71 in RTP1 , 2DKO . ( A ) Schematic depiction of OSN lineage tracing . We crossed a mouse carrying M71–IRES- Cre with Rosa26-lox-stop-lox-tdTomato . In the progeny , the expression of Olfr151 ( M71 ) in an OSN will drive the expression of Cre , leading to the permanent production of tdTomato by the removal of the transcriptional stop sequence . Larger numbers of tdTomato positive OSNs that do not express Olfr151 would indicate unstable gene expression ( right ) . On the other hand , if tdTomato OSNs largely stained positive for Olfr151 ( double positive , shown in yellow ) , it would indicate stable OR expression . ( B ) Representative images from the wild-type and RTP1 , 2DKO OE stained with antibody against Olfr151 ( green ) and tdTomato ( red ) . Arrow heads indicate double positive OSNs and the arrows show tdTomato positive and Olfr151 negative OSNs . ( C ) Each point represents the ratio of the number of tdTomato and Olfr151 double positive OSNs to the number of only tdTomato positive OSNs in one mouse . RTP1 , 2DKO mice have significantly lower OR gene choice stability compared to both wild-type and het mice . p<0 . 05 , , Fisher’s Exact test , n = 4 mice . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 02010 . 7554/eLife . 21895 . 021Figure 9—figure supplement 1 . OSNs expressing Olfr151 do not switch to Olfr143 . ( A ) Representative images from the wild-type and RTP1 , 2DKO OE stained against Olfr143 ( green ) and tdTomato ( red ) . ( B ) The percent Olfr143 positive cells is greater in RTP1 , 2DKO mouse and tdTomato expressed under the control of the M71 locus shows a reduction in the number of percent positive cells in RTP1 , 2DKO mouse . ( C ) The number of neurons that are only tdTomato positive in wild type and in RTP1 , 2DKO ( left ) . Number of neurons positive for both Olfr143 and tdTomato ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 021 To determine whether the Olfr151 gene switched to an oOR within the same locus we carried out a co-localization analysis between Olfr143 , an oOR within the Olfr151 locus and tdTomato under the control of M71-cre ( Figure 9—figure supplement 1 ) . We found no Olfr143 and tdTomato double positive OSNs in both RTP1 , 2DKO or their wild-type littermates , indicating that there is no higher likelihood of the gene switching between these ORs . Lastly , we investigated OSN axon targeting to the olfactory bulb ( OB ) in RTP1 , 2DKO . Olfactory axons entered the OB and innervate to the glomerular layer based on our OMP immunostaining in RTP1 , 2DKO ( Figure 10A ) . However , using M71-IRES-tau GFP mice we found that Olfr151 expressing OSNs were unable to converge in the OB in RTP1 , 2DKO while their wild-type littermates had two Olfr151 glomeruli in each of their OBs as expected ( Figure 10B–C ) . To investigate whether the axon targeting defect was ubiquitous to all receptors , we used β2AR-IRES-tauLacZ ( Feinstein et al . , 2004b ) ( Figure 6B ) . Both the wild-type and RTP1 , 2DKO mice formed glomeruli , however the mutant mice had ectopic glomeruli for OSNs expressing this GPCR ( Figure 10B–C ) . These data suggest the RTP1 , 2DKO mice do not have a complete set of glomeruli , but retain the ability to form them . We observed tdTomato-positive axons forming small glomeruli in RTP1 , 2DKO;M71-IRES-cre; Rosa26-lox-stop-lox-tdTomato mice ( 2 out of 3 mice examined had 1 glomerulus each ) ( Figure 10B–C ) . This may indicate that OSNs initially expressing Olfr151 switch and/or stabilize expression of a specific OR , presumably an oOR , and axons from these OSNs can converge in the OB . 10 . 7554/eLife . 21895 . 022Figure 10 . RTP1 , 2DKO mice are able to form glomeruli . ( A ) OMP staining shown in red and nuclear staining in cyan . Both wild-type and RTP1 , 2DKO mice have OMP in the glomerular layer . Scale bar = 25 μm . ( B ) A whole mount GFP fluorescence from axons expressing M71 from M71-IRES-tauGFP mice ( left ) , tdTomato fluorescence from M71-IRES-Cre; Rosa26-lox-stop-lox-tdTomato ( middle ) and LacZ positive axons from β2AR-IRES-tauLacZ mice ( right ) . RTP1 , 2DKO OBs lack Olfr151 ( M71 ) glomeruli but have tdTomato and LacZ positive ones , while labeled glomeruli are observed in wild-type with M71-IRES-tauGFP , M71-IRES-Cre; Rosa26-lox-stop-lox-tdTomato and β2AR-IRES-tauLacZ . Only the dorso-lateral OB are visible for β2AR-IRES-LacZ in our preparation . Scale bar = 25 μm . ( C ) Quantification of the total number of glomeruli observed in wild-type and RTP1 , 2DKO OBs . Each dot represents one mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 022 The absence of RTP1 and RTP2 leads to the underrepresentation of nearly half of the ORs , while about 10% of the ORs are significantly overrepresented . This translates into a change in the numbers of OSNs choosing each OR . How do the RTPs regulate the probability of OR gene choice ? Our data indicate that protein sequences of ORs differentially influence OR gene choice , which is linked to trafficking of ORs to the cell surface . Our attempts to identify protein motifs , domains , or features specific to either uORs or oORs were so far unsuccessful . This fits with recent reports where large-scale mutational analysis of Olfr151 in heterologous cells failed to identify any specific amino acids or domains that regulate its cell surface expression ( Hague et al . , 2004; Jamet et al . , 2015 ) . Nevertheless , this study provides a large set of sequence information of ORs that will allow us to conduct future structure-function studies by testing uOR/oOR chimeras and/or searching for hidden features within uORs or oORs , which may in turn give us clues as to why nearly half of the ORs are retained in the ER in the absence of the RTPs . Previous studies have suggested the UPR protein nATF5 is induced once an OSN starts actively expressing an OR and is lost when OR expression is stabilized ( Dalton et al . , 2013 ) . Our results show an expanded expression of nATF5 in RTP1 , 2DKO OSNs , suggesting persistent UPR during OR gene choice . Importantly , the frequency of co-localization between nATF5 and uORs increases in RTP1 , 2DKO whereas this increase is not observed for oORs . This suggests that the persistent UPR phenotype observed in RTP1 , 2DKO is due to the OSNs that express uORs resulting in unstable OR gene choice for these OSNs . This is reinforced by our observation that OSNs initially expressing Olfr151 , a uOR , are more likely to terminate their expression in RTP1 , 2DKO mice . We present a model for the role played by RTP1 and RTP2 in the gene choice made by an OSN ( Figure 11A ) . In our model , the RTPs suppress UPR response by allowing ORs to exit the ER and be transported to the plasma membrane . uORs are not trafficked to the cell surface in the absence of the RTPs , giving rise to persistent UPR in the OSNs that express them; as a consequence , destabilization of the initial OR gene choice leads to cell death or to the stabilization of another OR . In contrast , oORs are trafficked to the cell surface as functional proteins in the absence of the RTPs , allowing these OSNs to terminate UPR and stabilize gene expression . Although we cannot rule out that RTPs themselves play a role in the elimination of the UPR response , the lack of increase in nATF5 observed in OSNs expressing oORs in RTP1 , 2DKO suggests interaction between the RTPs and the UPR pathway through ORs . Even for OSNs expressing oORs and β2AR , UPR is likely to be induced in the initial stage ( Dalton et al . , 2013 ) . Recent reports utilizing single cell RNA-Seq suggest that immature OSNs express as many as 12 ORs at low levels but mature OSNs only show the expression of one dominant OR ( Saraiva et al . , 2015; Tan et al . , 2015; Hanchate et al . , 2015 ) . Low-level expression of uORs precedes the expression of the RTPs and may be sufficient to trigger UPR in the developing OSNs . Alternatively , both oORs and uORs induce UPR at the initial stage . 10 . 7554/eLife . 21895 . 023Figure 11 . Model for the role of RTP1 and RTP2 in OR gene choice . ( A ) A model showing that in the wild-type ( left ) , ORs are transported to the surface of the cell in conjunction with the RTPs . In the absence of the RTPs ( right ) , oORs reach the surface of the cell and these OSNs survive , whereas uORs are not trafficked to the surface and OSNs expressing them show a persistent nATF5 expression . These uOR-expressing OSNs undergo OR choice switching and the OSNs that switch to an oOR are able to survive , leading to oOR overrepresentation . Those that are unable to switch to an oOR undergo cell death . DOI: http://dx . doi . org/10 . 7554/eLife . 21895 . 023 RNA-Seq data in the wild-type suggests that oORs as a group tend to be more frequently chosen . It could be that initial expression of RTP1 and RTP2 in the developing OSNs is not stable or abundant enough , causing oORs to be stabilized . Alternatively oORs tend to ‘win the competition’ ( Nguyen et al . , 2007; Abdus-Saboor et al . , 2016 ) even in the presence of RTPs probably because of its ability to suppress UPR . Both change in probability of OR gene choice and biased cell death could alter OR population representation both in the level of transcripts and in the number of OSNs . The relative contributions of cell death and gene switching to the differential representation of ORs can be further clarified using Bax knockout mice where cell death in developing neurons is suppressed ( Robinson et al . , 2003 ) . OR genes that did not significantly change in RTP1 , 2DKO mice showed lower abundance in wild-type mice suggesting that these ORs are chosen less frequently . Deeper sequencing and/or increased sample sizes will help classify these ORs as underrepresented , overrepresented , or not changed . In our lineage tracing experiment , we were unable to distinguish tdTomato-positive , Olfr151-negative OSNs between those express Olfr151 as one of the low-level ORs , and the others express Olfr151 as the dominant OR later switch to express another OR . Irrespective of this our data show greater gene instability for Olfr151 expression without the RTPs . Consistent with multiple OR gene expression in developing OSNs , our lineage tracing results suggest that only 31% of OSNs that initially expressed Olfr151 went on to stabilize its expression in the wild-type . Curiously , the vast majority of OSNs that initially expressed Olfr1507 ( MOR28 ) show stable expression in a similar experiment ( Shykind et al . , 2004; Dalton et al . , 2013 ) , suggesting that some ORs are more likely to be stabilized than other ORs . The frequency with which any OR is chosen is different and the underlying cause for this difference remain unknown . Our results show that the number of OSNs expressing β2AR expressed from the Olfr151 locus is lower than the number expressing Olfr151 . Our data call for future experiments to test whether protein sequences of ORs differentially influence initial OR gene choice . Our data suggest that mice without RTP1 and RTP2 had diminished but not abolished responses to odorants . Even though we see a clear reduction in the number of mature OSNs and dramatically diminished responses to odors in RTP1 , 2DKO mice , functional OSNs are not completely eliminated . Previous studies suggest that anosmic mice often die in the first few days after birth ( Brunet et al . , 1996 ) , but the RTP1 , 2DKO mice seem to have the sufficient olfactory ability for postnatal development . Our data show that OSNs expressing oORs are likely to mature and be functional in RTP1 , 2DKO , explaining the residual responses observed in the mutant . These OSNs become more abundant in the mutant OE as the mouse ages indicating that these ORs help maintain the olfactory ability of these mice . It will be interesting to assess olfactory-mediated behaviors of the RTP1 , 2DKO mice in which only a minor fraction of the ORs are functional , since this could address a fundamental question of why most mammals have so many ORs . No GFP-positive glomeruli were observed in RTP1 , 2DKO;M71-IRES-tauGFP indicating OSNs expressing Olfr151 are unable to converge their axons without RTP1 and RTP2 . Yet we observed small tdTomato-positive glomeruli in RTP1 , 2DKO; M71-IRES-cre; Rosa26-lox-stop-lox-tdTomato . These tdTomato-positive glomeruli could be formed by OSNs that initially chose Olfr151 and then switched and/or stabilized the expression of the same OR , presumably an oOR . Previous reports have shown that OR gene switching tend to take place within the same gene locus ( Roppolo et al . , 2007; Pacifico et al . , 2012 ) leading us to test the hypothesis that the tdTomato-positive axons forming these glomeruli in RTP1 , 2DKO mice could be stabilizing the expression of Olfr143 , an oOR near the Olfr151 gene locus . However , we found no tdTomato-positive OSNs that also expressed Olfr143 . This suggests that at least a portion of tdTomato positive OSNs stabilize the expression of the same OR , which is probably an oOR other than Olfr143 . Further investigation of the identities of tdTomato-positive OSNs could further our understanding of the OR gene switching mechanism . A number of reports have shown that ORs are expressed in various organs outside the olfactory system ( Griffin et al . , 2009; Feldmesser et al . , 2006; Flegel et al . , 2013; Kang and Koo , 2012 ) , whereas expression of RTP1 and RTP2 appears to be confined to the peripheral olfactory tissues . One well-established example of a functional OR outside the olfactory system is Olfr78 ( also known as MOR18-2 ) and its human ortholog OR51E2 , which have been reported to function in the prostate , airway and kidney as well as the carotid body ( Aisenberg et al . , 2016; Chang et al . , 2015; Wang et al . , 2006; Pluznick et al . , 2013 ) . This receptor is an oOR and is trafficked to the cell surface in heterologous cells without the RTPs ( Zhou et al . , 2016; Pluznick et al . , 2011; Neuhaus et al . , 2009 ) . It is also interesting to note that ORs expressed in the bladder and thyroid ( Olfr544 , Olfr558 , and Olfr1386 ) are all oORs ( Kang et al . , 2015 ) . Together , it is tempting to speculate that oORs expressed outside the OE are more likely to play chemosensory roles ( Feingold et al . , 1999 ) . In conclusion , our study suggests the importance of OR trafficking by the RTP family members in OSN function and the probability of OR gene choice . Our studies contribute to a broader understanding of how cells discern the presence of a GPCR on their cell surface post translationally and link protein trafficking to epigenetic modifications that give rise to changes in the cell’s expression profile . In the future , it will be interesting to ask whether the UPR pathway that links the functional trafficking of receptors to the cell surface with epigenetic modifications is utilized by other tissue types . The strategies for generating RTP1 and RTP2 knockout mice are illustrated in Figure 1 . The coding regions of RTP1 and RTP2 were replaced by loxP and puromycin ( Pac ) cassette , respectively . Fragments used for the left and right arms were amplified by PCR using BAC clone derived from C57BL/6 mice as templates . For RTP1 , ACN cassette and DT-A cassette were used for positive and negative selection , respectively . A targeting vector was electroporated into the ES cell line EF-1 , which is 129/B6 hybrid ( Bronson et al . , 1996 ) . Colonies were picked up in G418-containing medium . Genomic DNA was digested by Acc65 I and hybridized with a 500 nt external probe on Southern blots . To generate RTP1&2 double knockouts , the RTP2 targeting vector was electroporated into two ES cell lines in which RTP1 is replaced by loxP-Cre-neo-loxP cassette . Colonies were picked up in puromycin -containing medium . Genomic DNA was digested by Bgl II and hybridized with a 500 nt external probe on Southern blots . Targeted ES cell clones were injected into C57BL/6 blastocysts . Chimeric mice were bred with C57BL/6 mice . Mice with RTP1 , 2 mutant allele were maintained by backcrossing with C57BL/6 . Primers for genotyping are: 5’-cggaattcatgtcaggctgcaacttc-3’ and 5’-gggccgatattgggttaggag-3’ for WT allele; 5’-agccagctcttaagtccttac-3’ and 5’-gctcgagatctagatatcgataccgt-3’ for mutated allele . Primers for genotyping RTP2 KO are: 5’-ccctgaagagtctcacccgctc-3’ and 5’-cacatataccccaacttctagg-3’ for WT allele; 5’-caaacagacgaaccctagcaattcccactg-3’ and 5’-cttcattctcagtattgttttgccaagttc-3’ for mutated allele . The procedures of animal handling and tissue harvesting were approved by the Institutional Animal Care and Use Committee of Duke University . All experiments were carried out on both male and female mice . The following mouse strains were all obtained from The Jackson Laboratory: Olfr151tm14 ( Adrb2 ) Mom/MomJ ( β2 Adrenergic Receptor-IRES-LacZ ) ( Feinstein et al . , 2004b ) , Stock no . 006691 ( RRID:IMSR_JAX:006691 ) . B6;129P2-Olfr151tm26Mom/MomJ ( M71-IRES-tauGFP ) ( Feinstein et al . , 2004a ) , Stock no . 006676 ( RRID:IMSR_JAX:006676 ) . Olfr151tm28 ( cre ) Mom/MomJ ( M71-IRES-Cre ) ( Li et al . , 2004 ) , Stock no . 006677 ( RRID:IMSR_JAX:006677 ) . B6;129S6-Gt ( ROSA ) 26Sortm9 ( CAG-tdTomato ) Hze/J ( ROSA26-loxP-stop-loxP-tdTomato ) ( Madisen et al . , 2010 ) , Stock no . 007905 ( RRID:IMSR_JAX:007905 ) . OR ORFs were cloned into pCI ( Promega ) with a Rho tag at the N terminal . All plasmids were verified using Sanger’s sequencing ( 3100 Genetic Analyzer , Applied Biosystems ) . Mice aged 2–4 months were sacrificed by anesthesia followed by rapid decapitation . After removing the skin , the skull was hemisected along the midline with a razor blade , exposing the turbinates . Hemisections were stabilized with pins in a custom Sylgard chamber for recording . Electroolfactograms were measured using glass pipettes filled with ACSF ( tip size15–20µm , resistance ~0 . 5 MΩ ) . For some recordings , electrode tips were filled with 0 . 5% agarose . Electrodes were placed on the anterior surface of turbinate II and referenced to an Ag/AgCl wire placed on the surrounding bone . Signals were filtered ( 0 . 1Hz - 100 Hz ) and amplified ( 1000X ) using a differential amplifier ( DAM-80 , WPI ) . High-pass filtering introduced a slight rebound above baseline for strong responses . Odors were delivered using a custom olfactometer at a final dilution of 0 . 01% ( 1:1000 vol/vol in mineral oil , and a further 1:10 via airflow by combining flow from odorant headspace with a moisturized deodorized airstream ) and a total flow rate of 100 ml/min . Odorants were obtained from Sigma at the highest purity available and diluted in mineral oil . Epithelia were kept moisturized at all times . The blanks are an average of multiple interleaved trials interspersed within the series . This averaged blank is re-displayed for each different odorant for comparison . Mice were weaned at three weeks and their OE was dissected out and flash frozen . 20–25 micron thick sections were cut onto slides which were stored at −80°C and were subsequently thawed and fixed in 4% PFA for 20 min , washed for 1 min with 0 . 5% triton-x-100 in PBS and then rinsed twice in PBS . They were blocked for 1 hr ( See table for blocking reagent and antibody concentration ) in PBS with 0 . 1% triton-x-100 and then kept overnight at 4°C in primary antibody solution made in the same blocking reagent . The antibody solution was washed 3 times for 5 min in PBS and then subject to the secondary antibody for an hour . 0 . 001% bisbenzimide , used to visualize the nucleus , was added to the slides for 1 min followed by 3 × 5 min washes in PBS . The coverslip was mounted in 5–6 drops of Mowiol . No . TargetSourceDilutionCompanyBlockingCatalog no . RRID1GFPRabbit1:4005% milk2OMPGoat1:400Wako5% milk019-22291AB_6646963pS6Rabbit1:200ThermoFisher5% milk44-923GAB_25337984Caspase 3Rabbit1:1000Cell Signaling5% milk9661AB_23411885LSD1Rabbit1:800abcam4% Donkey serumab17721AB_4439646ATF5Goat1:1000Santa Cruz4% Donkey serumsc-46934AB_20587617M71Guinea Pig1:3000Barnea et al . 4% Donkey serum8LacZMouse1:3000Promega4% Donkey serumZ3781AB_4308779Td TomatoRabbit1:10 , 000Rockland4% Donkey serum600-401-379AB_2209751 Candidate ORs which had less than 80% identity with all other ORs were chosen from the RNA-Seq data . Probes against their ORFs were prepared by the addition of the T3 start site to the 3’ end of the pCI plasmid primer followed by incorporation of digoxigenin ( DIG ) using T3 RNA polymerase ( Promega ) and alkaline degradation to get labeled probes of around 200 bp . Slides were prepared as described above and fixed in 4% PFA for 15 min and washed twice in PBS . They were acetylated in 1 . 2% triethanolamine and dropwise addition of acetic acid . They were washed in PBS for 5 min and prehybridized in buffer for an hour in large plates soaked in 5XSSC and 50% formamide at 58°C . 1 μl of the aforementioned labeled probe in 200 μl of the prehybridization buffer was pipetted on to the slides and covered with a layer of parafilm and kept overnight at 58°C . The slides were thoroughly washed in 5XSSC and then in 0 . 2XSSC twice for 30 min , 5 min in PBS and then blocked for an hour . The slides were stained with alkaline phosphatase conjugated antibody against DIG ( Roche ) for an hour and then kept in development buffer for 5 min before being subject to NBT-BCIP in development buffer overnight . Slides were then stained with bisbenzimide and mounted in Mowiol as described above . Slides were prepared and treated using the in situ hybridization protocol until the antibody staining step . Horse radish peroxidase ( HRP ) conjugated antibody against DIG was used instead of the alkaline phosphatase conjugate . Antibody staining amplification was carried out using tyramide signal amplification ( TSA ) using cy3 as the flourophore ( PerkinElmer ) followed by antibody staining against nATF5 as described in the immunofluorescence section above . Three-week-old mice were sacrificed and the entire head was dissected and kept in 4%PFA for 30 min . The tissue was washed in buffer A ( 100 mM phosphate buffer [pH 7 . 4] , 2 mM MgCl2 , and 5 mM EGTA ) once for 30 min and then for 5 min in buffer A and buffer B ( 100 mM phosphate buffer [pH7 . 4] , 2 mM MgCl2 , 0 . 01% sodium desoxycholate , and 0 . 02% Nonidet P40 ) and then kept in buffer C ( buffer B , with 5 mM potassium-ferricyanide , 5 mM potassium-ferrocyanide , and 1 mg/ml of X-Gal ) containing X gal overnight at 4°C . Whole mount tissue was then washed in PBS and imaged under a 5x objective ( Mombaerts et al . , 1996 ) . The medial glomeruli were not observed using this method . The whole olfactory mucosa was dissected out of 3-week-old mice ( 2 males and 1 female sex matched littermates ) and homogenized in trizol . This solution was centrifuged at maximum speed for 10 min and the supernatant collected was treated with 0 . 2 ml chloroform for every 1 ml of trizol , hand shaken for 3 min and spun for 15 min at maximum speed . The aqueous phase was collected and added to isopropanol ( 500 μl per 1 ml of trizol ) and shaken and kept for 5 min at room temperature and then spun for 10 min . The RNA pellet thus collected was washed once in 75% ethanol by adding 500 μl per 1 ml of trizol and centrifuged for 2 min and then again in 180 μl of 75% ethanol . The pellet was then briefly air-dried before being suspended in 50 μl of water and the concentration of RNA was determined using a spectrophotometer by taking 1 μl of the sample and diluting it with 9 μl of water . RNeasy cleanup kit was then used to process the sample . Library generation was carried out using Illumina TruSeq Stranded RNA-Seq kit and HiSeq Illumina sequencing was carried out at the Duke Sequencing and Genomic Technologies Core . Reads were mapped , using kallisto ( Bray et al . , 2016 ) , to the mouse transcripts which were downloaded from UCSC Genome Browser ( https://genome . ucsc . edu ) and whose OR genes were replaced with extended OR gene annotations . Reads assigned on each gene were counted by kallisto . The read count table thus generated was analyzed using EdgeR . DESeq was used to calculate the size factors of individual libraries , FDR ( False Discovery Rate ) was used to adjust multiple comparisons between the ORs as previously published ( Jiang et al . , 2015 ) . HEK293T cells were grown in Minimal Essential Medium ( MEM ) containing 10% FBS ( vol/vol ) with penicillin-streptomycin and amphotericin B at 37°C , saturating humidity and 5% CO2 . These cells were authenticated using polymorphic short tandem repeat ( STR ) at the Duke DNA Analysis Facility using GenePrint 10 ( Promega ) and shown to share profiles with the reference ( ATCC ) . No mycoplasma infection was detected . HEK293T cells were grown to a 100% confluence before being trypsinized and resuspended and seeded onto 35 mm plates at 25% confluency . These plates were cultured overnight then transfected with rho tagged ORs in the plasmid pCI for 18 to 20 hr along with GFP expression vector to monitor the transfection efficiency . The cells were stripped with cell stripper and triturated before being kept in 5 mL round bottom polystyrene ( PS ) tubes ( Falcon 2052 ) on ice . The cells were spun down at 4°C and resuspended in PBS containing 10 mM HEPES , 15 mM NaN3 , and 2% FBS to wash the cell stripper . They were subjected to 30 min in primary antibody ( mouse anti Rho [Laird and Molday , 1988] ) and then washed , stained with phycoerythrin ( PE ) -conjugated donkey anti-mouse antibody ( Jackson Immunologicals ) in the dark . 7-Amino-actinomycin D ( Calbiochem ) , a fluorescent , cell-impermeant DNA binding agent that selectively stains dead cells , was added to eliminate dead cells . The cells were analyzed using BD FACSCanto II FACS with gating allowing for 10 , 000 GFP positive , single , spherical , viable cells and the results were analyzed using Flowjo ( Dey and Matsunami , 2011 ) . All images were captured on Zeiss Axioskop two fluorescent microscope using Q capture pro . The images were then analyzed using ImageJ . For OR bias , phosphorylated S6 and caspase three experiments , nuclear staining was quantified by selecting the OSN layer in the OE and using the maxima function in ImageJ to select and count all cells followed by quantification of OR or Caspase three positive OSNs by hand scoring using the cell counter function . Percent positive cells were calculated as ( Positive cells/ total number of cells ) *100 . ATF5 was quantified by selecting the OSN layer and digitally straightening it using ImageJ followed by manually selecting the ATF5 positive cells and using the measure function to get the X and Y co-ordinates to measure the height from the basal end of the epithelium . All colocalization experiments were manually scored by selecting the OR positive cell’s nucleus and measuring the pixel intensity for the same selection . The pixel intensity of the neighboring area was subtracted to remove background and determine positive cells . The area occupied by OSNs in the OE was hand selected using nuclear staining in image J . The thickness of the sections was measured by straightening the OE followed by measuring the height of the straightened section in four places . The average height for every position was compared between the two genotypes . The OMP or ACIII or GAP43 positive area was selected using imageJ thresholding and these areas were measured and expressed as percentages in the figures . Images were taken at five roughly equivalent positions in the OE using the VNO and OB as landmarks ( anterior VNO , middle VNO , posterior VNO , anterior OB and middle OB , Figure 2—figure supplement 1 shows an example of matching OE sections using the anterior OB ) . We further analyzed all the sections from RTP1 , 2DKO mice at 1 day , 21 day and 6 months and compared the difference in percent area occupied by the OMP positive layer , ACIII positive layer and GAP43 positive layer with that of their littermates . Each data point is obtained from one image from one matched section . The data sets were compared using a paired student t test or Mann Whitney U test as indicated in the figure legend . Percent positive cells were calculated by hand scoring positive cells and calculating percentage based on the total number of OSNs in the image counted using nuclear staining . Each individual section was counted as an individual data point . Height of positive cells was calculated by straightening OE sections and obtaining the Y co-ordinates of hand scored ATF5 positive cells . Multiple comparison data from our ANOVA analysis is included in supplementary file 3 . Each OSN positive for OR FISH signal was selected in image J . These selections were used to measure the pixel intensity of pS6 staining . The average pixel intensity of the entire OE selection was used as background . The average intensity of each cell was normalized by the background followed by its subtraction .	Olfaction , or the sense of smell , is perhaps the most complicated and least understood of the five basic senses . Olfactory neurons in the nose can detect and distinguish between tens of thousands of different odor producing substances . They do so by using hundreds of unique sensors called olfactory receptors , each of which responds to a specific type of odor . During development , each olfactory neuron “chooses” to produce only one type of olfactory receptor . Once the neuron recognizes that the functional receptor is being generated and transported to the cell surface , it will stop making all the other olfactory receptors . Chaperone proteins are responsible for transporting many olfactory receptors to the cell surface . To investigate how the loss of these chaperones affects how the olfactory system develops , Sharma et al . studied mice that were unable to produce the olfactory chaperone proteins . In these mice , developing neurons that chose to produce a type of olfactory receptor that depends on chaperone protein transport could not fully shut off other olfactory receptor genes . This led either to the neuron attempting to produce another type of receptor , or the death of the neuron . As a result , more neurons than usual produced receptors that do not require chaperone proteins to transport them to the cell surface . The olfactory neurons therefore produced only a fraction of all possible olfactory receptors , which decreased the ability of the mice to respond to odors . In the future , it will be important to understand what determines whether an olfactory receptor can be transported to the cell membrane in the absence of chaperone proteins . Olfactory receptors are G protein-coupled receptors ( GPCRs ) , which are the largest molecular class of drug targets for cancer and diseases that affect the brain and heart . Thus , results presented by Sharma et al . will also be relevant to researchers who study how GPCR malfunction causes diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2017	Olfactory receptor accessory proteins play crucial roles in receptor function and gene choice
Planar cell polarity ( PCP ) and neural tube defects ( NTDs ) are linked , with a subset of NTD patients found to harbor mutations in PCP genes , but there is limited data on whether these mutations disrupt PCP signaling in vivo . The core PCP gene Van Gogh ( Vang ) , Vangl1/2 in mammals , is the most specific for PCP . We thus addressed potential causality of NTD-associated Vangl1/2 mutations , from either mouse or human patients , in Drosophila allowing intricate analysis of the PCP pathway . Introducing the respective mammalian mutations into Drosophila Vang revealed defective phenotypic and functional behaviors , with changes to Vang localization , post-translational modification , and mechanistic function , such as its ability to interact with PCP effectors . Our findings provide mechanistic insight into how different mammalian mutations contribute to developmental disorders and strengthen the link between PCP and NTD . Importantly , analyses of the human mutations revealed that each is a causative factor for the associated NTD . Neural tube closure defects ( NTDs ) are common congenital malformations in humans , affecting approximately 1 in 1000 births . During embryogenesis , the neural plate undergoes shaping , before neural folds develop and elevate , contact , and then fuse at the midline to create the closed tube structure that will form the spinal cord and brain ( Copp et al . , 2003 ) . Studies in Xenopus , zebrafish , and subsequently mouse , revealed a role for planar cell polarity ( PCP ) signaling in this process . Further , it was demonstrated that Looptail ( Lp ) mice , with mutations in the mVangl2 locus ( see below ) , fail to establish a neural plate width that will allow for bending and closure ( Etheridge et al . , 2008; Wallingford and Harland , 2002; Wang et al . , 2006a; Ybot-Gonzalez et al . , 2007 ) . Consistent with a role in neurulation , Vangl2 ( an orthologue of the Drosophila PCP core gene Van Gogh/Vang ) is expressed broadly in the neural plate prior to , during , and after closure ( Kibar et al . , 2001 ) . Interestingly , the second orthologous mouse gene , Vangl1 , is also expressed in this tissue; however , its expression is more restricted , showing a complementary pattern with Vangl2 ( Doudney et al . , 2005; Torban et al . , 2008; Torban et al . , 2007 ) . Lp was originally defined by two independent mutations ( D255E and S464N ) in mVangl2 that map to the C-terminal tail . Both display an identical phenotype , with homozygous mice developing craniorachischisis , a completely open neural tube ( Kibar et al . , 2001; Murdoch et al . , 2001 ) . At the molecular level , the mutations were found to disrupt binding between Vangl2 and effectors and also showed reduced membrane localization , decreased protein levels , and defective phosphorylation of Vangl2 ( Devenport and Fuchs , 2008; Gao et al . , 2011; Guyot et al . , 2011; Murdoch et al . , 2014; Song et al . , 2010; Torban et al . , 2004; Torban et al . , 2007 ) . Overall , this led to suggestions that the mutants are loss-of-function ( LOF ) , and that Vangl2 is a rate-limiting component in a dosage-sensitive pathway . However , this is in contrast to the earliest genetic studies , which suggested a semi-dominant mutation with incomplete penetrance . This was due to the presence of a looped tail phenotype in the majority of , but not all heterozygotes , caused by delay of posterior neuropore closure ( Copp et al . , 1994; Strong and Hollander , 1949 ) . Since the original identification of the Vangl2Lp alleles , other PCP mutant mice were found to develop NTDs , implicating the pathway defects as a disease risk factor ( Andre et al . , 2012; Curtin et al . , 2003; Etheridge et al . , 2008; Hamblet et al . , 2002; Merte et al . , 2010; Murdoch et al . , 2003; Wang et al . , 2006a; Wang et al . , 2006b ) . Due to this strong correlation , efforts began to investigate whether human NTD patients might also exhibit mutations in PCP components . A number of mutations were identified in patient populations in VANGL1 and VANGL2 , as well as additional core PCP and associated proteins ( reviewed in Juriloff and Harris , 2012 ) . In each case , the mutation was heterozygous and , interestingly , familial mutations could be found to result in different disease severity among family members . However , this is perhaps not surprising given the complex etiology of NTDs , which are thought to result from a mixture of genetic and environmental factors . Furthermore , digenic mutations were also discovered among PCP genes , consistent with a multifactorial hypothesis , whereby additional factors would be required to achieve a threshold for disease progression ( Allache et al . , 2012; Beaumont et al . , 2019; Chen et al . , 2018b; De Marco et al . , 2012; Merello et al . , 2015; Wang et al . , 2018 ) . This is additionally supported by chimeric mouse studies using the Lp allele that showed an all-or-nothing effect in developing craniorachischisis ( Musci and Mullen , 1990 ) . Despite the progress in identifying mutations within PCP components , data is lacking as to whether these mutations are in fact pathological , with only a few studies addressing this question to date ( Iliescu et al . , 2014; Kibar et al . , 2007; Lei et al . , 2010; Reynolds et al . , 2010 ) . Furthermore , evidence has emerged that the mVangl2Lp allele may in fact be a dominant negative mutation ( Song et al . , 2010; Yin et al . , 2012 ) , indicating there is more to investigate surrounding its molecular behavior . In addition , a recessive mVangl2Lp mutation was discovered ( with substitution R259L ) that gave an apparently normal phenotype in heterozygotes . In this case , only 47% of homozygote animals displayed a looped tail and 12% developed spina bifida , a milder NTD as compared to craniorachischisis , leading to the suggestion that this Lp mutation is a hypomorphic Vangl2 LOF allele ( Guyot et al . , 2011; Wang et al . , 2006b; Yin et al . , 2012 ) . Polarization of epithelial cells , and cells in general , is critical for the morphogenesis and function of mature tissues , with perturbation of cellular polarity and tissue organization implicated in numerous diseases . Epithelial cell polarity can be derived in two axes , apical-basal and orthogonal to the plane of the epithelium , which is referred to as planar cell polarity ( PCP ) . PCP establishment is governed by members of the conserved non-canonical Wnt/Frizzled-PCP pathway . Besides the four-pass trans-membrane protein Vang ( Vangl1 and Vangl2 in mammals , see above ) , which was - like all other core PCP factors - originally discovered in Drosophila ( Taylor et al . , 1998 ) , a . k . a . strabismus/stbm ( Wolff and Rubin , 1998 ) , they include the atypical cadherin Flamingo ( Fmi; Celsr in mammals ) , the seven-pass transmembrane protein Frizzled ( Fz; Fzd in vertebrates with several family members ) , and the cytoplasmic proteins Dishevelled ( Dsh; Dvl in mammals ) , Diego ( Dgo; Inversin/Diversin in vertebrates ) , and Prickle ( Pk ) . The pathway is synonymous with the asymmetric localization of these core members , which form into two sub-complexes on opposite sides of a given cell , creating an intracellular bridge to convey polarity across the tissue . The complexes also direct spatially restricted downstream signaling through tissue-specific effectors , leading to cytoskeletal rearrangement in the majority of cases ( Adler , 2012; Goodrich and Strutt , 2011; Humphries and Mlodzik , 2018; Singh and Mlodzik , 2012; Vladar et al . , 2009; Yang and Mlodzik , 2015 ) . Asymmetric localization of PCP complexes is directly observable in cells of the Drosophila wing where they align to the proximal-distal axis . Molecular interactions promote the formation of stable complexes at proximal ( Fmi-Vang-Pk ) and distal ( Fmi-Fz-Dsh-Dgo ) cell membranes , with Fmi forming a homotypic interaction across cells . In the wing , signaling leads to the formation of a single-actin rich hair at the distal vertex of each cell . While in other tissues displaying visible PCP features , phenotypes can include fate specification or coordinated cell movement ( Adler , 2012; Goodrich and Strutt , 2011; Humphries and Mlodzik , 2018; Singh and Mlodzik , 2012 ) . For example , in the Drosophila eye , PCP signaling is responsible for determining the differential fate specification of photoreceptors R3 and R4 , which also directs rotation and subsequent orientation of the photoreceptor clusters , referred to as ommatidia . These signaling events in turn lead to a mirror-image arrangement of ommatidia across the dorso-ventral midline , or equator . Disruption of PCP in the wing leads to misorientation of cellular hairs or multiple wing hairs ( Adler , 2012; Goodrich and Strutt , 2011; Humphries and Mlodzik , 2018; Singh and Mlodzik , 2012 ) , while in the eye chirality or rotation defects of the ommatidia are observed ( Mlodzik , 1999; Strutt and Strutt , 1999 ) . The importance of PCP during vertebrate development and disease has also become widely recognized ( Goodrich and Strutt , 2011; Humphries and Mlodzik , 2018; Simons and Mlodzik , 2008; Wang and Nathans , 2007 ) . The PCP pathway is highly conserved in vertebrates , with different tissues displaying PCP readouts or defects upon misregulation ( Goodrich and Strutt , 2011; Wang and Nathans , 2007; Yang and Mlodzik , 2015 ) . One example of this is the above mentioned mVangl2Lp , and the first example of a link between PCP signaling and NTDs ( Kibar et al . , 2001; Murdoch et al . , 2001 ) . Based on the fact that several human patient-derived VANGL1/2 mutations exist , and the availability of comparable mouse alleles , we set out to investigate how these alleles impact PCP signaling functionally in a uniform defined genetic background . For this we utilized Drosophila . Overall , we focused on six mutations located in the C-terminal tail of Vang , where all molecular interactions have been thus far mapped . We observed that all resulted in aberrant PCP signaling and phenotypes , however , to varying degrees . In the majority of cases , the mutation was dominant , either hypermorphic gain-of-function or dominant negative , leading to altered protein localization of both alleles in vivo . We were also able to show that one mutation was indeed hypomorphic in its behavior . Taken together , our analyses demonstrate in molecular detail the nature of the different VANGL1/2 alleles . These alleles display defective PCP signaling in vivo , indicating their causative association with the NTD phenotypes of the human patient mutations . The mammalian core PCP genes Vangl1/2 , and Vang in Drosophila , play essential roles in planar polarity signaling . Vang interacts genetically with each PCP core gene . It has further been demonstrated that it also can physically interact with all other members of the PCP complex , and the interaction sites with the cytoplasmic PCP core factors have been mapped to its C-terminal domain ( Bastock et al . , 2003; Darken et al . , 2002; Das et al . , 2004; Humphries and Mlodzik , 2018; Jenny et al . , 2005 ) . While mutations have been observed along the entire span of the VANGL1/2 genes in human NTD patients , mutations known to be causative for NTD in the mouse all map to the C-terminal tail ( Chen et al . , 2013; El-Hassan et al . , 2018; Guyot et al . , 2011; Kibar et al . , 2001; Murdoch et al . , 2001 ) . We were interested to explore whether the different mutations observed in human patients could impact PCP phenotypically . Furthermore , we set out to investigate how mutations found in both the mouse and human patients affected signaling at a molecular and functional level . For this , we turned to Drosophila as it allows for intricate and quantitative analyses , and is a genetically simpler system with less redundancy - there is only one Vang gene in flies - that , importantly , shows functional and mechanistic conservation ( Adler , 2002; Adler , 2012; Goodrich and Strutt , 2011; Singh and Mlodzik , 2012 ) . We chose to focus our attention on mutations within the C-terminal tail due to the functional importance of this region . A number of C-terminal mutations have been discovered to date in both mice and humans and are detailed in Figure 1—source data 1 . For our study , we selected two mutations from the mouse and four mutations from human patients for further analysis , these include D255E and R259L identified in mouse Vangl2 ( Guyot et al . , 2011; Kibar et al . , 2001 ) , M328T and R517H found in human VANGL1 ( Kibar et al . , 2007; Merello et al . , 2015 ) , and R270H and R353C from human VANGL2 ( Kibar et al . , 2011; Lei et al . , 2010; Figure 1 ) . These mutations were selected for varying reasons . Firstly , we were interested to compare mouse D255E and R259L in a genetically uniform background , as the mutations are only a few residues apart , and strikingly the more conservative mutation ( D255E ) displays a more severe phenotype as compared to R259L in mouse models ( Guyot et al . , 2011; Kibar et al . , 2001 ) . Secondly , R270H was selected , as the residue is absolutely conserved across all species , and it covers two independent mutational events , as a mutation in VANGL1 was also observed at the equivalent residue ( see Figure 1B ) in a digenic combination along with a CELSR1 mutation ( Chen et al . , 2018a; Kibar et al . , 2007 ) . Thirdly , M328T was chosen because information to date suggests this mutation may be a LOF allele ( Reynolds et al . , 2010 ) , and R353C as it was suggested that effector binding may be reduced ( Lei et al . , 2010; Figure 1B ) . Finally , we selected R517H , a highly conserved residue with a conservative mutation where the bioinformatic prediction of its tolerance was unclear ( Figure 1—source data 1 ) . Overall , we analyzed mutations from both VANGL1 and −2 , investigated mutations spread along the length of the C-terminal tail ( Figure 1A ) , compared conservative as well as drastic substitutions , and focused on well conserved sites across species . To compare mutations in vivo in Drosophila , we took advantage of the Gal4/UAS system to allow for regulated expression of our transgenes ( Brand and Perrimon , 1993 ) . We also included a Flagx3-tag and an attB ( bacterial attachment ) site . To enable site-specific recombination , constructs were then introduced into a fly strain containing an attP ( phage attachment ) docking site , along with PhiC31 integrase activity to mediate integration between the attB and P sites ( Bischof et al . , 2013; Bischof et al . , 2007 ) . Thus , by using the same attP docking site , we ensured insertion of the transgenes at a constant genetic locus and thus equivalent expression for accurate comparison . We chose to introduce mutations into the Drosophila Vang gene , rather than use mouse Vangl2 or human VANGL1/2 , to preserve native signaling and to reduce genetic complexity . As the equivalent residues have different numbers in Vang , as compared to the mouse or human orthologues , the mutations we investigated are referred by their numbers in Drosophila , here and throughout the paper , and are as follows: D317E , R321L , R332H , V391T , K418C and K577H ( see Figure 1B for comparison to mammalian residue numbers ) . In order to first test the contribution of individual mutations , we performed an overexpression experiment . Driving expression of wild-type ( wt- ) Vang with actin-Gal4 ( ac >WT ) led to a moderate hair reorientation phenotype ( Figure 2C and D ) . This allowed us to compare whether mutations showed a gain- or loss-of-function phenotype with respect to the positive ( wt-Vang ) and negative ( Vang-/- ) control . A change or an increase in phenotype quality or strength as compared to wt-Vang overexpression could suggest a dominant negative effect or a gain-of-function in activity , while a decrease in strength would suggest a reduction in protein function or stability . Upon overexpression of wt-Vang , we observed consistent hair reorientation in specific regions of the wing . We therefore chose two regions that would be suitable for qualitative and quantitative analyses of the different mutants , with region 1 showing a more severe PCP phenotype as compared to region 2 ( Figure 2A ) . Different mutations displayed different patterns of hair reorientation within these regions , confirming the system had an appropriate degree of sensitivity for this type of analysis . In region 1 , D317E , R321L and K577H showed a distinct phenotype as compared to wt-Vang ( over ) expression . In fact , their hair pattern was more similar to that of Vang-/- wings or a genetically wild-type wing ( w1118 ) , included as a reference control that has no phenotype . In contrast , R332H , V391T and K418C all showed hair reorientation patterns similar to the gain-of-function ( GOF ) effect of wt-Vang overexpression ( Figure 2C and Figure 2—figure supplement 1B ) . We also observed comparable changes in hair reorientation within region 2 . In this case , D317E and K577H mirrored the phenotype of Vang-/- , while R321L showed similarity to the reference control ( w1118 ) . R332H , V391T and K418C again showed similar patterns to the GOF overexpression of wt-Vang ( Figure 2D and Figure 2—figure supplement 1B ) . To quantify hair patterns , we utilized FijiWingsPolarity ( Dobens et al . , 2018 ) , which determines the orientation of each hair within a given region , relative to the proximal-distal ( P-D ) axis . For visualization purposes , hair orientation angles are represented through arrows and a color gradient ( Figure 2C and D ) . Through quantification of hair orientation angles , we were able to show that in region 1 all mutations with the exception of V391T had a significant alteration in hair reorientation as compared to wt-Vang overexpression , while in region 2 all mutations showed significant divergence from the wt-Vang ( Figure 2B ) . To allow for statistical analysis , the data were pooled into three bins of angle orientation; normal orientation as determined by the reference control w1118 ( green ) , as well as a more anterior reorientation ( yellow ) or posterior reorientation ( blue ) . The full spectrum of angles in 30° segments also reflected the changes in pattern ( Figure 2—figure supplement 1A ) . We confirmed that similar levels of protein expression were observed for each transgene ( due to our experimental design ) . However , interestingly , D317E , R321L and K577H all showed a notable change in protein mobility ( Figure 2E ) . This mobility shift was also observable in samples from Vang-/- tissue and in transfection experiments in S2 cells ( Figure 2—figure supplement 1C and Figure 4C ) . It has previously been demonstrated that Vang is phosphorylated leading to a mobility change and associated band shift on gels ( Kelly et al . , 2016; Strutt et al . , 2019 ) . Accordingly , phosphatase treatment altered wt-Vang mobility , but it did not impact any of the mutants suggesting that their phosphorylation is reduced or lost ( Figure 2—figure supplement 1C ) . As Vang phosphorylation is associated with membrane localization and function , this suggests that the D317E , R321L and K577H mutants are defective in one or both of these regards ( Kelly et al . , 2016; Strutt et al . , 2019 ) . To further investigate the effect of the respective mutation in the overexpression system , we analyzed protein localization in the pupal wing at ~25 hr APF , by which time wt-Vang is predominantly localized to the plasma membrane ( Figure 3B ) . Besides comparing the localization of the mutants themselves , we also assessed any dominant effects by examining their effect on the localization of co-expressed wt-Vang . We expressed wt-Vang directly via the actin promoter ( ac-Vang-GFP ) within physiological levels throughout the whole wing blade/animal and simultaneously used engrailed ( en ) -Gal4 to overexpress the transgenes in the posterior compartment of the wing ( Figure 3A and Figure 3—figure supplement 1A ) . This allowed for comparison of the localization of wt-Vang-GFP in the anterior ( control ) vs . posterior ( transgene expression ) compartments . Overexpression of wt-Vang in the posterior compartment showed distinct membrane localization ( as expected ) , visualized through Flag staining , while no signal was observed in the anterior compartment ( Figure 3B ) . Vang-GFP ( expressed from ac-Vang-GFP ) also showed distinct membrane staining with levels reduced in the posterior compartment as compared to the anterior , suggestive of competition from the overexpressed Vang-Flag for membrane recruitment ( Figure 3B ) . The junctional marker PatJ was unaffected by Vang expression ( Figure 3B ) . Consistent with their overexpression phenotypes , a number of changes to localization were observed for the different mutants . D317E and K577H were largely localized to the cytoplasm , and strikingly also altered the localization of wt-Vang-GFP in the posterior compartment ( Figure 3C ) . In contrast , R321L behaved similarly to wt-Vang in both its localization and effect on Vang-GFP . R332H and K418C displayed a less distinct membrane localization that was echoed by diffuse Vang-GFP localization in the posterior compartment ( Figure 3C ) . V391T showed distinct membrane localization with slightly increased membrane levels of Vang-GFP , as compared to controls ( Figure 3B and C ) . In all cases , Vang-GFP and PatJ staining were unaffected in the anterior compartment ( Figure 3—figure supplement 1B ) . Overall , the localization data is consistent with our observed hair reorientation phenotypes , as well as protein mobility data ( Figure 2 ) . D317E and K577H behave like dominant negative , antimorph alleles . They cause hair reorientation patterns similar to Vang-/- ( Figure 3D and Figure 3—figure supplement 1B ) . The similarity in pattern to Vang-/- can be explained , as in these like in Vang-/- the protein is lacking from the membrane ( Figure 3C ) . However , these mutants do not represent ‘simple’ LOF alleles , as they are dominantly affecting the localization of wt-Vang . R321L also displayed changes to protein mobility but did not lead to significant hair reorientation when overexpressed; its largely normal membrane localization and lack of effect on wt-Vang-GFP suggest it is a hypomorphic allele with reduced protein function ( Figures 2 and 3C ) . R332H and K418C also appear to act in a dominant fashion , but in this case the protein is recruited to the membrane , where it could have a dominant effect on signaling ( Figure 3C ) . In contrast , the staining of V391T and slight increase in accompanying Vang-GFP membrane localization suggest it may act as a mild GOF to enhance Vang signaling ( Figure 3C ) . To confirm that the localization of the mutant proteins was not due to interference from endogenous Vang , we next assayed its localization in a Vang-/- background . Here , we used actin-Gal4 driven expression with which , due to the higher than endogenous expression levels of the transgenes ( via the Gal4 amplification ) , we observed a gain-of-function phenotype with wt-Vang ( Figure 4A ) . Despite this effect , we observed hair reorientation patterns in the absence of endogenous Vang that were consistent with the experiments described above . D317E and K577H displayed hair reorientation patterns similar to Vang-/- , while R332H , V391T and K418C showed a similar phenotype of hair orientation swirls as compared to overexpressed wt-Vang . Furthermore , R321L showed a more ‘complete rescue’ , supporting its behavior as a hypomorph in protein function and thus a ‘weaker GOF’ phenotype , which resembles a better rescue of the Vang-/- background ( Figure 4A and Figure 4—figure supplement 1B ) . Analysis of Flag staining for the transgenes revealed similar localization patterns as compared to the overexpression experiment in a wt control background ( Figure 3 ) . Wt-Vang localized distinctly at the membrane , indicated by an overlap with cortical actin stain ( Figure 4B ) , while R321L and V391T showed slightly reduced and enhanced membrane localization , respectively . Furthermore , R332H and K418C displayed a diffuse membrane localization , and D317E and K577H failed to localize to the membrane ( Figure 4B ) . We again confirmed that expression levels of each transgene were equivalent , and expected changes in protein mobility were detectable ( Figure 4C ) . Taken together , our data are consistent with the conclusions that D317E and K577H act as dominant-negative mutations with reduced membrane localization , while R332H and K418C act as dominant mutations that exert their effect at the membrane . R321L behaves like a hypomorph with reduced protein function , and V391T is a mild GOF enhancing protein function . To confirm and further refine the suggested mechanistic effects of mutations in altering Vang function , we analyzed their phenotype in the absence of endogenous Vang in an additional tissue . For this , we used the Drosophila eye , which shows generally weaker GOF effects with high expression levels of Vang as compared to the wing . For consistency , we used actin-Gal4 ( ac >Vang ) to express the transgenes , and we also performed experiments with sep-Gal4 , an eye specific driver related to and based on the sevenless/sev enhancer and promoter , that gives lower levels and spatially restricted expression ( Fanto et al . , 2000 ) . PCP phenotypes in the eye are characterized by changes in ommatidial chirality as well as orientation ( Figure 5A ) . Introduction of ac >Vang Flag in the Vang-/- background resulted in a near perfect rescue with very minor chirality and orientation defects . As this phenotype was trending toward an overexpression phenotype of Vang , this suggested that the levels were just marginally too high for a complete rescue ( Figure 5B ) . In contrast , due to its weaker expression sep >Vang Flag showed only a partial rescue ( Figure 5B ) . By comparing phenotypes with both Gal4 drivers , we were thus able to better ascertain how the mutants affected Vang function . With both drivers , D317E and K577H showed very limited rescue in regards to chirality and only partial rescue in orientation , overall displaying a similar phenotype to Vang-/- ( Figure 5C , D and Figure 5—figure supplement 1 ) . This is consistent with their antimorphic nature; however , the partial rescue we observe suggests the mutants retain some activity . With actin-Gal4 , R321L displayed a significantly better rescue of chirality ( as compared to wt-Vang control ) , while with sep-Gal4 the rescue was less efficient both in terms of chirality and orientation ( Figure 5C , D and Figure 5—figure supplement 1 ) . Together this confirmed that R321L is a mild hypomorphic LOF allele and that the better rescue level observed with ac-Gal4 , as compared to wt-Vang , is due to the higher expression levels . R332H showed a significant enhancement in chiral defects , as compared to control wt-Vang , with the actin-driver , as well as more significant orientation defects with the sep-driver . K418C showed a phenotype similar to wt-Vang ( Figure 5C , D and Figure 5—figure supplement 1 ) . While , V391T trended towards more severe chiral defects with the actin-driver , with a significantly better rescue of orientation with the sep-driver ( Figure 5C , D and Figure 5—figure supplement 1 ) . Taken together with the wing data , this supports the notion that V391T is a mild hypermorph with GOF activity . Due to the defects associated with the different mutants , we wished to determine how they might affect effector binding . As many of the mammalian studies have focused on the interaction between Dishevelled ( Dvl ) and Vangl , we first examined the binding between Dsh and Vang , the Drosophila orthologues . We expressed wt-Vang-Flag and mutant constructs in S2 cells and tested for their interaction with Dsh-GFP by pull-down . D317E , R321L and K577H all showed a reduced ability to interact with Dsh-GFP ( Figure 6A ) , while normal binding , comparable to wt-Vang , was observed for R332H , V391T and K418C ( Figure 6B ) . Next we tested binding between the different Vang mutants and Prickle ( Pk ) , the main effector and cytoplasmic interaction partner of Vang in vivo . The same pattern of binding was observed , with a reduction in binding to D317E , R321L and K577H , but no change with R332H , V391T and K418C ( Figure 6C and D ) . This is consistent with the affected membrane association of the first three Vang mutations and our previous functional and phenotypic assertions ( summarized in Figure 6E ) . Furthermore , these interaction data suggest a trend whereby D317E , R321L and K577H all lead to a general reduction in effector binding . Interestingly , analyzing the location of these mutants along the C-terminal tail , they map in close proximity to regions of specific effector binding sites ( Figure 6G ) . K557H is situated near to the PDZ binding motif which is involved in Scribble ( Scrib ) binding ( Scrib is required for PCP and has been shown to interact with Vang in both mice and Drosophila [Courbard et al . , 2009; Montcouquiol et al . , 2003] ) . We were able to show that Dgo binding requires amino acids 304–323 ( Figure 6F ) and D317E and R321L map within this region . As demonstrated , these mutations do not show specificity for a particular effector but instead a general reduction in binding . This correlation suggests that sites of effector binding along the C-terminal tail highlight regions of high sensitivity for Vang functionality , not just for binding to a particular effector , but in general for the integrity of the entire protein . The detailed analyses of the NTD-associated mutations revealed specific functional features . For example , the R321L mutation ( all numbers refer to the residues in Drosophila Vang , see Figure 1 for the respective numbers in mouse and human , unless otherwise indicated ) behaves like a mild hypomorphic LOF allele , as demonstrated largely by a lack of phenotype upon over-expression and through the level of rescue observed . This is consistent with data from the Vangl2Lp-R259L mouse , where only mild NTD defects were observed , along with no difference in membrane localization or levels ( Guyot et al . , 2011 ) . In our system , besides reduced membrane localization , we also observed its aberrant mobility in western blot assays , an indication that functionality of the protein was decreased , as well as diminished ability to bind effectors . Differences in protein mobility , localization , and effector binding were also observed for D317E and K577H . In contrast to R321L , however , both mutants failed to localize to the membrane , which was reflected in a hair orientation phenotype reminiscent of a Vang-/- wing . Furthermore , both mutants showed largely no rescue in the respective assays . D317E displayed a stronger phenotype as compared to K577H , which correlated with the severity of subcellular mislocalization , highlighting the sensitivity and accuracy of the Drosophila assays . Expression of D317E and K577H transgenes also led to mislocalization of wild-type Vang , revealing mechanistic insight into the dominant nature of these mutations . For D317E , this is consistent with data from the mouse ( D255E ) , with our data further demonstrating that the mutation does indeed act in a dominant negative manner . It was previously demonstrated that Sec24b promotes the selective sorting of Vangl2 into vesicles for ER to Golgi transport , and that the Vangl2Lp ( D255E ) allele fails to undergo this process ( Merte et al . , 2010 ) . It has been demonstrated that the Vangl2Lp mouse shows reduced membrane protein levels , thought to result from the protein becoming trapped in the ER and subsequently targeted for degradation . While this might suggest a ‘simple’ LOF phenotype , the ability of Vangl1 and Vangl2 to interact complicates matters ( Yin et al . , 2012 ) . Further , it was demonstrated that Vangl1 is missing from the membrane in Vangl2Lp mice but not in a Vangl2 knock-out , with Vangl2Lp also displaying a more severe phenotype ( Song et al . , 2010; Yin et al . , 2012 ) . Additionally , a dominant effect was observed in regard to the phosphorylation of wild-type Vangl2 , which was reduced upon co-expression of the Vangl2Lp mutant ( Gao et al . , 2011 ) . This is consistent with - and reminiscent of - our dominant relocalization observations and , due to the hair reorientation pattern caused in our experiments , our data support the rationale that the original Vangl2Lp alleles are dominant negative mutations and not LOF . This is in alignment with the original genetic studies performed in the mouse that proposed semi-dominance ( Copp et al . , 1994; Strong and Hollander , 1949 ) . The patient derived K577H mutation had not previously been investigated . Here , we demonstrate that it is causative for PCP defects , and interestingly showed mechanistic similarity to D317E ( the original Vangl2Lp mutation , see above ) , functioning like a dominant negative . Therefore , our data provide a link between the NTD mouse mutations and a human mutation and support the idea that mutations that affect PCP signaling in distinct ways could all contribute to NTD pathology . In line with this notion , we also observed that R332H and K418C displayed dominant activity in functional assays , consistent with the hypothesis that they are also causative in NTD . However , their phenotypic impact was different . These mutants retained effector binding and were localized to the membrane , even if not as distinctly as wild-type Vang . Additionally , the hair re-orientation phenotype observed was reminiscent of a GOF effect of Vang , which results from the non-polar localization of core PCP proteins at the membrane . As their interaction with cytoplasmic effectors and phosphorylation appear to be retained , this points to R332H and K418C behaving as hypermorphs with an effect on the PCP signaling complexes at the membrane . However , due to the relatively high expression levels of our transgenes we were not able to find conditions to analyze changes to protein asymmetry . Nonetheless , taken together and most importantly , the above-mentioned human patient mutations , cause dominant PCP defects , K577H with a dominant negative function , and R332H and K418C as hypermorphic alleles , and thus are likely causative of the NTDs . Finally , we obtained data suggestive of V391T functioning as a mild hypermorph GOF . Although in the majority of assays V391T behaved similarly to R332H and K418C , its membrane localization appeared more robust , and rescue experiments were suggestive of the mutation leading to a milder GOF , and not dominant behavior , as was seen with R332H and K418C . Nonetheless , the data for this human mutation is again consistent with it also being causative of NTD . While there are caveats of using any specific system , including Drosophila , the ability to phenocopy and refine results suggested from the mouse , for example in the case of the original Vangl2Lp , and to demonstrate sensitivity through quantitative analyses , highlights its utility in exploring the functionality and potential causative nature of human NTD-associated mutations . In fact , our assays were both able to reveal additional functional insight into mutations that had been studied previously by other means and to define functional and mechanistic behavior of previously unstudied human mutations . A popular system to investigate the different mutations is to analyze their localization in MDCK cells . In this assay , the mouse Vangl2Lp allele , D255E , was detected within the ER , and showed reduced stability and proteosomal degradation , echoing phenotypes from the mouse ( Gravel et al . , 2010 ) . The same assay also showed defective localization for S464N , R259L and R274Q ( equivalent to R332H in our assay ) , with all mutations displaying very similar phenotypes ( Iliescu et al . , 2014; Iliescu et al . , 2011 ) . While a useful assay to reveal potentially damaging effects upon Vangl1/2 , it does not capture the true nature of the mutation as our in vivo assays do . For example , R259L is milder in phenotype as compared to S464N and displays membrane localization in vivo , which is not captured in the MDCK assay . In contrast , our assays capture these features in the Drosophila in vivo system . Furthermore our results define that R274Q/R270H in hVANGL1/2 , respectively ( R332H in Drosophila ) , likely has dominant negative activity . The result in MDCK cells for both R259L and R274Q is also at odds with experiments performed in zebrafish . Both mutants behaved similarly to wild-type protein in overexpression and rescue experiments ( Guyot et al . , 2011; Reynolds et al . , 2010 ) . The lack of phenotype from the zebrafish experiments suggests that , while suitable for mutations that have a strong pathological effect , this system may not be sensitive enough to reveal phenotypic and functional insight with milder mutations . In line with this hypothesis , the zebrafish system did show phenotypic behavior for the Vangl2Lp mutations , mouse D255E and S464N , and additionally human M328T , with all performing as LOF in assays in the fish ( Reynolds et al . , 2010 ) . Our data also revealed that mouse D255E ( D317E in Drosophila ) displayed LOF characteristics , for example diminished effector binding . However , our assays allow us to conclude further , as discussed above , that the mutant displays dominant negative function . In contrast , our data are not consistent with the zebrafish findings for human M328T . The equivalent Drosophila mutant , V391T , displayed a mild GOF phenotype . This residue was the least conserved among the mutations tested , and while it is methionine in human and mouse Vangl1 , the equivalent residue is valine in mouse , human and frog Vangl2 , as well as in Drosophila Vang . The differences in phenotypic outcome may be due to this discrepancy , and thus investigating the phenotypic behavior of Vangl2 with an equivalent mutation in a mammalian assay could provide further insight . This result also implies , as is to be expected , that the most conserved residues between systems will give the most reliable functional insight . Nonetheless , we were able to define that a mutation at this residue causes a dominant PCP defect , suggesting its importance for overall Vang function in PCP in general and NTD in particular . As discussed above , our data set reveals graded phenotypes in vivo , with different mutations displaying more or less severe PCP defects . However , we could not find a pattern to correlate the severity observed with features of the different residues , for example , whether they occurred originally in VANGL1 vs . VANGL2 , if the substitution was more conservative or drastic , or whether familial or sporadic . This suggests that it is the location of the residue that exhibits a mutational change as the most important factor in determining phenotypic strength . A prime example of this is the comparison of the two mouse Vangl2Lp alleles with a conservative substitution D255E and more drastic substitution R259L . While they are located in a similar region , the closer proximity of D255E downstream of the transmembrane domain , which also correlates with the Dgo interaction region , suggests it is a functionally more important residue and site , for protein integrity . Consistent with this notion , our observations that the mutations that display LOF characteristics , D255E , R259L and K577H , all map close to regions of effector binding , suggest that these regions are important for the general functional and structural integrity of Vang family proteins . While we cannot rule out that the reduced binding observed for D317E and K577H is due to their mislocalization and inaccessibility to effector proteins , this is clearly not the case for R321L which is localized to the membrane . We therefore favor the above hypothesis and believe these mutants lead to a general structural defect . While many of the NTD mutations may interfere with Vang/Vangl function by affecting its structure , it is also possible that they hit as yet unknown motifs or sites of post-translational modification . For example , R274/270 , in human VANGL1/2 , respectively ( R332 in Drosophila ) , forms part of a putative D-box ubiquitination motif ( Iliescu et al . , 2014 ) , and two mutations found in the N-terminus of human patients are mapped to phosphorylation sites important for Vang function ( Gao et al . , 2011; Kibar et al . , 2009; Lei et al . , 2010 ) . Overall , this suggests that there is much more to be revealed in determining both the contribution of PCP mutants to NTD , and that analyzing such NTD-associated mutations will reveal important mechanistic insight into Vang/Vangl function in general , in both developmental and disease contexts . Flies were raised on standard medium , and maintained at 25°C unless otherwise stated . To generate UAS-Vang-Flagx3 transgenic flies , Vang-Flagx3 ( Kelly et al . , 2016 ) was PCR-amplified using phusion high-fidelity DNA polymerase ( Thermo Scientific ) and cloned into the pUAST-attB vector ( Bischof et al . , 2007 ) using NotI-XbaI sites . Point mutants were created using site-directed mutagenesis , the primers used can be found in the key resources table . Plasmids were verified by Sanger sequencing ( GENEWIZ ) and sent to BestGene Inc for insertion into BDSC stock number 9752 - PBAC{yellow[+]-attP-3B}VK00037 . S2 cells were grown according to standard procedures in Schneider’s Medium ( Gibco ) supplemented with 10% heat-inactivated Fetal Bovine Serum ( Gibco ) . Effectene ( QIAGEN ) was used to transfect plasmids into S2 cells according to manufacturer’s protocols . For details of constructs used please refer to the key resources table . Cells were transfected for ~48 hr before lysis in buffer containing 50 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 1 mM EDTA and 1% Triton-X-100 . For pull-down experiments with Flag , 10 μl of anti-Flag M2 affinity gel was used per sample ( Sigma Aldrich ) . Lysates were incubated with affinity gel at 4°C , followed by two washes with buffer containing 50 mM Tris-HCl pH 7 . 5 , 350 mM NaCl , 1 mM EDTA and two washes with buffer containing 50 mM Tris-HCl pH 7 . 5 , 500 mM NaCl , 1 mM EDTA , 0 . 1% SDS . For GFP pull-down , GFP-Trap agarose was used ( Chromotek ) . In this case , cells were lysed in buffer containing; 10 mM Tris-HCl pH7 . 5 , 150 mM NaCl , 0 . 5 mM EDTA , and 1% Triton-X-100 . This was diluted prior to agarose incubation as per the manufacturer’s protocol . Lysates were incubated with 20 μl agarose at 4°C followed by four washes with buffer containing 10 mM Tris-HCl pH7 . 5 , 350 mM NaCl , 0 . 5 mM EDTA . For both pull-downs , samples were eluted through boiling at 95°C in 5x final sample buffer . Wing discs were dissected from third instar larvae and lysates prepared by boiling collected discs at 95°C in 5x final sample buffer . To perform phosphatase treatment , lysates were exposed to lambda protein phosphatase ( NEB ) for 30 min at 30°C . Lysates were resolved by polyacrylamide gel electrophoreses and transferred to nitrocellulose membrane . The primary antibodies and concentration used for immunoblotting can be found in the key resources table . HRP conjugated secondary antibodies were used at 1:5000 ( Jackson ImmunoResearch Laboratories ) . Adult wings were collected in PBS containing 0 . 1% Triton-X-100 ( PBST ) and incubated for 1 hr at room temperature before mounting in 80% glycerol in PBS . Pupal wings were dissected and fixed in 4% paraformaldehyde containing 0 . 1% Triton-X-100 for 45 mins–1 hr . Tissue was washed with PBST twice and incubated in 5% donkey or goat serum containing PBST for 15 min . Primary antibodies were added and tissue incubated overnight at 4°C . Tissue was washed three times with PBST before incubation with fluorescent secondary antibodies and phalloidin diluted in 5% serum PBST for 2–4 hr at room temperature . Samples were washed four times with PBST and mounted in Vectashield media ( Vector Labs ) . Primary antibodies used are listed in the key resources table . Secondary antibodies were from Jackson ImmunoResearch Laboratories ( 1:200 ) and Phalloidin from Molecular Probes ( 1:1000 ) . Eye sections were prepared as previously described ( Gaengel and Mlodzik , 2008 ) , and eyes were sectioned near the equatorial region for analysis . Imaging of adult wings and eye sections was performed on a Zeiss Axioplan microscope , imaging of pupal wings was carried out on either a Leica SP5 or Zeiss 880 confocal microscope . Hair reorientation angles were quantified using the FijiWingsPolarity plugin , details of which can be found in Dobens et al . ( 2018 ) . To perform the quantification , equivalent regions were cropped from three wings in each genotype . Angles of polarity were determined utilizing the plugin and the angles from each genotype combined . For statistical analysis , angles were divided into three categories: anterior reorientation , wild-type , and posterior reorientation . This satisfied the conditions of a valid Chi-squared test . Adult eye sections were assigned chirality by hand and orientation angle was defined using ImageJ . For statistical analysis , a Chi-squared test was performed . In all cases , experiments were performed on at least 3 distinct occasions to ensure technical replication . To ensure biological replication for experiments involving Drosophila tissue , 3–10 individual flies were examined for phenotypic similarity . All images and blots within the study were selected as the most representative of the population or findings after this analysis . This approach and sample size is consistent with previous studies performed in our laboratory that have generated reproducible data . In all cases , unless otherwise stated a Chi-square test was performed . This analysis was performed as it is an accepted test for differences between binned distributions as is the case for our data , and so our analysis examined whether we could disprove to a certain level of significance , the null hypothesis that two data sets are drawn from the same population distribution function . For analysis of actin rotation due to the low values in the medium and severe categories these categories were combined and a Fisher’s exact test was performed . The Fisher’s exact test was performed in place of the Chi-square test as the low sample size in specific categories meant conditions of the latter would not be satisfied leading to inaccurate analysis . All analyses were done using Prism software and N value and p values are detailed in tables below .	As an embryo develops , its cells must work together to build mature tissues and organs . During the formation of the nervous system , for example , a sheet of cells destined to become the brain and spinal cord folds up into a tube spanning the length of the embryo . Normally , this tube – known as the ‘neural tube’ – zips up , and the cells that will eventually become skin and other surrounding tissues close in over it . If the neural tube does not close completely , different parts of the spinal cord or brain can remain unprotected . This can cause diseases called neural tube defects , such as spina bifida , which is characterized by holes in the backbone exposing the spinal cord and surrounding membranes . Patients with neural tube defects can have similar genetic mutations , for example , in the genes controlling a process called “planar cell polarity” , or PCP for short . Cells arranged in flat sheets use the PCP process to sense direction , and it is this process that allows structures , such as the scales on a fish or the hairs on a mouse , to all point in the same direction . PCP is also important in embryonic development: sheets of cells that can sense direction correctly can therefore move collectively to complete complex tasks ( such as closing the neural tube ) . However , no-one knew whether the specific PCP gene mutations implicated in neural tube defects in humans actually affected the cells’ ability to sense direction , or indeed whether they were even involved in causing the diseases . Humphries et al . set out to find out more about these mutations using fruit flies as a model system . The fruit fly is widely used to study the genes and signals involved in direction sensing , especially PCP . Problems with PCP produce easily measurable changes in the wing and eye , showing what went wrong and how badly . Humphries et al . genetically engineered fruit flies to have the same mutations as human patients and revealed that these mutations did indeed alter cells’ ability to sense direction . These experiments also showed that each mutation did so in a different way , and with varying severity . This explained why the same mutations caused different levels of neural defects in mice ( which are commonly used to study human diseases ) and suggests that they might contribute to neural tube disorders in humans . These results show potential connections between neural tube defects and direction sensing in cells . In the future , this study and follow-up work could help researchers to understand what types of mutation have the most impact , which may eventually allow doctors to better predict who is most at risk of being affected by these conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2020	Mutations associated with human neural tube defects display disrupted planar cell polarity in Drosophila
A widely accepted model for the evolution of cave animals posits colonization by surface ancestors followed by the acquisition of adaptations over many generations . However , the speed of cave adaptation in some species suggests mechanisms operating over shorter timescales . To address these mechanisms , we used Astyanax mexicanus , a teleost with ancestral surface morphs ( surface fish , SF ) and derived cave morphs ( cavefish , CF ) . We exposed SF to completely dark conditions and identified numerous altered traits at both the gene expression and phenotypic levels . Remarkably , most of these alterations mimicked CF phenotypes . Our results indicate that many cave-related traits can appear within a single generation by phenotypic plasticity . In the next generation , plasticity can be further refined . The initial plastic responses are random in adaptive outcome but may determine the subsequent course of evolution . Our study suggests that phenotypic plasticity contributes to the rapid evolution of cave-related traits in A . mexicanus . A major problem in modern biology is understanding how organisms adapt to an environmental change and how complex , adaptive phenotypes originate . Phenotypic evolution can result from standing genetic variation , new mutations , or phenotypic plasticity followed by genetic assimilation , but these processes are often difficult to distinguish in slowly changing environments . This difficulty can be overcome by studying adaptation to more abrupt environmental changes , such as the dramatic transition from life on the Earth’s surface to subterranean voids and caves . A unifying feature of subterranean environments is complete darkness ( Culver and Pipan , 2009; Pipan and Culver , 2012 ) . Cave-adapted animals have evolved a range of unusual and specialized traits , often called troglomorphic traits , which enable survival in challenging conditions of the subsurface . In cave dwelling animals , visual senses and protection from the effects of sunlight are unnecessary , and consequently eyes and pigmentation are usually reduced or absent . To compensate for lack of vision , other traits , especially those related to chemo- and mechano-receptor sensations , are enhanced . Circadian rhythms that fine-tune organismal physiology with day-night cycles are also distorted , and light dependent behaviors , as well as the neural and endocrine circuits controlling these behaviors , are modified . Because photosynthetic organisms are not present in caves , primary productivity is absent and nutrient availability is usually limited . Survival under conditions of reduced and/or sporadic food resources is possible due to the evolution of modified feeding behaviors and adaptive changes in metabolism , such as lower metabolic rate , increased starvation resistance , and changes in carbohydrate and lipid metabolism ( Culver and Pipan , 2009 ) . The ancestors of cave dwelling animals originally lived on the surface . Regardless of whether the pioneering animals entered the subsurface accidently ( by capture of surface waters or by falling into a pit ) or purposely ( to take refuge from harsh competition or changing climate conditions ) , they were suddenly exposed to darkness and subterranean living conditions . How did they overcome the challenges of life in darkness , such as orientation in the absence of light , survival during long periods without food , and detection of mates , in order to successfully establish underground lineages ? The preadaptation hypothesis suggests that the ancestors of cave adapted animals were pre-adapted to darkness by already leading semi-nocturnal lifestyles in environments with low-light penetration , such as rotting logs , leaf litter , or in the soil , simplifying their transition to life in caves ( Vandel , 1965 ) . However , many cave dwelling animals , such as the cavefish Astyanax mexicanus , are descended from diurnal surface-dwelling animals; and the cavefish Poecillia mexicana , is derived from photophilic surface ancestors ( Parzefall et al . , 2007 ) with no obvious pre-adaptations to low light . For these animals , the transition to dark subterranean habitats could have posed significant challenges . In the present investigation , we used the teleost Astyanax mexicanus as a model system to understand the molecular , physiological and morphological basis of cave colonization . A . mexicanus is an excellent system for studying adaptation to cave life because it consists of both the ancestral-proxy form and multiple derived forms: surface dwelling ( surface fish , SF ) and cave-dwelling conspecifics ( cavefish , CF ) respectively . Surface fish live in rivers exposed to day-night cycles with abundant food availability . In the karstic Sierra Madre Oriental del Norte mountain range in Mexico , SF have colonized about 30 caves ( Gross , 2012 ) , and their cave dwelling descendants have adapted to permanently dark , mostly food restricted subterranean waters . Recent studies revealed that cavefish lineages are much younger than previously thought , from less than 20 , 000 ( Fumey et al . , 2018 ) to about 200 , 000 ( Herman et al . , 2018 ) years old , implying that the evolution of numerous cave-related phenotypes may have occurred over relatively short time periods . Furthermore , troglomorphic phenotypes are maintained in A . mexicanus cavefish despite the lack of complete isolation between surface fish and cavefish populations ( Bradic et al . , 2012; Chakraborty and Nei , 1974; Herman et al . , 2018; Wilkens and Hüppop , 1986 ) . A remarkably recent origin of troglomorphic traits , also involving likely immigration from surface populations , has also been proposed for the evolution of cave-related traits in other cave dwelling animals , including invertebrates and vertebrates ( Behrmann-Godel et al . , 2017; Klaus et al . , 2013; Niemiller et al . , 2008; Schilthuizen et al . , 2005; Villacorta et al . , 2008; Zhang and Li , 2013 ) . What are the mechanisms that enable the rapid evolution of various distinct cave-related phenotypes , and how are they preserved despite the homogenizing process of gene flow from surface populations ? Clearly , strong selection must be involved in these cases , but in order for selection to act , it is essential to have phenotypic variants with differences in fitness in the cave environment . The conspecific surface and cave morphs of A . mexicanus are capable of hybridization in the laboratory and thus have been the subjects of extensive genetic analysis ( O'Quin et al . , 2013; Protas et al . , 2008; Yoshizawa et al . , 2012 ) . However , not many mutations have been uncovered in CF , even in the presumably dispensable genes associated with eye degeneration , such as crystallin genes ( Hinaux et al . , 2013; Ma et al . , 2014 ) . A more immediate mechanism for survival is likely necessary when new colonizers first enter caves . This suggests the presence of standing genetic variation and/or phenotypic plasticity followed by genetic assimilation ( Waddington , 1953 ) , rather than the complete reliance on the accumulation of new mutations to produce the necessary variability . An instance of standing genetic variation has been reported in A . mexicanus CF with respect to the evolution of eye reduction ( Rohner et al . , 2013 ) , but phenotypic plasticity has not been studied extensively in this system ( but see Reyes , 2015 ) . This is surprising since phenotypic plasticity plays an important role in the adaptation of many different organisms to changing environments and the colonization of new habitats reviewed in Fox et al . ( 2019 ) ; Morris ( 2014 ) . To replicate the colonization of the subterranean environment , we placed A . mexicanus SF in completely dark conditions for up to two years . By assaying different traits , we discovered that dark-raised SF show numerous phenotypes that are normally associated with CF adaptations to dark environments . The results imply that some cave-adapted traits may have appeared rapidly by phenotypic plasticity in A . mexicanus . Our results provide a basis for the evolution of adaptations in the cavefish lineage by genetic assimilation . In this study , we exposed SF to D/D from early developmental stages until up to 2 years after spawning , and compared them to L/D controls . As a first step we compared body shape , length and width , eye size , and pigmentation in SF raised under D/D and L/D conditions ( Figure 1A ) . We found no consistent differences in body parameters or eye sizes between D/D and L/D SF ( Figure 1B ) . However , when we compared thickness of retinal layers between the two groups , 5 out of 7 layers were significantly different: D/D fish showed a thinning of the two plexiform layers and a thickening of the two nuclear layers and the photoreceptor layer ( Figure 1D , Figure 1—figure supplement 1 ) . Surprisingly , we also observed a significantly higher number of melanophores in the flank of the trunk and below the dorsal fin in D/D compared to L/D SF , although similar levels of pigmentation were noted in other regions of the body ( Figure 1C ) . To understand the effects of exposure to constant darkness at the molecular level , a comparative transcriptomic analysis was conducted . We used three , 7 month-old surface fish placed in either D/D or L/D conditions within one dpf ( day post fertilization ) . RNA sequencing yielded 21 . 9 to 27 . 6 million reads from each fish , representing a total of 25 , 194 genes . Basic statistical data are included in Supplementary file 2 . We found 356 differentially expressed genes at a significance threshold of padj < 0 . 1 ( Figure 2—source data 1 ) . Of these , 210 were up-regulated and 146 were down-regulated in the D/D fish; the set of differentially expressed genes contained 67 genes with unconfirmed functions . We were particularly interested in the genes related to known aspects of the CF phenotype . Genes involved in circadian regulation , locomotor rhythm and visual perception were down-regulated in D/D fish , whereas lipid metabolism was the main functional category enriched in the up-regulated gene set . Two genes involved in pigmentation were also changed but their expression was higher in D/D , consistent with the increase in pigment cells described above . We also found significant changes in gene expression which were not predicted from known CF-associated phenotypes . These genes function in oxidation-reduction processes , hormone activity , hemostasis , aromatic amino acid metabolism , gene expression , metabolism of proteins , and signal transduction ( Figure 2 ) . To validate the RNAseq results , we examined the levels of several differentially expressed genes by real time PCR ( RT-PCR ) ( Figure 3 ) . We confirmed the down-regulation of the hormone related genes somatostatin 1 tandem duplicate 2 ( sst1 . 2 ) and inhibin beta B ( inhbb ) , ros involved dual oxidase ( duox ) , the circadian rhythm genes period circadian clock 2 ( per2 ) and cryptochrome-1-like ( cry3b ) , the vision related genes retinoschisin 1a ( rs1a ) and tubby like protein 1a ( tulp1a ) , as well as genes involved in the regulation of metabolism pancreatic and duodenal homeobox 1 ( pdx1 ) , and deptor in D/D compared to L/D SF . However , some genes , most notably aromatic amino acid metabolism genes hydroxyphenylpyruvate dioxygenase a ( hpda ) , tryptophan 2 , 3-dioxygenase ( tdo2a ) , aralkylamine N-acetyltransferase 1 ( aanat1 ) , and tryptophan hydroxylase 1a ( tph1a ) showed down-regulation according to RT-PCR results , whereas they were up-regulated in the transcriptome . This discrepancy may be related to differences in the age or condition of fish at the time of sampling . In addition , we quantified some genes which did not show significant differential expression according to RNAseq results but are important for the cavefish phenotype: heat shock protein 90α ( hsp90aa1 . 2 ) , and the de novo DNA methyltransferases dnmt1 , dnmt3aa , dnmt3ab , dnmt3ba and dnmt3bb . Hsp90aa1 . 2 , dnmt1 and dnmt3bb showed changes in expression related to photoperiod in Astyanax . The RNAseq results and subsequent RT-PCR validation showed changes in the expression of differentially expressed genes relative to photoperiod in both SF and PA CF . Expression of some genes changed in the same direction in both SF and CF ( e . g . inhbb , per2 , sst1 . 2 ) , whereas others changed in one and not the other fish type , or changed in the opposite directions in different fish types ( e . g . duox , nr1d1 , tph1a ) . Except for duox and dnmt3bb . 1 , all of the significantly changed genes show the same direction of changes in L/D vs . D/D SF and in SF vs . PA: hsp90aa1 . 2 , rs1a , tdo2a , dnmt1 , per2 , sst1 . 2 , tulp1a , aanat1 , and tph1a . A major challenge facing animals that colonize caves is low food availability due to limited or absent primary productivity . To determine how a surface ancestor may have coped with this difficulty , we raised SF and PA larvae in D/D and L/D beginning < 24 hpf ( hours post fertilization ) without feeding ( N = 36 larvae/group ) . D/D and L/D controls were fed daily portions of brine shrimp from seven dpf , when larvae normally lose their reliance on yolk and begin feeding . By 15 dpf , about 25% more SF and over 65% more PA unfed larvae were alive in D/D compared to L/D conditions . By 18 dpf , only a few unfed embryos were still alive , and they were all from the D/D conditions ( 8 PA and 2 SF ) ( Figure 4 ) . In control conditions , SF survived better than PA , and both fish types had higher survival in D/D conditions . By 18dpf , 35 SF D/D , 17 SF L/D , 20 PA D/D and 12 PA L/D larvae were still alive . The results show that dark raised SF and CF are more resistant to starvation than siblings raised under a normal photoperiod and that PA larvae survive starvation better than SF larvae regardless of the lighting conditions . We hypothesized that one of the reasons why fish in D/D survived starvation longer was because of lower energy expenditure . To test this possibility , we exposed SF and PA to D/D vs . L/D conditions within first 24 hpf after spawning and measured O2 consumption at 2 . 5 dpf and 7 . 5 dpf . The results showed that SF and PA larvae raised in D/D have decreased metabolic rates when compared to SF and PA larvae raised in L/D conditions . This result was found to be significant in SF each time this experiment was conducted ( three replicates ) for both 2 . 5 and 7 . 5-day-old larvae . Furthermore , D/D PA showed a decrease in O2 consumption at 2 . 5 but not at 7 . 5 dpf when compared to larvae raised in L/D . In addition , PA and SF metabolic rates were not significantly different at 2 . 5 dpf , whereas at 7 . 5 dpf PA has a lower metabolic rate than SF larvae raised in L/D conditions . At 2 . 5 dpf metabolic rate is similar between SF and PA , and darkness caused it to be reduced in both types of fish . However , by 7 . 5 dpf the metabolic rate in PA was reduced compared to L/D SF , and PA did not show a plastic response to darkness , whereas at the same age SF metabolic rate was still affected by darkness ( Figure 5 ) . In summary , the results support the hypothesis that D/D fish survive starvation longer because of lower metabolic rates . Improved survival in darkness also could be mediated by other factors . We hypothesized that exposure to constant darkness may represent a chronic stressor and tested this possibility by comparing cortisol levels in both L/D and D/D SF and PA . Surface fish raised in D/D conditions had significantly higher cortisol levels than L/D SF ( Figure 6 ) . In contrast , PA did not show a significant change in cortisol levels after exposure to D/D . This result was confirmed in three independent experiments using fish of different ages and different periods of dark exposure and shows that dark raised SF exhibit higher cortisol levels than SF raised on a normal photoperiod . Our RNAseq results show that genes involved in many aspects of fat metabolism were up-regulated in D/D versus L/D reared SF . To confirm this on a phenotypic level , we quantified triglyceride content in SF and PA raised to adulthood under D/D and L/D conditions . In fish raised under L/D conditions , PA had higher triglyceride levels than SF . Surface fish raised in D/D conditions had markedly higher triglyceride levels than L/D SF . Triglyceride levels in D/D PA were also higher than levels measured in L/D PA , although the difference is more modest than that observed for SF ( Figure 7 ) . These results suggest that dark raised SF have higher levels of triglyceride metabolism that SF raised on a normal photoperiod . Because of morphological changes in the pituitary and thyroid glands in dark raised SF ( Rasquin , 1949 ) and differential expression of some of the genes associated with pituitary hormones ( e . g . sst1 . 2 , inhbb , ghrl ) in the dark-raised SF transcriptome , we quantified the levels of the pituitary hormones thyroid stimulating hormone and growth hormone in SF and PA adults raised in either L/D or D/D conditions . Thyroid stimulating hormone levels were lower in D/D SF when compared to L/D SF and higher in D/D PA when compared to L/D PA , although the difference was not statistically different in each experimental replicate . PA had higher thyroid stimulating hormone levels than SF regardless of lighting conditions ( Figure 8A ) . When thyroid stimulating hormone levels were compared between SF and three populations of CF , there was a trend for higher thyroid stimulating hormone levels in TI ( p=0 . 07 ) , but not MO ( p=0 . 16 ) , and thyroid stimulating hormone levels were significantly higher in PA ( Figure 8B ) . The levels of growth hormone were higher in both SF and PA raised in D/D conditions when compared to fish raised in L/D ( Figure 9A ) . All three populations of CF had higher growth hormone levels than SF ( Figure 9B ) . According to the RNAseq analysis , the tryptophan metabolism pathway was significantly up-regulated in D/D versus L/D SF , while real time qPCR suggested lower expression of the tdo2a , aanat1 and tph1a genes in dark–raised SF . Therefore , we compared serotonin ( 5-HT ) levels in D/D and L/D SF and PA adults by HPLC . Since serotonin levels fluctuate according to a daily rhythm ( Fingerman , 1976 ) , samples for this assay were collected between 11 AM and 3 PM ( day ) or 11 PM and 3 AM ( night ) regardless of rearing conditions . Serotonin concentrations were lower in the brains ( Figure 10A ) and the bodies ( Figure 10B ) of D/D compared to L/D SF . Likewise , the 5-HT metabolite 5-HIAA was lower in the brain of D/D SF ( Figure 10—figure supplement 1 ) . It was not possible to quantify 5-HIAA in the body because it was masked by several interfering peaks . Serotonin levels in the SF brain , but not the body , were significantly lower at night compared to samples collected during the day . In PA brains collected during the day , 5-HT levels were lower in fish raised in D/D compared to L/D conditions but 5-HT levels were similar in D/D and L/D brains at night . Modest , but significant , changes in body 5-HT were evident in PA with increases in D/D relative to L/D collected during the day , and the opposite change occurred in bodies collected during the night . Serotonin levels were significantly lower in PA brains and bodies when compared to SF regardless of experimental conditions ( D/D , L/D , day/night ) . Comparison of SF brain 5-HT with TI , MO and PA fish showed that all three cave populations have lower brain 5-HT ( Figure 10C ) . In contrast to adults , 5-HT was significantly higher in PA compared to SF at seven dpf , and darkness did not affect 5-HT levels in either larval fish types ( Figure 10D ) . To test whether the plastic changes in the fish reared in the dark during the first generation are maintained and transferred to the next generation , we induced spawning in D/D reared SF . We measured starvation resistance and metabolic rate in G1 embryos ( dSF ) developing in D/D or in L/D during the second generation , as well as the offspring of L/D reared controls exposed to L/D or D/D conditions . The G1 larvae reared in L/D showed no differences in starvation resistance ( Figure 11 ) or metabolic rate ( Figure 12 ) , regardless of whether they were the progeny of L/D or D/D parents . Their cohorts raised in darkness showed a modification in these phenotypes: dSF larvae derived from D/D parents showed slightly enhanced plasticity and a minor shift toward lower metabolic rate and higher starvation resistance ( although not statistically significant ) than larvae raised from L/D parents . These results suggest that the plastic changes that appeared in the dark during the first generation may be subject to refinement during darkness in the second generation . In addition to the widely accepted view that cave-adaptations result from long-term genetic processes ( Barr , 1968; Culver , 1982; Juan et al . , 2010 ) , our results indicate that some cave-related traits can appear within a single generation by phenotypic plasticity . Exposure to constant darkness can trigger rapid metabolic , neurological , morphological and molecular changes necessary for survival of A . mexicanus SF colonizers , which may have allowed for the adaptations evident in extant CF . The initial plastic responses can be adaptive or non-adaptive . Subsequently , in the case of an adaptive response , plasticity would be further selected for , or in the case of non-adaptive or mal-adaptive responses , plasticity would be selected against . Our findings strongly implicate phenotypic plasticity as an important mechanism of cave colonization and rapid evolution of cave-related traits and open the possibility that genetic assimilation may be an underlying mechanism of adaptive evolution in A . mexicanus cavefish . Phenotypic plasticity is often the first response of organisms exposed to an environmental change . The ability to quickly respond to novel environmental conditions is essential for initial survival and , eventually , for adaptation to occur on the genetic level ( Fox et al . , 2019; Lande , 2009; Morris , 2014 ) . Darkness is a hallmark of the subterranean realm , the initial stress new cave colonizers encounter , and the only environmental component shared by all underground habitats ( Pipan and Culver , 2012 ) . Building on the work of Rohner et al . ( 2013 ) , we hypothesized that stress caused by perpetual darkness may be an important facilitator of the rapid appearance of novel and beneficial traits that enabled ancestral SF to endure the environmental shift associated with their entry and ultimately their adaptation to life in caves . To test this hypothesis , we used contemporary A . mexicanus SF as a proxy of the ancestral SF colonizers , and exposed them to complete darkness to simulate the most important change that the original SF ancestors of CF experienced when colonizing caves . Indeed , we found an increase in cortisol level in D/D exposed SF supporting our initial premise of darkness activating a stress response . We expected to find plastic changes in D/D reared SF involving pituitary and thyroid hormonal regulation , adipose tissue , and body shape based on the previous work on dark-raised A . mexicanus SF , Chica CF , and Los Sabinos CF ( Rasquin , 1949 ) . Furthermore , experiments on Poecillia mexicana , another teleost with cave adapted lineages , showed that darkness caused degenerate changes in the spine and promoted sexual isolation between adjacent surface and cave lineages ( Riesch et al . , 2016; Riesch et al . , 2011; Torres-Dowdall et al . , 2018 ) . Although our study was unable to replicate the changes in body shape seen by Rasquin ( 1949 ) , we confirmed her observation of increased body fat . We also confirmed changes in pituitary and thyroid hormone metabolism , both in gene expression and hormone levels . Down-regulation of sst1 . 2 , a homolog of somatostatin - growth hormone-inhibiting hormone , is consistent with increased growth hormone in D/D SF . In addition , the changes in thyroid stimulating hormone and growth hormone levels we observed in dark raised SF are consistent with the reduction of basophils and the predominance of acidophils in the pituitary gland of D/D SF reported previously ( Rasquin , 1949 ) . These results suggest that one of the immediate responses to cave colonization and darkness may be changes in the endocrine system . In fact , endocrine signaling , along with epigenetic changes , activation of heat shock proteins , and transcriptional regulation , are the major molecular mechanisms involved in regulating plastic responses ( Aubin-Horth and Renn , 2009; Beldade et al . , 2011; Kelly et al . , 2012 ) . Our transcriptome analyses revealed that genes and GO terms involved in all of the above processes showed significant changes in response to darkness . In addition to hormonal signaling , there are changes in genes involved in the stress response , several different signal transduction pathways , protein metabolism , including protein folding and posttranslational modifications , and genes showing transcription factor or epigenetic regulator activities . Furthermore , using RT-PCR we found that the expression of hsp90aa1 . 2 , dnmt1 and dnmt3bb , genes related to stress or epigenetic processes ( Gore et al . , 2018 ) , were changed in D/D SF . These results suggest that multiple molecular mechanisms underlying phenotypic plasticity may be mobilized by exposure of SF to darkness . Our RNAseq results contained functional categories enriched in both the up- and the down-regulated dataset: hormone activity ( as mentioned above ) and oxidation-reduction processes . Changes in oxidation-reduction processes may be related to increased stress following the exposure to a changed environment . Alternatively , these changes may be light-dependent since light activates the production of reactive oxygen species ( ROS ) , which , in turn , act as messengers between photoreception and the circadian clock ( Hirayama et al . , 2007; Pagano et al . , 2018 ) . Pathways involved in the response to oxidative stress , photoreception , and circadian regulation were down-regulated . Thus , the absence of light may directly reduce ROS production , photoreception , and circadian control . Our study revealed many traits that change in SF upon exposure to darkness . Strikingly , all these traits are associated with the cave lifestyle , and almost all of them change to resemble CF adaptive phenotypes . For example , compared to SF , CF have higher starvation resistance ( Aspiras et al . , 2015 ) , a lower metabolic rate ( Hüppop , 1986; Moran et al . , 2014 ) , higher basal cortisol levels ( Gallo and Jeffery , 2012 ) , and lower serotonin levels ( but see Elipot et al . , 2014 ) . When exposed to darkness , in addition to mimicking higher growth hormone and triglyceride levels in CF ( see above ) , we found that SF show higher starvation resistance , lower metabolic rates , higher cortisol levels , and lower serotonin levels . The results demonstrate that for these traits SF begin to resemble CF in the same generation following exposure to complete darkness . Because the plastic phenotypes are under genetic control in multiple independently evolved CF populations and irreversible when CF are transferred to light in the laboratory ( Gross , 2012; Jeffery , 2001 ) , the results are consistent with the possibility that the initial dark-induced plastic responses in SF became fixed in CF-like descendants via genetic assimilation . However , not all CF traits could be replicated by exposing SF to darkness . Therefore , although important , plasticity caused by constant darkness is not the only path for the appearance of cavefish traits . Other mechanisms , including plasticity caused by environmental factors other than darkness ( e . g . changes in nutrient availability ) , selection on the standing genetic variation , new mutations , or any combination of these processes , might also be involved . In addition , standing genetic variation for differences in plasticity in an ancestral population could result in the evolution of these traits . Although darkness is an ecologically relevant cue , the induction of changes in SF by the absence of light does not necessarily imply that they are adaptive for cave life . Plasticity can result in adaptive , non-adaptive as well as maladaptive responses ( Ghalambor et al . , 2007; Langerhans and DeWitt , 2002 ) . It is self-explanatory how increased triglyceride content , starvation resistance or decreased metabolic rate can increase fitness in the ( food-limited ) cave environment , but the effects of neurotransmitter or hormone changes are less intuitive . Therefore , we hypothesized that , if a trait is adaptive , it would be present in multiple independently evolved cavefish populations ( Prevorcnik et al . , 2004 ) . As a proof of principle , higher starvation resistance and triglyceride content ( Aspiras et al . , 2015 ) conform to this criterion . We tested neurotransmitter and hormone levels and showed that growth hormone and serotonin change in multiple CF concordantly to the initial SF plastic response . Conversely , thyroid stimulating hormone showed the opposite direction of change , suggesting that the initial SF plastic response was not adaptive . Also opposite to the expected direction of change , melanophore numbers increase in D/D SF . The increase of melanophores in dark-raised SF was modest and confined to specific regions of the body , but consistent with up-regulation of some melanosome-related genes in the transcriptome . This result was unexpected but considered reliable , since SF x CF hybrids can also show hyper-melanization ( Gross et al . , 2016 ) , and neutral mutations are a likely cause of melanophore reduction in A . mexicanus CF ( Protas et al . , 2007 ) . We found a surprising result in another classic troglomorphic trait , eye degeneration: although visual perception was a GO term associated with down-regulated genes in D/D SF transcriptome , we were unable to detect changes in eye size . However , D/D SF exhibited well-developed retinal layers , and we even recorded thickening of the inner and outer nuclear layer , and the photoreceptor layer . These regions are cell proliferation and intercalation centers so it is possible that darkness activates mitotic activity in these retinal layers as a compensation for defective transmission of light from the lens and induced retinal stress ( Alunni et al . , 2007 ) . The finding of increased eye size under our experimental conditions of complete darkness is puzzling , but has also been observed in a natural Astyanax population that has recently colonized a cave in Texas ( McGaugh et al . , 2019 ) . Our results are consistent with studies showing that teleosts change body pigmentation and retinal phenotypes in response to light levels ( Epp , 1972; Kay et al . , 2001; Tarboush et al . , 2016 ) . For example , Astyanax F2 hybrids with small or absent eyes also show hyperpigmentation , which has been explained as background adaptation due to the lack of visual input ( Gross et al . , 2016 ) . Overall , our results suggest that plasticity lacks the propensity to produce adaptive variants , and non-adaptive , even maladaptive , effects seem to be as likely as adaptive outcomes . Importantly , pigmentation and eyes are absent in PA and other fully troglomorphic CF . The fact that they showed non- or mal-adaptive plastic response in SF exposed to darkness , suggests that natural selection acting against plasticity may be involved in the evolution of these traits in CF . A novel aspect of the evolution of classical troglomorphic traits , the reduction of eyes and pigmentation , may be that they are a consequence of strong selection against non- or mal-adaptive plastic responses during colonization of caves by surface ancestors . To test if the inheritance of some of the plastic phenotypic changes described here occur during the next generation , we induced spawning in D/D SF and investigated starvation resistance and metabolic rate in the offspring ( dSF ) , traits shown to exhibit plasticity in the parental D/D generation . Neither of these traits showed differences between offspring spawned in the light or dark when maintained in L/D . However , when maintained in D/D , dSF showed continued plasticity of both traits , as well as potential adaptive-tuning of the character states , suggesting that plasticity is evolvable and can itself be an evolutionary adaptation . Increased plasticity was previously suggested to be an important stepping stone in the process of adaptations to extreme environments ( Lande , 2015; Lande , 2009 ) . Therefore , consistent with the Baldwin effect ( Crispo , 2007 ) , the evolution of cave related traits may have proceeded through initial selection for continued plasticity and refinement of the characters , followed by selection of individuals carrying the most improved variants of the phenotype . Higher growth hormone and triglyceride levels , as well as lower body serotonin levels in PA than D/D SF , also imply refinement of the initial SF plastic response in CF . In nature SF are periodically swept into caves and plasticity in many traits that enable survival may be critical to extend their lifespan long enough to leave offspring . According to our results , the next generation may be slightly better equipped for coping with the cave environment by increased plasticity and small adaptive switches in traits . One of the reasons the role of plasticity in adaptive evolution has been contested is its randomness in the production of outcomes where non-adaptive and mal-adaptive responses are as likely as adaptive responses . However , our study suggests that even the initial non-adaptive plastic responses may eventually produce an adaptive outcome . Some traits , including the hallmarks of troglomorphic adaptations ( the loss of eyes and pigmentation ) initially show non-adaptive plastic responses but the complete loss of these traits in CF indicates strong selection against plasticity in SF colonizers . Non-adaptive plasticity may have a greater role in evolution than previously appreciated because it may increase the strength of selection against plasticity , and perhaps , in some cases , may affect the whole trait itself . Conversely , in cases where plasticity produces adaptive outcomes , initial selection will act to enhance plasticity , and plasticity will be maintained in the derived lineage for many generations . Overall , plasticity has a critical role in producing variability ( adaptive or not ) on which selection can act ( for or against ) to produce individuals with optimal fitness in the new environment . We also uncovered two cases in which the plasticity of a single trait changes within an individual over the course of development . Both metabolic rate and serotonin showed shifts in plasticity depending on developmental stage . Metabolic rate initially ( at 2 . 5 dpf ) changes in response to the light regime but later ( by 7 . 5 dpf ) this response is lost in PA . In contrast , serotonin levels are plastic in adults but not in 7 dpf larvae . Interestingly , switches in plasticity occur at the same time as switches in character state . At 2 . 5 dpf there is no difference in metabolic rate between SF and PA , but by 7 . 5 dpf , the metabolic rate in PA is lower than in SF , agreeing with previous studies done on adults ( Hüppop , 1986; Moran et al . , 2014 ) . In the case of serotonin , character states show opposite relationships at different stages: they are lower in larvae but higher in adult SF compared to PA . Therefore , for metabolic rate and serotonin levels , both the trait and its plasticity change concordantly , suggesting that the underlying mechanisms of character development may also influence the ability to respond to environmental stimuli . These two examples show that plasticity is not an intrinsically fixed property of a trait but can develop by itself . In our examples , plasticity develops concordantly with the traits it affects , and it can develop in a direction of expansion or reduction . The development of metabolic rate in PA is accompanied by canalization of the same trait . In the case of serotonin , development of the 5-HT system is accompanied by an increase in its plasticity . Phenotypic plasticity can fill several gaps in the current model of cave colonization and the evolution of cave dwelling organisms . Our study shows how a colonizing surface ancestor , having no adaptations to caves and presented with the challenges of low food availability , orientation , and reproduction in complete darkness , could have overcome a transition to the extreme subterranean environment and establish a cave-adapted lineage . Further , due to plasticity , numerous adaptive traits arise within a single generation in response to only one , albeit the most relevant , environmental cue: darkness . Finally , plasticity may enable the maintenance of distinct phenotypes in the face of gene flow from ancestral surface-dwelling populations . The incorporation of phenotypic plasticity as one of the mechanisms underlying cave colonization and adaptive evolution provides a solution to dilemmas associated with drift or selection ( Cartwright et al . , 2017 ) acting alone to induce cave-related phenotypes in A . mexicanus . Because multiple successful transitions by the same surface ancestors are not unique to Astyanax ( Carlini et al . , 2009; Verovnik et al . , 2004 ) , some animal groups have colonized caves repeatedly in different times and geographical regions , and several unique cave dwellers are closely related to highly invasive ( and plastic ) surface dwelling species ( Bilandžija et al . , 2013; Kupriyanova et al . , 2009 ) , it is possible that plasticity may be a general phenomenon in the colonization and adaptation to cave environments . The study groups included Astyanax mexicanus SF and CF primarily from Pachón Cave ( PA ) . In some experiments , CF from Tinaja ( TI ) and Molino ( MO ) caves were also used . Animals were obtained from the Jeffery Laboratory colony at the University of Maryland , and had undergone 3–5 generations in captivity since their original collection from the caves in 2002 ( PA , TI ) or 2006 ( MO ) . Fish in the main colony were kept in 40 L tanks in cohorts of 8–15 animals and exposed to a 14 hr light/10 hr dark photoperiod . Spawning was induced every two weeks by extra feeding a few days prior to increasing the water temperature ( Jeffery et al . , 2000 ) . Temperature regime changes were as follows: fish were normally kept in 22 °C , on the first day of spawning temperature was raised to 24 °C , on the second day raised to 26 °C , on the third day returned to 24 °C , and on the fourth day returned to 22 °C . Embryos were spawned in the night and collected in the morning by washing them from the breeding nets . Dead embryos were removed and the remainder transferred to clean fish system water containing methylene blue . The fish were raised and handled according to established University of Maryland and NIH guidelines and all experiments conform to the regulatory standards . Fish embryos spawned at the same time from different parent tanks were mixed in order to minimize the effects of specific genetic backgrounds and mimic the situation in nature where fish are randomly swept into caves . Mixed embryo clutches were divided in half and randomly assigned to different experimental conditions . One group was initially exposed to darkness either as embryos or young fry ( from 1 to 2 hr to 3 days after spawning ) and then raised in complete darkness ( D/D ) for up to two years . The other group , originating from the same brood , was reared in a normal 14 hr light/10 hr dark photoperiod ( L/D ) . All other conditions were identical between the D/D and L/D study groups . Fish were reared in plastic tanks with no filtration or running water . Depending on their size and age , fish were reared in 1 , 3 or seven liter tanks and then transferred to larger tanks as they increased in size . Once the fish reached approximately 1 cm in length they were transferred to 3L tanks , and once they grew to over 2 cm they were transferred to 7L tanks . Densities of fish among different groups depended on survival but were approximately similar and ranged between 5 to 10 fish per tank . Water was changed approximately once a month at the same time for every tank in the experiment . Fish in each tank were fed the same volume of brine shrimp once per day , approximately 0 . 5 mL of concentrated brine shrimp was added to smaller fish in 1L tanks , 1 mL in larger 3L tanks , and 2 mL in the largest 7L tanks . All procedures followed standard procedures for Astyanax fish husbandry in the Jeffery Laboratory . Dim red light ( 25 watts ) was used when it was necessary to feed or handle the fish in the dark ( Romero , 1985 ) . Dead fish were removed when they were observed . Fish of different ages were used throughout the experiments and were all adults ( unless otherwise noted ) . Within every experiment fish groups of comparable ages were used for comparisons . Fish from both sexes were included in each experiment , and fish used in each experiment were collected at the same time of the same day . For the G1 generation experiments , SF placed in D/D as embryos and raised for 2 years were spawned using standard procedures as explained above . The spawned embryos ( dSF ) were divided in two groups within the first 24 hr post-fertilization ( hpf ) , and one group was left in D/D conditions and the other group was placed in L/D photoperiod within first 24 hpf . A control group of SF spawned from L/D raised parents was also divided into two groups and raised in L/D or D/D during the period of experimentation . Information on sample sizes , time lapsed in experimental conditions , and the ages of fish used in specific experiments is provided in the sections corresponding to the description of specific experimental methods . The following procedure was used for collecting , and when necessary pulverizing , adult fish for the respective experiments: fish were sacrificed with 0 . 4 or 0 . 5 mg/L MS222 , tricaine methanesulfonate ( Western Chemical Inc , Ferndale , WA , USA ) . Fish were maintained in the dark until they expired and then photographed , weighed , and pulverized in liquid nitrogen using a mortar and pestle . The pulverized material was divided into several Eppendorf tubes and stored at −80 °C until further processing . For the brain chemistry experiment , whole brains were rapidly removed , placed in 1 . 5 mL Eppendorf tubes , immediately frozen on dry ice and stored at −80 °C until analysis . After photography , the body length ( total: rostrum to tail tip , fork: rostrum to tail fork , and standard: rostrum to beginning of the tail ) , the dorsoventral width of the body ( at the level of the operculum and at the beginning of the dorsal fin ) , the eye diameter , and the pupil diameter of 16 D/D fish and 15 L/D from each experimental condition were measured using ImageJ software . Five fish from each group , kept under the experimental conditions described above for 9 months beginning two hpf , were used for melanophore quantification . Fish were fixed for 1 hr in 4% paraformaldehyde , washed three times in PBS , and examined under a stereomicroscope . Pigment cells were counted as in Bilandžija et al . ( 2018 ) in four different places on the left side of each fish: in the proximal tail fin stripe , below the dorsal fin , along the dorsal flank of the trunk , and on the chin . Eyes of four SF kept in D/D and three L/D controls kept in experimental conditions for 1 . 5–2 . 5 years were embedded in paraplast blocks , sectioned , stained , and morphometric analyses was performed on retinal layers as described previously ( O'Quin et al . , 2013 ) . To account for intra-retinal variation caused by changes in location or sectioning , we measured retinas in the middle of the region between the optic nerve and the ora serrata on fifteen test fields randomly chosen across five sections for every fish . In order to compare these retinal measurements among individual fish for statistical analysis , we used eye diameters as independent parameters for normalization of our data ( Collery et al . , 2014 ) . Three surface fish kept in D/D and three in L/D experimental conditions for 7 months post-fertilization were used for RNA sequencing . Prior to RNA extraction , the stomach and liver were removed from the fish because they compromised RNA quality . Total RNA was isolated from samples stored in TRIzol reagent according to the manufacturer’s instructions ( Invitrogen , Carlsbad , CA , USA ) . The fragment size , concentration , RNA integrity number ( RIN ) and 28S/18S ratio of RNA extracts were determined using an Agilent 2100 Bioanalyzer ( Agilent , Santa Clara , CA , USA ) . The cDNA libraries were constructed using TruSeq mRNA Library Prep Kit ( Illumina , San Diego , CA , USA ) . Fragment sizes and concentrations of libraries were verified using an Agilent 2100 Bioanalyzer and ABI StepOnePlus Real-Time PCR machine ( Thermo Fisher Scientific , Waltham , MA , USA ) . The samples were sequenced in a single Hiseq X-ten lane with a strategy of 150 bp paired-ends ( PE ) , resulting in about 8 Gb raw data for each sample . To avoid potential adapter contamination , the last 60 bp of all reads were trimmed . Subsequently , the reads were removed if they contained: 1 ) more than 50% low quality ( Q < 35 ) bases; 2 ) more than 10 Ns , and 3 ) were PCR duplications . Cleaned read quality was confirmed in FASTQC ( Andrews , 2010 ) , and a total of 230 , 437 , 086 cleaned reads were retained for transcriptomic analysis . The reads have been submitted as bioproject PRJNA557727 ( accession numbers SRX6631237 - SRX6631242 ) . A local database was built on the Carbonate Cluster implemented by the National Center for Genome Analysis Support ( Stewart et al . , 2017 ) using the Astyanax mexicanus draft genome v102 . 93 ( GCA_000372685 . 1 ) and associated gene model annotation files . The index of the reference genome was built using Bowtie2 v2 . 3 . 2 ( Langmead and Salzberg , 2012 ) , and paired-end clean reads were aligned to the reference genome using TopHat v2 . 1 . 1 , with max-intron-length set to 1000 ( Kim et al . , 2013 ) . Transcript coverage and differential expression were determined following the Cufflinks pipeline ( Trapnell et al . , 2010 ) . Briefly , mapped reads for each sample were assembled using Cufflinks v2 . 1 . 1 with the parameters --frag-bias-correct , --multi-read-correct , --upper-quartile-norm , and --compatible-hits-norm . Transcripts were merged using default Cuffmerge parameters , and Cuffquant with --frag-bias-correct and --multi-read-correct was implemented to quantify transcript expression . Significant changes between transcripts were calculated using Cuffdiff ( Trapnell et al . , 2012 ) with parameters --compatible-hits-norm , --frag-bias-correct , and --multi-read-correct for light and dark treatments . Results were visualized using the package CummeRbund ( Goff et al . , 2014 ) implemented in R . Studio v1 . 0 . 136 ( R Studio T , 2015 ) using R v3 . 5 . 1 ( R Development Core Team , 2018 ) . Significantly differentially expressed genes ( padj < 0 . 1 ) were subjected to Gene Ontology ( GO ) and Kyoto Encyclopedia of Genes and Genomes ( KEGG ) pathway enrichment analysis as implemented on DAVID Bioinformatics Resources ( Huang et al . , 2009a; Huang et al . , 2009b ) and pathways were investigated more closely using Reactome ( Fabregat et al . , 2016 ) . Total RNA was isolated from 10 to 12 month old adult SF and PA placed in D/D or L/D conditions within the first few hours after fertilization . After the liver and stomach were removed , the RNA from the remaining body was isolated with Trizol Reagent ( Thermo Fisher Scientific , Waltham , MA , USA ) according to the manufacturer’s instructions and treated with DNase to remove genomic DNA if present . Equal amounts of RNA ( 1 . 5 µg ) were reverse transcribed using Superscript III or IV ( Invitrogen , NY , USA ) . The cDNA concentration was adjusted to 12 . 5 ng/µL and up to 50 ng were used in each reaction . We tested 12 different HKG ( housekeeping genes ) : ( ube2a , rpl13a , eef1a1l1 , nek7 , tbcb , rnf7 , rpl27 , ndufa6 , ap5s1 , mob4 , lsm12a , actc1b ) using RefFinder , which is located on the Cotton EST Database webpage ( Xie et al . , 2011 ) , and integrates currently available major computational programs ( geNorm , Normfinder , BestKeeper , and the comparative ΔΔCt method ) to compare and rank the candidate reference genes . The geometric mean of the three most stable HKG ( tbcb , mob4 and rnf7 ) was used for normalization . Primers for genes of interest were designed across exon/intron boundaries except for pdx1 , where this was not possible ( Supplementary File 1 ) . Primer efficiencies were calculated in Excel from standard curves , and those with efficiencies between 90% and 110% were used for quantifications . The resulting PCR products were sequenced to confirm the targeted genes . Relative expression was calculated according to a formula ΔCt = E ˆ ( Ct reference – Ct target ) ( E = 1+primer efficiency/100 ) , which enables the comparison of normalized expression of a gene between multiple samples , instead of ratios between two samples , taking primer efficiency into account . Statistical significance of observed expression differences was calculated by ANOVA and adjusted by Bonferroni index . SF and PA embryos were placed in L/D vs . D/D experimental conditions within 24 hpf . At 7 dpf , the first day larval fish were fed brine shrimp , a group of 36 larvae from L/D and 36 from D/D per fish type was fed . Groups of 36 larvae from each experimental condition and each fish type were not fed . The fish were identically treated in all other aspects , and the number of living fish was recorded daily . Dead fish were removed each day . The experiment was repeated 3 times independently . Statistical analyses for survival analysis were carried out using R version 3 . 5 . 3 ( R Development Core Team , 2018 ) . Differences in starvation resistance were compared with a Cox proportional hazards model using the function coxph ( ) from the R package ‘survival’ ( Therneau and Grambsch , 2000 ) . For the parental generation , the model included the covariates fish ( PA vs . SF ) , light status ( D/D vs . L/D ) , and fish x light status . For the G1 generation , the model included the covariates fish ( dSF vs . SF ) , light status ( D/D vs . L/D ) and fish x light status . If the null hypothesis that all β = 0 was rejected , post-hoc pairwise comparisons of contrasts were carried out using the function glht ( ) from the package multcomp with the specification mcp = ‘Tukey’ ( Hothorn et al . , 2008 ) . Proportionality assumptions for Cox proportional hazards models were tested using the coxphz function ( Therneau and Grambsch , 2000 ) . Unhatched SF and PA embryos or 7 dpf larvae were placed individually in air-tight glass vials filled to the top with fish system water that contained no air bubbles . Half of the vials from each type were kept in L/D and the other half were wrapped in aluminum foil and kept in the same way in D/D . After two days , the O2 remaining in each vial was measured using a Membrane Inlet Mass Spectrometer ( MIMS ) Machine ( Bay Instruments , Easton , MD , USA ) . The amount of O2 consumed per fish was calculated by subtracting the measured amount of O2 in the vials with fish from the mean amount of O2 measured in blank vials without fish . This experiment was repeated twice with embryos and once with larvae . The same methods were used in all replications . We quantified cortisol from adult SF and PA that were maintained in D/D or L/D experimental conditions for 1 . 5 to 2 years beginning before 3 dpf . Cortisol was extracted using previously developed procedures ( Canavello et al . , 2011; Gallo and Jeffery , 2012 ) . Briefly , tissue samples stored at −80 °C were thawed and homogenized in PBS after which 5 mL of diethyl ether was added . Following centrifugation for 5 min at 3500 x g , the cortisol-containing top layer was removed and evaporated in a fume hood overnight . The cortisol was reconstituted in PBS overnight at 4 °C and quantified using the Cortisol ELISA Kit ( Item № 500360 , Cayman Chemical , Ann Arbor MI , USA ) following the kit protocol . Cortisol samples from fish within each group were pooled and assayed in triplicate . The recorded absorbance was compared to a standard curve to determine the cortisol concentration , and each sample was standardized to protein concentration determined by Pierce BCA Protein Assay Kit ( Thermo Fisher Scientific , Waltham , MA , USA ) . The experiment was repeated 3 times on independent batches of fish . Pulverized fish were weighed and 20 µl per mg of 6% NP-40 ( Abcam , Cambridge , UK ) was added to each sample . Samples were heated to 80–100°C for 2–5 min , and the process was repeated 2–3 times . Insoluble material was removed by centrifugation for 2 min at 13 , 000 rpm , and equal volumes of supernatants from all samples of the same fish type were pooled and diluted 20x in ddH2O before ELISA . For ELISA , 50 µL of each sample was assayed in triplicate using the Triglyceride Quantification Assay Kit ( ab65336 ) , ( Abcam , Cambridge , UK ) according to the manufacturer's instructions . The procedure followed a previously published method ( Aspiras et al . , 2015 ) . Concentrations were determined from the standard curve and standardized to protein concentration as described above . The experiment was repeated on 3 independent batches of fish . Levels of growth hormone ( CSB-E12121Fh ) and thyroid stimulating hormone ( CSB-EQ02726Fl ) were quantified by ELISA using commercially available kits from Cusabio ( Wuhan , China ) . Adult SF kept in D/D for 1 . 5 years and PA kept in the experiment for almost 2 years were used . For comparison of hormone levels between different populations , 3 to 4 month old SF , PA , TI , and MO were used . The fish tissue was weighed , and 3 µl PBS/mg tissue was added to each sample . After two freeze-thaw cycles and homogenization , the homogenate was centrifuged at 5000 x g for 5 min at 4 °C . Equal volumes of supernatant from each fish type were pooled and immediately loaded on the plate , 50 µl per well , in triplicate . Assays were performed following the manufacturer’s instructions and concentrations were determined according to the standard curve . Results were standardized to protein concentration as described above . Each experiment was repeated 3 times independently . Brains were dissected from PA and SF kept in L/D or D/D experimental conditions for 1 . 5 to 2 years beginning at or before 3 dpf , and from adult SF ( N = 8 ) , PA ( N = 5 ) , TI ( N = 5 ) , and MO ( N = 6 ) kept in the main fish system and HPLC was used to quantify neurotransmitter levels as in Bilandžija et al . ( 2018 ) . In addition , we quantified serotonin ( 5-HT ) in samples of pooled five larvae ( 7 dpf ) placed in the L/D or D/D experimental conditions within first few hours after fertilization , N = 6 ( SF L/D ) , 5 ( SF D/D ) , 5 ( PA L/D ) , and 4 ( PA D/D ) . Serotonin and its metabolite , 5-hydroxyindoleacetic acid ( 5-HIAA ) , were analyzed using HPLC with electrochemical detection as described previously with minor modifications ( Renner and Luine , 1986 ) . Brains were placed into 100 µL of sodium acetate buffer ( pH 5 . 0 ) containing the internal standard alpha-methyl dopamine ( αMDA; Merck and Co . , Inc , Kenilworth , NJ ) . Fish bodies were placed in either 200 to 400 µL of acetate buffer containing αMDA based on the amount of tissue present . Both tissue types were disrupted by sonication using a 4710 Ultrasonic Homogenizer ( Cole-Parmer Instrument Co . , Chicago IL ) and stored at −80°C . Prior to analysis , the sonicated brain samples were thawed , 4 μL of 1 mg/mL ascorbate oxidase ( Sigma-Aldrich , St . Louis , MO , USA ) was added to each sample and the samples were centrifuged at 17 , 000 g for 15 min . Fish body samples were treated the same way except that the supernatant was centrifuged a second time through a 0 . 2 µm filter . The filtered supernatant was removed and a Waters Alliance e2695 separation module was used to inject 50 µL of the supernatant onto a C184 µm NOVA-PAK radial compression column ( Waters Associates , Inc Milford , MA ) held at 30°C . The initial mobile phase ( pH 4 . 1 ) was prepared using 8 . 6 g sodium acetate , 250 mg EDTA , 14 g citric acid , 80 mg octylsulfonic acid , and 80 mL methanol in 1 L of distilled water ( monoamine standards and chemicals were purchased through Sigma-Aldrich ) and adjusted with small additions of octylsulfonic acid , glacial acetic acid , and methanol to optimize the separation . Electrochemical detection was accomplished using an LC four potentiostat and glassy carbon electrode ( Bioanalytical Systems , West Lafayette , IN , USA ) set at a sensitivity of 0 . 5 nA/V ( brain samples ) or 1 nA/V ( body samples ) with an applied potential of +0 . 7 V versus an Ag/AgCl reference electrode . The pellet was solubilized in 400 µL of 0 . 4 N NaOH and protein content was analyzed using the Bradford method ( Bradford , 1976 ) . A CSW32 data program ( DataApex Ltd . , Czech Republic ) was used to determine 5-HT and 5-HIAA concentrations in the internal standard mode using peak heights calculated from standards . Injection versus preparation volumes were corrected and amine concentrations were normalized by dividing pg amine by µg protein . Data were tested for differences using a Three Way Analysis of Variance ( SigmaStat version 3 . 5 , Systat Software Inc , San Jose , CA ) . In analyses that revealed a significant effect between groups , the Holm-Sidak method was used to conduct pairwise comparisons . All data were tested for the presence of outliers using the Grubb’s test ( Rohlf and Soka , 1981 ) . Based on this analysis two brain ( PA D/D night , SF L/D day ) and three body samples ( PA D/D night , PA L/D day , SF D/D day ) were deleted from the data set . In the end a total number of fish used in the analysis for brains was 9 ( SF L/D day ) , 10 ( SF D/D day ) , 10 ( PA L/D day ) , 10 ( PA D/D day ) , 7 ( SF L/D night ) , 6 ( SF D/D night ) , 4 ( PA L/D night ) , 4 ( PA D/D night ) and for the bodies 10 ( SF L/D day ) , 9 ( SF D/D day ) , 8 ( PA L/D day ) , 10 ( PA D/D day ) , 7 ( SF L/D night ) , 7 ( SF D/D night ) , 4 ( PA L/D night ) , 5 ( PA D/D night ) . We measured metabolic rate and starvation resistance in the G1 offspring of dark raised ( dSF ) and control SF using the methods described above . Statistical analysis was also done using the methods described above with N = 24 larvae per group in starvation resistance experiments and N = 10 ( L/D SF ) , 13 ( D/D SF ) , 12 ( L/D dSF ) and 15 ( D/D dSF ) in metabolic rate experiments .	The Mexican tetra is a fish that has two forms: a surface-dwelling form , which has eyes and silvery grey appearance , and a cave-dwelling form , which is blind and has lost its pigmentation . Recent studies have shown that the cave-dwelling form evolved rapidly within the last 200 , 000 years from an ancestor that lived at the surface . The recent evolution of the cave-dwelling form of the tetra poses an interesting evolutionary question: how did the surface-dwelling ancestor of the tetra quickly adapt to the new and challenging environment found in the caves ? ‘Phenotypic plasticity’ is a phenomenon through which a single set of genes can produce different observable traits depending on the environment . An example of phenotypic plasticity occurs in response to diet: in animals , poor diets can lead to an increase in the size of the digestive organs and to the animals eating more . To see if surface-dwelling tetras can quickly adapt to cave environments through phenotypic plasticity , Bilandžija et al . have exposed these fish to complete darkness ( the major feature of the cave environment ) for two years . After spending up to two years in the dark , these fish were compared to normal surface-dwelling and cave-dwelling tetras . Results revealed that surface-dwelling tetras raised in the dark exhibited traits associated with cave-dwelling tetras . These traits included changes in the activity of many genes involved in diverse processes , resistance to starvation , metabolism , and levels of hormones and molecules involved in neural signaling , which could lead to changes in behavior . However , the fish also exhibited traits , including an increase in the cells responsible for pigmentation , that would have no obvious benefit in the darkness . Even though the changes observed require no genetic mutations , they can help or hinder the fish’s survival once they occur , possibly determining subsequent evolution . Thus , a trait beneficial for surviving in the dark that appears simply through phenotypic plasticity may eventually be selected for and genetic mutations that encode it more reliably may appear too . These results shed light on how species may quickly adapt to new environments without accumulating genetic mutations , which can take hundreds of thousands of years . They also may help to explain how colonizer species succeed in challenging environments . The principles described by Bilandžija et al . can be applied to different organisms adapting to new environments , and may help understand the role of phenotypic plasticity in evolution .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2020	Phenotypic plasticity as a mechanism of cave colonization and adaptation
The de novo emergence of new genes has been well documented through genomic analyses . However , a functional analysis , especially of very young protein-coding genes , is still largely lacking . Here , we identify a set of house mouse-specific protein-coding genes and assess their translation by ribosome profiling and mass spectrometry data . We functionally analyze one of them , Gm13030 , which is specifically expressed in females in the oviduct . The interruption of the reading frame affects the transcriptional network in the oviducts at a specific stage of the estrous cycle . This includes the upregulation of Dcpp genes , which are known to stimulate the growth of preimplantation embryos . As a consequence , knockout females have their second litters after shorter times and have a higher infanticide rate . Given that Gm13030 shows no signs of positive selection , our findings support the hypothesis that a de novo evolved gene can directly adopt a function without much sequence adaptation . The evolution of new genes through duplication-divergence processes is well understood ( Chen et al . , 2013; Kaessmann , 2010; Long et al . , 2013; Tautz and Domazet-Lošo , 2011 ) . But the evolution of new genes from non-coding DNA has been little considered for a long time ( Tautz , 2014 ) . However , with the increasing availability of comparative genome data from closely related species , more and more cases of unequivocal de novo gene emergence have been described ( McLysaght and Hurst , 2016; Schlötterer , 2015; Tautz , 2014; Tautz and Domazet-Lošo , 2011 ) . These analyses have shown that de novo gene emergence is a very active process in all evolutionary lineages analyzed . A comparative analysis of closely related mouse species has even suggested that virtually the whole genome is ‘scanned’ by transcript emergence and loss within about 10 million years of evolutionary history ( Neme and Tautz , 2016 ) . But unlike the detection of the transcriptional and translational expression of de novo genes , functional studies of such genes have lacked behind . In yeast , the de novo evolved gene BSC4 was found to be involved in DNA repair ( Cai et al . , 2008 ) and MDF1 ( Li et al . , 2010; Li et al . , 2014 ) was found to suppress mating and to promote fermentation . Knockdown of candidates of de novo genes in Drosophila have suggested effects on viability and male fertility ( Chen et al . , 2010; Reinhardt et al . , 2013 ) . Male fertility was also found to be affected for Pldi in mice , which codes for a lncRNA . In this case the knockout was shown to affect sperm motility and testis weight ( Heinen et al . , 2009 ) . There is generally a tendency to focus on male testis effects for newly evolved genes . However , considering that the mammalian females have complex reproduction cycles , including morphology , physiology and behavior relating to mate choice , pregnancy , and parenting , de novo genes in mammals should also be expected to have a function in female-specific organs and affect female fertility and reproductive behavior as well . Here , we have first generated a list of candidate genes that have evolved in the lineage of mice , after they split from rats . We have analyzed ribosome profiling and mass spectrometry data for these and find that most of them are translated . From this list , we have then chosen a gene specifically expressed in the female reproductive system to address the question of the role of de novo gene evolution in this as yet little studied context . We used a knockout line for the reading frame of the gene , created through CRISPR/Cas9-mediated frameshift mutagenesis , and subjected it to extensive molecular and phenotypic analysis . We conclude that it functions in the oviduct and affects female fertility cycles and that its emergence may have been driven by an evolutionary conflict situation . Given that we find no measurable acceleration of sequence evolution in the gene , we conclude that it became directly functional after its open reading frame became functional . These results support the notion that random protein sequences have a good probability for conveying evolutionarily relevant functions ( Neme et al . , 2017 ) . To identify candidates for recently evolved de novo genes , we have applied a combined phylostratigraphy and synteny-based approach . We were able to identify 119 predicted protein-coding genes from intergenic regions that occur only in the mouse genome , but not in rats or humans . We re-assembled their transcript structures and estimated their expression levels using available ENCODE RNA-Seq data in 35 tissues from the mouse ( Figure 1 , Figure 1—source data 1 ) . To validate that their predicted open reading frames ( ORFs ) are indeed translated , we have searched ribosome profiling and peptide mass spectrometry datasets ( Figure 1—source data 1 ) . We found for 110 out of the 119 candidate genes direct evidence for translation . Expression of these genes is found throughout all tissues analyzed , with notable differences . Testis and brain express the relatively largest abundance of these candidate de novo genes , while the digestive system and liver express the lowest ( Figure 1A ) . Expression levels of these genes are generally lower than those of other protein-coding genes ( FPKM medians: 0 . 63 vs . 8 . 18; two-tailed Wilcoxon rank sum test , p-value<2 . 2 × 10−16; Figure 1B ) . Most overall molecular patterns are similar to previous findings ( Neme and Tautz , 2013; Schmitz et al . , 2018; Wilson et al . , 2017 ) . They have fewer exons ( medians: 2 vs . 7; two-tailed Wilcoxon rank sum test , p-value<2 . 2 × 10−16 ) and fewer coding exons than other protein-coding genes ( medians: 1 vs . 6; two-tailed Wilcoxon rank sum test , p-value<2 . 2 × 10−16 ) . The lengths of their proteins are shorter than those of other proteins ( medians: 125 vs . 397; two-tailed Wilcoxon rank sum test , p-value<2 . 2 × 10−16 ) . However , their proteins are predicted to be less disordered than other proteins ( medians: 0 . 20 vs . 0 . 27; two-tailed Wilcoxon rank sum test , p-value=0 . 0024; Figure 1C ) and equally hydrophobic to other proteins ( medians: 0 . 56 vs . 0 . 57; two-tailed Wilcoxon rank sum test , p-value=0 . 52; Figure 1D ) , but note that the two sets of values show a broad distribution . To study the function of a gene expressed in the female reproductive tract , we picked Gm13030 ( Figure 2 ) from the above list for in-depth analyses , including evolutionary history , reading-frame knockout , transcriptomic studies and phenotyping . According to the ENCODE RNA-Seq data , Gm13030 is only expressed in two tissues , the ovary of 8 weeks old females ( FPKM 0 . 135 ) , as well as the subcutaneous adipose tissue of 8 weeks old animals ( FPKM 0 . 115 ) ( Figure 1—source data 1 ) . Given that the ovary is a small organ , with closely attached tissues , such as oviduct and gonadal fat pad , there could be contamination between these different tissue types . Hence , we were interested whether there is specificity for one of them . We used reverse transcription PCR on RNA from the respective carefully prepared tissue samples , to trace the expression of Gm13030 and a control gene ( Uba1 ) . We found that Gm13030 is not expressed in the ovary , but predominantly in the oviduct with only a weak signal from the adjacent fat pad ( Figure 2B ) . To trace the evolutionary emergence of Gm13030 , we used available whole genome information of different mouse species to generate alignments , combined with Sanger sequencing data of PCR fragments from mouse populations , subspecies , and related species from the genus Mus . We found the respective genomic region covering the ORF in all mouse species analyzed . It is not possible to identify an unequivocal orthologous region in the rat , because the unique genomic region in the mouse matches with multiple diverged genomic fragments in the rat reference genome , and all these fragments overlap only marginally with the mouse region . The alignments for the whole coding region allowed us to infer mutations that have led to the opening of the reading frame ( enabler mutations ) , as well as further substitutions and secondary disablers along the tree topology ( Figure 3 , Figure 3—figure supplement 1 ) . The most distant species in which we can trace the orthologous genomic region , M . pahari , lacks part of the coding region . Two further outgoup species , M . matheyi and M . caroli have an orthologous genomic region that spans the whole reading frame , but harbor stop codons at position 204 and 258 of the alignment ( Figure 3 , Figure 3—figure supplement 1 ) . At position 258 we find a change from TGA to TGC in all ingroup species , that is this is a clear enabler mutation . The same change is seen at position 204 , but some of the ingroup species that show also secondary disablers ( see below ) retain the TGA . But since both enabler mutations are at least seen in M . spicilegus , we place the emergence of the Gm13030 ORF at this node , that is between 2–4 million years ago . Figure 3 includes all coding and non-coding substitutions that have occurred beyond this node . This includes secondary disablers in M . spretus , as well as M . m . domesticus . Most notably , all three M . m . domesticus populations carry a 17nt deletion that leads to a disruption of the reading frame . They share also several other substitutions , not only among them , but also with M . spretus and M . spicilegus , suggesting a secondary introgression effect ( Figure 3 , Figure 3—figure supplement 1 ) . Hence , after the emergence of the Gm13030 ORF , only the M . m . musculus and M . m . castaneus populations have retained it . When focusing on the substitutions that occurred within the lineage towards M . m . musculus , we find a total of 7 coding and six non-coding substitutions . Hence , the total number of substitutions is slightly higher than the 6–7 expected for approximately neutral substitutions from a genomic average between these populations ( indicated on top of Figure 3 ) , but there is no bias towards coding mutations . Overall , there are too few mutations to apply a dN/dS test and the ratios of non-coding to coding mutations are all non-significant ( Figure 3—figure supplement 3 ) . Hence , we conclude that there is no traceable signal of positive selection on the protein after the emergence of the ORF . For the further functional characterization of Gm13030 , we obtained a knockout line with a frameshift in the ORF through CRISPR/Cas9 mutagenesis . The knockout line is from a laboratory strain that is nominally derived from Mus musculus domesticus ( C57BL/6N ) . However , as stated above , Mus musculus domesticus populations have disabling mutations . But C57BL/6N is known to carry also alleles from Mus musculus musculus ( Yang et al . , 2011 ) and the Gm13030 allele represents indeed the non-interrupted version that is found in M . m . musculus and M . m . castaneus . The CRISPR/Cas9 treatment introduced a 7 bp deletion at the beginning of the ORF ( position 41–47 ) causing a frameshift and a premature stop codon in exon 2 ( Figure 2C ) . The CRISPR/Cas9 experiment to generate our knockout line might have generated potential off-target mutations . In order to rule out this possibility , we performed whole genome sequencing on both animals of our founding pair . The female and male of our founding pair were selected from the first-generation offspring of the mating among mosaic and wildtype mice which were directly developed from the zygotes injected . Each of them contained the 7 bp deletion allele described above and a wildtype allele . If there were any off-target sites , they should exist as heterozygous or homozygous indels or single nucleotide variants . However , in our genome sequencing results , we found no variant located in the 100 bp regions around the genome-wide 343 predicted off-target sites . Further , we manually checked the reads mapped to the regions around the top 20 predicted sites in both samples and none of them yielded an indication of variants . The Gm13030 knockout line is homozygous viable and fertile . We were therefore interested in studying the impact on the transcriptional network in the tissue in which Gm13030 is predominantly expressed . Given the observation that Gm13030 is specifically expressed in adult oviducts , we focused the RNA-Seq analysis on the oviducts of 12 homozygous knockout and 12 wildtype females ( 10–11 weeks old ) . There were on average 75 . 9 million unique mapped reads per sample ( range from 57 . 5 to 93 . 0 million reads; Figure 4—figure supplement 1 ) . The genotypes of the 24 samples were further confirmed by the reads covering the sites in which the 7-bps deletion locates ( Figure 4—figure supplement 1 ) . In the initial analysis involving all samples , we found no differentially expressed gene between knockouts and wildtypes . However , given that the expression in oviducts should be fluctuating according to estrous cycle , we clustered the transcriptomes of the individuals based on both principle component analysis ( PCA ) and hierarchical clustering methods , which allowed to distinguish three major clusters ( Figure 4A and B ) . To confirm that these correspond to three different phases of the estrous cycle , we analyzed the expression of three known cycle dependent genes in the respective clusters , progesterone receptor ( Pgr ) and estrogen receptors ( Esr1 and Gper1 ) . We found that these genes change indeed in the expected directions , both in the wildtype as well as the knockout animals ( Figure 4C–E ) . Based on this finding , we performed the differential expression analysis on the three clusters separately . We found 21 differentially expressed genes in cluster 1 ( DESeq2 , adjusted p-value≤0 . 01; fold changes range from 0 . 75 to 1 . 59; Table 1 ) , but still none for clusters 2 and 3 . The 21 differentially expressed genes in cluster 1 do not include the genes neighboring Gm13030 ( Pla2g2e and Pla2g5 ) . This suggests that Gm13030 acts during the phase of high progesterone receptor and estrogen receptor 1 expression , and low G protein-coupled estrogen receptor 1 expression , corresponding to proestrus or the starting of estrus , that is , the phase where females start to become receptive for implantation . The top three differentially expressed genes belong all to a single young gene family , namely Dcpp1 , Dcpp2 and Dcpp3 , all three of which were significantly up-regulated in the knockout samples ( DESeq2 , fold changes: 1 . 45 for Dcpp1 , 1 . 47 for Dcpp2 , and 1 . 59 for Dcpp3 , Figure 4—figure supplement 2 ) . These genes are expressed in female and male reproductive organs and the thymus , and were previously found to function in oviducts to stimulate pre-implantation embryo development ( Lee et al . , 2006 ) . Given the special importance of the expression differences for the Dcpp genes , we confirmed them by a quantitative PCR assay ( Figure 4—figure supplement 2 ) . The fourth gene in the list of significantly changed expression is Rxfp1 , the receptor for the pregnancy hormone relaxin . Relaxin signaling is involved in a variety of cellular processes ( Valkovic et al . , 2019 ) , whereby the regulation of the reproductive cycle is one of them ( Anand-Ivell and Ivell , 2014 ) . Given that the Dcpp genes are more highly expressed in the knockouts , one could predict a higher implantation frequency of embryos , as it has been shown through experimental manipulation of Dcpp levels ( Lee et al . , 2006 ) . We assessed the litters of pairs that were produced from our normal breeding stocks and found that the first litters from homozygous knockout females were produced after the same time as those from wildtype or heterozygous females ( medians: 23 vs . 22 days , Figure 5—source data 1 ) . However , we saw a major difference with respect to the second litter . Homozygous knockout females tended to produce this faster than wildtype or heterozygous females ( medians: 23 vs . 38 days , Figure 5—source data 1 ) . To test this observation directly , we set up additional 10 mating pairs of homozygous knockout females with wildtype males and 10 wildtype pairs for control , all at approximately the same age at the start ( 8–9 weeks old ) . We found that the knockout and wildtype pairs had their first litter after the same time ( medians: 23 vs . 22 days , Figure 5—source data 1 ) , while the knockout females had their second litter after a shorter time ( medians: 24 vs . 36 days , Figure 5—source data 1 ) , thus confirming the initial observation . However , the data have to be seen in the context of the non-continuous nature of pregnancy , caused by the ovulation cycles of females . Females can ovulate within a day of giving birth , but if no successful mating occurs at that time , ovulation is suppressed while the female is lactating . This results in a delay in the timing of the next pregnancy . Figure 5 shows that this pattern is also evident in our data . We found that the times to the second litter were either smaller than or equal to 25 days ( early group ) or larger than or equal to 35 days ( late group ) for both the homozygous knockout females and the wildtype or heterozygous females . But in the homozygous knockouts , we saw more in the early group , leading to the median values having a big difference . When using the two-tailed Wilcoxon rank sum test which does not require the assumption of a normal distribution , we found that this difference is significant when calculated across all breeding data ( p-value=0 . 042 ) . Interestingly , we found not only a timing difference for the second litter but also infanticide in about a quarter of the litters ( 4 out of 16 ) from homozygous females , but none in wildtype or heterozygous females ( two-tailed Fisher's exact test , p-value=0 . 031 , Figure 5—source data 1 ) . This could indicate that when the second litter follows too quickly , the females may be under stronger postpartum stress resulting in partial killing of pups . In addition , one could also have expected to see homozygous knockout females having larger litter sizes than those of wildtype or heterozygous females , but they were almost the same ( medians: 6 . 5 vs . 7 . 0 for littler 1 and 6 . 5 vs . 7 . 5 for litter 2 , Figure 5—source data 1 ) . One possible explanation is that considering the high infanticide rate for homozygous knockouts , more pups from homozygous knockout females were eaten before being observed . These results suggest that the loss of Gm13030 should be detrimental to the animals in the wild . Still , we see that the M . m . domesticus populations have secondarily lost this gene ( Figure 3 ) . Intriguingly , when inspecting the copy number variation data that we have produced previously ( Pezer et al . , 2015 ) , we found that Dcpp3 was also lost in M . m . domesticus populations ( Figure 4—figure supplement 3 ) . Under the assumption that this results in an overall lowered expression of Dcpp RNAs , it could be considered to compensate for the loss of Gm13030 . The knockout line did not show an overt phenotype , but we considered this also as a priori unlikely , given that a de novo evolved gene is expected to be only added to an existing network of genes ( Zhang et al . , 2015 ) . But given the observed transcriptome changes in the oviducts , we were encouraged to apply the fertility test . We identified a possible direct link between the identified phenotype of a shorter interval to second birth in the knockouts and the transcriptomic changes . We found that the expression level of all three copies of Dcpp genes in C57BL/6N mice is enhanced in the Shj knockout animals . Dcpp expression is induced in the oviduct by pre-implantation embryos and is then secreted into the oviduct . This in turn stimulates the further maturation of the embryos and eventually the implantation ( Lee et al . , 2006 ) . Hence , this is a system where a selfish tendency for Dcpp expression favoring embryo implantation could develop , in expense of the interest of the mother that wants to build up new resources first . Accordingly , Shj could have found its function in controlling this expression , that is , ‘defending’ the interests of the mother . Intriguingly , the secondary loss of Shj in M . m . domesticus populations is accompanied by a loss of Dccp3 in the same populations . This is compatible with the notion that an evolutionary conflict of interest exists for these interactions , whereby it remains open whether the loss of Dcpp3 preceded the loss of Shj or vice versa . We note that Shj inactivation alleles segregate also in the populations of the other subspecies ( M . m . musculus and M . m . castaneus ) in low frequency , implying that the evolutionary process of fully integrating this new gene is still ongoing . There has so far been much focus on de novo genes and other new genes to have male-biased expression and to affect male fertility ( Chen et al . , 2013; Ellegren and Parsch , 2007; Heinen et al . , 2009; Kaessmann , 2010; Long et al . , 2013; Reinhardt et al . , 2013; Zhao et al . , 2014 ) . Only recently , one of a pair of duplicated genes in Drosophila , Arts , has been shown to have high expression in the ovary and to affect fertility ( VanKuren and Long , 2018 ) . Here we have shown that a de novo gene specifically expressed in the female reproductive tract affects the female fertility cycle . Female reproduction should be subject to accelerated evolution patterns , especially in mammals which have high complexities in female reproduction , including mate choice , pregnancy , and parenting , which has been neglected so far . One reason is that the estrous cycle in females adds to the complexity of the analysis . Our clustering analysis of the transcriptomic data , which considers the stages of estrous cycle , provides an approach for studying biased gene expression in female mammals as well . Another reason for the current focus on males is the large number of new genes that are transcribed in testis . However , this is due to the promiscuous phase of expression in meiotic cells , where many genes use alternative promotors ( Kleene , 2001 ) . These meiotic cells are abundant in testis , but are difficult to analyze in ovaries . Hence , it is still open whether there might be a similar phase of over-expression of new genes in female meiotic stages as well . Shj exerts its effects in somatic cells , that is , independent of a possible expression in meiosis , but in the context of a possible selfish gene conflict situation , which has so far been ascribed mostly to the male reproductive system ( Kleene , 2005 ) . Hence , we expect that a better analysis of female-specific expression of genes should reveal more evolutionary interesting insights in the future . It has long been assumed that the emergence of function out of non-coding DNA regions must be rare , and if it occurs , the resulting genes would be far away from assuming a function . Our results do not support these assumptions . It is possible to find many well supported transcripts that could be considered to be true de novo genes . We have shown here that Shj has functions on the transcriptome and the phenotype . In fact , we have initial data for two additional de novo genes expressed in the brain and limbs , where knockouts produce an effect on the transcriptome and show subtle phenotypes ( data available on bioRxiv doi . org/10 . 1101/510214 ) . However , since lacZ replacement constructs were used instead of CRISPR-induced knockouts , it remains still open whether the effects are due to the new ORFs or to chromatin effects caused by the deletion constructs . This will need further analysis . The Shj ORF has acquired only a small number of additional substitutions , both coding and non-coding after it emerged . This suggests that it did not need additional adaptation of the protein sequence to become functional . This is in line with a similar analysis on a larger set of de novo ORFs in the mouse ( Ruiz-Orera et al . , 2018 ) . Hence , this raises the question whether we should necessarily expect signatures of positive selection around de novo genes as part of proof that it is a true gene ( McLysaght and Hurst , 2016 ) . Alternatively , given the observation that a large set of expressed random sequences can exert phenotypes ( Bao et al . , 2017; Neme et al . , 2017 ) , it would seem more likely that the conversion of a non-coding region into a coding one would already be sufficient to create a gene function . In the early phase of evolution , such genes would likely be frequently subject to secondary loss ( Palmieri et al . , 2014 ) , but they could eventually also become fixed and then further evolutionarily optimized . The mouse studies were approved by the supervising authority ( Ministerium für Energiewende , Landwirtschaftliche Räume und Umwelt , Kiel ) under the registration numbers V244-71173/2015 , V244-4415/2017 and V244-47238/17 . Animals were kept according to FELASA ( Federation of European Laboratory Animal Science Association ) guidelines , with the permit from the Veterinäramt Kreis Plön: 1401–144/PLÖ−004697 . The respective animal welfare officer at the University of Kiel was informed about the sacrifice of the animals for this study . We modified previous phylostratigraphy and synteny-based methods to identify Mus-specific de novo protein-coding genes from intergenic regions . Note that while the phylostratigraphy based approach was criticized to potentially include false positives ( Moyers and Zhang , 2015 ) , we have shown that the problem is relatively small and that it is in particularly not relevant for the most recently diverged lineages within which de novo gene evolution is traced ( Domazet-Lošo et al . , 2017 ) . We started with mouse proteins annotated in Ensembl ( Version 80 ) ( Zerbino et al . , 2018 ) ( 1 ) with protein length not smaller than 30 amino acids , ( 2 ) with a start codon at the beginning of the ORF , ( 3 ) with a stop codon at the end of the ORF , ( 4 ) without stop codons within the annotated ORF . For the phylostratigraphy-based strategy , in order to save computational time , we first used NCBI BLASTP ( 2 . 5 . 0+ ) to align low complexity region masked mouse protein sequences to rat protein sequences annotated in Ensembl ( Version 80 ) and filtered out the mouse sequences having hits with E-values smaller than 1 × 10−7 . Next we used NCBI BLASTP ( 2 . 5 . 0+ ) to align the remaining low complexity region masked sequences to NCBI nr protein sequences ( 10 Nov . 2016 ) ( O'Leary et al . , 2016 ) and filtered out the mouse sequences having non-genus Mus hits with E-values smaller than 1 × 10−3 according to Neme and Tautz ( 2013 ) . The genes remaining after these filtering steps are the candidates for the de novo evolved genes . In order to deal also with proteins having low complexity regions , we further applied a synteny-based strategy on the rest proteins by taking advantage of the Chain annotation from Comparative Genomics of UCSC Genome Browser ( http://genome . ucsc . edu/ ) ( Kent et al . , 2002 ) . We filtered out the proteins encoded on unassembled scaffolds because their chromosome information is not compatible between Ensembl and UCSC annotations . We only compared rat and human proteins with mouse proteins because their genomes are well assembled and genes are well annotated . We performed the same procedures on rat and human data separately , and used ‘mm10 . rn5 . all . chain’ and ‘rn5ToRn6 . over . chain’ from UCSC and gene annotation from Ensembl ( Version 80 ) for rat , and ‘mm10 . hg38 . all . chain’ from UCSC and gene annotation from Ensembl ( Version 80 ) for human . For each mouse gene , if its ORF overlaps with any ORFs in the rat or human mapping regions in Chain annotation , we aligned its protein sequence to those protein sequences with program water from EMBOSS ( 6 . 5 . 7 . 0 ) ( Rice et al . , 2000 ) ; if one of the alignment scores is not smaller than 40 , we filtered out the protein . The remaining 119 genes are the candidates for the following analysis and the pool for us to select the gene for further functional experiments . We downloaded the raw read files of 135 strand-specific paired-end RNA-Seq samples generated by the lab of Thomas Gingeras , CSHL from ENCODE ( ENCODE Project Consortium , 2012; Sloan et al . , 2016 ) including 35 tissues from different organs and different developmental stages , and each of them had multiple biological or technical replicates . We trimmed the raw reads with Trimmomatic ( 0 . 35 ) ( Bolger et al . , 2014 ) , and only used paired-end reads left for the following analyses . We mapped the trimmed reads to the mouse genome GRCm38 ( Waterston et al . , 2002; Zerbino et al . , 2018 ) with HISAT2 ( 2 . 0 . 4 ) ( Kim et al . , 2015 ) and SAMtools ( 1 . 3 . 1 ) ( Li et al . , 2009 ) , and took advantage of the mouse gene annotation in Ensembl ( Version 80 ) by using the --ss and --exon options of hisat2-build . We assembled transcripts in each sample , and merged annotated transcripts in Ensembl ( Version 80 ) and all assembled transcripts with StringTie ( 1 . 3 . 4d ) ( Pertea et al . , 2015 ) . Then we estimated the abundances of transcripts , FPKM values , in each sample with StringTie ( 1 . 3 . 4d ) . For each tissue , we summarized the FPKM values of each transcript by averaging the values from multiple biological or technical replicates; and if a gene has multiple transcripts , we assigned the summary of the FPKM values of the transcripts as the transcriptional abundance of the gene . We downloaded the datasets that included both strand-specific ribosome profiling ( Ribo-Seq ) and RNA-Seq experiments of the same mouse samples from Gene Expression Omnibus ( Barrett et al . , 2013 ) under accession numbers GSE51424 ( Gonzalez et al . , 2014 ) , GSE72064 ( Cho et al . , 2015 ) , GSE41426 ( Djiane et al . , 2013 ) , GSE22001 ( Guo et al . , 2010 ) , GSE62134 ( Diaz-Muñoz et al . , 2015 ) , and GSE50983 ( Castañeda et al . , 2014 ) , which corresponded to brain , hippocampus , neural ES cells , heart , skeletal muscle , neutrophils , splenic B cells , and testis . Ribo-seq datasets were depleted of possible rRNA contaminants by discarding reads mapped to annotated rRNAs , and then the rest reads were mapped to GRCm38 ( Waterston et al . , 2002; Zerbino et al . , 2018 ) with Bowtie2 ( 2 . 1 . 0 ) ( Langmead and Salzberg , 2012 ) . RNA-Seq reads were mapped to the mouse genome GRCm38 with TopHat2 ( 2 . 0 . 8 ) ( Kim et al . , 2013 ) . Then we applied RiboTaper ( 1 . 3 ) ( Calviello et al . , 2016 ) which used the triplet periodicity of ribosomal footprints to identify translated regions to the bam files . Mouse GENCODE Gene Set M5 ( Ensembl Version 80 ) ( Mudge and Harrow , 2015 ) was used as gene annotation input . The Ribo-seq read lengths to use and the distance cutoffs to define the positions of P-sites were determined from the metaplots around annotated start and stop codons as shown below . SampleRead lengthsOffsetsBrain29 , 3012 , 12Hippocampus29 , 3012 , 12Neural ES cells27 , 28 , 29 , 3012 , 12 , 12 , 12Heart29 , 3012 , 12Skeletal muscle29 , 3012 , 12Neutrophils25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 3312 , 12 , 12 , 12 , 12 , 12 , 12 , 12 , 12Splenic B cells30 , 3112 , 12Testis2812 All mouse peptide evidence from large-scale mass spectrometry studies was retrieved from PRIDE ( 09 Aug . 2015 ) ( Vizcaino et al . , 2016 ) and PeptideAtlas ( 31 Jul . 2015 ) ( Desiere , 2006 ) databases . We performed the same procedures on PRIDE and PeptideAtlas data separately following the method described in Xie et al . ( 2012 ) . In brief , if the whole sequence of a peptide was identical to one fragment of the tested de novo protein sequence , and had at least two amino acids difference compared to all the fragments of other protein sequences in the mouse genome , the peptide was considered to be convincing evidence for the translational expression of the respective de novo protein . The exon number of a gene was assigned as the exon number of the transcript having highest FPKM value among all the transcripts of the gene . The intrinsic structural disorder of proteins was predicted using IUPred ( Dosztányi et al . , 2005 ) , long prediction type was used . The intrinsic structural disorder score of a protein was assigned as the average of the scores of all its amino acids . The hydrophobic clusters of proteins were predicted using SEG-HCA ( Faure and Callebaut , 2013 ) , and then the fraction of the sequence covered by hydrophobic clusters for each protein was calculated . ‘Other’ genes used to compare against the de novo protein-coding genes were the protein-coding genes annotated in Ensembl ( Version 80 ) excluding the de novo genes . The ovaries , oviducts , uterus , and gonadal fat pad from the females from the Gm13030 line were carefully collected and immediately frozen in liquid nitrogen . Total RNAs from those tissues were purified using QIAGEN RNeasy Microarray Tissue Mini Kit ( Catalog no . 73304 ) , and the genomic DNAs were removed using DNase I , RNase-free ( Catalog no . 74106 ) . The first strand cDNAs were synthesized using the Thermo Scientific RevertAid First Strand cDNA Synthesis Kit ( Catalog no . K1622 ) by targeting poly-A mRNAs with oligo dT primers . Two pairs of primers targeted on the two junctions of Gm13030 gene structure and a pair of primers targeted on a control gene Uba1 were used . The sequences of the primers are shown below . PCR was done under standard conditions for 38 cycles . Primer nameSequence ( 5’>3’ ) junc1_FGGACACAGGCCAGGGAAATGjunc1_RCCTTAGGCCTTGCGAAGGAAjunc2_FGCCTGCTTTCACCATTTCAGGjunc2_RTATGAAAGGCTGGGTGAGGTGUba1_FGAAGATCATCCCAGCCATTGUba1_RTTGAGGGTCATCTCCTCACC The genomic sequences from wild mice M . spretus ( eight individuals ) , M . m . castaneus ( TAI , 10 individuals ) , M . m . musculus from Kazakhstan ( KAZ , eight individuals ) , M . m . musculus from Afghanistan ( AFG , six individuals ) , M . m . musculus from Czech Republic ( CZE , eight individuals ) , M . m . domesticus from Iran ( IRA , eight individuals ) , M . m . domesticus from Germany ( GER , 11 individuals ) , and M . m . domesticus from France ( FRA , eight individuals ) were retrieved from the whole genome sequencing data in Harr et al . ( 2016 ) . The genomic sequences from mouse strains CAROLI/EiJ ( M . caroli ) and PAHARI/EiJ ( M . pahari ) were retrieved from the whole genome sequencing data in Thybert et al . ( 2018 ) . For all these sequences , we manually checked and corrected the substitutions based on the original mapped reads . The genomic sequences from wild mice M . mattheyi ( four individuals ) and M . spicilegus ( four individuals ) were determined by Sanger sequencing of the PCR fragments from the genomic DNAs purified with salt precipitation . The PCR primers listed below were designed according to the whole genome sequencing data in Neme and Tautz ( 2016 ) . FragmentDirectionSequence ( 5’>3’ ) 1ForwardCAATATACAGACTTATACCAATGAAAACCReverseTGGGATCCTTAAGGTTCATTGTG2ForwardCCAGAGACCTCTGGATTTGCReverseAAGGCACATCTCAAAGTAAAAGC Whole genome sequencing data in Harr et al . ( 2016 ) and Neme and Tautz ( 2016 ) were used to obtain the average distances for the taxa in this analysis . For each individual , the mean mapping coverage was calculated using ANGSD ( 0 . 921–10-g2d8881c ) ( Korneliussen et al . , 2014 ) with the options ‘-doDepth 1 -doCounts 1 -minQ 20 -minMapQ 30 -maxDepth 99999’ . Then , ANGSD ( 0 . 921–10-g2d8881c ) was used to extract the consensus sequence for each population accounting for the number of individuals and the average mapping coverage per population ( mean + three times standard deviation ) with the options “-doFasta 2 -doCounts 1 -maxDepth 99999 -minQ 20 -minMapQ 30 -minIndDepth 5 -setMinDepthInd 5 -minInd X1 -setMinDepth X2 -setMaxDepthInd X3 -setMaxDepth X4’ . X1 , X2 , X3 , and X4 are listed below . The consensus sequences of the mouse populations were used to calculate the Jukes-Cantor distances for 10 , 000 random non-overlapping 25 kbp windows from the autosomes with APE ( 5 . 1 , ‘dist . dna’ function ) ( Paradis et al . , 2004 ) . The average distances obtained in this way are provided in Figure 3—figure supplement 2 . The expected distances for Gm13030 were calculated by multiplying the length of the gap-free alignment with the average distances . The observed values were retrieved from the distance table of the alignments using Geneious ( 11 . 1 . 2 ) . Pairwise substitution comparisons for the Gm13030 reading frame were calculated with DnaSP ( Librado and Rozas , 2009 ) . For this , indels were excluded , stop codons were treated as 21st amino acid following the settings of the program . The results are included in Figure 3—figure supplement 3 . PopulationMean coverageStandard deviation of coverageX1X2X3X4M . mattheyi23 . 30483 . 02815273273M . spicilegus25 . 13824 . 62715100100M . spretus24 . 88514 . 2164206854M . m . castaneus14 . 0157 . 57352537370M . m . musculus from Afghanistan17 . 76858 . 55131559354M . m . musculus from Kazakhstan25 . 12315 . 97542074592M . m . musculus from Czech Republic24 . 33814 . 10342067536M . m . domesticus from Iran20 . 2499 . 82042050400M . m . domesticus from Germany21 . 63910 . 51842054432M . m . domesticus from France21 . 49910 . 02742052416 Gm13030 was originally targeted by the Knock-Out Mouse Project ( KOMP ) , but the line was lost . Hence , we obtained a custom-made CRISPR/Cas9 line from the Mouse Biology Program ( MBP ) . The guide RNA was designed to target the beginning of the ORF in the second coding exon and away from the splicing site ( genomic DNA target: 5’ TGCTCCATCTGCTTTTCAGG 3’ ) . We obtained three mosaic frameshift knockout mice ( genetic background: C57BL/6N ) . Then we mated them with the wildtypes from the same litters to have heterozygous pups , and selected one female and one male with a heterozygous 7 bp deletion ( chr4:138 , 873 , 545–138 , 873 , 551 ) as the founding pair for further breeding and experiments . Primers for genotyping are listed below . Allele ( Fragment length ) DirectionSequence ( 5’>3’ ) KO ( 502 bp ) ForwardCCTACCACATTGGGGCCATCReverseTACAAGCCATAAAACCTCCTGGATWT ( 353 bp ) ForwardTTTTCTGCTCCATCTGCTTTTCAReverseAGTCACAGAGAAGGGGACGA The genomic DNAs from the founding pair were purified with salt precipitation . Then the samples were prepared with Illumina TruSeq Nano DNA HT Library Prep Kit ( Catalog no . FC-121–4003 ) , and sequenced on HiSeq 2500 with TruSeq PE Cluster Kit v3-cBot-HS ( Catalog no . PE-401–3001 ) and HiSeq Rapid SBS Kit v2 ( 500 cycles ) ( Catalog no . FC-402–4023 ) . The reads were 2 × 250 bp in order to have good power to detect indels . We followed GATK Best Practices ( Van der Auwera et al . , 2013 ) to call variants . Specifically , we mapped the reads to mouse genome GRCm38 ( Waterston et al . , 2002; Zerbino et al . , 2018 ) with BWA ( 0 . 7 . 15-r1140 ) ( Li and Durbin , 2009 ) , and marked duplicates with Picard ( 2 . 9 . 0 ) ( http://broadinstitute . github . io/picard ) , and realigned around the indels founded in C57BL/6NJ line ( Keane et al . , 2011 ) with GATK ( 3 . 7 ) , and recalibrated base quality scores with GATK ( 3 . 7 ) using variants founded in C57BL/6NJ line ( Keane et al . , 2011 ) to get analysis-ready reads . We assessed coverage with GATK ( 3 . 7 ) and SAMtools ( 1 . 3 . 1 ) ( Li et al . , 2009 ) , and the coverage of female was 35 . 48 X and the one of male was 35 . 09 X . High coverages also provided good power to detect indels . We called variants with GATK ( 3 . 7 ) , and applied generic hard filters with GATK ( 3 . 7 ) : "QD <2 . 0 \|\| FS >60 . 0 \|\| MQ <40 . 0 \|\| MQRankSum <−12 . 5 \|\| ReadPosRankSum <−8 . 0 \|\| SOR > 3 . 0’ for SNVs and ‘QD <2 . 0 \|\| FS >200 . 0 \|\| ReadPosRankSum <−20 . 0 \|\| SOR > 10 . 0’ for indels . We found 80375 SNVs and 73387 indels in the female and 81213 SNVs and 71857 indels in the male . 347 potential off-target sites were predicted on http://crispr . mit . edu:8079/ based on mouse genome mm9 . 343 of them still existed in mouse genome mm10 ( GRCm38 ) after converting by liftOver ( 26 Jan . 2015 ) ( Kent et al . , 2002 ) , and the four missing sites were ranked low anyway: 131 , 132 , 143 , and 200 . GATK ( 3 . 7 ) was used to look for variants found in the whole genome sequencing in the 100 bp regions around the 343 sites . In addition , the reads mapped to the regions around the top 20 sites were manually checked in both samples . The oviducts of 10–11 weeks old females from the Gm13030 line were carefully collected and immediately frozen in liquid nitrogen . Then , total RNAs were purified using QIAGEN RNeasy Microarray Tissue Mini Kit ( Catalog no . 73304 ) , and prepared using Illumina TruSeq Stranded mRNA HT Library Prep Kit ( Catalog no . RS-122–2103 ) , and sequenced using Illumina NextSeq 500 and NextSeq 500/550 High Output v2 Kit ( 150 cycles ) ( Catalog no . FC-404–2002 ) . All procedures were performed in a standardized and parallel way to reduce experimental variance . Raw sequencing outputs were converted to FASTQ files with bcl2fastq ( 2 . 17 . 1 . 14 ) , and reads were trimmed with Trimmomatic ( 0 . 35 ) ( Bolger et al . , 2014 ) . Only paired-end reads left were used for following analyses . We mapped the trimmed reads to mouse genome GRCm38 ( Waterston et al . , 2002; Zerbino et al . , 2018 ) with HISAT2 ( 2 . 0 . 4 ) ( Kim et al . , 2015 ) and SAMtools ( 1 . 3 . 1 ) ( Li and Durbin , 2009 ) , and took advantage of the mouse gene annotation in Ensembl ( Version 86 ) by using the --ss and --exon options of hisat2-build . We counted fragments mapped to the genes annotated by Ensembl ( Version 86 ) with HTSeq ( 0 . 6 . 1p1 ) ( Anders et al . , 2015 ) , and performed differential expression analysis with DESeq2 ( 1 . 14 . 1 ) ( Love et al . , 2014 ) . Principle component analysis and hierarchical clustering with Euclidean distance and complete agglomeration method on the variance stabilized transformed fragment counts were also performed using DESeq2 ( 1 . 14 . 1 ) to assign the 24 samples into three clusters . The relative expression levels of three Dcpp genes ( Dcpp1 , Dcpp2 , and Dcpp3 ) in the six cluster 1 samples of the oviducts were further validated by droplet digital PCR . For each sample , 20 µl first strand cDNA solution was obtained using the Thermo Scientific RevertAid First Strand cDNA Synthesis Kit ( Catalog no . K1622 ) by targeting poly-A mRNAs with oligo dT primers from 1 μg RNAs . Then the cDNA samples were diluted with water 1:400 for the PCR reactions . The information of probes and primers for three Dcpp genes , and Uba1 ( the reference gene ) are listed below . The sequences of the probe and primers for Dcpp genes were carefully designed to target all three genes at the same time . All PCR reactions were run with the same master mix and in the same plate . The PCR reaction mixture was prepared from 12 . 5 μL Bio-Rad ddPCR Supermix for Probes ( Catalog no . 1863010 ) , 1 . 25 μL oligo mix ( 5 μM probes , 18 μM forward primers , and 18 μM reverse primers ) for Dcpp genes , 1 . 25 μL oligo mix for Uba1 , and 10 μL cDNA dilution . The oil droplets containing 20 μL of the reaction mixture for each sample were generated by Bio-Rad QX100 Droplet Generator ( Catalog no . 1863002 ) . After droplet generation , the plate was sealed with a pierceable foil heat seal using Bio-Rad PX1 PCR Plate Sealer ( Catalog no . 1814000 ) and then placed on a thermal cycler for amplification . The thermal cycling conditions were: 95°C for 10 min ( one cycle ) , 94°C for 30 s and 56°C for 60 s ( 40 cycles ) , 98°C for 10 min ( one cycle ) . After PCR , the 96-well PCR plate was loaded into Bio-Rad QX100 Droplet Reader ( Catalog no . 1863003 ) which reads the signals in the droplets . Raw data were analyzed with Bio-Rad QuantaSoft analysis software provided with the Bio-Rad QX100 Droplet Reader . The relative expression level of Dcpp genes in each sample was calculated by dividing the concentration of Dcpp genes by the concentration of Uba1 . For each sample , two independent technical replicates were performed . GeneOligo5’ modificationSequence ( 5’>3’ ) 3’ modificationThree Dcpp genesProbeFAMGGACGGTCAAGTGTATGGCTBHQ1Forward primerGATTATCATGGTCCAGAAGTTGGAReverse primerATGTGCTCTTCCTTAGACAGTCTGUba1ProbeHEXCTGAACCTCTTGCTGCACCTBHQ2Forward primerGAAGATCATCCCAGCCATTGReverse primerTTGAGGGTCATCTCCTCACC In addition to using the fertility data from the stock breeding of the Gm13030 animals , dedicated mating pairs were set up for the fertility test . The female and male in each pair were 8–9 weeks old when the mating was started . All the males were wildtype , and 10 females were homozygous knockout and the other 10 were wildtype . The time ( days ) until having the first and second litters , the numbers of pups of the first and second litters , and whether the pups were eaten later for each mating pair were carefully observed and recorded by animal caretakers who were blind for the genotypes . Figure 5—source data 1 provides the details of the mice , the individual phenotype scores and the notes on the losses of litters , both for the stock breeding , as well as the specifically set up pairs .	Different species have specific genes that set them apart from other species . Yet exactly how these species-specific genes originate is not fully known . The traditional view is that existing old genes are duplicated to make a ‘spare’ copy , which can change through mutations into a new gene with a new role gradually over time . Despite there being lots of evidence supporting this theory , not all new genes found in recent years can be traced back to older genes . This led to an alternative view – that recently evolved genes can also appear ‘de novo’ , and come from regions of random DNA sequences that did not previously code for a protein . So far , the possibility of genes forming de novo during evolution has largely been supported by comparing and analyzing the genomes of related species . However , very little is known about the biological role these de novo genes play . Now , Xie et al . have generated a list of recently evolved de novo mouse genes , and carried out a detailed analysis of one de novo gene expressed in females at the time when embryos implant into the uterus wall . To study the role of this gene , Xie et al . created a strain of knock-out mice that have a defunct version of the protein coded by the gene . Loss of this protein caused female mice to have their second litter after a shorter period of time and increased the likelihood that female mice would terminate their newborn pups . This suggests that this newly discovered de novo gene is involved in regulating the female reproductive cycles of mice . Further analysis showed that this de novo gene counteracts the action of an older gene that promotes the implantation of embryos . This gene has therefore likely evolved due to the benefit it offers mothers , as it protects them from experiencing the increased physiological stress caused by a premature second pregnancy . These findings support the idea that genes which have evolved de novo can have an essential biological purpose despite coming from random DNA sequences . This establishes that de novo evolution of genes is the second major mechanism of how new genes with significant biological roles can form in the genome .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2019	A de novo evolved gene in the house mouse regulates female pregnancy cycles
Hedgehog ligands activate an evolutionarily conserved signaling pathway that provides instructional cues during tissue morphogenesis , and when corrupted , contributes to developmental disorders and cancer . The transmembrane protein Dispatched is an essential component of the machinery that deploys Hedgehog family ligands from producing cells , and is absolutely required for signaling to long-range targets . Despite this crucial role , regulatory mechanisms controlling Dispatched activity remain largely undefined . Herein , we reveal vertebrate Dispatched is activated by proprotein convertase-mediated cleavage at a conserved processing site in its first extracellular loop . Dispatched processing occurs at the cell surface to instruct its membrane re-localization in polarized epithelial cells . Cleavage site mutation alters Dispatched membrane trafficking and reduces ligand release , leading to compromised pathway activity in vivo . As such , convertase-mediated cleavage is required for Dispatched maturation and functional competency in Hedgehog ligand-producing cells . Hedgehog ( Hh ) ligands are produced as precursor proteins that undergo autocatalytic processing whereby a carboxyl-terminal intein-like domain cleaves itself , in a cholesterol-dependent manner , from an amino-terminal signaling domain ( Guy , 2000; Lee et al . , 1994 ) . The resulting ~20 kDa signaling protein is covalently modified by cholesterol on its new carboxyl-terminal cysteine and by a long chain fatty acid on its amino-terminus ( Porter et al . , 1996a; Chamoun et al . , 2001; Pepinsky et al . , 1998; Long et al . , 2015 ) . These lipid modifications contribute to physiological Sonic Hh ( Shh ) activity by governing ligand distribution across developing tissues , and influencing ligand potency toward target cells ( Long et al . , 2015; Taylor et al . , 2001; Li et al . , 2006; Porter et al . , 1996b; Burke et al . , 1999 ) . Distribution is controlled by Shh lipid modifications conferring high membrane affinity to the mature ligand , thereby anchoring it to the producing cell surface . In order to reach long-range target cells , Shh must be deployed from producing cell membranes through a process that is dependent upon Dispatched 1 ( Disp ) , a predicted twelve-pass transmembrane protein that shares homology with the bacterial Resistance , Nodulation and Division ( RND ) Transporter superfamily ( Burke et al . , 1999; Caspary et al . , 2002; Ma et al . , 2002; Kawakami et al . , 2002; Amanai and Jiang , 2001 ) . Disp1 knockout mice phenocopy animals lacking the essential Shh signal transducing component Smoothened ( Smo ) , underscoring the importance of Disp for pathway activity during early development ( Caspary et al . , 2002; Ma et al . , 2002; Kawakami et al . , 2002 ) . In vertebrates , Disp functions with the secreted glycoprotein Scube2 to facilitate Shh membrane extraction ( Ma et al . , 2002; Creanga et al . , 2012; Tukachinsky et al . , 2012 ) . The precise mechanism by which Disp and Scube2 mobilize Shh from the producing cell membrane is not yet clear . However , Disp contains a sterol sensing domain ( SSD ) that is thought to interact with the Shh cholesterol modification to position the ligand for transfer to Scube2 ( Creanga et al . , 2012; Tukachinsky et al . , 2012 ) . Despite this advance in understanding the Disp-Scube2 functional relationship , little is known about how Disp activity is regulated . Biochemical and cell biological analyses have shown Disp must organize into trimers and localize to the basolateral cell surface to release Shh ( Etheridge et al . , 2010 ) . Genetic studies in Drosophila suggest a crucial role for Disp-mediated endosomal recycling during Hh deployment , demonstrating that apically localized Hh must be internalized in a Disp-dependent manner , and then retargeted to the cell surface to exit ligand-producing cells ( D'Angelo et al . , 2015; Callejo et al . , 2011 ) . Loss of Disp function triggers apical accumulation of Hh and disruption of long-range signaling ( D'Angelo et al . , 2015; Callejo et al . , 2011 ) , suggesting the ability of Disp to appropriately traffic with Hh is imperative for ligand release . The regulatory processes influencing Disp membrane targeting and recycling have not yet been established . Herein , we demonstrate that Disp membrane targeting and recycling is dependent upon convertase-mediated cleavage . Cleavage occurs at an evolutionarily conserved site in the predicted first extracellular loop of Disp ( EC1 ) by the proprotein convertase Furin . Mutation of the EC1 cleavage site prevents Disp processing and disrupts Shh deployment , consistent with convertase cleavage being an essential step in Disp functional maturation . Results suggest that Disp is clipped at the cell surface and that the resulting amino-terminal fragment and processed carboxyl domain are differentially trafficked post-processing . Disruption of processing by cleavage site mutation results in altered membrane distribution of Disp , leading to compromised pathway activity in vivo . Combined , these results establish cleavage as an essential step for Disp functionality , and provide novel mechanistic insight into control of Disp function in ligand-producing cells . To begin biochemical and cell biological analysis of Disp regulation , we generated a carboxyl-terminally HA epitope-tagged murine Disp ( DispHA ) expression vector . All commercial and custom anti-Disp antibodies tested failed to detect the murine Disp protein , necessitating use of the epitope-tagged expression vector . Western blot of cell lysates from NIH3T3 cells transfected with plasmid encoding DispHA revealed two distinct protein bands detected by anti-HA antibody , one running near the predicted molecular weight of 175 kDa , hereafter referred to as Disp175 , and a second with an apparent molecular weight of ~145 kDa , Disp145 ( Figure 1A ) . Because membrane and secreted proteins are commonly modified by addition of N-linked glycans , we tested whether the size difference of the two species resulted from differential N-glycan modification . Lysates from cells expressing DispHA were treated with Endo H or PNGase F enzymes , and their migration on SDS-PAGE gels was assessed . Treatment with Endo H , which removes simple N-glycans added in the endoplasmic reticulum ( ER ) , resolved a Disp protein species from Disp175 , indicating a fraction of the upper band was ER-localized ( Figure 1B lane 2 , arrowhead ) . The lower band was resistant to Endo H . However , PNGase F , which strips both simple and complex post-ER glycans , significantly altered migration of Disp145 , indicating post-ER localization of the smaller protein species ( lane 3 , arrow ) . PNGase F treatment collapsed Disp175 to a size similar to its Endo H-sensitive fraction , consistent with the larger protein species containing both ER and post-ER fractions ( lane 3 , arrowhead ) . The observation that the Disp145 fraction was highly enriched for EndoH-resistant glycosylation , suggested Disp175 might be cleaved to generate a truncated protein after ER exit . To test this , a V5 epitope tag was inserted in the amino-terminal region of the predicted EC1 , as determined using TMPred and HMMTOP 2 . 0 secondary structure prediction tools ( Figures 1C and 2B ) . The V5 insertion site was chosen based upon the apparent molecular weight difference of the two DispHA protein species . Double tagged V5DispHA was expressed in NIH3T3 cells , and cell lysates were assessed by western blot . Disp175 was detected by both HA and V5 antisera ( Figure 1C ) . Conversely , Disp145 was detected only by HA , suggesting loss of the V5 epitope from the Disp145 species . Accordingly , a ~ 30 kDa fragment was detected by V5 antisera , confirming that Disp protein is clipped to produce Disp145 . To determine whether Disp175 was processed before or after reaching the plasma membrane , cell surface biotinylation experiments were performed ( Figure 1D ) . V5DispHA-expressing NIH3T3 cells were incubated in biotin-containing culture medium for 30 min at 4°C prior to lysis , and biotinylated proteins were captured from cell lysates on streptavidin-coated beads . Bound ( surface ) and unbound ( intracellular ) fractions were examined by western blot against the V5 tag to detect the unprocessed protein and the ~30 kDa V5 cleavage fragment . Combined densitometry analysis of four independent experiments revealed that unprocessed Disp175 enriched in the non-biotinylated intracellular fraction . Approximately 60% of total Disp175 signal was detected in the unlabeled intracellular fraction with ~40% present on the cell surface ( Figure 1D . lane 4 vs . 6 , arrowhead and densitometry summary , purple ) . Although Disp175 represented the lesser pool of surface-labeled V5DispHA , its presence on the surface argued against Disp145 conversion occurring prior to it reaching the plasma membrane . Consistent with this hypothesis , the processed 30 kDa V5Disp fragment was significantly enriched on the cell surface , accounting for ~76% of the total V5Disp30 signal ( Figure 4D , arrow and densitometry summary , green ) . Combined with the above deglycosylation analysis , these results suggest Disp145 is likely generated from Disp175 at the cell surface . Its generation in the ER or Golgi is unlikely given the low percentage ( ~23% ) of Disp30 in the intracellular , non-biotinylated fraction . We next sought to identify the exact cleavage site in EC1 . To do so Disp-Flag was expressed in HEK293T cells , and Disp175 and 145 proteins were purified on Flag beads . Disp145 was excised and subjected to Edman degradation to identify its amino-terminal residues ( Figure 2A , arrow ) . EVDWNF , which maps to amino acids 280–285 in EC1 of the murine protein , was identified as the amino-terminal sequence ( Figure 2B , red line ) . This is directly adjacent to a dibasic amino acid proprotein convertase ( PC ) cleavage motif that is conserved in Disp proteins from Drosophila , zebrafish , chick , mouse and human ( Figure 2B , yellow box ) ( Seidah et al . , 2013 ) . Consistent with this being a functional cleavage site , disruption of the murine Disp convertase motif by R279A , E280A mutation ( DispCS ) appreciably reduced Disp175 to Disp145 conversion , evidenced by nearly undetectable Disp145 or Disp 30 signal in DispCS cell lysates ( Figure 2C , lane 3 and 2D , lane 3 ) . The 175 kDa fraction of DispCS was predominantly Endo H resistant , indicating that failure to cleave was not likely due to the mutant protein being retained in the ER ( Figure 2C , lane 8 ) . Notably , CS mutation resulted in pronounced accumulation of a large molecular weight band ( ~250 kDa ) that was also evident for the wild-type protein , albeit at reduced intensity ( Figure 2C lane 3 compared to 2 , bracket ) . The 250 kDa fraction of DispCS was largely resistant to deglycosylating enzymes , suggesting the molecular weight shift was not due to alteration of N-glycosylation status ( Figure 2C , lanes 7–9 ) . Moreover , Disp250 accumulated for both wild type and CS DispHA proteins at equal intensities following proteasome inhibition by MG132 treatment ( Figure 2E ) . As such , Disp250 may represent a Disp175 species marked for proteosomal degradation by post-translational modifications such as ubiquitination , neddylation or sumoylation . To confirm a proprotein convertase was responsible for Disp processing , DispHA-expressing NIH3T3 cells were treated with cell-permeable Furin Inhibitor I , which blocks activity of convertase family members Furin , PCSK1 , PCSK2 , PACE4 , PCSK5 and PCSK7 . Treatment with increasing concentrations of drug dose-dependently reduced Disp145 levels ( Figure 3A , and green in bottom panel ) . Despite this , steady state levels of the Disp175 precursor species did not increase ( magenta ) . Instead , as was observed following mutation of the cleavage site ( Figure 2C ) , chemical inhibition of Furin family proteases triggered accumulation of the ~250 kDa fraction ( Figure 3A , bracket , and bottom panel , black ) . Conversion of Disp175 to this larger species likely accounts for the lack of Disp175 accumulation following cleavage inhibition . Furin inhibitor I sensitivity , combined with results suggesting cleavage occurs after Disp reaches the cell surface ( Figure 1 ) , indicated a dibasic amino acid-specific convertase such as Furin , PCSK5 , PACE4 or PCSK7 ( Seidah et al . , 2013; Seidah and Prat , 2012 ) . To identify the specific PC facilitating Disp cleavage , CRISPR/Cas9 technology was used to generate knockout MEF lines for each of these genes . V5DispHA was expressed in two independent clonal lines knocked out for each of these genes , and examined for cleavage by western blot for the 30 kDa V5 fragment ( Figure 3B and Figure 3—figure supplement 1 ) . Knockout of Furin , which targets substrates in the trans-Golgi , at the cell surface and in recycling endosomes ( Seidah et al . , 2013 ) , blocked formation of V5Disp30 . Knockout of PCSK5 , PACE4 and PCSK7 did not , identifying Furin as the candidate convertase responsible for Disp cleavage . Accordingly , over-expression of epitope-tagged Furin-Myc with V5DispHA enhanced cleavage of the wild-type protein , and induced low-level cleavage of the CS mutant , suggesting that by increasing Furin protein levels compensatory and/or off-site cleavage can occur ( Figure 3C–D ) . To test for Furin-Disp association , V5DispHA was co-expressed with Furin-Myc in HEK293T cells and anti-Myc immunoprecipitation experiments were performed ( Figure 3D ) . Both wild-type and CS mutant V5DispHA proteins were detected in anti-Myc immunoprecipitates , consistent with an interaction occurring between Furin and Disp ( right panel , lanes 5 and 7 ) . V5DispHA was not collected by anti-Myc in the absence of Furin-Myc expression , confirming specificity of the immunoprecipitation ( right panel , lanes 1–4 ) . To further test for a specific requirement for Furin in facilitating Disp cleavage , V5DispHA was expressed in Furin-deficient colorectal adenocarcinoma-derived LoVo cells , and generation of the V5Disp30 cleavage fragment was assessed ( Takahashi et al . , 1993 ) . When expressed in control HCT-15 colorectal cells , both 175 kDa and 30 kDa protein species were evident ( Figure 3E , lane 1 ) . Conversely , murine V5DispHA protein expressed in LoVo cells failed to produce V5Disp30 , and instead migrated in a manner similar to the CS mutant ( lanes 2–3 ) . Cleavage disruption was specific to Furin loss because its re-expression in LoVo cells rescued V5Disp30 production ( lane 4 ) . Combined with the above , these results support that Disp is cleaved by Furin . Proprotein convertases such as Furin typically act on inactive proproteins to remove inhibitory or regulatory domains as a requisite step in functional maturation of the substrate . However , a small number of substrates are inactivated by convertase cleavage ( Seidah et al . , 2013 ) . To test how Furin cleavage affected Disp , Shh transcriptional reporter assays were performed by co-culturing Shh-responsive LightII reporter cells with Disp-/- mouse embryonic fibroblasts ( MEFs ) engineered to stably express Shh ( Ma et al . , 2002; Taipale et al . , 2000 ) . Shh-expressing Disp-/- MEFs were transiently transfected with vectors encoding GFP control , wild-type DispHA , DispHA-CS or a published nonfunctional Disp mutant , DispHA-TM . This mutant harbors mutations of conserved residues in the predicted transporter motifs in TM domains 4 and 10 ( Figure 2B [Ma et al . , 2002] ) . Ligand-producing cells were co-cultured with LightII reporter cells for ~48 hr , and reporter induction was measured ( Figure 4A ) . Co-culture of LightII cells with Disp-/- MEFs expressing GFP +Shh failed to induce a significant change in reporter gene activity over that of the GFP control . Conversely , co-culture of reporter cells with Shh-expressing MEFs transiently expressing wild-type DispHA induced a statistically significant reporter response , consistent with Shh deployment being rescued by re-expression of wild-type DispHA protein . Expression of DispCSHA in Shh-expressing Disp-/- MEFs affected LightII cell reporter activity similarly to the GFP control . This reduced signaling level was also similar to what was observed in LightII cells co-cultured with Shh-stable Disp-/- MEFs expressing nonfunctional DispTM-HA ( Ma et al . , 2002 ) . As such , attenuation of Disp cleavage likely compromises Shh deployment to target cells . To directly test the ability of DispCS to deploy Shh , ligand release into culture media of Shh-expressing Disp-/- MEFs was examined . In vertebrates , Disp functions with the secreted glycoprotein Scube2 to promote ligand release ( Creanga et al . , 2012; Tukachinsky et al . , 2012 ) . Therefore , wild type or DispCSHA proteins were co-expressed with increasing Scube2-Flag in Shh-stable Disp-/- cells , and ligand accumulation in culture media was monitored by western blot and densitometry analysis of protein-normalized media samples ( Figure 4B ) . In the absence of Scube2 , Shh was not detected in culture media of GFP and Shh-expressing Disp-/- cells ( Figure 4B , lane 2 , media and light gray in densitometry analysis ) . Co-expression of Scube2-Flag was unable to bolster Shh release from GFP-transfected cells , despite efficient Scube2-Flag secretion from the Shh-stable Disp-/- cells ( lanes 3–5 , media and light gray ) . Low-level re-expression of wild-type DispHA in MEFs modestly increased Shh release into culture media over that of the GFP control ( lane 7 compared to 2 , media and dark gray in densitometry analysis ) . Consistent with Scube2 partnering with Disp to facilitate ligand extraction from the membrane , co-expression of increasing levels of Scube2-Flag with DispHA prompted a dose-dependent increase in Shh protein detectable in conditioned culture media ( lanes 8–10 and dark gray ) . Conversely , cleavage-deficient DispCS failed to effectively promote Shh release into conditioned media when expressed alone or in combination with Scube2-Flag ( Figure 4B , lanes 12–15 and black ) . Shh release was similarly affected by genetic elimination of Furin ( Figure 4C ) . Whereas control MEFs released Shh into culture media , CRISPR/Cas9 generated Furin-/- MEF clones , which failed to effectively cleave V5DispHA , were compromised in their ability to release ligand ( lanes 1–2 vs 3–4 ) . Combined , these results support that Disp cleavage is necessary for Shh deployment . To assess whether cleavage site disruption would compromise Disp activity in vivo , we turned to the Drosophila system , which is a robust and genetically tractable model for Hh signal transduction ( Jiang and Hui , 2008; Lee et al . , 2016 ) . We first confirmed processing of endogenous Drosophila Disp ( dDisp ) in cultured fly cells using a polyclonal antibody raised against predicted EC4 of dDisp ( Figure 5A–A’ ) . Wing imaginal disc-derived Clone 8 ( Cl8 ) cells were treated with control or 5’dispUTR dsRNA to assess endogenous dDisp , or transfected with a dDispHA expression vector to assess over-expressed protein . The dDisp antibody detected two distinct bands , one migrating at the predicted molecular weight of ~150 kDa ( dDisp150 ) and a second species with an approximate molecular weight of ~110 kDa ( dDisp110 ) . The intensity of both bands decreased following disp dsRNA treatment and increased with over-expression of epitope-tagged dDispHA ( Figure 5A ) . However , the ratio of the fractions shifted from being approximately equal for the endogenous protein to the upper band being predominant when over-expressed ( Figure 5A , lane 1 vs . 3 ) . Similar to what was observed for mouse Disp , a ~ 30 kDa V5 fragment was released from double-tagged V5dDispHA expressed in Cl8 cells ( Figure 5B ) . To test for processing of endogenous dDisp protein in vivo , dDisp was immunoprecipitated from wing imaginal disc lysate prepared from third instar larvae using the dDisp antisera ( Figure 5A’ ) . Both bands were evident at equal levels in anti-Disp immunoprecipitates , but not in IgG control immunoprecipitates , confirming that endogenous dDisp protein is processed in cultured fly cells and in vivo in wing imaginal discs . In addition to the dibasic convertase cleavage motif at 237/238 of the fly protein that aligns with the mouse cleavage site ( Figure 2B ) , we identified additional consensus motifs at amino acids 209 and 218 . Mutation of each of the three sites on their own did not block cleavage ( not shown ) . We therefore engineered an in frame deletion to remove sequence encompassing all three putative cleavage sites ( Δ206–238 ) , and tested for processing of the ΔCS mutant in Cl8 cells . Deletion of the three putative cleavage sites ablated generation of dDisp110 ( Figure 5C lane 4 compared to 1 ) . Endo H and PNGase F sensitivity analysis revealed that like the mouse protein , dDisp110 of the wild type protein harbored complex N-linked glycans , indicative of post-ER localization ( Figure 5C , lane 3 ) . The dDispΔCS mutant showed both ER and post-ER fractions , indicating loss of cleavage did not result from ER retention ( Figure 5C lanes 4–6 ) . Hh patterns the Drosophila wing by controlling gene expression in the wing imaginal disc ( De Celis , 2003 ) . Alteration of Hh signaling during wing development triggers phenotypes in the adult wing , providing a robust system for monitoring changes in pathway activity in vivo . Wild type or ΔCS UAS-dispHA transgenes were expressed in the dorsal compartment of wing imaginal discs using the apterous-GAL4 driver , and adult wings were screened for phenotypes . Consistent with the established positive role of dDisp in Hh release ( Burke et al . , 1999 ) , over-expression of wild-type dDispHA triggered obvious blistering of the adult wing ( Figure 5E compared to D ) . This phenotype is similar to what is observed in response to dorsal wing disc over-expression of activating mutants of the Hh signal transducing component Smo ( Marada et al . , 2013 ) . Blistering is indicative of overgrowth of the dorsal face of the wing blade , likely resulting from over-proliferation of dorsal compartment cells in response to enhanced Hh release by over-expressed dDispHA . By comparison , over-expression of dDispΔCS triggered modest wing curling , but did not induce pronounced blistering ( Figure 5F ) . These results suggest compromised in vivo activity by the cleavage-deficient mutant . To directly test for the effect of dDisp cleavage disruption on Hh export , a transgene encoding a Hh protein with an internal GFP that is retained post ligand processing ( HhGFP [Torroja et al . , 2004; Hartman et al . , 2013] ) was expressed alone or in combination with wild type or ΔCS V5dDispHA proteins in Drosophila salivary glands ( Figure 5G–I ) . Salivary glands were chosen because they are large and do not express endogenous disp ( modENCODE ) . These characteristics allowed for clear visualization of dDispHA effects on HhGFP without compensation by the endogenous protein . UAS-hhGFP and UAS-V5dispHA transgenes were recombined onto the same chromosome , and then expressed under control of SGS-GAL4 . In control non-dDisp expressing salivary glands , Hh accumulated in large puncta on the basal surface of salivary gland cells ( Figure 5G ) . Expression of wild-type V5dDispHA in salivary gland cells resulted in a mostly uniform membrane localization of HhGFP with distinct puncta evident throughout basolateral optical sections ( Figure 5H and Figure 5—figure supplement 1 ) . HhGFP was largely depleted at the basal surface but was evident in a small number of distinct puncta . Conversely , cells expressing V5dDispHAΔCS showed an overt increase in HhGFP puncta throughout basolateral optical sections with accumulation of large puncta at the basal surface that resembled puncta observed in the absence of V5dDispHA expression ( Figure 5I and Figure 5—figure supplement 1 ) . This punctate organization is similar to that observed for Hh protein expressed in embryonic and larval tissues ( Callejo et al . , 2011; Gallet et al . , 2003 ) , suggesting the puncta could represent ‘packaged’ HhGFP that has been primed for release . Accumulation of these puncta at the basal surface of DispΔCS-expressing salivary glands , where the majority of DispΔCSHA signal was detected , suggests that although HhGFP may appropriately package for release , it cannot effectively deploy when Disp cleavage is compromised . Thus , dDisp cleavage is required for ligand deployment in vivo . Having established an evolutionarily conserved requirement for Disp processing for effective ligand release , we next wanted to examine the mechanism by which processing impacted Disp functionality . Disp is predicted to assemble into functional trimers ( Etheridge et al . , 2010 ) , raising the possibility that cleavage might control oligomerization . To determine whether cleavage disruption attenuated Disp trimer assembly , wild type and CS murine DispHA proteins were expressed in NIH3T3 cells , and lysates were examined by native gel electrophoresis and western blot ( Figure 6A ) . A ~ 480 kDa fraction consistent with the predicted molecular weight of the Disp trimer was evident for both wild type and cleavage deficient DispHA proteins , indicating that blocking cleavage did not block trimer formation ( Figure 6A , top ) . Moreover , larger molecular weight fractions ( ~750 kDa ) were present at equal intensities for both wild type and CS proteins , suggesting cleavage disruption did not prevent murine Disp from forming higher order assemblies . Having confirmed DispCS was not deficient in trimer formation , we next assessed its ability to bind Shh . Disp is thought to bind Shh through a sterol sensing domain ( SSD ) -mediated association with the carboxyl-terminal Shh cholesterol modification , which is subsequently transferred to Scube2 ( Tukachinsky et al . , 2012 ) . Disp EC1 , which contains the processing site , is situated directly adjacent to the SSD ( Figure 2B ) . As such , Disp cleavage could potentially influence Shh association by governing SSD access . To test this , wild type and CS mutant V5DispHA proteins were co-expressed with Shh in Disp-/- cells , and the ability of Shh to co-immunoprecipitate with DispHA from cellular lysates was examined . Similar amounts of Shh co-immunoprecipitated on anti-HA beads with both WT and CS mutant DispHA proteins ( Figure 6B , lanes 3 and 4 ) . Shh-DispHA binding was specific because Shh failed to bind HA beads in the absence of DispHA ( lane 2 ) . These results suggest that Disp cleavage does not regulate ligand binding . Disp predominately localizes to basolateral membranes in polarized epithelial cells , but a minor sub-apical , vesicular pool has been reported ( Etheridge et al . , 2010; Callejo et al . , 2011 ) . Cholesterol-modified Hh ligand enriches apically , placing the majority of Disp protein and its target ligand in non-overlapping membrane domains ( D'Angelo et al . , 2015; Callejo et al . , 2011 ) . In Drosophila , dDisp achieves ligand release by capturing apical Hh in recycling endosomes , which subsequently retarget to the plasma membrane for ligand deployment ( D'Angelo et al . , 2015; Callejo et al . , 2011 ) . Because convertase-mediated cleavage can impact protein function by affecting intracellular trafficking ( Constam , 2014 ) , we hypothesized Disp cleavage might affect its membrane targeting . To assess its subcellular localization in polarized cells in vivo , we tested V5dDispHA subcellular localization in ovarian follicle cells , which are large and polarized , making them ideal for monitoring protein trafficking and subcellular localization . Wild type and ΔCS V5dDispHA proteins were expressed using the follicle cell driver C204-GAL4 , and localization of V5 ( amino ) and HA ( carboxyl ) epitope tags was examined in stage 10 ovaries ( Figure 6C–D ) . Colocalization between amino-V5 ( magenta ) and carboxyl-HA ( green ) , indicative of the unprocessed protein , was evident for wild type V5dDispHA in basal vesicles and along basolateral membranes ( Figure 6C , white; F-Actin marks apical , blue ) . Consistent with processing removing the amino-terminal fragment from dDisp110 , a clear separation of the two signals was observed . The released amino-terminal V5 fragment , evidenced by V5 signal not colocalized with HA , localized to apical and basolateral membrane and in vesicles throughout the cell ( Figure 6C , magenta ) . Notably , dDisp110HA depleted apically , enriching on basolateral membrane and in basally-localized vesicles ( Figure 6C , green ) , suggesting differential trafficking of the two dDisp domains post-cleavage . Although we cannot rule out a functional role for the amino-terminal fragment post-cleavage , we do not think it contributes to signaling because its over-expression in wing disc-derived Cl8 cells did not alter induction of Hh-dependent luciferase reporter gene activity ( Figure 6E ) . We were unable to directly confirm activity of DispΔN lacking the amino-terminal prodomain due to it being retained in the ER ( not shown ) . However , the functional pool of Disp protein is thought to enrich basolaterally in both murine and Drosophila systems , which is consistent with what we observed for dDisp110HA in follicle cells ( Etheridge et al . , 2010; Callejo et al . , 2011 ) . Visualization of cleavage-deficient V5dDispΔCSHA revealed a strikingly altered localization from that of the wild-type protein ( Figure 6D ) . The cleavage site mutant showed pronounced accumulation on both apical and basolateral membranes , along with uniform vesicular distribution throughout the cell , suggestive of altered membrane trafficking upon cleavage loss . Notably , signal intensity of cleavage-deficient dDisp was increased compared to wild type , potentially indicating that dDisp protein turnover might be affected by altered membrane recycling . To directly test whether dDisp membrane recycling was altered by cleavage disruption , we expressed wild type and ΔCS V5dDispHA proteins in S2 cells , and tested for dDisp colocalization with the early endosomal marker Rab5 by immunofluorescence confocal microscopy ( Figure 6F , white ) . Imaris image analysis software was used to perform colocalization analysis of ~50 V5dDispHA-expressing cells per condition across three independent experiments . In cells expressing wild-type V5dDispHA , approximately 50% of the HA signal was colocalized with endogenous Rab5 signal . Cleavage site deletion lessened V5dDispHA-Rab5 colocalization , reducing the percent of HA signal colocalized with Rab5 signal to ~35% ( Figure 6G ) . These results are consistent with compromised membrane recycling , and taken together with in vivo experiments , suggest that Disp cleavage is necessary for proper membrane trafficking . Disp was first identified as a crucial regulator of Hh ligand deployment in 1999 through a genetic screen conducted in Drosophila ( Burke et al . , 1999 ) . A number of vertebrate genetic studies subsequently established the importance of Disp in Shh morphogen gradient formation and activity during tissue development ( Caspary et al . , 2002; Ma et al . , 2002; Kawakami et al . , 2002; Nakano et al . , 2004 ) . Owing to the comparatively small number of cell biological and biochemical interrogations of Disp activity ( Creanga et al . , 2012; Tukachinsky et al . , 2012; Etheridge et al . , 2010 ) , mechanistic insight into its regulation and functionality has remained limited . The study presented here improves understanding of Disp regulation by revealing an evolutionarily conserved cleavage event that influences the ability of Disp to deploy Hh family ligands from ligand-producing cells . We report that Disp is cleaved at a conserved processing site in its predicted first extracellular loop by the proprotein convertase Furin . Cleavage site mutation compromises Disp-mediated ligand deployment in vitro and in vivo , leading to reduced pathway activation in target cells . As such , this study is the first to provide mechanistic insight into a process promoting functional maturation of Disp for its role in ligand deployment . Proprotein convertase-mediated cleavage of substrate proteins typically promotes their maturation by removing inhibitory prodomains , revealing active domains , releasing bioactive fragments , priming substrates for cleavage by additional proteases , or by influencing substrate subcellular localization ( Seidah et al . , 2013; Seidah and Prat , 2012 ) . How prodomains affect substrate trafficking is not yet fully understood , but a logical hypothesis is that cleavage regulates association with trafficking molecules and/or tethering proteins along secretory or endosomal recycling routes . Such a model has been proposed for the convertase substrate Nodal , which accumulates on the cell surface following processing inhibition ( Constam , 2014; Blanchet et al . , 2008a , 2008b ) . Because cleavage disruption altered full-length dDisp membrane localization in polarized epithelial cells to mimic what was observed for the processed 30 kDa V5 fragment , we hypothesize that like Nodal , Disp membrane trafficking is regulated in cleavage-dependent manner . The observed colocalization of amino- and carboxyl-terminal dDispWT epitope tags on basolateral membranes of Drosophila follicle cells suggests cleavage occurs after basolateral targeting of unprocessed Disp . Consistent with this hypothesis , Furin has been demonstrated to traffic basolaterally in polarized epithelial cells ( Simmen et al . , 1999 ) . Intriguingly , whereas Disp is predominantly observed to localize to basolateral membrane , Hh enriches on apical membrane , from which it must be endocytosed in a Disp-dependent manner to facilitate its release upon plasma membrane recycling ( D'Angelo et al . , 2015; Callejo et al . , 2011; Gallet et al . , 2003 ) . Our observations that cleavage-deficient Disp ( 1 ) accumulated uniformly along apical and basolateral membranes of Drosophila follicle cells and ( 2 ) showed reduced colocalization with the Rab5 endosomal marker when expressed in S2 cells suggest that its endosomal trafficking is likely compromised by cleavage disruption . As such , we suggest a testable model in which basolateral Disp cleavage activates the protein for endosomal recycling , allowing it to capture , recycle and deploy apically localized Hh . We do not believe Disp cleavage is required to interact with ligand because both wild type and cleavage-deficient murine Disp proteins co-immunoprecipitated with Shh . Cleavage-deficient Disp was also capable of forming multimers , diminishing the likelihood that EC1 clipping regulates self-association . Results obtained using both murine and Drosophila experimental systems demonstrate that Disp processing is evolutionarily conserved . However , whereas mutation of a single consensus cleavage motif in EC1 of mouse Disp was sufficient to disrupt cleavage , multiple predicted sites had to be targeted in Drosophila Disp EC1 . That three predicted motifs had to be deleted to block Drosophila Disp cleavage suggests cleavage site redundancy in the fly protein . Multiple redundant motifs may indicate increased reliance upon Disp cleavage for function in the Drosophila system . Notably , Drosophila lack a Scube2-like protein that partners with dDisp to extract Hh from ligand-producing cells ( Creanga et al . , 2012; Tukachinsky et al . , 2012 ) . Multiple redundant sites might serve as fail-safes to assure dDisp cleavage and efficient Hh membrane release in the absence of Scube2-mediated assistance . Another possible explanation is that Drosophila Disp is cleaved by additional or alternative proteases with different cleavage site preferences or efficiencies . It has been reported that although Drosophila convertases can share substrate specificity with their vertebrate counterparts , cleavage efficiency will often vary between the two systems ( De Bie et al . , 1995 ) . In vertebrates , genetic loss-of-function of proprotein convertases such as Furin , PCSK5 and PACE4 triggers developmental defects leading to embryonic lethality ( Seidah et al . , 2013; Roebroek et al . , 1998; Essalmani et al . , 2008; Constam and Robertson , 2000a; Constam and Robertson , 2000b ) . Although Furin and Disp1 knockout mice both show axial rotation and heart looping defects that lead to death at or before embryonic days ~ E9 . 5-10 . 5 , their phenotypes are not indistinguishable ( Caspary et al . , 2002; Ma et al . , 2002; Kawakami et al . , 2002; Roebroek et al . , 1998 ) . Most notably , whereas Disp1 mutant embryos show clear disruption of left-right asymmetry , Furin mutants do not ( Ma et al . , 2002; Kawakami et al . , 2002; Roebroek et al . , 1998 ) . This could be due to functional compensation by other convertases in vivo . Consistent with this notion , functional redundancy between Furin , PACE4 , PCSK5 and PCSK7 has been reported ( Seidah et al . , 2013; Roebroek et al . , 1998 ) . It is also possible that Furin-mediated Disp cleavage occurs in temporal or tissue-specific manners to scale Shh release efficiency commensurate with increased need . In such a scenario , Disp cleavage would be predicted to occur during later developmental stages or in larger developing tissues to bolster Shh deployment for a growing population of target cells . Future in vivo studies using vertebrate model systems will be required to explore these hypotheses , and to determine how Disp cleavage disruption impacts Shh-dependent developmental patterning . NIH3T3 ( CRL-1658 ) , HEK293T ( CRL-11268 ) , LoVo ( CCL-229 ) , HCT-15 ( CCL-225 ) and LightII ( JHU-68 ) cells were obtained from ATCC , S2 cells from ThermoFisher ( R690-07 ) , and Cl8 cells ( CME W1 Cl . 8+ ) were obtained from DGRC . Disp-/- knockout MEFs were obtained from P . Beachy and A . Salic ( Ma et al . , 2002; Tukachinsky et al . , 2012 ) . Furin-/- , Pcsk5-/- , Pace4-/- , and Pcsk7-/- cell lines were generated using CRISPR/Cas9 technology . C57BL/6 MEF cells were transiently transfected with 3 . 5 µl of Cas9 RNP ( Cas9 ( Berkeley Macrolab ) , 40 pmole; sgRNA ( Synthego , Redwood City , CA ) , 156 pmole ) via nucleofection ( Lonza , 4D-NucleofectorTM X-unit , Basal , Switzerland ) using solution P3 , program DS-150 in small cuvettes according to the manufacturers recommended protocol . sgRNAs used were: Furin: 5’- TCTGTAGCCGGCTGTGCCGC; Pcsk5: TGGAAAGAAACCTTGGTACT; Pace4: TACCACATGTTAGACCAAAT; Pcsk7: TTGTGGTTGCCAGTGGTAAT . Cells were single-cell sorted by flow cytometry 3 days post-nucleofection , clonally expanded and verified for disruption of the endogenous locus via western blot for protein expression if antibodies were available , and/or targeted deep sequencing to identify frameshift mutations . All cell lines were routinely validated by functional assay and western blot as appropriate , and screened monthly for mycoplasma contamination by PCR . Commercially available cell lines are re-ordered quarterly . Cells were cultured as described below . Plasmids , transgenes , Drosophila embryo injection , protein expression and antibody generation pCDNA3-Disp was generated by introducing Disp1 cDNA from RIKEN ( Wako , Japan , ( http://dna . brc . riken . jp/ ) ) into Not1-Xba1 sites in pCDNA3 ( Invitrogen ) . The HA tag was introduced as an annealed oligo into the Xho1-Xba1 site in pCDNA3 using primers ( forward 5’ tcgagtacccctacgatgtgcccgattatgcatacccatacgatgttccagattacgctgtttaat and reverse 5’ ctagattaaacagcgtaatctggaacatcgtatgggtatgcataatcgggcacatcgtaggggtac ) . pCDNA3-Scube2Fg was generated from pFLC-I-Scube2 ( SourceBiosciences , E30016G2 ) by sub-cloning into the Not1/Xho1 site . To generate double-tagged Disp , V5 epitope tag coding sequence was introduced behind amino acid Alanine106 of pCDNA3-DispHA using the primers ( forward 5’ gaggctggccttgcaggtaagcctatccctaaccctctcctcggtctcgattctacggcctcccccgctttg and reverse 5’ caaagcgggggaggccgtagaatcgagaccgaggagagggttagggataggcttacctgcaaggccagcctc ) . Mutagenesis of the cleavage site was performed using the Quickchange II XL kit ( Agilent ) using forward 5’ gatcaccatgagagagagagaGCAGCAgtggactggaacttccagaaag and reverse primer 5’ ctttctggaagttccagtccactgctgctctctctctctcatggtgatc . Drosophila disp was amplified from pFLC-I-disp cDNA ( DGRC ) , and inserted in frame with an HA epitope tag into the pAc5 . 1 vector ( Invitrogen ) to generate pAc-dispHA . The cleavage site deletion ( Δ206–238 ) mutant was generated by Quickchange mutagenesis ( Agilent ) of pAc-dispHA . To make the double tagged construct , sequence encoding the V5 epitope tag was introduced following V108 to generate pAc-V5dispHA . To generate transgenic Drosophila , V5dispHA and V5dispHAΔCS were sub-cloned from pAc5 . 1 into pUAS-attB ( Bischof et al . , 2007 ) . Transgenes were targeted to landing site 68E1 on chromosome 2 . Embryo injections were performed by Best Gene , Inc . For salivary gland analysis , UAS-HhGFP and UAS-V5dispHA transgenes were recombined using standard methods . To generate antisera against Drosophila Disp , the coding region of the predicted fourth extracellular loop ( amino acids: 694–959 ) was introduced into pET-28b in frame with a carboxyl terminal 6X His tag . Protein was expressed in BL-21 cells and affinity purified on nickel resin by standard methods . Antisera were produced in rabbits using the Covance custom antibody service . For insect cell transfections , approximately 3 × 106 Clone 8 ( Cl8 ) cells were plated in M3 insect media ( Sigma ) plus 10% fetal bovine serum ( FBS ) and 2% fly extract in 60 mm dishes the day before transfection . The following morning , cells were transfected with 2 μg of pAc5 . 1 expression vectors for Disp or Hh proteins using Lipofectamine 2000 ( Invitrogen ) . DNA content was normalized with empty pAc5 . 1 vector . For mammalian cell transfection , HEK293T , NIH3T3 , LoVo , HCT-15 , Furin-/- or Disp-/- cells were seeded at a density of 1 × 106 cells/60 mm dish in DMEM plus 10% bovine calf serum or DMEM plus 10% FBS form MEFs . Empty pCDNA3 ( 2 μg ) , pCDNA3-DispHA ( 2 µg ) , pCMV6-huFurin ( 1 μg , Origene ) , pCDNA3-Shh ( 1 μg ) and/or pCDNA3-GFP ( 1 μg ) constructs were transfected into NIH3T3 , LoVo , HCT-15 , Furin-/- or Disp-/- cells using Lipofectamine 2000 or 3000 ( Invitrogen ) . For immunofluorescence analysis of Drosophila ovaries , tissue was dissected from 2 to 3 day old C204 >V5 dispHA ( WT or ΔCS ) females using standard methods . Samples were imaged on a Leica TCS SP8 confocal microscope with a 1 . 4NA 63X objective and 0 . 7 AU pinhole using spatial sampling matching nyquest criteria . Images were deconvolved using Huygens Professional software ( theoretical PSF , Classic Maximum Likelihood Estimation ( CMLE ) algorithm , with five iterations , max ) and processed using LAS X and Adobe Photoshop CS4 . V5 and HA epitope tags were detected using Anti-V5 ( 1:500; Life Technologies ) along with AlexaFluor 488 ( 1:1000; Life Technologies ) and anti-HA ( 1:250; Roche ) along with AlexaFluor 555 ( 1:1000; Life Technologies ) respectively . Rab5 was detected using anti-Rab5 ( 1:100 , Abcam ) and AlexaFluor secondary antibody ( 1:1000 , Life Technologies ) . Phalloidin conjugated with AlexaFluor 633 ( 1:100; Life Technologies ) was used to mark F-actin . For Rab5 colocalization analysis , confocal images were acquired in z-stacks ( 3–5 slices with a slice interval of 0 . 25 µm ) using a Zeiss LSM780 microscope . Colocalization analysis was done using Imaris image analysis software . The ‘Spots’ function was used to define HA and Rab5 puncta on all slices of each cell . The ‘Colocalize Spots’ function was used to identify the number of HA spots colocalized with Rab5 spots , using preset Imaris parameters . Fifty V5dDispHA-expressing cells per condition were selected at random , and analyzed over three independent experiments to determine percent signal colocalization . Significance was determined using Student’s t-test . For dDisp expression analyses in insect cells , membrane fractions were isolated from Cl8 cells in modified HK Buffer ( HK Buffer ( 20 mM Hepes , 10 mM KCl; pH 7 . 9 ) +5% Glycerol+150 mM NaCl ) as described ( Ogden et al . , 2003 ) . For Shh release and Disp expression analyses in Disp-/- or Furin-/- cells , transfected cells were washed twice in serum-free DMEM , then incubated for 6 hr in serum-free DMEM with three media changes during incubation . Shh conditioned media was collected by incubating washed cells in 2 mL serum-free DMEM for ~48 hr . Conditioned media was centrifuged at 4°C for 1 hr at 9000 x g . The resulting supernatant was centrifuged an additional hour at 16 , 000 x g . Supernatant was TCA precipitated for six hours at 4°C before pelleting and re-suspending in TCA Resuspension Buffer ( 2% w/v SDS , 0 . 42M Tris-HCl , pH 7 . 4 , 4% v/v glycerol , 0 . 01% w/v Bromphenol Blue and 0 . 05M DTT ) as described ( Goetz et al . , 2006 ) . Protein concentration was determined by BCA assay ( Pierce , Waltham , MA ) and equal total protein amounts for each sample analyzed by SDS-PAGE on Criterion gels ( Biorad , Hercules , CA ) and western blot . For NIH3T3 cell lysis , cells were washed twice in 1X PBS , harvested in 1% NP-40 Lysis Buffer ( 50 mM Tris-HCl , pH 8 . 0 , 150 mM NaCl , 1% NP-40 , 0 . 1% SDS , 1X Protease Inhibitor Cocktail and 0 . 5 mM DTT ) and incubated for 30 min at 4°C . Extracts were cleared by centrifugation at 14 , 000 x g at 4°C for 45 min and analyzed as above . For western blotting , SDS-PAGE samples were transferred onto Protran Nitrocellulose ( GE ) or Immobilon-P PVDF ( Millipore ) using Tris/Glycine/SDS Buffer ( Biorad ) at 100V for one hour at 22°C . Membranes were blocked with 5% milk and 0 . 1% Tween-20 in Tris-buffered saline ( TBS ) for 1 hr at room temperature . Nitrocellulose membranes were immunoblotted for 1 hr at 22°C using anti-HA ( 1:5000; Covance , Princeton , NJ or 1:3000; Roche ) , anti-V5 ( 1:5000; Life Technologies ) , anti-Hh ( 1:1000; SCBT ) , Drosophila Kinesin ( 1:10 , 0000; Cytoskeleton Inc . , Denver , CO ) , Actin ( 1:10 , 000; Millipore ) , Mouse Kinesin ( anti-Kif5B , 1:5000; Abcam ) , and/or Tubulin ( 1:10 , 000; Cell Signaling ) followed by three 5-min washes in secondary milk ( primary milk diluted to 25% with TBS ) . Corresponding HRP-conjugated secondary antibodies ( Jackson Immuno , West Grove , PA ) were incubated for 1 hr at RT at a 1:10 , 000 concentration . Infrared antibodies ( Li-Cor ) were used at a 1:10 , 000 concentration with HRP-conjugated antibodies when duplexing . Blots were developed on film or by using an Odyssey Fc ( Li-Cor ) with ECL Prime ( GE , Pittsburgh , PA ) . For murine co-culture reporter assays , Disp-/- cells stabling expressing MSCV-Hygro or MSCV Hygro-Shh were seeded at a density of 1 × 106 cells per 60 mm plate in DMEM-10% Fetal Bovine Serum complete media . The following day , pCDNA3-GFP ( 2 µg ) , pCDNA3-DispHA ( 2 µg ) , pCDNA3-DispCSHA ( 2 µg ) and pCDNA3-DispTM4/TM10HA ( 2 µg ) were transfected into Disp-/- cells expressing vector or Shh using Lipofectamine 3000 . The following day , FlexiPerm discs ( Sarstedt , Germany ) were sterilized in 70% ethanol , dried and placed in the center of each well in a 6-well dish , creating a barrier between the inner ring of the well and the outer ring of the well . LightII reporter cells were seeded at 1 × 106 cells per well on the outer ring of the FlexiPerm disc in DMEM-10% BCS complete media . Disp-/- stable cells transfected with the indicated Disp expression vectors were seeded at a density of 1 × 105 cells per well in the inner ring of the FlexiPerm disc . Cells were allowed to recover for 4 hr . Media was removed from the cells and the FlexiPerm discs were removed creating a cell-free barrier between the LightII and Disp-/- cells . Cells were washed with PBS and then DMEM Serum Free Complete media . DMEM Serum-Free Complete media was added back to each well and allowed to incubate for 2 hr . Washing was carried out over 6 hr repeating the above wash steps . After 6 hr , 3mls of DMEM Serum Free Complete Media was added to each well and the cells were allowed to incubate for ~36 hr . Reporter assay were carried out according to Dual Luciferase Reporter Assay Kit instructions ( Promega ) . Experiments were repeated four times in triplicate or quadruplicate , and all data pooled . Error bars indicate s . e . m . Significance was determined using a one-way ANOVA . For insect cell reporter assays , Cl8 cells were plated in 60 mm culture dishes the day before transfection and grown to ~70% confluency and transfected using Lipofectamine 2000 . Twenty-four hours post-transfection , reporter cells transfected with ptcΔ136-luciferase ( 600 ng ) reporter construct and pAc-renilla ( 60 ng ) normalization control were combined in a 1:3 ratio with ligand-producing cells transfected with pAc-hh ( 1 μg ) and wild type or V5 fragment pAc-disp ( 1X = 500 ng ) in 12-well culture dishes . Cells were co-cultured for ~48 hr and processed using the Dual Luciferase kit ( Promega ) . Proteins of interest were expressed in NIH3T3 or HEK293T cells . Cell lysates were prepared ~48 hr post-transfection using RIPA lysis buffer ( Millipore , Burlington , MA ) . Co-immunoprecipitation assays were performed as described ( Marada et al . , 2015 ) with the following modifications . EZview Red Anti-HA Affinity Gel ( Sigma ) and EZview Red Anti-c-Myc Affinity Gel ( Sigma ) were used to immunoprecipitate HA and Myc epitope-tagged proteins respectively . Immunoprecipitates were analyzed by western blot using the following antibodies: Anti-HA ( 1:2000 , Roche ) , anti-Shh ( 1:2000 , SCBT ) , anti-Kin/mKif5B ( 1:10 , 000 , Cell Signaling ) , anti-Myc ( 1:1000 , Roche ) , anti-Flag ( 1:2000 , Sigma ) , anti-Furin ( 1:1000 , SCBT ) and anti-Tub ( 1:10 , 000 , Cell Signaling ) . Wing imaginal discs from six to eight wild type ( Oregon R ) third instar larva were homogenized in RIPA lysis buffer ( 0 . 05M Tris-HCl , pH 7 . 4 , 0 . 15M NaCl , 0 . 25% deoxycholic acid , 1% NP-40 , 1 mM EDTA and 0 . 1% SDS ) . Lysates were centrifuged for 10 min at 2000 x g and supernatant was precleared with 30 μL of a 50% A/G plus agarose slurry for 30 min . Supernatants were incubated with 10 μg anti-Disp antibody or rabbit IgG for 2 hr with gentle rocking at 4°C . Immune complexes were collected on 30 μL of A/G bead slurry for 60 min at 4°C . Beads were washed twice in lysis buffer and associated proteins were eluted by boiling for 5 min in 2x sample buffer ( 2% w/v SDS , 2 mM DTT , 4% v/v glycerol , 0 . 04 M Tris-HCL , pH 6 . 8% and 0 . 01% w/v Bromphenol blue ) and analyzed by SDS-PAGE and western blot . Deglycosylation , biotinylation and densitometry analysis were preformed exactly as previously described ( Marada et al . , 2013 , 2015 ) . Transfected cells were treated with MG132 ( 50 μM ) , or Furin Inhibitor I ( 10 , 25 , 50 , 75 and 100 mM ) in serum-free DMEM for ~6 ( MG132 ) or ~8 ( Furin Inhibitor I ) hours prior to cell lysis . NIH3T3 cells were transfected with pCDNA-DispHA , pCDNA-DispCSHA or empty vector control . Lysates were processed as previously described ( Etheridge et al . , 2010 ) with slight modifications . Approximately 48 hr post transfection cells were lysed for 30 min on ice in 1x NativePAGE sample buffer containing protease inhibitor cocktail ( Roche ) and 1% n-dodecyl-B-d-maltoside ( DDM ) . Lysates were treated with Benzonase nuclease ( Sigma ) for 30 min at room temperature followed by centrifugation at 12 , 000xg for 30 min at 4°C . Supernatants were collected and run on a 4–20% NativePAGE Bis-Tris gel and transferred to PVDF membrane . A fraction of the lysates was run on a 7 . 5% Tris-HCl gel ( BioRad ) using denaturing settings and transferred to nitrocellulose membrane . Blots were probed using anti-HA antibody ( Roche ) and protein sizes were determined using NativeMark on native gels and Precision plus protein standard ( Biorad ) on denaturing gels . NativePAGE sample buffer , DDM , NativePAGE gels and NativeMark molecular weight standard were purchased from Life Technologies . HEK293T cells were seeded in thirty 100 mm plates at a density of 8 × 106 cells/plate in DMEM plus 10% Fetal Bovine Serum and transfected the following morning with 10 μg of pcDNA3-Disp-Flag per plate according to FuGene HD ( Promega ) instructions . Cells were incubated for an additional 48 hr prior to lysis in 1% Triton X-100 Buffer ( 50 mM Tris-HCl , pH 8 . 0 , 150 mM NaCl , 1% Triton X-100 , and 1X Protease Inhibitor Cocktail ) . Lysates were pooled and centrifuged at 14 , 000 x g at 4°C for 45 min . Supernatant was pre-cleared with 400 µl of 50% EZ View Red Protein A Affinity Gel ( Sigma ) for 1 hr at 4° C . Pre-cleared supernatant was transferred to a new tube and incubated with 400 µl of EZview Red ANTI-FLAG M2 Affinity Gel ( Sigma ) for 3 hr at 4° C . Beads were washed with 1% Triton X-100 Lysis Buffer with increasing amounts of NaCl ( 0 . 25M , 0 . 5M , 0 . 75M , and 0 . 150M ) before eluting with 3x Flag Peptide ( Sigma ) according to instructions . Protein was TCA precipitated for 6 hr at 4°C before pelleting and resuspending in TCA Resuspension Buffer ( 2% w/v SDS , 0 . 42M Tris-HCl , pH 7 . 4 , 4% v/v glycerol , 0 . 01% w/v Bromphenol Blue and 0 . 05M DTT ) . Samples were electrophoresed on a NuPage gel using NuPage MOPS running buffer ( Invitrogen ) then transferred to Immobilon-PSQ PVDF ( Millipore ) with NuPage Transfer Buffer without Methanol ( Invitrogen ) . The PVDF membrane was stained with coomassie blue and allowed to air dry . The band of interest was excised and sent to Tufts University Core Facility ( http://tucf . org/protein-f . html ) for protein sequence identification . Wings from male apterous-GAL4;UAS-disp flies were mounted and imaged using a Zeiss ICc3 camera and processed using Adobe Photoshop . Multiple male and female progeny from at least two independent crosses were analyzed . Representative wings were imaged . For salivary gland analysis , salivary glands were dissected from SGS-GAL4; UAS-HhGFP , UAS-V5dispHA third instar larvae using standard methods and immunostained as described ( Carroll et al . , 2012 ) . Disp was detected using anti-HA ( 1:2000 ) and AlexaFluor anti-mouse secondary ( 1:10 , 000 , Thermofisher ) . Multiple salivary glands from male and female larva were examined . A representative image is shown .	As an embryo develops , its cells divide many times until they specialize to become distinct cell types that make up the tissues and organs . To do so , the cells need to communicate , and some send signals by making and releasing proteins that travel to more distant cells . One such signaling pathway is called Hedgehog signaling . This pathway is necessary so that the tissue and organs develop properly . If faulty , it can stop the embryo from developing properly and even lead to diseases such as cancer . Hedgehog signaling is initiated by the Hedgehog protein , which needs to be released from the cells that produce the message to transport the signal to the target cells . A protein called Dispatched helps Hedgehog to get free and travel to its destination . Without Dispatched , Hedgehog cannot be released and the embryos will not develop . Now , Stewart , Marada et al . wanted to find out if and how Dispatched itself is controlled by studying embryo cells of mice . The results showed that a protein called Furin activates Dispatched by splitting it at a specific point . When the break-point on Dispatch was genetically modified , Furin could no longer cleave the protein . As a consequence , Dispatched did not reach the correct location within cells to help Hedgehog move away from signal-releasing cells . This suggests that Furin is an essential protein of the Hedgehog pathway . A next step will be to see if this is also the case in humans . Some cancer cells can produce large amounts of Hedgehog protein , which makes tumors grow faster . A better understanding of Hedgehog signaling may help to find new cancer therapies that can block this pathway in cancer cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2018	Cleavage activates Dispatched for Sonic Hedgehog ligand release
Lateral olivocochlear ( LOC ) efferent neurons modulate auditory nerve fiber ( ANF ) activity using a large repertoire of neurotransmitters , including dopamine ( DA ) and acetylcholine ( ACh ) . Little is known about how individual neurotransmitter systems are differentially utilized in response to the ever-changing acoustic environment . Here we present quantitative evidence in rodents that the dopaminergic LOC input to ANFs is dynamically regulated according to the animal’s recent acoustic experience . Sound exposure upregulates tyrosine hydroxylase , an enzyme responsible for dopamine synthesis , in cholinergic LOC intrinsic neurons , suggesting that individual LOC neurons might at times co-release ACh and DA . We further demonstrate that dopamine down-regulates ANF firing rates by reducing both the hair cell release rate and the size of synaptic events . Collectively , our results suggest that LOC intrinsic neurons can undergo on-demand neurotransmitter re-specification to re-calibrate ANF activity , adjust the gain at hair cell/ANF synapses , and possibly to protect these synapses from noise damage . In our daily lives , we are routinely exposed to a constantly changing acoustic environment , for example while walking from the office to the cafeteria or from home to our commute . Changes in the acoustic scene require rapid analysis of relevant acoustic information and adjustment of sound input gain . The auditory efferent system is critical for top-down modulation of sensory flow . Sound information is collected by cochlear hair cells in the inner ear and transmitted to the brain via auditory nerve fibers ( ANFs ) . The olivocochlear ( OC ) system is the final and mandatory step for direct modulation of hair cell and ANF activity in the peripheral hearing organ , the cochlea . In rodents , one subset of OC neurons , the lateral olivocochlear ( LOC ) neurons , originate in and around the lateral superior olive ( LSO ) in the brainstem and send axons that synapse onto the unmyelinated dendrites of ANFs underneath the inner hair cells ( IHCs ) ( Figure 1A; Warr and Guinan , 1979 ) . These synapses are strategically located before the spike initiation zone on the myelinated peripheral ANF axons ( Hossain et al . , 2005 ) , allowing them to directly modulate the hair cell transmitted postsynaptic activity in the ANFs and thereby affect firing rates and neural coding in the auditory nerve . LOC neurons utilize a diverse cohort of neurotransmitters and neuromodulators , including γ-aminobutyric acid ( GABA ) , calcitonin gene-related peptide ( CGRP ) , opioid peptides , acetylcholine ( ACh ) and dopamine ( DA ) ( Ciuman , 2010; Darrow et al . , 2006b; Eybalin , 1993; Reijntjes and Pyott , 2016; Sewell , 2011; Vetter et al . , 1991 ) . However , even for the better investigated cholinergic and dopaminergic LOC pathways , there is only limited and sometimes contradictory knowledge available regarding their function and little is known about the underlying mechanisms for modulating afferent activity ( Arnold et al . , 1998; d'Aldin et al . , 1995; Felix and Ehrenberger , 1992; Garrett et al . , 2011; Maison et al . , 2012; Maison et al . , 2010; Niu and Canlon , 2006; Nouvian et al . , 2015; Oestreicher et al . , 1997; Ruel et al . , 2001; Sun and Salvi , 2001 ) . LOC neurons have been divided into two subgroups , based on morphological criteria ( Figure 1A and B; Brown , 1987; Vetter and Mugnaini , 1992; Warr et al . , 1997 ) . In mice , the somata of LOC ‘shell’ neurons are located in the periolivary regions around the LSO . Their axons usually bifurcate upon entering the organ of Corti and travel extensively along the cochlear spiral , forming sparse terminals along the way . The somata of LOC ‘intrinsic’ neurons reside within the LSO . When reaching the cochlea , their axons usually turn in one direction , and form a patch with a high density of bouton terminals along the cochlear coil . The majority of LOC intrinsic neurons are cholinergic ( Maison et al . , 2003; Safieddine and Eybalin , 1992; Figure 1B ) . In mice , it is believed that dopaminergic LOC neurons form a separate neurochemical group and are mainly shell neurons ( Darrow et al . , 2006b; Figure 1B ) . However , in guinea pig , dopaminergic neurons overlap with cholinergic LOC intrinsic neurons ( Safieddine et al . , 1997 ) . Several studies have perfused transmitters into the cochlea and recorded ANF activity in vivo . These studies suggest that ACh can increase and dopamine can decrease ANF firing rates ( Arnold et al . , 1998; d'Aldin et al . , 1995; Felix and Ehrenberger , 1992; Nouvian et al . , 2015; Oestreicher et al . , 1997; Ruel et al . , 2001; Ruel et al . , 2006 ) . It has been proposed that individual neurotransmitters differentially modulate the ‘set point’ of ANFs in one or the other direction and contribute to generating a continuum of spontaneous activities ( Ciuman , 2010; Le Prell et al . , 2003; Nouvian et al . , 2015 ) . Several pieces of indirect evidence suggest that LOC neurons respond to sound ( Adams , 1995; Drescher et al . , 1983; Eybalin et al . , 1987; Thompson and Thompson , 1991 ) . It has been proposed that they form a three-neuron feedback loop ( Figure 1A ) , namely the ‘LOC acoustic reflex’ ( Guinan , 2011 ) . In particular , dopaminergic LOC neurons have been shown to change their innervation density after sound conditioning in guinea pig , which is thought to provide protection against ANF damage induced by noise exposure ( Niu and Canlon , 2002 ) . Additionally , mice deficient for specific dopamine receptors were found to be more vulnerable to noise exposure compared to wildtype mice , again suggesting a protective effect ( Maison et al . , 2012 ) . Thus , the dopaminergic LOC efferents , which are the focus of the study here , provide an interesting candidate for investigating sound environment-dependent modulation of peripheral inputs to the auditory pathway . Such regulatory feedback may be important for better detection of signals in background noise , and additionally may protect synapses from excitotoxicity as a result of excessive noise exposure . Here we present compelling quantitative evidence that tyrosine hydroxylase ( TH ) , an essential enzyme for the synthesis of dopamine , hence a marker for dopaminergic fibers , is dynamically regulated according to the animal’s recent history of sound exposure . The assumption is that upregulation of TH results in increased dopamine synthesis and release . In response to sound exposure , TH is upregulated specifically in central and peripheral components of cholinergic LOC intrinsic neurons in a frequency and sound level-dependent manner . These results imply that the same LOC neurons at times may co-release ACh and DA onto ANFs , most likely resulting in complex changes at multiple sites of the IHC afferent synapses that influence firing rates and dynamic range of ANFs . Electrophysiological data show that DA reduces ANF firing rate by two mechanisms: 1 ) by reducing the presynaptic release rate , and 2 ) by reducing the EPSC amplitude and area , thereby most likely reducing the percentage of EPSPs that activate APs . Taken together , these data suggest that LOC neurons can dynamically adjust their dopamine release to change the gain at hair cell/ANF synapses . To investigate the extent of dopaminergic efferent inputs to the cochlea , dopaminergic LOC fibers were labeled in whole-mount preparations of C57BL/6J mouse cochleas by immunostaining against tyrosine hydroxylase ( TH ) , an enzyme essential for the synthesis of dopamine . As described in previous publications , three types of TH+ neurons were found in the cochlea: sympathetic fibers ( Hozawa et al . , 1989; Spoendlin and Tachtensteiger , 1967; Terayama et al . , 1966 ) , apical type II afferent neurons ( Vyas et al . , 2017 ) and a subset of LOC efferent fibers in the inner spiral bundle ( ISB ) below the IHCs ( Darrow et al . , 2006b; Figures 2 and 3 ) . Curiously , consistent with previous descriptions in CBA/CaJ mice ( Darrow et al . , 2006b ) , TH+ LOC efferent bouton terminals were not homogeneously distributed in the ISB along the cochlear spiral , but appeared in distinct patches ( Figure 2A , B ) , here called ‘terminal regions’ . Besides TH+ fibers with bouton terminals , TH+ fiber bundles with no obvious terminal varicosities , except for occasional en passant swellings , were present throughout the cochlear spiral . These fiber bundles were best identified in-between the terminal regions , here called ‘spiral regions’ ( Figure 2C ) . This pattern of alternating terminal and spiral regions is established during postnatal development , between postnatal weeks 1 and 3 ( Figure 2—figure supplement 1 ) . The tonotopic frequency map along the mouse cochlear spiral ranges from ~3 kHz in the apex to ~75 kHz in the base ( Müller et al . , 2005 ) . To investigate whether the distribution of TH+ terminal regions follows any systematic pattern , TH+ terminal and spiral regions were mapped along the cochlear coil in 1–3 month-old mice ( n = 6 cochleas , six mice ) and compared using ‘line plots’ ( Figure 2D ) . In these line plots , upper lines represent terminal regions and lower lines spiral regions . The apical cochlear tip was set at 0% , and the basal tip at 100% of cochlear length . Below the linear axis representing the cochlear length , as reference , a logarithmic map of ANF characteristic frequency is plotted , based on Müller et al . ( 2005 ) . As reflected in the average line plot of all six cochleas ( Figure 2D , bottom ) , terminal regions covered the base of the cochlea ( 80–100% of cochlear length ) with a higher probability than the 0–80% of cochlear length . Otherwise , terminal patches seemed to appear randomly , with no systematic pattern regarding location or length of individual patches . Cholinergic and dopaminergic LOC fibers with bouton endings have different cochlear innervation patterns: dopaminergic bouton terminals appear in patches , whereas cholinergic bouton endings cover the entire cochlear spiral ( Maison et al . , 2003 ) , suggesting that dopaminergic and cholinergic fibers either constitute two separate systems , or that dopaminergic fibers with bouton terminals represent a subset of cholinergic fibers . The appearance of individual TH+ terminal regions is reminiscent of the cochlear innervation by individual cholinergic LOC intrinsic neurons ( Warr and Boche , 2003; Figure 1B ) . Therefore , experiments were performed to test if cholinergic and dopaminergic LOC fibers overlap . TH immunostaining was performed on cochlear tissue with genetically labeled cholinergic LOC efferents . Choline acetyltransferase ( ChAT ) is one of the enzymes necessary for the synthesis of acetylcholine . In the knock-in ChatiresCre mouse crossed to the Cre-dependent reporter line Ai3 , the fluorescent marker EYFP is expressed in cholinergic neurons , including the cholinergic LOC intrinsic neurons , as confirmed by co-immunolabeling with an antibody against ChAT in the LSO ( Figure 3—figure supplement 1A–C ) . TH immunostaining in ChatiresCre; Ai3 cochleas showed the typical dopaminergic LOC innervation pattern , with distinct terminal regions . In this preparation , TH+ terminals clearly constitute a subset of the cholinergic terminals ( n = 6 cochleas , five mice ) ( Figure 3A; two brackets point to TH+ terminal regions amongst cholinergic terminals ) . Co-immunostaining of TH with the vesicular acetylcholine transporter ( VAChT ) ( n = 6 cochleas , six mice ) ( Figure 3B ) further confirmed the cholinergic identity of TH+ bouton terminals . To determine if the morphology of individual TH+ LOC neurons is consistent with that of cholinergic LOC intrinsic neurons , a knock-in Th2A-CreER mouse ( Abraira et al . , 2017 ) crossed with the Cre-dependent reporter line Ai9 was used to sparsely label TH+ neurons by the administration of a low dose of tamoxifen ( Feil et al . , 2009 ) . The morphology of individual TH+ fibers compares well with the known morphology of cholinergic intrinsic neurons: their axons travel some distance in one direction along the inner spiral bundle under the IHCs before forming a patch of dense bouton endings ( Brown , 1987; Vetter and Mugnaini , 1992; Warr et al . , 1997 ) ( n = 8 fibers in seven cochleas , six mice ) ( Figure 3C–D ) . Consistently , in brainstem sections of ChatiresCre; Ai3 mice , TH labeled a subset of cholinergic LOC intrinsic neurons ( Figure 3—figure supplement 1D and Figure 5C ) . Besides TH+/ChAT+ LOC intrinsic neurons , TH+/ChAT- LOC shell neurons were also observed around the LSO ( Darrow et al . , 2006b; Figure 3—figure supplement 1D ) , and LOC fibers with a shell neuron morphology ( bifurcating axons with diffuse and sparse terminals ) were found in sparsely labeled Th2A-CreER; Ai9 mice in the cochlea ( Figure 3—figure supplement 2A ) . Additionally , TH+/ChAT- fibers of an unreported morphology were observed in the cochlea ( Figure 3—figure supplement 2B ) . Nonetheless , neither of these TH+/ChAT-fibers form terminals that cluster into discrete patches of dense bouton endings . Therefore , they were not further investigated in this study . Together , these data suggest that TH+ bouton endings in terminal regions are formed by TH-expressing cholinergic LOC intrinsic neurons . Results described so far ( Figures 2 and 3 ) were based on cochleas harvested from mice that were raised in an institutional vivarium with a highly variable and generally noisy sound environment ( Lauer et al . , 2009; Figure 4—figure supplement 1A ) . Previous studies in guinea pig have shown that sound conditioning induces an increase in TH+ fibers in the IHC region of the cochlea ( Niu and Canlon , 2002 ) . The highly variable TH+ LOC efferent innervation patterns across the genetically homogenous WT mice therefore could result from differences in the acoustic experience of individual animals . Thus , we hypothesized that the distribution of TH+ terminal patches in individual cochleas is dynamically regulated by sound . To test this hypothesis , mice were raised in a ‘low noise’ vivarium with lower ambient sound levels ( Figure 4—figure supplement 1A; Lauer et al . , 2009 ) , and their cochleas were immunostained for TH at the age of 8 weeks . Line plots show the coverage of TH+ terminal and spiral regions in an example cochlea ( Figure 4A , control ) and for the average of 22 cochleas ( 11 mice ) ( Figure 4B , control ) . In contrast to mice raised in the institutional vivarium ( Figure 2D ) , cochleas from mice raised in the ‘low noise’ vivarium showed only a few or no terminal regions in the apical half ( no terminal patches < 30 kHz for n = 19/22 cochleas ) . Most of the identified terminal regions were concentrated in the most basal , high frequency region of the cochlea . To test if sound exposure increases the coverage of TH+ LOC terminal regions along the cochlear coil , mice raised in the ‘low noise’ vivarium were exposed to a 12 kHz-centered one-octave noise band at 110 dB SPL for 2 hr at the age of 7 weeks ( see Materials and methods ) . The effects of this sound exposure protocol on hearing were evaluated with auditory brainstem responses ( ABRs ) measured on a separate set of animals . This sound exposure paradigm resulted in irreversible ABR threshold shifts to levels > 85 dB SPL on average for the tested frequency range ( clicks and 8–32 kHz tones ) ( Figure 4—figure supplement 1B ) . 7–10 days after noise exposure , cochleas were immunostained for TH ( n = 20 cochleas , 11 mice ) . The delay of at least one week after noise exposure for immunostaining was chosen to allow adequate time after noise exposure for the synthesis , transport , and accumulation of TH proteins in the axonal terminals of LOC neurons , to a level that can be detected by immunostaining in the cochlear epithelium . Compared to control , after sound exposure , the apical half of the cochlear coil at and above the frequency range of the noise band showed a significantly increased likelihood to be covered by TH+ terminal regions ( Figure 4A , B , exposed ) . Interestingly , there was a paradoxical significant decrease in the number of terminal patches in the basal part of the cochlea , at frequencies > 60 kHz ( Figure 4B , Figure 4—source data 1 ) . Nevertheless , the total percentage of cochlear spiral covered by TH+ terminal regions increased significantly , about 5-fold , in sound exposed mice compared to control littermates ( median ± IQR , control: 0 . 1 ± 0 . 06 , exposed: 0 . 47 ± 0 . 27 ) ( Figure 4C ) . Though multiple potential mechanisms could account for increased cochlear coverage of TH+ terminal regions after sound exposure , the most straightforward explanation is an increase in the fraction of cholinergic LOC intrinsic neurons that express TH . To test this hypothesis , parallel to analyzing the effects of sound exposure in the periphery ( Figure 4 ) , the number of TH+ LOC neurons in the LSO was quantified in brainstem sections from the same set of mice . Compared to control littermates , the number of TH+ LOC intrinsic neurons in sound exposed mice increased significantly , about 5-fold ( median ± IQR per LSO , control: 11 ± 7 , exposed: 47 ± 35 ) ( Figure 5A , B and D ) . This 5-fold change is remarkably comparable to the increase in the percentage of cochlear spiral covered by TH+ terminal regions , suggesting that the LOC neurons that became TH+ due to sound exposure , covered additional length along the ISB in the cochlea . Secondly , to verify that after sound exposure TH was in fact upregulated in cholinergic LOC intrinsic neurons , a subset of sound exposure experiments was performed on mice with genetically labeled cholinergic neurons ( ChatiresCre; Ai9 mice ) . Sound exposure did not impact the number of cholinergic LOC intrinsic neurons . The median number of genetically labeled ChAT+ LOC neurons was not significantly different after sound exposure ( 422 , n = 6 cochleas , 3 mice ) compared to control littermates ( 433 , n = 8 cochleas , 4 mice ) ( Mann-Whitney U test , p=0 . 491 ) . After sound exposure , immunolabeled TH+ neurons represented a subset ( 16 ± 4% ) of the cholinergic LOC neurons . Again , similar to the whole dataset ( Figure 5A , B , D ) , the percentage of TH+ cholinergic LOC neurons in sound exposed mice was ~5 times higher compared to control ( 16 ± 4% versus 3 ± 1% ) ( Welch t-test , p<0 . 0005 ) ( n = 6 LSOs , 3 mice; Figure 5C ) . In comparison , after sound exposure , no obvious change was observed in the number of TH+/ChAT- LOC shell neurons ( not quantified ) . To test if TH expression affects the cholinergic identity of LOC terminals , after sound exposure , we immunostained for ChAT , in a ChatiresCre; Ai9 mouse . Most TH+ bouton terminals were positive for ChAT , both by genetic labeling ( which indicates their original identity ) and immunolabeling ( which represents expression at the time of tissue harvesting ) , suggesting that when TH expression is upregulated , the cholinergic phenotype of existing ChAT+ LOC neurons was maintained , for at least 7–10 days after sound exposure ( Figure 5—figure supplement 1A–B ) . In summary , sound exposure induces the expression of TH in previously TH- cholinergic LOC intrinsic neurons . The schematic of cochlear innervation by subtypes of LOC fibers has been modified to reflect these results ( Figure 5E ) . To test whether TH expression in cholinergic LOC intrinsic neurons can be dynamically regulated by sound , a less damaging , and partially reversible sound exposure protocol was applied . 7-week-old mice were exposed to a 12 kHz-centered one-octave noise band at 90 dB SPL for 12 hr each day for 5 consecutive days . One week after starting sound exposure , ABRs showed significant threshold shifts to ~53 dB SPL for clicks and 55–90 dB SPL for 12–32 kHz tones ( no significant change at 8 kHz ) ( Figure 4—figure supplement 1C ) . Three weeks after starting sound exposure , the ABR threshold shift had reversed completely for clicks and 12–16 kHz tones , and partially for 24 and 32 kHz tones ( Figure 4—figure supplement 1C ) . At one week after the sound exposure protocol had been initiated , the number of TH+ LOC intrinsic neurons increased significantly in sound exposed mice compared to control littermates by about 2-fold ( median ±IQR , control: 15 ± 11 , exposed: 31 ± 46 ) ( Figure 5D ) . This is about half of the increase that was found earlier with the more damaging sound exposure paradigm . Interestingly , two weeks later ( three weeks after initiating the sound exposure protocol ) , the number of TH+ LOC intrinsic neurons in sound exposed mice had returned to levels that were not significantly different from those of control littermates ( median ±IQR , control: 15 ± 11 , exposed: 19 ± 9 ) ( Figure 5D ) . These results suggest that TH expression is dynamically regulated according to the animal’s acoustic experience . To demonstrate dynamic expression of TH within individual LOC intrinsic neurons over time , mouse genetic tools were used in combination with the variable acoustic environment in the institutional vivarium . Tamoxifen administration induces reporter protein ( EYFP ) expression in TH+ LOC neurons in Th2A-CreER; Ai3 mice , providing permanent labeling of LOC neurons that express TH around the time of tamoxifen administration . On the other hand , TH immunostaining , performed 1–3 weeks after tamoxifen injection will label TH-expressing LOC neurons at the time of tissue harvesting . Because of the highly variable acoustic environment in the institutional vivarium , it is expected that different sets of LOC intrinsic neurons express TH at different times . Indeed , we observed some LOC bouton terminals that were labeled for TH genetically , but not by immunostaining ( n = 4 cochleas , 3 mice ) ( Figure 5F ) , suggesting that these terminals expressed TH around the time of tamoxifen injection , but no longer at the time of tissue harvesting . However , genetic labeling and TH immunolabeling often overlapped at the base , even when the times of tamoxifen injection and tissue harvesting were several weeks apart ( Figure 5—figure supplement 1C ) , suggesting that the TH+ LOC intrinsic neurons that innervate the basal region of the cochlea have a more stable TH expression . These results suggest that the dynamic expression of TH can be found in a ‘common’ acoustic environment for laboratory mice . Previously , the effects of DA on ANF activity have been investigated with in vivo extracellular ANF recordings from guinea pig by perfusion of artificial perilymph containing DA into the inner ear ( Oestreicher et al . , 1997; Ruel et al . , 2001 ) . These studies suggest that DA reduces spontaneous and sound-evoked ANF activity . To investigate how DA modulates ANF activity at the cellular level , patch clamp recordings were performed at the bouton endings of ANFs directly underneath the IHCs in acutely excised rat apical cochlear coils ( Glowatzki and Fuchs , 2002; Grant et al . , 2010 ) . Such recordings monitor synaptic activity at individual glutamatergic hair cell ribbon synapses , representing all the peripheral input an ANF receives . For better success rates of these technically challenging recordings , rats were used instead of mice . Recordings were performed at an age range ( 15–31 postnatal days ) , when properties of subgroups of ANFs with low to high spontaneous rates have mostly developed ( Taberner and Liberman , 2005; Wu et al . , 2016 ) . This study provides several lines of evidence that sound exposure regulates TH expression in LOC intrinsic neurons in a level-dependent and dynamic manner . Mice raised in a ‘low noise’ vivarium showed significantly less TH+ LOC terminals in the cochlea compared to mice raised in the noisier institutional facility ( Figure 4 ) . Damaging sound exposure increased the number of TH+ LOC neurons about 5-fold , and a less damaging sound exposure protocol caused a smaller increase ( ~2 fold ) in TH+ LOC neurons ( Figure 5D ) . In addition , 3 weeks after the less damaging sound exposure , the ABR thresholds of the exposed mice partially returned to control levels , along with the number of TH+ LOC neurons . In Th2a-CreER; Ai3 mice , a comparison of reporter protein expression and antibody labeling showed that individual neurons expressed different levels of TH at different points in time ( Figure 5E ) . Consistent with these results , previous studies have shown that dopamine metabolites increased in the rat cochlea after sound exposure proportionally to sound intensity ( Gil-Loyzaga et al . , 1993 ) . In addition , an increase in TH immunoreactivity was observed in the guinea pig cochlea after non-damaging sound conditioning ( Niu et al . , 2004; Niu and Canlon , 2002 ) . However , in contrast to what we observed , studies in guinea pig showed that damaging sound exposure down-regulated TH immunoreactivity in both cochlea and brainstem . This discrepancy might be due to species differences or to the use of anesthesia during the damaging sound exposure , but not during the conditioning sound exposure by Niu and Canlon , whereas sound exposures in the study here were performed on awake animals . Suppressive effects of anesthesia on efferent function have been demonstrated at least for MOC neurons ( Chambers et al . , 2012 ) . For both , mice raised in the institutional and in the ‘low noise’ vivarium , TH+ terminal regions covered more prominently the base where high frequency sounds are represented ( Figure 2D , Figure 4B , control ) . This is most likely due to a larger number of TH+ LOC intrinsic neurons projecting to the base , rather than due to an increase in the length of the terminal bouton patches formed by individual LOC intrinsic neurons that innervate the base of the cochlea . Reconstructions of single LOC intrinsic neuron axons do not show any systematic variation in the length of the terminal arbor along the cochlear spiral ( Warr et al . , 1997; Warr and Boche , 2003 ) . Similarly , dopaminergic LOC neurons in the guinea pig cochlea preferentially innervate the high frequency region ( Mulders and Robertson , 2004 ) . However , sound level measurements both in the ‘low noise’ and in the institutional vivarium did not report high sound levels in the high frequency range ( Figure 4—figure supplement 1A ) . One possibility is that TH expression in LOC neurons in the high frequency range is activated by mouse ultrasonic vocalizations ( Lahvis et al . , 2011 ) . Another possibility is that LOC neurons are denser in the medial , high-frequency processing LSO limb ( Kaiser et al . , 2011; Radtke-Schuller et al . , 2015 ) , which could result in a higher probability of TH+ LOC efferent terminals to appear in the base . Notably , after damaging noise exposure , a paradoxical decrease of innervation by TH+ terminal in the basal cochlear coil was observed ( Figure 4B ) . This could be due to noise-induced damage to ANF terminals and subsequent loss of efferent innervation . In the excised rat cochlea in vitro , dopamine reduces the firing rate of ANFs on average by ~40% . These data are qualitatively similar to previous results reporting reduced ANF firing when dopamine was perfused into the guinea pig inner ear in vivo ( Oestreicher et al . , 1997; Ruel et al . , 2001 ) . The study here describes two mechanisms by which dopamine reduces ANF firing rate: ( 1 ) by a reduction in the synaptic event rate and ( 2 ) by a reduction in EPSC amplitude and area . ( 1 ) The reduction in event rate suggests a presynaptic mechanism; hair cell release is affected by dopamine . Such a presynaptic downregulation of afferent activity provides an unexpected but highly effective potential strategy for avoiding glutamate-induced excitotoxic effects on afferent endings in response to sound exposure ( Kujawa and Liberman , 2009; Le Prell et al . , 2003; Nouvian et al . , 2015 ) . IHC release could be affected if DA were to act in a paracrine fashion by diffusing and binding to hypothetical dopamine receptors on the IHCs . Direct protein-protein interaction between dopamine D1A receptor and components of the vesicular exocytosis machinery , including otoferlin , have been demonstrated ( Selvakumar et al . , 2017 ) , suggesting that dopamine receptors could potentially directly influence the release process . However , dopamine receptor expression has not been found in mouse or rat IHCs ( Inoue et al . , 2006; Maison et al . , 2012 ) , though they have been described in fish hair cells ( Drescher et al . , 2010; Perelmuter et al . , 2019; Toro et al . , 2015 ) . An alternative mechanism of affecting IHC release rate could be through a dopamine receptor-dependent release of retrograde messengers from ANFs . ( 2 ) The reduction in EPSC amplitude and area by dopamine highly likely reduces the EPSP amplitude/area and thereby a lower percentage of EPSPs may activate APs . The effect on EPSC amplitude/area could originate pre- and/or postsynaptically , for example by reducing the quantal size or influencing glutamate receptor function . Additionally , dopamine could affect the excitability of ANFs by modulating multiple ion channels in afferent dendrites . For example , studies on guinea pigs and rats suggested that DA could affect ANF firing by decreasing sodium currents ( Oestreicher et al . , 1997; Valdés-Baizabal et al . , 2015 ) . Finally , besides dopamine receptors on the ANFs , the LOC efferent terminals may contain D2 receptors ( Inoue et al . , 2006; Maison et al . , 2012 ) . Dopamine acting on D2 auto-receptors at the LOC efferent terminals may alter the release of dopamine and other LOC efferent neurotransmitters , further complicating the effects of dopamine . Future studies are needed to dissect the different mechanisms responsible for reducing ANF firing rates by dopamine released from LOC efferent fibers . In CBA/CaJ mice , dopaminergic LOC neurons were mainly identified as non-cholinergic shell neurons , located just outside of the boundaries of the LSO , suggesting that dopaminergic and cholinergic LOC neurons exist as separate groups ( Darrow et al . , 2006b ) . The study here also reports non-cholinergic dopaminergic shell neurons in C57BL/6J mice . Surprisingly , TH expression is additionally found in a small percentage of cholinergic intrinsic neurons in the LSO . Cholinergic bouton endings of LOC fibers are found all along the cochlea , however , in ‘patches’ these endings co-express TH . Even more TH+ cholinergic patches can be found after sound exposure . The co-expression of TH and ChAT suggests that potentially both ACh and DA are being released from the same neurons . These results drastically change our thinking about how LOC intrinsic neurons may modulate spike rates of individual ANFs . In vivo experiments in the guinea pig showed that cochlear perfusion of ACh increases and DA decreases ANF firing rate ( Arnold et al . , 1998; d'Aldin et al . , 1995; Felix and Ehrenberger , 1992; Garrett et al . , 2011; Nouvian et al . , 2015; Oestreicher et al . , 1997; Ruel et al . , 2001 ) . The study here supports that DA decreases ANF firing rates , however , there are no consistent in vitro data available yet for effects of ACh or ACh/DA combined on ANF activity . Assuming that the two neurotransmitter systems act in divergent ways , it has been hypothesized that the LOC efferent system sets the sensitivity of the ANFs by manipulating a ‘set point’ ( Nouvian et al . , 2015; Ruel et al . , 2006 ) . The current study extends our understanding of how this might work . With the ability to actively adjust its relative dopaminergic versus cholinergic output , the LOC system may be able to fine-tune ANF activity in response to the sound environment . Thus , we hypothesize that cholinergic outputs dominate in a low noise environment and provide a basal level of positive modulation of ANF activity as cholinergic innervation is found all along the cochlear coil , independent of the sound environment; in animals raised in quiet as well as after sound exposure . In response to sound exposure , DA is released additionally from cholinergic terminals in a frequency-dependent manner , to tune down ANF activity and adjust their sensitivity , and , in the case of damaging sound exposure , to possibly prevent excitotoxic effects on ANF dendrites . The dynamic changes occurring in the neurochemical profile of the LOC efferent system based on the sound environment could explain the puzzling contradictory results of several classic LOC lesion studies . The immediate effect of such lesions could either result in an enhancement ( Darrow et al . , 2007 ) or in depression ( Le Prell et al . , 2005 ) of ensemble ANF activity , which is probably due to differences in the output profile of LOC efferents at the time of the experiments . Chronic LOC lesioning reduced the basal positive tone provided by LOC efferents resulting in an overall decreased ANF spontaneous activity ( Liberman , 1990 ) . Finally , when 1-methyl-4-phenyl-1 , 2 , 3 , 6-tetrahydropyridine ( MPTP ) was used to disrupt dopaminergic LOC innervation in guinea pig , spontaneous ANF activity was also reduced ( Le Prell et al . , 2014 ) . This result could be explained by MPTP also targeting dopaminergic/cholinergic LOC neurons , and thereby reducing the cholinergic excitatory input onto ANFs . The existence of multi-transmitter neurons has been increasingly recognized in the central nervous system , including the combinations of GABA/ACh , GABA/Glutamate , GABA/DA , DA/Glutamate ( Granger et al . , 2017; Hnasko and Edwards , 2012; Tritsch et al . , 2016 ) . The use of multiple neurotransmitters expands the range of synaptic coding that individual neurons can provide . LOC fibers in the auditory pathway now add DA/ACh to this increasing repertoire of transmitter combinations . Here , sensory stimulation modulates dopamine release from otherwise cholinergic efferent fibers . This process adds an additional layer of complexity to the behavior of these multi-transmitter LOC neurons , as the co-existence of the transmitter systems are stimulus and time-dependent . Neurotransmitter switching has been reported in several brain areas of the adult nervous system , however , in most instances , it was artificially induced or occurred in a diseased state ( Spitzer , 2015; Spitzer , 2017 ) . The first example of adult neuron neurotransmitter switching in response to sensory stimuli has been reported in the paraventricular and periventricular nucleus of the hypothalamus , where neurons switch their neurotransmitter from somatostatin to dopamine ( TH+ ) after exposure to extended photoperiods ( Dulcis et al . , 2013; Meng et al . , 2018 ) . The current study provides another example of sensory induced changes in neurotransmitter identity . However , here instead of a complete pre- and postsynaptic switch from one neurotransmitter system to another , the LOC efferent terminals that gain TH expression seem to retain their cholinergic identity ( Figure 5—figure supplement 1A ) and postsynaptic dopamine receptors seem to be expressed constantly on most ANF endings ( Maison et al . , 2012 ) . This arrangement allows the ANFs to respond immediately whenever DA is released . Therefore , rather than switching , LOC fibers operate with gain and loss of an additional neurotransmitter . Nevertheless , it remains to be investigated whether the expression of dopamine receptors will increase when dopamine release from the LOC fibers is chronically elevated . Such a mechanism has been observed in the midshipman fish inner ear , where seasonal changes of dopamine receptor expression on hair cells provide the main regulator for dopaminergic modulation of ANF activities ( Perelmuter et al . , 2019 ) . Future studies are needed to further elucidate the complex feedback modulation of the first synapse in the auditory pathway . All the experiments were performed on mice except for electrophysiological recordings that were performed on rats . Mice ( WT strain C57BL/6J , Jackson Laboratory ) and Rats ( Sprague Dawley , Charles River Laboratories ) of either sex were used in the experiments indiscriminately . For initial experiments ( Figures 2 and 3 ) , mice were raised in the institutional vivarium . For sound exposure experiments ( Figures 4 and 5 ) , mice were raised in a ‘low-noise’ satellite vivarium . Rats were raised in the institutional vivarium ( Figure 6 ) . Generation and genotyping of Cre driver lines ChatiresCre ( Rossi et al . , 2011 ) , Th2A-CreER ( Abraira et al . , 2017 ) and Cre-dependent reporter mouse lines ( Ai3 and Ai9 , Allen Brain Institute ) ( Madisen et al . , 2010 ) have been previously described . The tamoxifen injection procedure for the inducible Cre-loxP system when using the Th2A-CreER mouse line is described below . All transgenic mouse lines were either obtained on pure C57BL/6J background or bred in-house for at least nine generations with C57BL/6J WT mice before breeding for sound exposure experiments . Cochleas of one-week to three-month-old mice were harvested , perfused through the round and oval windows with 4% paraformaldehyde ( Electron Microscopy Sciences ) , rinsed in phosphate buffered saline ( PBS ) and fixed for ~1 hr at room temperature ( RT ) . Cochleas were carefully microdissected in PBS . Due to the presence of bones in cochleas from older mice , some partially dissected cochleas were decalcified in 0 . 2 M ethylenediaminetetracetic acid ( EDTA ) in PBS for 1–2 days before further processing . To achieve better penetration of primary antibodies , a fast freeze/thaw step in 30% sucrose was included occasionally . Whole-mount cochlear preparations were first incubated at RT in a blocking and permeabilizing buffer ( PBS with 10% normal donkey serum , 0 . 5% Triton X-100 ) for 1–2 hr . Preparations were then incubated in primary antibody diluted in PBS containing 5% normal donkey serum , 0 . 25% Triton X-100% and 0 . 01% NaN3 for ~42 hr at RT . Samples were rinsed three times with PBS before incubation with the appropriate secondary antibody diluted 1:1000–2000 in PBS containing 5% normal donkey serum , 0 . 25% Triton X-100 at RT for 2 hr . Preparations were again rinsed three times in PBS before mounting on glass slides in FluorSave mounting medium ( Calbiochem ) . All incubations and rinses were performed on a rocking table at RT . See Key Resources Table for a list of primary antibodies used in this study . The locations of ‘terminal regions’ with TH+ bouton endings were mapped along the cochlear coil using the Measure_line ImageJ plugin ( Massachusetts Eye and Ear Infirmary ) by reconstructing the whole cochlear spiral from dissected pieces of cochlear coils . The logarithmic axis representing ANF characteristic frequency ( f ) was constructed using the formula d ( % ) =100 – ( 156 . 5 + 82 . 5 × log ( f ) ) , based on Müller et al . ( 2005 ) . For cochleas shown in Figure 2 , the terminal regions were identified by examining the reconstructed whole cochlea image made from tiles of confocal images taken with a 40x objective . For sound exposure experiments , terminal regions were identified by examining panorama confocal images of individual cochlear pieces taken with a 10x objective lens . In initial trials , the terminal regions were identified by using a higher magnification objective lens while viewing at the microscope and marking the location on the images taken with the 10x objective lens . Both methods gave comparable results . Mice were given an overdose of 50 mg/ml sodium pentobarbital and perfused transcardially with PBS followed by 4% paraformaldehyde . The brain tissue was then removed from the skull bones and post-fixed in 4% paraformaldehyde overnight at 4°C . The caudal portion of the brain tissue was trimmed off to allow it to sit flat on its coronal plane and placed in a well coated with petroleum jelly . 5 ml of gel albumin mixed with 0 . 4 ml of 5% glutaraldehyde and 1 ml of 37% paraformaldehyde was placed into the well containing the brain tissue and allowed to harden for 30 s . The brain was mounted on the vibrating microtome ( Vibratome 1000 Plus ) with superglue and sectioned in the coronal plane into 50 µm slices . The brain slices were first incubated at RT in a blocking and permeabilizing buffer ( PBS with 10% normal donkey serum , 0 . 5% Triton X-100 ) for 1–2 hr . The brain slices were then incubated with primary antibodies diluted in PBS containing 5% normal donkey serum , 0 . 25% Triton X-100 for two days in a cold-room at 4°C on a shaker . After washing with PBS , the brain slices were incubated with secondary antibodies diluted in PBS containing 5% normal donkey serum , 0 . 25% Triton X-100 for 1–2 hr at RT on a shaker . The brain slices were then washed again with PBS , before mounting on glass slides in Fluoromount-G mounting medium ( SouthernBiotech ) . TH+ LOC intrinsic neurons were quantified by visual inspection of confocal stacks of each brainstem slice . As a test , we circled all the TH+ LOC intrinsic neurons found on two adjacent brainstem slices . Overlaying the images from two consecutive brainstem slices showed non-overlapping neurons , suggesting that it is unlikely to count one TH+ LOC intrinsic neuron twice on consecutive brainstem slices . Therefore , a simple summation of the numbers found on each brainstem slice is appropriate for quantifying the total number of TH+ LOC intrinsic neurons . Fluorescence images were acquired using a LSM 700 confocal microscope ( Zeiss ) with a Fluar 10x/0 . 50 M27 objective , a LCI Plan-Neofluar 25x/0 . 8 Imm Korr DIC M27 objective and a Fluar 40x/1 . 30 Oil M27 objective using the ZEN black 2011 software . The pinhole was set at one airy unit . The size of optical sections was determined by stepping at half the distance of the theoretical z-axis resolution ( the Nyquist sampling frequency ) . Images were acquired in a 1024 × 1024 raster . Images are presented as maximum intensity projections through a subset of the collected optical stacks . Some images were processed in ImageJ or Fiji without deconvolution , filtering , or gamma correction . Tamoxifen freebase ( Sigma-Aldrich ) was prepared in corn oil ( Sigma-Aldrich ) at 10 mg/ml and sonicated at RT for 0 . 5–1 hr until no precipitations were visible . This solution was stored at 4°C for 5–7 days protected from light . Tamoxifen solution was administrated intraperitoneally ( i . p . ) or through gavage for a total amount of 0 . 2–2 mg for each injection . To achieve sparse labeling of TH-expressing neurons , a single dose of 0 . 2–0 . 5 mg tamoxifen was administrated . Octave-band noise was generated with digital-to-analog converters and gated with electronic switches ( Tucker-Davis Technologies ) . Stimulus levels were controlled by programmable attenuators ( Tucker-Davis Technologies ) and an audio amplifier ( Crown Audio ) . Stimulus waveforms were transduced by two overhead high-frequency speakers ( Pyramid ) . The overhead location of the speaker minimized the effects of head orientation on sound energy propagating to the tympanic membrane . Speakers were calibrated with the same microphone and software used for the auditory brainstem response measurements ( see below ) . Additional checks on the sound levels were performed using a sound level meter fitted with a ½” free field microphone ( Larson-Davis ) just prior to each sound exposure . During the presentation of noise , mice were placed inside a mesh cage with food pellets and water gel ad libitum . The cage was situated on top of a rotating platform making counter-clock or clockwise rotations 30 s/step to make sure even exposure to the sound field . Each step was 1 . 8 degrees and happened abruptly . Sound levels at test frequencies varied by no more than 10 dB within the listening area . The whole platform was isolated from extraneous environmental sounds by a sound-attenuation chamber . The inner walls of the chamber were lined with anechoic foam ( Sonex ) . For most of the trials , control littermates were placed inside a mesh cage on top of a simultaneously rotating platform with the same built outside of the sound-attenuation chamber and exposed to ambient noise levels in the procedure room during the sound exposure session . For two trials , control littermates were either placed in the shared housing cage outside the sound-attenuation chamber or placed on the rotating platform inside the sound-attenuation chamber during daytime without noise exposure . Mice were exposed to a 1-octave band noise centered at 12 kHz . The 2 hr 110 dB SPL exposure began with an initial moderate level for 15 s , followed by an interim level for 15 s to finally reach the maximum level of ~110 dB SPL . The maximum sound exposure level was maintained for 2 hr . The exposure was performed during the daylight cycle at the vivarium . A less damaging sound exposure used a 1-octave band noise centered at 12 kHz at ~90 dB SPL . The exposure lasted 12 hr each session . All of the mice were exposed five consecutive nights ( approximately 9 PM – 9 AM ) , except for one that was exposed for three consecutive nights , though the whole dataset was collectively referred to as 5 × 12 hr 90 dB SPL exposure . For sound exposure and control experiments , both wild-type and transgenic mice were bred on a pure C57BL/6J background and were raised in a low-noise satellite vivarium . The sound exposure started at 7–8 weeks of age . Mice were sacrificed 7–10 days after the beginning of the sound exposure . Cochlear and brainstem samples were harvested for immunostaining experiments . Mice used in the sound exposure or control experiments were not used in any previous procedures or sound exposures . Sound levels in the ‘low noise’ vivarium and the large-capacity institutional vivarium were recorded using a data-logging sound level meter with 1/3 octave band measurement capabilities and ½” free field microphone ( Larson-Davis ) . Ultrasound range was measured in each room using a M500-384 USB Ultrasound Microphone ( Pettersson Elektronik ) linked to a laptop running the BatSound Touch Lite software . 5 min recordings were made simultaneously with the sound level meter and the ultrasound microphone . Such recordings were made on 4–5 days at slightly different locations inside each vivarium , where the mice were housed . All the recordings were made during daytime . Frequency analysis of the recorded sound was performed using Adobe Audition CC2018 ( Adobe ) . Recording procedures were similar to those previously described ( Lauer and May , 2011; Lina and Lauer , 2013; McGuire et al . , 2015 ) . Mice were anesthetized with 100 mg/kg ketamine and 20 mg/kg xylazine through intraperitoneal injection and placed on an electronically controlled heating pad inside a small sound-attenuating chamber 30 cm away from a Fostex speaker in front of the animal . Auditory brainstem responses ( ABRs ) were differentially recorded from the scalp using subcutaneous platinum needle electrodes ( G . R . A . S . ) placed over the left bulla and at the vertex of the skull , with a ground electrode inserted into the leg muscle . Responses were amplified 300 , 000 times ( ISO-80 , World Precision Instruments ) and bandpass filtered from 300 to 3 , 000 Hz ( Krohn-Hite ) . Auditory stimuli were generated at a 200 kHz sampling rate , attenuated to control presentation levels ( Tucker-Davis Technologies , PA5 ) , and amplified ( Parasound ) before being passed to a calibrated free-field speaker ( Fostex ) . Stimulus protocols were implemented on programmable real-time processors ( Tucker-Davis Technologies , RX6 ) using a custom MATLAB program ( Ngan and May , 2001 ) . Responses were averaged over 300 stimulus presentations . Stimuli were clicks and 5 ms tone pips at 8 , 12 16 , 24 , 32 kHz with a rise/fall time of 0 . 5 ms played at a rate of 20 or 30 repetitions/s . Stimuli were calibrated with a ¼″ Bruel and Kjaer microphone placed at the location normally occupied by the mouse's head during testing using a custom MATLAB ( Mathworks ) based program . Thresholds and suprathreshold responses were measured by presenting a descending series of stimulus levels beginning with −10 dB of the maximum possible output of the speaker and continuing in 5- or 10 dB steps until no response could be discerned from the noise . Threshold was defined as the sound level at which the peak-to-peak ABR amplitude ( any wave ) was two standard deviations above the average level of a 5 ms window of baseline noise collected at the end of a 30 ms recording epoch . Responses to clicks were measured first to ensure an optimal signal-to-noise ratio and the presence of at least four distinct peaks . Responses to tones were measured in a pseudorandom order . Recordings were performed on acutely excised cochlear preparations from Sprague Dawley rats ( Charles River Laboratories ) at room temperature ( 22–25°C ) . Recording pipettes were fabricated from 1 mm borosilicate glass ( WPI ) . Pipettes were pulled with a multistep horizontal puller ( Sutter ) , coated with Sylgard ( Dow Corning ) and fire polished . Drug application was mediated by whole-bath perfusion or a gravity-driven flow pipette ( 100-µm-diameter opening ) placed near the row of IHCs and connected with a VC-6 channel valve controller ( Warner Instrument ) . Postnatal day ( P ) 15–31 SD rats were used for ANF recordings . Loose-patch extracellular and whole-cell patch clamp recordings in current and voltage clamp on the dendritic endings of ANFs were performed as described before ( Grant et al . , 2010; Wu et al . , 2016 ) . Pipette resistances were 9–15 MΩ for ANF recordings . ANF recordings were acquired using pCLAMP 9 . 2 or pCLAMP 10 . 2 software ( Molecular Devices ) in conjunction with a Multiclamp 700A or Multiclamp 700B amplifier ( Molecular Devices ) . The signal was low pass filtered at 10 kHz and digitized at 50 kHz with a Digidata 1322A ( Molecular Devices ) . Extracellular solutions for ANF recordings contained ( in mM ) : 5 . 8 KCl , 144 NaCl , 0 . 9 MgCl2 , 1 . 3 CaCl2 , 0 . 7 NaH2PO4 , 5 . 6 glucose , 10 HEPES , pH 7 . 4 ( NaOH ) , 300 mOsm . In a subset of ANF voltage-clamp recordings , K+ concentration is elevated to 15 mM in substitution of Na+ to increase presynaptic release . Pipette solution for extracellular loose-patch recordings contained the extracellular solution . Pipette solutions for whole-cell ANF recordings contained ( in mM ) : 20 KCl; 110 K-methanesulfonate; 5 MgCl2; 0 . 1 CaCl2; 5 EGTA; 5 HEPES; 5 Na2ATP; 0 . 3 Na2GTP; 5 Na2 phoshocreatine; pH 7 . 2 ( KOH ) , 290 mOsm or 135 KCl , 3 . 5 MgCl2 , 0 . 1 CaCl2 , 5 EGTA , 5 HEPES , 0–4 Na2ATP , 0–0 . 2 Na2GTP , pH 7 . 2 ( KOH ) , 290 mOsm . For ANF current clamp recordings , bridge balance and pipette capacitance neutralization were performed . For ANF voltage clamp recordings , holding potentials were between −99 and −84 mV . All recordings that had a leak current <350 pA , except for one recording that had a leak current up to 450 pA . Both current and voltage clamp recordings were corrected posterior for measured liquid junction potentials: 4 mV for potassium-based solution and by 9 mV for methanesulfonate-based solution . Series resistance Rs was not compensated for in voltage clamp recordings . DA solutions were prepared daily from dopamine hydrochloride powder ( Sigma ) and protected from light during the experiments . 1 mM DA was used in ANF loose-patch extracellular and current-clamp recordings , 1–2 mM DA was used in ANF voltage-clamp recordings . 0 . 01–0 . 1% ( w/v ) sodium ascorbate was supplemented in the DA solution as an antioxidative agent , except for two ANF recordings . For control experiments , DA but not sodium ascorbate was omitted in the Control solution . In a subset of voltage-clamp recordings , 1–2 µM TTX ( Tocris ) was present to block APs . 30–60 s are typically needed for a drug to wash into the tissue , and this was the minimal wait time before analyzing dopamine effects . For cochlear afferent recordings , the rate of EPSC/EPSP and AP fluctuates in a power-law fashion throughout the recording , even in control condition , as described in Wu et al . ( 2016 ) . Therefore , it is essential , to record for long enough in control , in DA , and after washout , to be able to separate random fluctuations from drug effects . This is why at least 30 s of recording were analyzed before , in and after DA . To study the effects of DA , segments of recordings were selected for analysis . For extracellular loose-patch recordings of ANFs , 2 min before DA application , 2 min before the end of DA application , and 2 min after 2 min of washout , were taken as the ‘before’ , ‘DA’ and ‘after’ time windows , respectively . The control experiments were performed and analyzed in the same way . The total application time for DA or Control solution was 3–5 min . For whole-cell recordings of ANFs , 30 s to 2 min of recording before DA application was selected for ‘before’ , 1 or 2 min of recording after at least 30 s of DA application was selected for ‘DA’ , 30 s to 2 min of recording after at least 1 . 5 min of wash-out was selected for ‘after’ . The total application time of DA varied between 1 . 5–6 min . Events in extracellular loose-patch and whole-cell current clamp recordings were detected using a routine in MiniAnalysis and subsequently accepted by eye . In current clamp , each peak detected was identified as an event for quantification of synaptic frequency . AP and EPSP were identified by their amplitude distribution . AP threshold was detected by several methods . The first method found the point of maximum slope of the first derivative of the rising phase of the AP using a customized MATLAB routine ( Wu and Chan , 2019 ) . For most of the APs in the ANF dendrite recordings , this method could successfully identify the AP threshold . In case this method failed to locate the AP threshold , a second method that found the maximum curvature point on the rising phase of the AP was used . If the second method still could not identify the AP threshold , the AP analysis routine in the MiniAnalysis ( Synaptosoft ) was used . Events in ANF voltage clamp recordings were detected by threshold search event detection method of Clampfit ( Molecular Devices ) , following a manual adjustment of baseline current . The end of an event was defined as the current that came back to the pre-set value from the baseline . Therefore , an event could contain several closely spaced peaks . Data were plotted using Excel 2016 ( Microsoft ) , Origin 9 . 1 ( OriginLab ) , RStudio , MATLAB ( MathWorks ) , and Illustrator ( Adobe ) . For representative traces , the data points were reduced by decimation to a 5 or 10 kHz sampling rate . Statistical analyses were performed with SPSS Statistics 25 ( IBM ) , STATA ( StataCorp ) or R studio ( R Core Team ) . Graphical representation of the quantification is defined in Figure Legend , with the definition of n and information about the statistical tests , unless otherwise states in the Result . Error bars can be either standard deviation ( SD ) or standard error of the mean ( SEM ) , as specified in Figure Legend . Statistical significance is defined as: n . s . ( not significant ) p>0 . 05 , p<0 . 05 , p<0 . 01 , **p<0 . 001 . For the quantification of the TH+ terminals and TH+ LOC neurons , the researchers that performed the majority of the analysis were blinded to the experimental condition during data quantification .	Every day , we hear sounds that might be alarming , distracting , intriguing or calming – or simply just too loud . Our hearing system responds to these acoustic changes by fine-tuning sounds before they enter the brain . For example , if a noise is too loud , the volume can be turned down by dampening the signals nerve fibers in the ear send to the brain . This is thought to reduce the damage loud sounds can cause to the sensory organ inside the ear . A set of nerve cells located at the base of the brain called the lateral olivocochlear ( LOC ) neurons coordinate this adjustment to different volumes and sounds . When these neurons receive information on external sounds , they signal back to the hearing organs and adjust the activity of auditory nerve fibers that communicate this information to the brain . LOC neurons use a diverse range of molecules to modify the activity of auditory nerve fibers , including the ‘feel-good’ neurotransmitter dopamine . But it is unclear what role dopamine plays in this auditory feedback loop . To find out , Wu et al . studied the hearing system of mice that had been exposed to different levels of sound . This involved imaging LOC neurons stained with a marker for dopamine and measuring the activity of nerve fibers in the inner ear . The experiments showed that LOC neurons in mice that had recently been exposed to sound were covered in an enzyme that is essential for making dopamine . The louder the sound , the more of this enzyme was present , suggesting that the amount of dopamine released depends on the volume of the sound . LOC neurons release another neurotransmitter called acetylcholine , which stimulates activity in auditory nerve fibers . Wu et al . found that dopamine and acetylcholine are released from the same group of LOC neurons . However , dopamine had the opposite effect to acetylcholine and reduced nerve activity . These findings suggest that by controlling the mixture of neurotransmitters released , LOC neurons are able to fine-tune the activity of auditory nerve fibers in response to acoustic changes . This work provides a new insight into how our hearing system is able to perceive and relay changes in the sound environment . A better understanding of this auditory feedback loop could influence the design of implant devices for people with impaired hearing .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Sound exposure dynamically induces dopamine synthesis in cholinergic LOC efferents for feedback to auditory nerve fibers
While targeted therapy against HER2 is an effective first-line treatment in HER2+ breast cancer , acquired resistance remains a clinical challenge . The pseudokinase HER3 , heterodimerisation partner of HER2 , is widely implicated in the resistance to HER2-mediated therapy . Here , we show that lapatinib , an ATP-competitive inhibitor of HER2 , is able to induce proliferation cooperatively with the HER3 ligand neuregulin . This counterintuitive synergy between inhibitor and growth factor depends on their ability to promote atypical HER2-HER3 heterodimerisation . By stabilising a particular HER2 conformer , lapatinib drives HER2-HER3 kinase domain heterocomplex formation . This dimer exists in a head-to-head orientation distinct from the canonical asymmetric active dimer . The associated clustering observed for these dimers predisposes to neuregulin responses , affording a proliferative outcome . Our findings provide mechanistic insights into the liabilities involved in targeting kinases with ATP-competitive inhibitors and highlight the complex role of protein conformation in acquired resistance . The epidermal growth factor receptor ( EGFR ) family of receptor tyrosine kinases plays a major role in proliferative signalling in a variety of cancers ( Baselga and Swain , 2009; Yarden and Pines , 2012 ) . Apart from EGFR ( also known as ErbB1 ) , the family consists of the orphan receptor HER2 ( ErbB2 ) , the pseudokinase HER3 ( ErbB3 ) , and HER4 ( ErbB4 ) . Overexpression of HER2 is an oncogenic driver in approximately 20% of all breast cancers ( Lovekin et al . , 1991; Owens et al . , 2004; Slamon et al . , 1987 ) . The high clinical relevance of these receptors has made them a target for directed therapy with both antibodies and small molecule kinase inhibitors . In the case of HER2+ breast cancer , the monoclonal antibody trastuzumab ( Herceptin ) and its cytotoxic drug-conjugated derivative trastuzumab-emtansine ( Kadcyla ) , the monoclonal antibody blocking HER2-HER3 dimerisation pertuzumab ( Perjeta ) , and the small molecule kinase inhibitor lapatinib ( Tykerb/Tyverb ) have been successful in the clinic ( Blackwell et al . , 2010; Cameron et al . , 2017; Diéras et al . , 2017; Geyer et al . , 2006; Krop et al . , 2017; CLEOPATRA Study Group et al . , 2015; EMILIA Study Group et al . , 2012 ) . While HER2 itself has no known ligand , HER3 binds the growth factor neuregulin ( NRG , also known as heregulin or HRG ) to induce heterodimerisation and signalling ( Sliwkowski et al . , 1994 ) . HER3 has been implicated in therapeutic resistance to HER2-targeted therapy through a variety of mechanisms , including receptor rephosphorylation , HER3 overexpression and increased NRG production ( reviewed in [Claus et al . , 2014] ) . In terms of cellular signalling in response to HER-family kinase inhibition , HER3-mediated buffering through the Akt/PKB signalling axis has been shown to be an important factor in therapeutic resistance ( Sergina et al . , 2007 ) . The dimerisation of EGFR family members is a fluid process mediated by interaction dynamics in practically every domain of the receptor . For EGFR , the ligand-bound , active dimer shows an upright , back-to-back extracellular domain ( ECD ) interaction where both receptors have bound ligand , although singly-bound dimers can also occur ( Garrett et al . , 2002; Liu et al . , 2012; Ogiso et al . , 2002 ) . Although HER2 has no known ligand , it natively adopts this upright , dimerisation-ready ectodomain conformation ( Garrett et al . , 2002 ) . On the intracellular side , formation of the active kinase domain dimer is critically affected by the conformation of the juxtamembrane domain ( JMD ) ( Jura et al . , 2009a; Thiel and Carpenter , 2007 ) . The kinase domains associate in an asymmetric dimer , which resembles the CDK/cyclin-like asymmetric dimer interface ( Jeffrey et al . , 1995; Zhang et al . , 2006 ) . In this canonical dimer , one kinase ( the ‘activator’ ) allows the dimerisation partner ( the ‘receiver’ ) to adopt an active conformation and become catalytically active . These various conformations have also been observed in near-complete receptors using negative stain electron microscopy ( Mi et al . , 2011 ) . Of note in these receptor dimer formations was the lack of active , asymmetrical kinase domain interactions when the receptor was bound to the ATP-competitive inhibitor lapatinib ( Mi et al . , 2011 ) . Although these interactions have mainly been described in the context of EGFR homodimerisation , they remain a template for the interactions of the rest of the EGFR family . The conformation of the active kinase domain interaction has been validated for EGFR-HER3 and HER2-HER3 ( Jura et al . , 2009b; Littlefield et al . , 2014; van Lengerich et al . , 2017 ) . A multitude of studies , using a variety of techniques , have confirmed that EGFR-family receptors can form higher order oligomers , and that the exact nature of these oligomers is modulated by a variety of conditions , including receptor density , ligand presence , ligand type and temperature-dependent membrane behaviour ( Clayton et al . , 2005; Clayton et al . , 2007; Huang et al . , 2016; Nagy et al . , 2010; Needham et al . , 2016; Saffarian et al . , 2007; van Lengerich et al . , 2017; Yang et al . , 2007; Zhang et al . , 2017 ) . Against the backdrop of such a multitude of association modes , it is clear that conformational dynamics and structural rearrangements are an integral regulator of protein behaviour in the EGFR family . We have shown previously that within a kinase , in this case PKCε , occupation of the nucleotide binding pocket with ATP ( or an inhibitor ) is a major determinant of protein behaviour , conferring the structural stability required for protein–protein interactions to occur and priming sites to be stably phosphorylated ( Cameron et al . , 2009 ) . Similar effects have been observed in several additional kinases , including PKB/Akt , IRE1 , and AMPK ( Okuzumi et al . , 2009; Papa et al . , 2003; Ross et al . , 2017; Wang et al . , 2012 ) . A notable example of nucleotide binding pocket occupation inducing behaviour independent of catalysis has been described for the RAF family , originally in cRAF , where the inhibitor SB 203580 paradoxically induced activity ( Eyers et al . , 1998 ) . More recently , a similar phenomenon has been shown in BRAF , where the small molecule kinase inhibitor vemurafenib blocks the oncogenic mutant V600E , but stabilises the wild type protein , promoting downstream proliferative signalling ( Hatzivassiliou et al . , 2010; McKay et al . , 2011; Poulikakos et al . , 2010; Thevakumaran et al . , 2015 ) . Within the EGFR family , we and others have shown previously that quinazoline inhibitors can cause homodimer formation of EGFR , and EGFR-MET heterodimerisation , by stabilising particular kinase domain conformers ( Arteaga et al . , 1997; Bublil et al . , 2010; Lichtner et al . , 2001; Ortiz-Zapater et al . , 2017 ) . The structural , conformational role that nucleotide pocket occupation can fulfil is particularly interesting in the context of pseudokinases , which have lost their catalytic activity . Sequence analysis shows that many pseudokinases retain several of the conserved residues involved in ATP-binding ( Boudeau et al . , 2006; Claus et al . , 2013 ) . In vitro analysis of the pseudokinome showed that many pseudokinases have nucleotide binding capability ( Murphy et al . , 2014 ) . In the case of these ATP-binding pseudokinases , where nucleotide binding does not elicit phosphotransfer , the structural stability conferred by ATP binding may be integral to protein function . This has been observed for the pseudokinase STRAD , which requires ATP binding to sustain a heterotrimeric complex with LKB and MO25 ( Zeqiraj et al . , 2009a; Zeqiraj et al . , 2009b ) . Similarly , in the pseudokinase FAM20A ATP-binding , albeit in a non-canonical orientation , is essential for stabilising the FAM20A/FAM20C complex ( Cui et al . , 2015; Cui et al . , 2017 ) . ATP binding is a structural requirement for the JAK2 JH2 V617F mutant to promote pathogenic signalling ( Hammarén et al . , 2015 ) . In the pseudokinase MLKL , ATP-binding pocket occupation is essential for membrane translocation and its role in necroptotic signalling ( Hildebrand et al . , 2014; Murphy et al . , 2013 ) . HER3 is able to bind ATP ( crystallised as PDB ID 3KEX , 3LMG ) , as well as the Src/ABL inhibitor Bosutinib ( PDB ID 4OTW ) ( Levinson and Boxer , 2014; Davis et al . , 2011; Jura et al . , 2009b; Murphy et al . , 2014; Shi et al . , 2010 ) . Considering the importance of HER3 as a conformational partner in the HER2-HER3 heterodimer , and the established importance of ATP-binding for complex formation in other pseudokinases , the role of nucleotide binding pocket occupation in HER3 function warrants investigation . Here , we have integrated the study of kinase-autonomous conformational effects of nucleotide binding pocket occupation with that of HER2-HER3 heterointeraction modalities and downstream proliferative phenotypes in response to drug treatment . We show that nucleotide pocket occupation in both HER2 and the pseudokinase HER3 is of great conformational importance for kinase domain heterodimerisation and subsequent proliferative signalling . In HER2+ breast cancer cells this leads to an unexpected synergy between the HER3 ligand NRG and the HER2 inhibitor lapatinib , by which their concomitant binding promotes proliferation in 2D and 3D culture systems . Lapatinib is able to promote heterodimerisation between the kinase domains of full-length HER2 and HER3 in cells . However , this dimer interface is different from the canonical active EGFR-family dimer , and it is necessary for the lapatinib/NRG combinatorial proliferative phenotype . Both the lapatinib-induced heterodimer and the cooperative proliferation effects depend strongly on the ability for the pseudokinase HER3 to bind ATP . Consistent with the model , occupying the pseudokinase HER3 with the Src/Abl inhibitor bosutinib stabilises the pseudokinase domain to the extent that it actually promotes HER2-HER3 heterodimerisation and downstream proliferation . The sensitivity of a variety of oncogene-addicted cell lines to small molecule kinase inhibitors can be counter-acted by the addition of growth factors ( Wilson et al . , 2012 ) . This includes the case of lapatinib-treated HER2+ breast cancer cell lines , where NRG is seen to mediate a rescue of drug toxicity ( Novotny et al . , 2016; Wilson et al . , 2012 ) . Using different experimental procedures , we have investigated further these competing effects of lapatinib and NRG on the proliferative behaviour of HER2+ breast cancer cells . In SKBR3 , BT474 , AU565 , and HCC1419 cells treated with a range of lapatinib concentrations for 72 hr , the addition of 10 nM NRG rescues the drug-induced cytotoxicity except at very high drug concentrations ( Figure 1a , Figure 1—figure supplement 1a–c ) . Interestingly , in the case of the SKBR3 , BT474 and AU565 cell lines , low concentrations of lapatinib ( ~40–400 nM ) are able to enhance proliferation in conjunction with 10 nM NRG by 25–30% compared to growth factor alone ( Figure 1a , Figure 1—figure supplement 1a–b ) . A partial response of this cooperative phenotype is observed in ZR75 and HCC1419 cells ( Figure 1—figure supplement 1c–d ) . This phenotype in SKBR3 cells , while observed previously , has gone unremarked ( Novotny et al . , 2016; Wilson et al . , 2012 ) . We corroborated our results with a cell counting assay , in which SKBR3 cells were treated for 72 hr with 250 nM lapatinib or vehicle ±10 nM NRG ( Figure 1b ) . The emergent effect of lapatinib plus NRG depends on lapatinib sensitivity . Two breast cancer cell lines with low lapatinib sensitivity , MCF7 and HCC1569 , show low inhibitor-growth factor cooperation ( Figure 1—figure supplement 1e–f ) . The growth phenotype in ZR75 may be partially explained by its HER4 expression , considering that NRG is also a ligand for HER4 ( Figure 1—figure supplement 1g ) . Although HER3 has been shown to bind lapatinib in vitro with very low affinity ( Kd = 5 . 5 μM ) ( Davis et al . , 2011 ) , the synergistic behaviour between lapatinib and NRG occurs in cells at a ~ 50 x lower dose than the in vitro Kd , indicating that any binding of lapatinib to HER3 would likely be minor under these conditions . Using a thermal shift assay ( TSA ) , which measures a shift in the thermal stability of a protein after ligand/inhibitor binding in vitro , we also show that lapatinib does not strongly bind HER3 as compared to ATP and a panel of other inhibitors ( Figure 2a , see further below ) . While EGF treatment rescued SKBR3 cells from the effects of low-concentration lapatinib treatment , synergistic growth effects such as those observed with lapatinib-NRG co-treatment were not observed for lapatinib-EGF co-treated SKBR3 or BT474 cells ( Figure 1—figure supplement 1h–i ) . Although NRG is also a growth factor ligand for HER4 , protein levels of HER4 in SKBR3 cells are very low ( Figure 1—figure supplement 1g ) . Additionally , lapatinib is a strong inhibitor of both EGFR and HER4 ( Davis et al . , 2011 ) . Taken together , these data seem to exclude a significant role for EGFR and HER4 in the synergistic growth observed for lapatinib-NRG co-treatment . Moreover , transient knockdown of HER3 with two different siRNA oligonucleotides shows a modest , but consistent reduction in the proliferative effect of ligand-inhibitor co-treatment , implicating HER3 as the relevant growth factor-binding receptor for this NRG response ( Figure 1—figure supplement 1j ) . The proliferative effects of lapatinib and NRG on SKBR3 cells were also observed in 3D spheroid cultures . As seen in 2D culture systems , in 3D spheroid culture the addition of NRG to lapatinib-treated cells rescues SKBR3 cells from lapatinib-induced cytotoxicity/cytostasis ( Figure 1c , Figure 1—figure supplement 1k–l ) . Lapatinib and NRG share a cooperative effect on the induction of proliferation in 3D spheroid cultures , where spheroid size is greater for inhibitor-ligand co-treatment conditions than for those treated with growth factor alone . The irreversible inhibitor neratinib binds the same inactive conformation as lapatinib and with similar binding affinity ( Davis et al . , 2011 ) . However , neratinib is an irreversible inhibitor and forms a covalent bond with HER2C805 , a residue conserved in EGFR and HER4 but not HER3 . Neratinib-NRG co-treatment did not show the synergistic proliferative phenotype observed with lapatinib-NRG , in either a cell counting assay , or in 3D spheroid formation ( Figure 1—figure supplement 2a–d ) . Similarly , the induction of HER2 and HER3 phosphorylation seen in western blot analysis of lapatinib-NRG co-treated 3D spheroids was absent in neratinib-NRG co-treatment ( Figure 1—figure supplement 1l , Figure 1—figure supplement 2d ) . This indicates that the proliferative phenotype observed for lapatinib is likely to necessitate a dynamic , reversible inhibitor binding . Collectively , the data from both 2D and 3D cultures show that there is a counterintuitive synergy between the HER2 inhibitor lapatinib and the HER3 ligand NRG in driving the proliferation of SKBR3 cells . This prompted us to examine the potential for novel allosteric regulation of HER2-HER3 heterotypic interactions by both ligand and inhibitors . To study the effects of ATP binding on HER3 function , we aimed to both stabilise and destabilise the pseudokinase nucleotide-binding pocket . This would allow us to investigate the importance of the structural role that nucleotide binding pocket occupation has been shown to play in several ( pseudo ) kinases . To separate the structural and trace catalytic roles that ATP-binding could fulfill in HER3 , we used the ATP-competitive Src/Abl inhibitor bosutinib , which has been shown to bind strongly to HER3 but not to other EGFR family members ( Levinson and Boxer , 2014; Davis et al . , 2011 ) . We compared bosutinib to a small panel of EGFR family inhibitors as well as an additional Src inhibitor , dasatinib , and measured HER3 thermal stability by TSA ( Figure 2a , Figure 2—figure supplement 1a ) . In line with previous observations , we confirmed that HER3 strongly binds bosutinib . Significantly , lapatinib was not able to provide a noticeable thermal shift , which corresponds to previously published results indicating HER3 does not bind lapatinib with high affinity ( Davis et al . , 2011 ) . While lapatinib was able to confer strongly increased thermal stability to HER2 , bosutinib was not ( Figure 2b ) . This is in line with previously published data that indicates HER2 is not a strong bosutinib binder ( Davis et al . , 2011 ) . We hypothesised that bosutinib might be able to aid proliferation in a cellular context by stabilising the nucleotide binding pocket of HER3 and helping sustain dimer formation , analogous to vemurafenib-bound behaviour of BRAF . In a 2D proliferation assay , SKBR3 cells treated with bosutinib over 72 hr show a dose dependent induction of proliferation without additional NRG stimulation ( Figure 2—figure supplement 1b ) . This proliferative effect is sustained in eight-day treatments in 3D spheroid cultures ( Figure 2c , Figure 2—figure supplement 1d , e ) . The ability of bosutinib to induce SKBR3 cell proliferation appears to be an EGFR-family mediated event , as lapatinib treatment can curtail its effects in a dose-dependent manner ( Figure 2—figure supplement 1e ) . In order to destabilise the HER3 nucleotide binding pocket we made the triple mutant HER3KGG . HER3K742 was mutated to methionine to hinder ATP α-phosphate coordination , which by itself has been shown to reduce HER3 mant-ATP binding affinity ( Shi et al . , 2010 ) . To obstruct ATP binding further , double aspartates were introduced in the glycine-rich loop ( HER3G716D/G718D ) to mimic the pseudokinase-specific aspartate residue observed in the glycine-rich loop of VRK3 ( Scheeff et al . , 2009 ) , adding a negative charge in the area where the ATP phosphates would normally sit . Introduction of this ATP-binding deficient HER3KGG mutant into MCF7 cells shows abrogation of ligand-induced trans-phosphorylation of HER3 by HER2 ( Figure 2d ) . SKBR3 cells ectopically expressing HER3wt or HER3KGG show a differential proliferative behaviour upon lapatinib ±NRG treatment . This indicates a critical role for HER3 ATP binding in order to sustain inhibitor-growth factor cooperative proliferation ( Figure 2—figure supplement 1f ) . The bosutinib binding of HER3wt , HER3KGG , and the proposed drug de-sensitised HER3T787M ( Levinson and Boxer , 2014; Dong et al . , 2017 ) , was investigated using an in-cell thermal shift assay ( CETSA ) ( Jafari et al . , 2014; Reinhard et al . , 2015 ) . Where wild type HER3 showed increased thermal stability in cells in the presence of 50 nM bosutinib , HER3KGG did not ( Figure 2—figure supplement 1g ) . Ectopic expression of wild type HER3 , but not HER3KGG or HER3T787M , enhances bosutinib-mediated proliferation , indicating this behaviour is driven by bosutinib binding to HER3 directly ( Figure 2e ) . Both HER3KGG and HER3T787M showed normal localization to the plasma membrane , as measured by flow cytometry , indicating that these mutations did not compromise the receptor and its traffic to the plasma membrane ( Figure 2—figure supplement 2 ) . The HER3KGG and bosutinib results indicate that nucleotide pocket occupation in HER3 is essential for its ability to sustain a proliferative signalling pathway under distinct circumstances: in the acute response to growth factor , in promoting ligand-inhibitor cooperative proliferation and even after treatment with a HER3-binding inhibitor . This indicates a critical structural role for HER3 ATP-binding pocket occupation in its ability to sustain heterointeractions and proliferation . Considering the proliferative effects observed with the HER3-binding inhibitor bosutinib , our results also suggest that any residual transferase activity HER3 retains does not appear to be important in these responses in vivo unless we invoke a hit-and-run mechanism of action for bosutinib on HER3 which would seem unlikely . The stability conferred to a protein kinase by small molecule inhibitor binding has been shown to play an important role in the promotion of protein–protein interactions . We investigated the potential role of lapatinib to similarly promote HER2-HER3 heterodimerisation by stabilising particular protein conformations in HER2 using a FRET-FLIM approach . We measured drug-induced heterodimerisation of HER2 and HER3 , as we have done previously in the case of drug-induced dimerisation of the EGF receptor ( Bublil et al . , 2010; Coban et al . , 2015 ) . At endogenous protein levels in SKBR3 cells , we observe lapatinib-driven HER2-HER3 heterodimerisation to levels similar to those seen with NRG ( Figure 3a ) . Interestingly , the lapatinib-induced dimerisation occurs in the absence of exogenously added NRG , indicating a HER2-HER3 dimer that is driven primarily through intracellular domain interactions . MCF7 cells , which express low levels of endogenous HER2 and HER3 compared to SKBR3 , also display lapatinib-induced heterodimerisation of ectopically expressed GFP-HER2wt and HA-HER3wt ( Figure 3b ) . As discussed above , occupation of the nucleotide binding pocket in HER3 is of importance for its ability to sustain proliferation . This is also reflected in the case of lapatinib-induced heterodimer formation , where the introduction of the nucleotide pocket compromised HER3KGG mutant strongly disrupts inhibitor-promoted heterodimerisation ( Figure 3c ) . In line with the proliferative effects described above , bosutinib was also able to directly promote heterodimerisation between HER2 and HER3 ( Figure 3d ) . Using stochastic optical reconstruction microscopy ( STORM ) , we analysed receptor clustering in SKBR3 cells . Treatment with either NRG , lapatinib , or bosutinib showed a shift in cluster population size compared to control , implying the formation of higher-order oligomers rather than dimers ( Figure 3e , f ) . The exact HER2-HER3 stoichiometry in these drug-treated oligomers remains elusive , because these experimental conditions allowed us to count only cluster size for either HER2 or HER3 , not both at the same time . Therefore , it is expected that the observed HER3 clusters also contain uncounted HER2 receptors , and vice versa , as evident in the FRET-FLIM data . The active signalling dimer in the EGFR family adopts an asymmetric orientation , in which there is a distinct division of labour in the activator-receiver pairing . One kinase ( the activator kinase ) does not phosphorylate substrates , but binds in a way that helps its heterodimerisation partner ( the receiver kinase ) in adopting an active conformation . The receiver kinase is then capable of substrate phosphorylation . Originally described for EGFR homodimerisation , and similar to the cyclin/CDK binding mode ( Jeffrey et al . , 1995; Zhang et al . , 2006 ) , this canonical active dimerisation interface has been reported across the EGFR family including the heterodimerisation of HER3 , which can only perform the activator role ( Jura et al . , 2009b; Littlefield et al . , 2014; van Lengerich et al . , 2017 ) . Mutations that disrupt this active interface in both the activator and receiver partner kinases are well-documented and are schematically highlighted ( Figure 4a , Figure 4—video 1 ) . In the case of the active , activator/receiver interface , HER3 buttresses the inward orientation of the HER2 α-C helix , leaving no space for the HER2 α-C helix to adopt the ‘out’ orientation characteristic of the inactive conformation . We modelled the potential effects of HER2 α-C helix positioning on lapatinib binding to test whether canonical activator/receiver orientation ( in which the HER2 α-C helix is pushed inwards ) would give sufficient space to still accommodate lapatinib . Our modelling showed that , for a HER2 α-C helix in the active , ‘in’ position , lapatinib binding results in a potential steric clash with HER2E770/M774 ( Figure 4—figure supplement 1a , b ) . A general decrease of the nucleotide binding pocket volume from 756 Å3 to 232 Å3 ( calculated using SURFNET v1 . 5 ( Laskowski , 1995 ) ) supports these predictions . To further test whether the lapatinib-induced HER2-HER3 is adopting the canonical activator/receiver orientation , we used FRET-FLIM to investigate lapatinib-induced dimer formation . The I714Q mutation in HER2 , which renders the receptor receiver-impaired , disrupted the lapatinib-driven HER2-HER3 association , indicating it is retained in the lapatinib-induced dimer interface ( Figure 4b ) . However , the reciprocal activator-impaired mutation in HER3 ( HER3V945R ) did not disrupt lapatinib-mediated heterodimerisation , although it efficiently suppressed the canonical active dimer after ligand-induced heterodimerisation ( Figure 4c ) . It is surmised that the inhibitor binding is able to robustly induce a heterodimer between HER2 and HER3 , which is distinct from the canonical active heterodimer induced after growth factor stimulation . The orientation of this non-canonical lapatinib-driven heterodimer retains HER2I714 in the dimer interface , giving us a starting point for in silico molecular modelling to investigate potential dimer conformations distinct from the well-described active dimer . In the case of type II kinase inhibitors such as lapatinib , the inhibitor stabilises an inactive conformation of the kinase domain , where the α-C helix is tilted outwards . As HER3 lacks the conserved glutamate residue in the α-C helix , HER3K742 is unable to form the salt bridge normally observed in active kinase domain structures ( Huse and Kuriyan , 2002 ) . The HER3 ATP-bound conformation therefore does not show a classical active conformation with the α-C helix tilted inward ( Jura et al . , 2009b; Shi et al . , 2010 ) , but instead resembles the inactive conformation seen in kinases bound to type II inhibitors such as lapatinib . Because lapatinib-bound HER2 and ATP-bound HER3 adopt similar conformations , there is a possibility that the lapatinib-induced , inactive dimer is oriented symmetrically . In the crystal lattices of EGFR and HER3 kinase domains , two different symmetrical interaction interfaces have been observed ( Jura et al . , 2009a; Jura et al . , 2009b ) . We used molecular modelling to investigate the potential for HER3 and lapatinib-bound HER2 to adopt either of these conformers ( Figure 5a–b , Figure 5—figure supplement 1a–b ) . HER2I714 is present in the interaction interface of both the EGFR-like , staggered orientation , as well as in the head-to-head , HER3-like orientation . This falls in line with the FRET-FLIM data in Figure 4 that suggests the retained presence of the HER2I714 residue in the lapatinib-induced dimer interface . On the basis of these models , we designed pairs of mutations in HER2 that would exclusively disrupt one of the potential heterodimer orientations ( Figure 5—figure supplement 1 , Figure 5—video 1 and 2 ) . For the EGFR-like , staggered dimer we substituted two hydrophobic residues on HER2 with two positively charged residues , HER2I748R/V750R , which should lead to repulsion from the positively charged residues , K998 and K999 , lying on the HER3 side of the interface . Likewise , for the HER3-like , head-to-head dimer , we predicted that the HER2N764R/K765F mutant would disrupt the dimerisation interface . The substitution of an asparagine residue ( HER2N764 ) with a positively charged arginine should lead to repulsion from a positively charged HER3 residue ( HER3R702 ) , lying within a radius of 4 Å and opposite to HER2N764 , therefore causing severe disruption of the HER3-like dimer interface . Furthermore , the substitution of a lysine residue ( HER2K765 ) with a bulky , hydrophobic residue such as phenylalanine should generate clashes at this HER2-HER3 interface . These dimer interface mutants were introduced into our FRET-FLIM assay for investigation of the lapatinib-induced heterodimerisation conformer ( Figure 5c ) . The HER2N764R/K765F mutant disrupted heterodimerisation upon lapatinib binding , whereas HER2I748R/V750R showed no difference in heterodimer formation . These mutational FRET/FLIM data is consistent with our model that the lapatinib-induced HER2-HER3 heterodimer adopts a symmetrical , head-to-head orientation , similar to the one observed in the HER3 kinase domain crystal lattice ( Jura et al . , 2009b ) ( Figure 5b ) . Having presented modelling and FRET/FLIM data consistent with an orientation of the lapatinib-induced HER2-HER3 dimer being distinct from the active activator/receiver dimer interface , we sought to identify which type of HER2-HER3 interaction caused the NRG-lapatinib co-stimulatory growth observed in 2D proliferation assays . In these assays , we did not ectopically introduce the HER2N764R/K765F mutant because , firstly , it might also disrupt the active , asymmetrical HER2-HER3 heterodimer interface and secondly , SKBR3 cells have vast numbers of endogenous HER2 receptors that would hinder analysis of the behaviour of ectopically expressed HER2N764R/K765F . Instead we identified HER3L700F as the reciprocal mutant to HER2N764R/K765F ( Figure 6a , Figure 6—video 1 ) . We introduced HER3L700F into SKBR3 cells to investigate the role of the head-to-head , symmetric dimer interface in the lapatinib-NRG synergistic proliferation described above . While the HER3V945R active dimer mutant did not disrupt drug-growth factor cooperative proliferation , the HER3L700F mutant did ( Figure 6d–e ) . Both HER3L700F and HER3V945R were expressed on the cell surface , as measured by flow cytometry ( Figure 2—figure supplement 2 ) . Combined , this indicates that the inhibitor-induced heterodimer of HER2 and HER3 is consistent with a head-to-head , symmetrical conformation , and it plays an important role in the synergistic proliferative effects of lapatinib and NRG . Although this conformation has been described from the HER3 kinase domain crystal lattice ( Jura et al . , 2009b ) , to our knowledge it is the first time a functional role has been ascribed to heterodimers consistent with this interface in cells . The conformational dynamics of HER2-HER3 heterodimerisation are an important consideration for evaluating existing and future targeted therapy intervention strategies against HER2+ breast cancer and other HER family driven cancers . Here , we show that the HER2 inhibitor lapatinib is paradoxically able to promote proliferative behaviour in HER2+ breast cancer cells when administered in the presence of the HER3 ligand NRG . The synergy between growth factor and inhibitor requires an intricate , multi-step cascade of conformational events . Lapatinib itself is able to promote heterodimerisation between the kinase domains of HER2 and HER3 , stabilising an orientation consistent with a symmetric , head-to-head kinase domain heterodimer that is distinct from the canonical , asymmetric , head-to-tail active kinase domain orientation that occurs throughout the EGFR family . An analogous interface has previously been observed in the HER3 kinase domain crystal lattice ( Jura et al . , 2009b ) ; here , we have provided modelling and cellular evidence of a heterodimer with an interface consistent to the one observed in the HER3 kinase domain crystal lattice . Sequestering HER2 and HER3 in these inactive , lapatinib-bound heterodimers was of benefit to NRG-mediated proliferative signalling . Our results , in which inhibitor binding drives dimer formation that boosts signalling and proliferation , shows some parallels with the inhibitor-induced signalling phenotypes in the RAF-family ( Eyers et al . , 1998; Hatzivassiliou et al . , 2010; McKay et al . , 2011; Poulikakos et al . , 2010; Thevakumaran et al . , 2015 ) While the FRET-FLIM analysis of the lapatinib-induced dimerisation was not able to differentiate between heterodimers or higher order oligomers , our clustering data shows that lapatinib is likely to induce higher order oligomers . Because of the modelled symmetrical nature of these lapatinib-induced dimers , in which both lapatinib-bound HER2 and HER3 would be conformationally available as ‘activator’ receptors for additional oligomerization partners , it is not inconceivable they may act as nucleation points for larger oligomeric signalling platforms . Such signalling arrays , in which mutual cooperativity increases signaling output , have been proposed for EGFR oligomers ( Huang et al . , 2016 ) . The addition of ligand potentially causes rearrangements within these platforms through the ligand-induced conformational ballet of multi-level interactions between the various extracellular and intracellular domains of EGFR family receptors ( reviewed in ( Lemmon et al . , 2014 ) ) . The formation of lapatinib-induced oligomeric platforms may facilitate a transition into active signalling heterodimers upon ligand binding , due to the availability of dimerisation partners in immediate proximity within these drug-induced oligomer platforms . Both the lapatinib-induced HER2-HER3 heterodimerisation and the downstream lapatinib-NRG synergistic effects on proliferation depended on the ability of HER3 to bind ATP . Although usually classified as a pseudokinase , HER3 has been shown to retain a measure of autophosphorylation activity ( not transphosphorylation ) under specific circumstances ( Shi et al . , 2010 ) . We show HER2-HER3 heterodimerisation and downstream proliferative effects can be elicited by the addition of the HER3-binding inhibitor bosutinib , indicating that nucleotide binding pocket occupation performs a structural role that is critical to HER3 function , and apparently independent of any retained catalytic activity . Observing increased heterointeractions and cellular proliferation due to inhibition of an activity-deficient kinase is a strong indication of the importance of ATP-binding in certain pseudokinases , and the necessity of pocket-occupied structural conformers in sustaining protein–protein interactions and subsequent downstream signalling output . Because of the importance of HER3 in HER2-targeted therapy resistance , its conserved ATP binding raised the possibility of targeting HER3 with ATP-competitive kinase inhibitors . Our data show , however , that stabilisation of the HER3 kinase domain with an ATP-competitive kinase inhibitor can have a stimulating effect on HER2+ breast cancer cell proliferation . This indicates that the development of small molecule targeted therapy against HER3 for use in HER2+ breast cancer needs to be directed away from stabilising the HER3 ATP binding pocket occupied conformer and rather towards stabilising the apo , inactive conformer . An exception to this might be the development of irreversible , adamantane-linked inhibitors of HER3 that target the receptor for proteosomal degradation ( Xie et al . , 2014 ) . The substantial effect that lower doses of lapatinib have on proliferation in the presence of growth factor may have an impact on the establishment of lapatinib-resistance in vivo . This is in accordance with the observation from xenograft models that resistance occurs much more readily if lapatinib is administered continuously at low doses than if it’s used intermittently at high dose ( Amin et al . , 2010 ) . Increased production of growth factors ( including NRG ) is a well-described resistance mechanism against HER2-targeted therapy ( reviewed in [Claus et al . , 2014] ) . NRG production by the microenvironment has also been shown to play a role in metastatic spread of ovarian cancer cells that express high levels of HER3 ( Pradeep et al . , 2014 ) . High expression levels of NRG in HER2+ breast cancer patients showed a strong correlation with disease recurrence ( Xia et al . , 2013 ) . Several somatic mutations in HER3 observed in cancer fall within the extracellular domain and have a potential effect on ligand-binding affinity ( Jaiswal et al . , 2013 ) . These mutations may exacerbate the inhibitor-growth factor synergistic behaviour reported here . Our results provide a potential molecular mechanism for the disappointing results observed in a recent Phase III study of lapatinib used in an adjuvant setting ( ALTTO trial ) ( Piccart-Gebhart et al . , 2016 ) . The lapatinib-only arm of this study was terminated prematurely , and the effects observed in the adjuvant setting for both lapatinib-trastuzumab co-treatment and trastuzumab treatment followed by lapatinib were not significant . These clinical results indicate there are complicating factors in hindering lapatinib efficacy in patients , which may involve the expression levels of HER3 and NRG stimulation by a complex tumour microenvironment . The complex relationships between distinct protein conformation dynamics , formation of oligomeric assemblies , the availability of ligand , and the various effects on downstream signalling all need to be considered when applying targeted therapy to avoid potentially unexpected enhanced cancer cell proliferation after inhibitor treatment . NRG1 was purchased from PeproTech . Lapatinib was a kind gift from Professor György Kéri ( Vichem Chemie Research Ltd Hungary ) . Bosutinib was purchased from LC Labs . Total HER2 , HER3 , PKB , HER2 pY877 , HER3 pY1289 , PKB pS473 and ERK1/2 pT202/pY204 antibodies were purchased from Cell Signaling Technology , anti-α-tubulin from Sigma , total ERK1/2 from Merck , and Alexa Fluor-488 conjugated anti-HER3 antibody from R&D systems . MCF7 and ZR75 cells were cultured in DMEM supplemented with 10% FCS , SKBR3 cells were grown in McCoys medium supplemented with 10% FCS . BT474 , AU565 , HCC1419 , and HCC1569 cells were grown in RPMI with 10% FCS . For BT474 cells 10 μg/ml bovine insulin was included in the culture medium . Cells were transfected with plasmid DNA using FuGENE 6 , FuGENE HD ( Roche ) , or Lipofectamine LTX ( Thermo Fisher Scientific ) according to the manufacturer’s protocol . All cell lines were sourced from the Francis Crick Institute's Cell Services facility , where they were tested negative for mycoplasma and authenticated via STR profiling . For 2D proliferation assays , cells were plated at 1 × 104 cells/well in a 96-well plate . The following day they were subjected to treatment for 72 hr , followed by addition of CellTiter-Glo reagent ( Promega ) and measured on an EnVision plate reader ( Perkin Elmer ) . CellTiter-Glo data were normalised to the growth factor-null/inhibitor-null untreated control . This caused some growth factor-treated plots to start at above-baseline levels , which is an indication of the proliferative effect that growth factor treatment had in these cells . For 3D spheroid assays , 3 × 103 cells were plated in a 96-well , round-bottom , ultra-low attachment plate ( Corning ) in the presence of 1% Matrigel ( Corning ) . After three days of growth , an equal volume of 2x media containing treatment conditions was added and refreshed every three days for a total of eight days of treatment . Phase contrast images were taken using a Zeiss Axiovert 40 CFL microscope with a Zeiss 5x A-plan objective and analysed using ImageJ . SKBR3 cells were transfected with RFP-HER3 mutants for 48 hr . Cells were pre-treated with 0 . 5 mM EDTA to facilitate removal from the substrate and stained for HER3 extracellular expression using Alexa Fluor-488 conjugated anti-HER3 antibody ( R&D systems , clone 66223 ) as per manufacturer’s instructions . Briefly , cells were blocked using mouse IgG ( Santa Cruz Antibodies ) for 15 min at room temperature , followed by incubation with conjugated antibody for 30 min at room temperature in the dark . Cells were washed in PBS , 0 . 5% BSA , 0 . 1% sodium azide three times before flow cytometric analysis using a BD Fortessa instrument ( BD ) . Results were analysed using the Flo-Jo software . Fluorescence resonance energy transfer ( FRET ) is used to quantitate direct protein–protein interactions and post-translational modifications . Processing of cells for FRET determination by FLIM has been previously described ( Barber et al . , 2009; Parsons and Ng , 2002 ) . FLIM was performed using time-correlated single-photon counting ( TCSPC ) with a multiphoton microscope system as described previously ( Peter et al . , 2005 ) . For experiments measuring endogenous protein , FRET pairs were Cy5-conjugated anti-HER2 IgG , and Alexa546-conjugated anti-HER3 IgG . For exogenous protein measurements , FRET pairs were HER2-GFP and HER3-HA with an anti-HA IgG , tagged with a Cy3 fluorophore . FRET efficiency between the donor and acceptor bound proteins was calculated with the following equation in each pixel and averaged per cell: FRET eff = 1-tau ( DA ) /tau ( control ) where tau ( DA ) is the lifetime displayed by cells co-expressing the donor and acceptor , whereas tau ( control ) is the mean donor ( GFP ) lifetime , measured in the absence of the acceptor . We modelled the HER2-HER3 dimer by comparative homology modelling using a multiple templates approach . The active , asymmetric HER2-HER3 dimer was modelled using the crystal structure of the active EGFR kinase domain ( PDB ID 2GS2 ) ( Zhang et al . , 2006 ) and one chain of the crystal structure of the HER3 homodimer ( PDB ID 3KEX ) ( Jura et al . , 2009b ) as templates . To build the EGFR-like , inactive , symmetric dimer we have used the crystal structure of the EGFR homodimer ( PDB ID 3GT8 ) ( Jura et al . , 2009a ) , the crystal structure of EGFR complexed with lapatinib ( PDB ID 1XKK ) ( Wood et al . , 2004 ) and only one chain of the crystal structure of the HER3 homodimer ( PDB ID 3KEX ) ( Jura et al . , 2009b ) . To build the HER3-like dimer , we have used the HER3 homodimer structure ( PDB ID 3KEX ) ( Jura et al . , 2009b ) , the crystal structure of EGFR lapatinib-bound ( PDB ID 1XKK ) ( Wood et al . , 2004 ) and the crystal structure of the inactive EGFR AMP-PNP bound ( PDB ID 2GS7 ) ( Zhang et al . , 2006 ) . The sequence alignment used to build the model has been created by using PRALINE with the homology-extended alignment strategy ( Simossis et al . , 2005 ) . We generated 200 three-dimensional models using the MODELLER package ( Sali and Blundell , 1993 ) . The selected models were chosen on the basis of the MODELLER objective function's DOPE score . The volume of the HER2 ATP binding pocket was calculated with the SURFNET 1 . 5 package ( Laskowski , 1995 ) , where the cavity regions in a protein are built up by fitting a probe sphere of 1 . 4 Å3 into the spaces between atoms . The structural alignment was performed using the multi-seq tool of the VMD 1 . 9 . 1 package ( Humphrey et al . , 1996 ) , and measurement of interaction surface buried residues was performed using POPScomp ( Kleinjung and Fraternali , 2005 ) . STORM imaging and cluster analyses are described in more detail at Bioprotocol ( Roberts et al . , 2018 ) . SKBR3 cells were treated with either 14 nM Lapatinib or 41 nM Bosutinib . HER2 and HER3 Affibody ligands were used to label the non-activated states of the receptors ( HER2 from Affibody Inc . and plasmid encoding the HER3 affibody was a gift from John Löfblom , protein made in house and shown to bind specifically to HER3 receptors ) and NRG-β1 ( Peprotech ) was used to stimulate the cells . The conjugation of dyes ( Invitrogen ) to HER2 and HER3 ligands was done in house and the ratio of dye:ligand was confirmed to be ~1:1 . The NRG-dye conjugate has been shown to be as active as the unlabelled protein . We incubated cells in 100 nM HER2Affibody-Alexa488 + 50 nM HER3Affibody-Alexa647 or 100 nM HER2Affibody-Alexa488 + 10 nM NRG-Alexa647 ±drug for 1 hr . Cells were chemically fixed using 4% paraformaldehyde ( EMS solutions ) +0 . 5% glutaraldehyde ( Sigma-Aldrich ) diluted into ice-cold PBS . Samples were imaged using a Zeiss Elyra super-resolution microscope to stochastically excite the Alexa488 and Alexa647 fluorophores bound to the receptors in the cells and to image single molecules . Imaging was done in TIRF mode using a 100x oil immersion objective lens . We used a 405 nm laser line to aid fluorophore blinking and 488 nm or 640 nm laser lines to excite the fluorophores , alternating the lasers to image the two receptors independently every 300 frames , over a total of ~10 , 000 frames . The exposure time was 20 ms . A minimum of two replicates of each sample were imaged generating at least 12 regions ( 25 . 6 μm x 25 . 6 μm ) covering at least one cell per region . The Zen software localised the single molecule spots in the cells , a threshold was set to discard background spots and the co-ordinates of the positive localisations ( typically 30 , 000 + for HER2 and 5 , 000 + for HER3 per region ) were passed into the Bayesian cluster identification algorithm ( Rubin-Delanchy et al . , 2015 ) The clustering algorithm expects the background and clusters to be uniformly distributed over a rectangular ROI ( Rubin-Delanchy et al . , 2015 ) . The analysed images mainly showed single cells . Of interest are the HER2 and HER3 receptors in the cell membrane , which were visible as a circular shape . In order to conform with the prerequisites of the clustering algorithms , rectangular regions have been manually selected that tightly cover the cell membrane using the most suitable angles ( assessed by visual inspection ) . The whole cell membrane has been covered in this way . The data selected by these regions have been rotated so that the sides of the rectangles became parallel to the coordinate axis . The result was used as input for the clustering algorithm and the algorithm was applied as described by the protocol . The complete lists of molecules per cluster that have been produced by the algorithm were used for the presentation . The baculoviral HER3 kinase domain construct was kindly provided by Prof . Mark Lemmon , University of Pennsylvania . Sf21 cells at 1 × 106 cells/ml were infected with P3 virus ( 7 × 107 pfu/ml ) at an MOI of 1 . 0 and allowed to grow for three days . The cells were lysed in lysis buffer containing protease inhibitors , 1 mM DTT and 2 mM BME . The lysate was clarified by centrifugation and incubated with NiNTA resin ( Qiagen ) for 30 mins at 4°C , after which the resin was washed extensively with buffer containing 50 mM Hepes ( pH 7 . 6 ) , 300 mM NaCl , 2 mM BME , 5% glycerol , 10 mM imidazole . HER3 was eluted in the same buffer with 200 mM imidazole added . Each elution was centrifuged at 10 , 000 rpm to remove any precipitate or resin and applied to a S200 gel filtration column in 50 mM HEPES ( pH 7 . 6 ) , 300 mM NaCl , 2 mM BME , 2 . 5% glycerol . Thermal shift assays were carried out as described in ( Niesen et al . , 2007 ) . Briefly , in a 96-well RT-PCR plate ( Life Technologies ) 1 μg HER3 kinase domain/well was incubated with 1 μM inhibitor or 200 μM ATP/10 mM MgCl2 ( as indicated ) for 30 mins at 4°C in the presence of Sypro Orange dye ( Sigma ) . HER2 TSA experiments were performed in a 384-well RT-PCR plate ( Thermo Fisher Scientific ) . 0 . 5 μg of HER2 kinase domain/well was incubated with 1 μM lapatinib , 1 μM bosutinib , or 200 μM ATP/10 mM MgCl2 for 20 mins at 4°C . HER3 measurements were taken on an Applied Biosystems 7500 Fast Real-Time PCR machine , and HER2 measurements on an Applied Biosystems Quant Studio 7 PCR machine . Data were trimmed and a Boltzmann sigmoidal curve fitted in GraphPad Prism 6 . The inflection point of the Boltzmann sigmoidal was taken as the Tm . Thermal shift ΔTm values were obtained by subtracting the Tm value of the kinase domain alone control . Cells were plated at 0 . 5 × 105 cells/well in 24-well plates . Cells were lysed in 1x sample buffer ( containing 1 mM DTT ) , sonicated and centrifuged . After centrifugation , the lysates were subjected to SDS-PAGE and analyzed by western blotting . CETSA was performed with COS7 cells transfected with HER3wt-RFP , HER3T787M-RFP or HER3KGG-RFP plasmids as described in ( Jafari et al . , 2014; Reinhard et al . , 2015 ) . Briefly , COS7 were treated with DMSO or 50 nM bosutinib for 1 hr at 37°C . Cells were washed with PBS , detached and washed again twice with cold PBS . Cell pellets were resuspended in cold PBS with protease inhibitors ( Roche ) and 100 μl of each cell suspension was transferred into 0 . 2 ml PCR tubes . PCR tubes were heated for 3 min at 42°C or 50°C in a thermal cycler ( DNA Engine DYAD , MJ research , Peltier thermal cycler ) and incubated at room temperature for 3 min . Tubes were then immediately transferred onto ice , 35 μl of cold PBS 1 . 4% NP-40 with protease inhibitors were added and tubes were snap-frozen . Samples were then subjected to two freeze-thaw ( at 25°C ) cycles and cell lysates were centrifuged at 20 , 000 g for 1 hr at 4°C . Supernatants were carefully removed and analysed by western blot .	Around 20% of breast cancers are caused because cells have too many copies of a receptor protein called HER2 on their surface . HER2 is responsible for telling the cell to divide . Cells with too many of these receptors – and breast cancer cells can have up to 1000 times too many – divide uncontrollably . This causes the cancer to grow . Several successful anti-cancer drugs , such as Herceptin and Kadcyla , are used in the clinic to block the signals produced by HER2 . Other drugs called kinase inhibitors prevent HER2 from building its faulty signals . However , a particular kinase inhibitor called lapatinib was not as successful in clinical trials as the medical community had hoped . Kinase inhibitors can have unexpected effects . While they can block specific signals in a cell , they can sometimes also cause new types of signals . Could this be one of the reasons behind the disappointing clinical trial results for lapatinib ? By performing experiments on breast cancer cells grown in the laboratory , Claus , Patel et al . found that lapatinib can counterintuitively boost the growth of breast cancer cells . This occurs because lapatinib causes HER2 receptors to cluster together like a daisy chain along with another protein receptor of the same family , called HER3 . These chains are primed to rapidly respond to a molecule called neuregulin , a growth factor that is commonly associated with breast cancer . The results presented by Claus , Patel et al . indicate that a particular subset of breast cancer patients – those whose cancer cells do not increase production of HER3 receptors – might better respond to lapatinib than others . The insights gained into what happens to HER2 when you try to block it should also influence the design of new drugs that target either HER2 or HER3 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2018	Inhibitor-induced HER2-HER3 heterodimerisation promotes proliferation through a novel dimer interface
The super-resolution microscopy called RESOLFT relying on fluorophore switching between longlived states , stands out by its coordinate-targeted sequential sample interrogation using low light levels . While RESOLFT has been shown to discern nanostructures in living cells , the reversibly photoswitchable green fluorescent protein ( rsEGFP ) employed in these experiments was switched rather slowly and recording lasted tens of minutes . We now report on the generation of rsEGFP2 providing faster switching and the use of this protein to demonstrate 25–250 times faster recordings . While for many decades lens-based optical microscopy could not resolve features finer than half the wavelength of light ( λ/2 > 200 nm ) , the recent development of fluorescence nanoscopy or super-resolution imaging methods boosted its resolution potential ( Hell , 2003 , 2009; Huang et al . , 2010 ) . In all these methods , the diffraction barrier is overcome by employing a fluorophore transition between two states , typically a fluorescent on- and a non-fluorescent off-state ( Hell , 2009 ) in order to separate neighboring features . The method called RESOLFT differs from stochastic single fluorophore on-off-switching methods by the fact that a doughnut or a line pattern is scanned across the sample , determining at any point in time the nanosized coordinate range where the fluorophores are in the on-state . Using long lifetimes of the on- and off-states reduces the light intensities required for optical switching by orders of magnitude over the related STED approach , making RESOLFT attractive for extended live-cell or large area imaging . Although it was suggested almost a decade ago ( Hell , 2003; Hell et al . , 2003 ) and its viability shown in principle ( Hofmann et al . , 2005; Dedecker et al . , 2007; Schwentker et al . , 2007 ) the wider applicability of RESOLFT nanoscopy to biological imaging has been demonstrated only recently ( Brakemann et al . , 2011; Grotjohann et al . , 2011; Rego et al . , 2012 ) . The reason is that fluorescent protein based RESOLFT critically relies on reversibly switchable fluorescent proteins ( RSFPs ) providing many on-off cycles , such as the recently designed rsEGFP ( Grotjohann et al . , 2011 ) . Unfortunately , in these demonstrations , the still relatively slow switching kinetics of rsEGFP entailed pixel dwell times of 10–20 ms at switching light intensities of ~1 kW/cm2 . Hence images of 10 µm × 10 µm in size and a pixel step size of 20 nm required recording times of about an hour . In this study we describe the generation , characterization and application of a novel RSFP , namely rsEGFP2 , which facilitates faster switching and , together with modifications of the initial switching scheme , enables 25–250 faster RESOLFT recordings in living cells . We observed that already the exchange of a single amino acid in the chromophore of the photostable and widely-used enhanced green fluorescent protein ( EGFP ) ( Patterson et al . , 1997; Tsien , 1998 ) , namely replacing threonine 65 by alanine , transformed EGFP into a fast switching RSFP , which exhibited a ‘negative switching mode’ . ‘Negative switching’ RSFPs switch from the on- into the off-state at the wavelength , here ~480 nm , that is used to elicit fluorescence . The switching from the off-state into the on-state is achieved by irradiation with a different wavelength , here ~405 nm . Protein solutions of EGFP ( T65A ) exhibited a high residual fluorescence ( ~50% of the on-state signal ) when the protein solution was switched into the off-state . However , analysis of EGFP ( T65A ) demonstrated fast switching and low switching fatigue , indicating that the different chromophore ( an Ala-Tyr-Gly chromophore instead of a Thr-Tyr-Gly chromophore ) might provide additional possibilities for the generation of fast switching RSFPs . Hence , we also introduced the mutation A206K to ensure that the protein is a true monomer ( Zacharias et al . , 2002 ) and introduced , alone or in combination , the four mutations that discriminate rsEGFP from EGFP . We screened this collection of EGFP variants for those exhibiting a high resistance against switching fatigue in combination with fast switching at light intensities previously used for RESOLFT ( few kW/cm2 ) . We found that introducing two mutations was sufficient to transform EGFP ( T65A , A206K ) into a novel RSFP with improved properties when compared to rsEGFP ( Figure 1 ) . We named this new variant ( EGFP[T65A , Q69L , V163S , A206K] ) ( Figure 1—figure supplement 1 ) reversibly switchable EGFP2 ( rsEGFP2 ) . Note that rsEGFP has a TYG chromophore , whereas rsEGFP2 has an AYG chromophore . 10 . 7554/eLife . 00248 . 003Figure 1 . Characteristics of rsEGFP2 . ( A ) Absorption ( black dashed line ) , excitation ( red dotted line ) , and emission ( green solid line ) spectra of rsEGFP2 in its equilibrium state at pH 7 . 5 . ( B ) Switching curves of rsEGFP2 ( blue ) and rsEGFP ( red ) . Switching was performed on purified proteins immobilized in a PAA-layer ( pH ~6 . 5 ) by alternating irradiation with 491 nm ( ~2 kW/cm² ) and 405 nm light ( ~2 kW/cm² , 40 µs ) . Fluorescence was recorded only during irradiation with light of 491 nm . Each curve is an average over 10 switching cycles . ( C ) Changes in the absorption spectrum of rsEGFP2 upon switching with light of 488 nm from the equilibrium to the off-state . The spectra were taken at the indicated time points and recorded on purified rsEGFP2 at pH 7 . 5 ( D ) Absorption spectra of equilibrium-state rsEGFP2 at different pH values . The absorption bands at 408 nm and 503 nm presumably correspond to the protonated and the de-protonated cis-chromophore , respectively . ( E ) Switching fatigue of rsEGFP2 ( blue ) and rsEGFP ( red ) . Switching was performed on living PtK2 cells expressing Vimentin-rsEGFP or Vimentin-rsEGFP2 by alternate irradiation with 405 nm ( 2 kW/cm² ) and 491 nm ( 5 . 7 kW/cm² ) light . Illumination times were chosen so that the fluorescence was fully switched to the minimum or maximum , respectively , in each cycle . Each plotted data point is the average ( over 100 cycles ) of the maximum fluorescence intensity in each cycle . The data points were fitted by a mono-exponential function , and the resulting curve was baseline corrected and normalized to 1 . ( F ) Comparison of the ensemble off-switching halftimes ( defined as the time after which the fluorescence reached 50% of its initial value ) of rsEGFP ( red ) and rsEGFP2 ( blue ) at different 491 nm light intensities . On-switching 405 nm light was kept constant ( 3 kW/cm2 ) . Data were collected on living PtK2 cells expressing Vimentin-rsEGFP or Vimentin-rsEGFP2 , respectively . Inset: Graph showing the ratio ( R ) of the off-switching halftime of rsEGFP divided by the off-switching halftime of rsEGFP2 against the 491 nm light intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 00310 . 7554/eLife . 00248 . 004Figure 1—figure supplement 1 . Alignment of the amino acid sequences of EGFP ( GenBank Accession #U55762 ) , rsEGFP ( GenBank Accession #JQ969017 ) , and rsEGFP2 . Differences are highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 00410 . 7554/eLife . 00248 . 005Figure 1—figure supplement 2 . Single-molecule brightness values of EGFP , rsEGFP , and rsEGFP2 measured in PBS buffer ( pH 7 . 5 ) . Average and standard deviation of >30 FCS measurements at various laser intensities between 5 and 100 kW/cm2 . Values normalized to EGFP . The error bars represent the error in the FCS experiments with the value for EGFP normalized to 1 in each individual experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 00510 . 7554/eLife . 00248 . 006Figure 1—figure supplement 3 . Off-switching speed of rsEGFP and rsEGFP2 . ( A ) Off-switching kinetics of rsEGFP and rsEGFP2 embedded in a PAA layer ( pH ~6 . 5 ) determined at different intensities of the 491 nm off-switching light . Each curve is an average of 100 measurements . In each measurement the proteins were switched into the on-state for 40 µs with 405 nm light ( 2 kW/cm² ) and subsequently the decay of fluorescence was recorded over time at the indicated 491 nm light intensities . ( B ) Dependence of the residual fluorescence ( off-state fluorescence ) in the ensemble off-state as a function of the off-switching light intensity . ( C ) Dependence of the residual fluorescence ( off-state fluorescence ) in the off-state of Vimentin-rsEGFP or Vimentin-rsEGFP2 in living Ptk2 cells as a function of the off-switching light intensity . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 006 Adopting the fluorescent state at equilibrium , rsEGFP2 has its fluorescence excitation and emission maximum at 478 nm and 503 nm , respectively ( Figure 1A ) . In the fluorescent state , a solution of rsEGFP2 exhibits an extinction coefficient of ε ≈ 61 , 000 M−1cm−1 and a fluorescence quantum yield of ΦFl = 0 . 3 ( Table 1 ) . Hence the brightness ( given by ε ΦFl ) of rsEGFP2 is similar to that of rsEGFP and about 60% of that of EGFP at low light intensity irradiation . Single-molecule spectroscopy measurements performed at higher light intensities ( 5 to 100 kW/cm2 ) showed that the single molecule brightness of rsEGFP and rsEGFP2 were ~52% and ~44% of that of EGFP , respectively ( Figure 1—figure supplement 2 ) . The slight difference between brightness values determined at low and high light intensities may be attributed to the increased population of unidentified dark states at higher intensities . rsEGFP2 exhibits a ‘negative’ switching mode , that is irradiation with light of around 480 nm induces fluorescence and , in a competing process , switches rsEGFP2 off . Subsequent irradiation with light of around 405 nm switches the protein back from the off- into the on-state ( Figure 1B ) . Irradiation of a purified rsEGFP2 solution with light of ~480 nm leads to a decrease of the 483 nm absorption band , which presumably corresponds to the anionic ( de-protonated ) cis-chromophore , and to the onset of an absorption band peaking at 408 nm , presumably corresponding to the neutral ( protonated ) trans-chromophore state ( Andresen et al . , 2005; Andresen et al . , 2007 ) ( Figure 1C ) . The single molecule brightness , that is the number of emitted photons per time unit , of rsEGFP and rsEGFP2 are similar . However , because a single switching cycle of rsEGFP2 is on average shorter than a single switching cycle of rsEGFP , rsEGFP2 emits fewer photons in a single cycle than rsEGFP . The pKa of the chromophore in the thermal equilibrium state is ~5 . 8 ( Figure 1D ) . Hence the pKa of rsEGFP2 is lower by 0 . 7 pH units than the pKa of rsEGFP . As reported previously for similar RSFPs , the light driven switching of rsEGFP2 is likely due to a cis/trans isomerization of the chromophore , accompanied by a change of the chromophoric protonation state ( Andresen et al . , 2005; Andresen et al . , 2007; Bourgeois and Adam , 2012 ) . The total number of switching cycles before bleaching is critical for the usability of a RSFP for RESOLFT microscopy ( Hell , 2009 ) . To compare rsEGFP2 with rsEGFP in this regard , we immobilized purified proteins in polyacrylamide layers ( PAA ) . Applying light intensities that have previously been used for RESOLFT ( few kW/cm2 ) , we recorded the fluorescence during 6 , 000 on-off switching cycles ( Figure 1E ) . The illumination times were adapted for the two proteins such that the signal reached a maximum or minimum in each cycle . We found that under these conditions the fluorescence was halved not before ~1 , 100 ( rsEGFP ) and ~2 , 100 ( rsEGFP2 ) cycles , demonstrating that rsEGFP2 accommodates even more switches than rsEGFP . In most negative switching RSFPs , including rsEGFP , the ensemble off-switching with blue light is slower by 2–3 orders of magnitude , compared to the on-switching with UV light at comparable light intensities , rendering the off-switching the time-limiting step in RESOLFT microscopy . We compared the off-switching half-time ( time after which the fluorescence signal is reduced to 50% ) of rsEGFP2 and rsEGFP as a function of the off-switching light intensity ( 491 nm ) ( Figure 1F ) . We found that the difference in the off-switching halftimes of rsEGFP2 and rsEGFP depends on the light intensities applied . The lesser the light intensities , the more pronounced is the speed advantage of rsEGFP2 . At a light intensity of 5 . 5 kW/cm2 the off-switching of rsEGFP2 was ~6 . 5 times faster than the off-switching of rsEGFP ( Figure 1F , inset ) . Presumably the switching speed advantage for rsEGFP2 is even higher at lower intensities . For undesirable intensities of >100 kW/cm2 , the differences between rsEGFP and rsEGFP2 are negligible ( Figure 1F , inset ) . These measurements also revealed that the level to which the fluorescence intensity of an ensemble of rsEGFP2 proteins can be reduced ( the off-state fluorescence ) , depends on the intensities for off-switching ( Figure 1—figure supplement 3 ) . Hence the light intensities and the irradiation times used for off-switching influence both the switching speed as well as the lowest residual fluorescence . Furthermore , we observed that the buffer conditions , most notably the pH , and the cellular environments also influence the absolute switching speed , potentially requiring adaptations of the illumination protocol to the observed samples . In mammalian cells , rsEGFP2 can be fused to histone H2B and alpha-tubulin , which require a truly monomeric fusion tag ( Figure 2 ) . Corroborating its monomeric nature , it migrates on semi-native gels in a single band of the expected size ( Figure 2—figure supplement 1 ) . rsEGFP2 maturates in vitro at 37°C with a halftime of ~20 min , which is shorter or comparable to most conventional fluorescent proteins ( Shaner et al . , 2008 ) including mCherry ( 15 min ) , tdTomato ( 60 min ) or TagRFP ( 100 min ) . Its maturation time is superior to both rsEGFP ( 3 hr ) ( Grotjohann et al . , 2011 ) and a maturation-improved Dronpa ( M159T ) variant ( Stiel et al . , 2007; Willig et al . , 2011 ) ( ~50 min ) . Presumably because of its fast maturation time and due to the fact that its linker is identical to that of EGFP ( which has been proven to be very suitable for the generation of fusion proteins ) rsEGFP2 is well suited to tag proteins in living mammalian cells ( Figure 2 ) . 10 . 7554/eLife . 00248 . 008Figure 2 . Expression of various functional rsEGFP2 fusion proteins in mammalian cells . ( A ) rsEGFP2-KDEL ( targeting to the ER ) , ( B ) Keratin19-rsEGFP2 , ( C ) Histone H2B-rsEGFP2 , ( D ) Vimentin-rsEGFP2 , ( E ) Pex16-rsEGFP2 , and ( F ) rsEGFP2-alpha-tubulin . Shown are single confocal sections ( C , F ) and maximum intensity projections of confocal images ( A , B , D , E ) recorded on living cells . Fluorescence was excited by simultaneous irradiation with light of 488 nm and 405 nm . ( A–E ) : PtK2 cells; ( F ) : Vero cell . Scale bars: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 00810 . 7554/eLife . 00248 . 009Figure 2—figure supplement 1 . Semi-native polyacrylamide gel electrophoresis of rsEGFP2 . Purified monomeric EGFP , dimeric dTomato , tetrameric DsRed , and rsEGFP2 were separated on a semi-native gel ( a two-phase polyacrylamide gel ) consisting out of a 12 . 5% separation gel ( 6 . 3 ml H2O , 5 ml 1 . 5 M Tris–HCl pH 8 . 8 , 8 . 3 ml Rotiphorese Gel 30 solution [Roth , Karlsruhe , Germany] , 200 μl 10% [wt/vol] sodiumdodecyl sulphate [SDS] , 200 μl 10% [wt/vol] ammonium persulfate ( APS ) , 20 μl Tetramethylethylendiamin [TEMED] ) and a 5% loading gel ( 5 . 6 ml H2O , 2 . 5ml 1 . 5 M Tris–HCl pH 6 . 8 , 1 . 7 ml Rotiphorese Gel 30 solution , 100 μl 10% [wt/vol] SDS , 100 μl 10% [wt/vol] APS , 10 μl TEMED ) . Images were taken with a custom-built gel documentation system . To detect green fluorescence ( EGFP and rsEGFP2 ) the gel was irradiated with light of 470 ± 5 nm and fluorescence was detected at 525 ± 30 nm . To detect red fluorescence ( dTomato and DsRed ) the gel was irradiated with light of 545 ± 10 nm and fluorescence was recorded at 617 ± 37 . Both images were overlaid and are represented in false colors . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 009 The possibility to switch rsEGFP2 faster than rsEGFP , its higher switching stamina , and its faster maturation kinetics , suggested that this protein would be superior for fast live-cell RESOLFT imaging . We first expressed rsEGFP2 fused to the C-terminus of Vimentin in mammalian PtK2 ( Potorous tridactylis ) cells . RESOLFT imaging was performed with a home-built setup maintaining the cells at 35°C . Because the filaments formed by Vimentin-rsEGFP2 are relatively immobile , we selected an irradiation scheme encompassing low switching light intensities in combination with extended irradiation times . The living cells were imaged pixel-by-pixel , by first irradiating with 405 nm light ( 2 kW/cm² ) for 40 µs to switch most proteins into their fluorescent on-state . After a short illumination break of 10 µs , the doughnut shaped 491 nm beam ( 10 kW/cm² , 300 µs ) was used to switch rsEGFP2 into the off-state , confining the on-state to the doughnut center . Finally , the rsEGFP2 fluorescence was probed for 30 µs by irradiation with 491 nm light ( 38 kW/cm² ) . Hence the pixel dwell time was 380 µs , which is 25- to 50-fold faster than previously reported using rsEGFP on similar structures ( Grotjohann et al . , 2011 ) . To enhance the image contrast , we also employed Richardson-Lucy restoration ( Richardson , 1972; Hofmann et al . , 2005 ) . Movements of Vimentin-filaments in a living cell were recorded in 10 µm × 10 µm frames of 20 nm pixel size , every 100 s ( Figure 3a; Figure 3—figure supplement 1; Movie 1 ) . After the 20th image , the overall fluorescence still amounted to 65% of its initial value , underscoring the bleaching resistance of rsEGFP2 . 10 . 7554/eLife . 00248 . 010Figure 3 . RESOLFT time lapse imaging using rsEGFP2 in living mammalian PtK2 cells . ( A ) Cells expressing Vimentin-rsEGFP2: initial confocal ( left ) and subsequent RESOLFT images taken every 100 s . Lower row: magnifications of the indicated areas . ( B ) Lateral resolution measurement: raw images of cells expressing Keratin19-rsEGFP2 recorded with a RESOLFT Quad P microscope ( Abberior Instruments GmbH , Göttingen , Germany ) with similar imaging conditions as in ( A ) ( on: 405 nm , 5 kW/cm² , 20 µs; off: 488 nm , 34 kW/cm² , 360 µs; read-out: 488 nm , 76 kW/cm² , 20 µs ) . From left to right: confocal raw image and corresponding raw RESOLFT image . Magnifications of the boxed areas in the RESOLFT image . The graphs show averaged line profiles across the indicated filaments ( i–iv ) within the respective boxes . The line profiles used for averaging were taken equidistant ( 20 nm ) along the whole respective indicated area . ( C ) , ( D ) rsEGFP2 targeted to the ER ( rsEGFP2-KDEL ) : ( C ) 10 µm × 10 µm initial confocal ( left ) and subsequent RESOLFT images recorded every 5 . 9 s , and ( D ) 2 . 8 µm × 3 . 2 µm RESOLFT image-series imaged at 2 Hz . ( E ) RESOLFT imaging of peroxisomes labeled by Pex16-rsEGFP2 fusion proteins . Pixel step sizes: 20 nm ( A , B ) and 40 nm ( C–E ) . Pixel dwell times: 380 µs ( A ) , 400 µs ( B ) , 75 µs ( C , D ) , and 120 µs ( E ) . In ( D ) and ( E ) pixels were interpolated to a size of 20 nm × 20 nm . The arrows indicate moving structures . Richardson Lucy restoration was used for all RESOLFT images except ( B ) . Scale bars: 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 01010 . 7554/eLife . 00248 . 011Figure 3—figure supplement 1 . Raw RESOLFT images of Figure 3A . No image processing was applied . Shown are PtK2 cells expressing Vimentin-rsEGFP2: initial confocal ( left ) and subsequent RESOLFT images taken every 100 s . Scale bar: 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 01110 . 7554/eLife . 00248 . 012Figure 3—figure supplement 2 . Lateral resolution in fast RESOLFT imaging . ( A ) , ( B ) Typical examples . Shown are raw images of cells expressing Keratin19-rsEGFP2 taken on a RESOLFT Quad P microscope ( Abberior Instruments GmbH , Göttingen , Germany ) with a pixel dwell time similar as in Figure 3D , E . Imaging conditions: ( on: 405 nm , 8 kW/cm² , 2 µs; off: 488 nm , 68 kW/cm² , 61 µs; read-out: 488 nm , 200 kW/cm² , 7 µs; pixel size: 40 × 40 nm ) . From left to right: confocal raw image and corresponding raw RESOLFT image . The graphs show averaged line profiles across the indicated filaments within the respective boxes . The line profiles used for averaging were taken equidistant ( 40 nm ) along the whole respective indicated area . Scale bar: 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 01210 . 7554/eLife . 00248 . 013Figure 3—figure supplement 3 . Comparison of rsEGFP and rsEGFP2 at RESOLFT imaging conditions . ( A ) Repeated imaging of peroxisomes labeled by Pex16-rsEGFP or Pex16-EGFP2 fusion proteins . Imaging conditions were as in Figure 3D . Pixel step size: 40 nm; on: 405 nm , 4 kW/cm² , 20 µs; off: 491 nm , 20 kW/cm² , 50 µs; read-out: 491 nm , 76 kW/cm² , 5 µs; pixel dwell time: 75 µs . Images were taken every 5 s . Shown are raw data . ( B ) Decay of the summed fluorescence intensities of the images shown in ( A ) . Note that rsEGFP photobleaches faster than rsEGFP2 . Scale bar: 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 01310 . 7554/eLife . 00248 . 014Movie 1 . Animated sequence of RESOLFT recordings of a living PtK2 cell expressing Vimentin-rsEGFP2 as shown in Figure 3A . 20 RESOLFT images were taken every 100 s . Image size: 10 µm × 10 µm . The movie is accelerated by a factor of 200 compared to the original recording speed . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 014 To determine the obtainable resolution , we imaged cells expressing Keratin19-rsEGFP2 filaments using similar illumination conditions as before . Evaluating the intensity profiles of these filaments in the focal plane revealed a resolution <50 nm in raw image data ( Figure 3B ) . Next , we targeted rsEGFP2 to the lumen of the endoplasmic reticulum ( ER ) . Since the ER is a fast moving structure we increased the recording speed further by doubling the light intensities ( on: 405 nm , 4 kW/cm² , 20 µs; off: 491 nm , 20 kW/cm² , 50 µs; read-out: 491 nm , 76 kW/cm² , 5 µs ) thus cutting the pixel dwell time down to 75 µs , which is ~250 times faster than reported previously using rsEGFP . We did not introduce an irradiation break within a single switching cycle . Because we aimed at capturing fast movements , we also increased the pixel size to 40 nm × 40 nm , thus covering a field of 100 µm² in 5 . 9 s ( Figure 3C ) . Repeated imaging revealed fast changes occurring in the range of seconds in the highly interconnected ER . To visualize movements of the ER in the sub-second range , we reduced the field of view to ~9 µm2 corresponding to 0 . 5 s frame times . Figure 3D shows six RESOLFT images out of a time lapse movie containing 100 images ( Movie 2 ) revealing changes in the ER structure occurring in <1 s . Since the ER-tubules are too large to determine the obtained resolution , we imaged again Keratin19-rsEGFP2 expressing cells using a similar dwell time and established <90 nm in raw data ( Figure 3—figure supplement 2 ) , meaning that speed was obtained at the expense of resolution . 10 . 7554/eLife . 00248 . 015Movie 2 . Animated sequence of RESOLFT recordings of a living PtK2 cell expressing rsEGFP2 targeted to the ER as shown in Figure 3D . 100 RESOLFT images were taken every 0 . 5 s . Image size: 2 . 8 µm × 3 . 2 µm . The speed of the movie corresponds to the imaging speed . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 015 Likewise , when rsEGFP2 was fused to the protein Pex16 to highlight the peroxisomes in living PtK2 cells , we could follow the movement of individual peroxisomes in the RESOLFT-mode in a 3 µm × 2 µm field of view at 2 Hz ( Figure 3E; Movie 3 ) . Since the fluorescent proteins are confined to individual peroxisomes and do not diffuse over large distances , labeled peroxisomes are well suited for a direct comparison of rsEGFP2 and rsEGFP bleaching . The comparison showed that the overall photobleaching of rsEGFP2 was substantially lower than that of rsEGFP when the above fast RESOLFT imaging conditions were applied ( Figure 3—figure supplement 3 ) . 10 . 7554/eLife . 00248 . 016Movie 3 . Animated sequence of RESOLFT recordings of a living PtK2 cell expressing Pex16-rsEGFP2 to highlight the peroxisomes as shown in Figure 3E . 20 RESOLFT images were taken every 0 . 5 s . Image size: 3 µm × 2 µm . The speed of the movie corresponds to the imaging speed . DOI: http://dx . doi . org/10 . 7554/eLife . 00248 . 016 Phototoxicity is a major concern in ( superresolution ) fluorescence microscopy of living cells . Among the established methods , STED-microscopy entails relatively large intensities , ( 5–200 MW/cm2 ) , depending on the desired resolution . However , since the wavelengths ( 560–800 nm ) used for STED are in the comparatively benign long-wavelength part of the spectrum ( Hell , 2009 ) , STED microscopy can image living cells and tissues , including neurons in the cerebral cortex of a living mouse ( Berning et al . , 2012 ) . In the RSFP-based RESOLFT microscopy demonstrated here , the employed light intensities ( 1–80 kW/cm2 ) are several orders of magnitude lower than in STED-microscopy and comparable to those used in live-cell confocal fluorescence microscopy . Stochastic single-molecule based approaches , such as the methods called PALM ( Manley et al . , 2008; Shroff et al . , 2008 ) , STORM ( Jones et al . , 2011; Shim et al . , 2012 ) , and GSDIM ( Fölling et al . , 2008 ) , typically use similar light intensities for imaging living cells ( 0 . 1–100 kW/cm² depending on the exposure time or camera frame rate ) , but they apply these intensities ( i . e . temporal and spatial photon densities ) continuously to all points in the imaged area . Since our RESOLFT approach has been implemented as a point-scanning system , the intensities employed are applied only for a brief duration on a small , sub-micrometer sized region of the imaged area . Hence any pixel is only illuminated during a small fraction of the recording time of the image . In the stochastic methods , the whole imaged area is irradiated for the entire time of recording , that is for a couple of seconds or minutes . Therefore , in the RESOLFT microscopy demonstrated here , the total light dose impinging on the cell is lower by 3–4 orders of magnitude compared to the stochastic single-molecule based approaches . Concretely , for recording the shown RESOLFT images , 2–10 J/cm² were applied for switching into the on-state , and 25–300 J/cm² for eliciting fluorescence and switching the protein off . PALM live-cell imaging reportedly requires light doses of 1 , 000–100 , 000 J/cm² for on-switching with UV ( 405 nm ) light and 25 , 000–300 , 000 J/cm² for fluorescence excitation ( Manley et al . , 2008; Shroff et al . , 2008 ) . Live-cell GSDIM experiments using a yellow fluorescent protein required even larger irradiation doses ( 100 , 000–900 , 000 J/cm² ) ( Fölling et al . , 2008; Testa et al . , 2010 ) . Although RESOLFT microscopy may induce phototoxicity after extended exposure , similarly to live-cell confocal microscopy , it currently is the superresolution method entailing the lowest light dose . Therefore , perhaps not surprisingly , RESOLFT microscopy has been used to image neurons in living organotypical hippocampal cultures over several hours without noticeable photodegradation ( Testa et al . , 2012 ) . Note that the comparatively low-dose/low-intensity requirement of the RESOLFT concept is due to the fact that it uses long-lived on- and off-states in combination with the fact that it does not require fast emission of many photons for establishing molecular coordinates ( localization ) . Since the applied intensities are largely determined by the on-off switching kinetics , the concept offers ample room for accommodating novel proteins with switching kinetics entailing even lower light doses and intensities . In conclusion , rsEGFP2 is a bright , monomeric , photostable , quickly maturating , and fast switching alternative to rsEGFP with comparatively low photobleaching . We expect it to outperform other green fluorescent RSFPs ( Ando et al . , 2007; Stiel et al . , 2007 ) because of its faster maturation and good usability for functional protein tagging . An implementation of quick illumination sequences allowed us to realize up to 250-fold faster recordings as compared to previous reports , thus facilitating live-cell RESOLFT nanoscopy with pixel dwell times down to 70 µs . Since one can adjust both the duration and the illumination intensity , as well as the optical switching scheme , RESOLFT nanoscopy allows one to adapt speed and resolution within a certain range , to the sample needs . Finally , we note that in the point-scanning scheme used here , the total recording time of the image scales with the area of recording . Parallelization of the scanning procedure with an array of doughnuts or lines ( so-called ‘structured illumination’ ) ( Gustafsson , 2005; Schwentker et al . , 2007; Rego et al . , 2012 ) overcomes this dependence on the field of view and cuts down the recording time by the degree of parallelization . Owing to the low-light level operation , the degree of parallelization can easily amount up to a factor of 100–1000 . For this reason , given the short pixel dwell times attained herein , parallelized RESOLFT versions should enable video-rate nanoscopy across the whole field of view of the objective lens . For site-directed mutagenesis , the QuikChange Site Directed Mutagenesis Kit ( Stratagene , La Jolla , CA ) or a multiple-site mutagenesis approach using several primers were used . The experimental procedures were essentially as described previously ( Grotjohann et al . , 2011 ) . In brief , proteins were expressed in the E . coli strain BL21-CP-RIL and purified by Ni-NTA affinity chromatography ( His SpinTrap , GE Healthcare ) , according to the manufacturer's instructions . The purified proteins were concentrated by ultrafiltration and taken up in 100 mM Tris–HCl , 150 mM NaCl , pH 7 . 5 . For the determination of the absorption , excitation and emission spectra of rsEGFP2 , a protein solution ( pH 7 . 5 ) was analyzed with a Varian Cary 4000 UV/VIS photospectrometer and a Varian Cary Eclipse fluorescence spectrometer , respectively . At this pH , the majority of the equilibrium-state rsEGFP2 chromophores are in the deprotonated cis-state ( see Figure 1D ) . To determine its emission spectrum , rsEGFP2 was excited at 460 nm; the excitation spectrum was determined by measuring fluorescence at 520 nm . The fluorescence quantum yields and the molar extinction coefficients at the respective absorption maximum were determined relative to the reported values of EGFP ( quantum yield ΦFL = 0 . 60 , molar extinction coefficient at 489 nm ε = 53 , 000 M−1 cm−1 ) ( Patterson et al . , 1997 ) . Irradiation-dependent changes in the absorption were quantified by illuminating the protein solution in a cuvette with a fiber coupled mercury lamp ( Lecia Microsystems , Wetzlar , Germany ) equipped with a ( 488 ± 5 ) nm excitation filter . For each measurement of the spectrum the irradiation was briefly interrupted . For the embedding of rsEGFP2 in a PAA layer , 24 . 5 µl of purified rsEGFP2 ( ~0 . 1 mM ) was mixed with 17 . 5 µl Tris–HCl pH 7 . 5 , 30 µl acrylamide ( Rotiphorese Gel 30 , Roth , Karlsruhe , Germany ) , 0 . 75 µl 10 % ammonium persulfate and 1µl 10 % TEMED . About 10 µl of this solution was placed on a glass slide and a cover slip was pressed onto the sample . After complete polymerization , the sample was sealed with silicon-based glue ( Picodent twinsil , Picodent , Wipperfürth , Germany ) . To determine the time required for chromophore maturation in rsEGFP2 , the E . coli cell strain TOP10 ( Invitrogen , Carlsbad , CA ) was transformed with the inducible expression plasmid pBad-rsEGFP2 and grown overnight at 37°C in LB-Amp medium . The overnight culture was used to inoculate 200 ml LB-Amp growth medium . At an OD600 of 0 . 5 to 0 . 6 , addition of arabinose to a final concentration of 0 . 2% induced the protein expression . The cultures were further incubated at 37°C for 2 hr . Cells were opened up by several freeze–thaw cycles and pelleted by centrifugation . rsEGFP2 was purified immediately from the supernatant using a His SpinTrap column ( GE Healthcare , Freiburg , Germany ) . The proteins were diluted in buffer ( final concentration: 20 mM NaH2PO4 , 500 mM NaCl , 30 mM imidazol , pH 7 . 5 ) . Care was taken that all preparation steps took place at 4°C . Finally , fluorescence emission spectra of rsEGFP2 were taken at several time points using a fluorescence spectrometer ( Varian Cary Eclipse ) while incubating the protein solution at 37°C . PtK2 ( Potorous tridactylis ) cells were cultured under constant conditions at 37°C and 5% CO2 in DMEM ( Invitrogen , Carlsbad , CA ) containing 5% FCS ( PAA , Pasching , Austria ) , 100 units per ml streptomycin , 100 µg/ml penicillin ( all Biochrom , Berlin , Germany ) , and 1 mM pyruvate ( Sigma , St . Louis , USA ) . For transfection , cells were seeded on cover glasses in 6-well plates . At the next day , cells were transfected with plasmid DNA using Nanofectin ( PAA , Pasching , Austria ) according to the manufacturer's instructions . After 24 hr the growth medium was replaced . Cells were imaged 24–72 hr after transfection . To generate the various fusion constructs of rsEGFP with Keratin19 , with the histone H2B , with Vimentin , or with the peroxisomal membrane protein Pex16 , rsEGFP was amplified ( forward primer: GATCCACCGGTCGCGGCGTGAGCAAGGGCGAGGAGCTG/reverse primer: ACAACTTAAGAACAACAATTGTTACTTGTACAGCTCGTCCATGCC ) . The PCR fragment was cloned into the gateway destination vector pMD-tdEosFP-N using the restriction sites AgeI and AflII , thereby replacing the tdEosFP coding sequence against the rsEGFP2 sequence . The final plasmids pMD-Ker19-rsEGFP2 , pMD-H2B-rsEGFP2 , pMD-Vim-rsEGFP2 and pMD-Pex16-rsEGFP2 were constructed by gateway vector conversion ( Invitrogen , Carlsbad , CA ) using the donor vectors pDONR223-Krt19 , pDONR223-Hist1H2BN , pDONR223-Vim and pDONR223-Pex16 , respectively ( Lamesch et al . , 2007 ) . Pex16-rsEGFP was cloned accordingly . To generate pMD-rsEGFP2-α-Tubulin , rsEGFP2 was amplified ( forward primer: GATCCGCTAGCGCTAATGGTGAGCAAGGGCGAGGAG/reverse primer: CACTCGAGATCTGAGTCCGGACTTGTACAGCTCGTCCATGCC ) and cloned into the vector pEGFP-Tub ( Clontech , Mountain View , CA ) using the restriction sites NheI and BglII replacing EGFP . To generate a construct that targets rsEGFP2 to the ER , the rsEGFP2 sequence was PCR-amplified ( forward primer: CTGCAGGTCGACATGGTGAGCAAGGGCGAGGA/reverse primer: TTCTG CGGCCGCCTTGTACAGCTCGTCCATGCCGCCGGT ) . The PCR product was ligated into the vector pEF/myc/ER ( Invitrogen , Carlsbad , CA ) using the SalI and NotI restriction sites . A home-built RESOLFT microscope ( Grotjohann et al . , 2011; Testa et al . , 2012 ) was adapted for imaging rsEGFP2 in living cells . The microscope utilized three separate beam paths for generating focal spots: two at 491 nm wavelength for excitation and off-switching and one at 405 nm for on-switching of the fluorophores . The two focal spots at 491 nm comprised: ( i ) a normal diffraction-limited focus with a Gaussian profile for reading out the fluorescence signal and ( ii ) a focus with a central intensity minimum ( ‘zero’ ) for off-switching at the focal periphery in the xy-plane , obtained by passing the beam through a vortex phase mask ( 463 nm mask , vortex plate VPP-A , RPC Photonics , Rochester , NY ) . The first two foci were both generated by the same laser diode ( 50 mW , Calypso 50 , Cobolt , Stockholm , Sweden ) . The third focal spot , again with a normal diffraction-limited Gaussian profile , was generated by a laser diode at 405 nm wavelength ( 30 mW , BCL-030-405-S , CrystaLaser , Reno , NV , USA ) and used for the on-switching of rsEGFP2 . The microscope was equipped with a glycerol-immersion objective lens ( PL APO , CORR CS , 63× , 1 . 3NA , glycerol; Leica Microsystems , Wetzlar , Germany ) . A piezo system ( ENV40/20 , Piezosystem Jena , Jena , Germany ) was used to move the objective lens along the optical axis . A separate piezo stage ( NV40 , Piezosystem Jena ) was implemented to translate the sample with nanometer precision in the xy-plane . The fluorescence signal was filtered by a band pass filter ( 532/70 nm ) and detected by an avalanche photo diode ( Perkin Elmer , Waltham , MA , USA ) ; fluorescence photons were only allowed to be counted when the 491 nm read-out beam was switched on . The individual laser beam paths were triggered either by an acousto-optic modulator ( MTS 130A3 , Pegasus Optik GmbH , Wallenhorst , Germany ) or by an acousto-optic tunable filter ( AOTF . nC/TN , Pegasus Optik GmbH ) . The pulse sequence and duration were defined by a pulse generator ( Model 9514 , QUANTUM COMPOSERS , Bozeman , MT , USA ) and triggered by a fast acquisition card ( MCA-3 Series/P7882 , FAST ComTec GmbH , Oberhaching , Germany ) pixel by pixel . Alternatively , we assembled the Abberior RESOLFT Quad P microscopy kit provided by Abberior Instruments GmbH , Göttingen , Germany , which used the same arrangement and wavelengths as the home-built system , except for the fact that scanning was accomplished by a galvanometer beam scanning system ( Quad scanner ) and the body of the microscope was an Olympus iX81 inverted microscope . Imaging was performed with a 100× Olympus oil immersion objective lens of 1 . 4 numerical aperture . Image acquisition was performed with the software Imspector ( www . imspector . de ) . Each image was recorded by applying a specific pulse scheme , pixel by pixel . The fluorescence signal was recorded only when the 491 nm read-out Gaussian shaped beam was on . Between each pixel pulse sequence ( pixel dwell times 75–380 µs ) a delay of 20 µs was inserted for synchronization , resulting in effective dwell times of 95–400 µs . The laser intensities used in our illumination scheme ranged between 1–100 kW/cm2 . The approximately 10% remaining switching background introduces some diffraction-limited components in the final raw image . To remove this background , we deconvolved the final image by Richardson–Lucy ( Richardson , 1972; Lucy , 1974 ) restoration with a 10% diffraction-limited PSF added to the RESOLFT PSF , as detailed previously ( Hofmann et al . , 2005 ) . 10 iterations were performed . All experiments were performed at 35°C except those presented in Figure 3B and Figure 3—figure supplement 2 . The single-molecule brightness of EGFP , rsEGFP and rsEGFP2 were determined using fluorescence fluctuation spectroscopy , specifically fluorescence correlation spectroscopy ( FCS ) ( Haustein and Schwille , 2003 ) and fluorescence intensity distribution analysis ( Chen et al . , 1999; Kask et al . , 1999 ) . Both methods analyze characteristic fluctuations δF ( t ) in the fluorescence signal F ( t ) in time t about an average value F ( t ) = <F ( t ) > + δF ( t ) by either calculating the second-order auto-correlation function G ( tc ) ( FCS , with correlation time tc ) or by building up a frequency histogram P ( n , ∆T ) of photon counts detected per time window ∆T ( FIDA , with number of photons n ) . Fluctuations in F arise for example from diffusion of the fluorescent proteins in and out of the confocal detection volume or by transitions into and out of a dark state such as the triplet , other metastable dark or the switch-off state . FCS and FIDA data were analyzed using common theory . As outlined in detail previously ( Eggeling et al . , 2007 ) , the analysis most importantly resulted in three characteristic molecular parameters of the fluorescent proteins: the single-molecule brightness ( or count-rate per particle ) q ( from FIDA measurements with ∆T = 10 µs ) , the observation time τobs ( from FCS measurements ) , and the average population of a µs-long-lived dark state ( probably the triplet state of the fluorophore , from FCS measurements ) . In Figure 1—figure supplement 2 the q values in relation to the normalized q value of EGFP are shown . Without saturating the excitation , the brightness q ~ ΦFL ε scales with the fluorescence quantum yield ΦFL and the extinction coefficient ε ( compare Table 1 ) . For EGFP , the observation time τobs is given by its average transit time through the focal spot , while for rsEGFP and rsEGFP2 it is given by both the transit time and—if faster—the average switch-off time ( Eggeling et al . , 2007 ) . The fluorescence fluctuation data were recorded on a FCS reader ( Insight , Inovation GmbH , Osnabrück , Germany ) , applying a water immersion objective ( 60× UPLSAPO , NA 1 . 2 , Olympus , Japan ) . Data was recorded for different powers of the 491 nm excitation laser ( Viper , Qioptiq , Hamble , UK ) and for the fluorescent proteins in aqueous solution ( PBS buffer , pH 7 . 5 ) . The observation times τobs were ~225 µs for EGFP ( in accordance to the expected focal transit time ) , while those of rsEGFP and rsEGFP2 were shorter , reaching a value of ~40 µs at excitation powers >50 µW ( 15 kW/cm2 ) for rsEGFP and ~10–15 µs for rsEGFP2 . The shorter observation times in the case of rsEGFP and rsEGFP2 result from a fast population of >200 µs-lived dark states ( for details see ( Eggeling et al . , 2007 ) ) .	For decades it was assumed that the diffraction of light meant that optical microscopy could not resolve features that were smaller than about the half the wavelength of the light being used to create an image . However , various ‘super-resolution’ methods have allowed researchers to overcome this diffraction limit for fluorescence imaging , which is the most popular form of microscopy used in the life sciences . This approach involves tagging the biomolecules of interest with fluorescent molecules , such as green fluorescent protein ( GFP ) , so that they can be identified in cells . An excitation laser then drives the fluorescent molecule , which is also known as a fluorophore , into an excited state: after a short time , the fluorophore can return to its ground state by releasing a fluorescence photon . Images of the sample are built up by detecting these photons . In STED super-resolution microscopy a second laser is used to instantly send the molecules from their excited or ‘on’ states back to their ground or ‘off’ states before any fluorescence can occur . The second laser beam is usually shaped like a doughnut , with a small region of low light intensity surrounded by a region of much higher intensity . STED microscopy is able to beat the diffraction limit because the second laser turns all the fluorophores ‘off’ except those in the small sub-wavelength region at the centre of the doughnut . The image is build up by scanning both lasers over the sample so that the small region in which the fluorophores are ‘on’ probes the entire cell . RESOLFT is a similar technique that employs fluorescent molecules with ‘on’ and ‘off’ times that are much longer than those used in STED microscopy . In particular , RESOLFT uses fluorescent molecules that can be rapidly switched back and forth between long-lived ‘on’ and ‘off’ states many times by the two lasers . The fact that both these states are long-lived states means that RESOLFT requires much lower laser intensities than STED , which makes it attractive for imaging biological samples over large areas or long times . RESOLFT demonstrated its suitability for bioimaging for the first time last year , with a protein called rsEGFP ( reversibly switchable enhanced GFP ) being employed as the fluorophore . However , the time needed to switch this protein between the ‘on state’ and the ‘off state’ was relatively long , and it took about an hour to record a typical image . Now , Grotjohann et al . have modified this protein to make a new fluorophore called rsEGFP2 with a shorter switching time , and have used it to image various structures—including Vimentin , a protein that forms part of the cytoskeleton in many cells , and organelles called peroxisomes—inside live mammalian cells . They were able to record these images some 25–250 times faster than would have been possible with previous RESOLFT approaches . The combination of RESOLFT and rsEGFP2 should allow researchers to image a wide variety of structures and processes in living cells that have not been imaged before .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2012	rsEGFP2 enables fast RESOLFT nanoscopy of living cells
Cytochromes c are ubiquitous heme proteins in mitochondria and bacteria , all possessing a CXXCH ( CysXxxXxxCysHis ) motif with covalently attached heme . We describe the first in vitro reconstitution of cytochrome c biogenesis using purified mitochondrial ( HCCS ) and bacterial ( CcsBA ) cytochrome c synthases . We employ apocytochrome c and peptide analogs containing CXXCH as substrates , examining recognition determinants , thioether attachment , and subsequent release and folding of cytochrome c . Peptide analogs reveal very different recognition requirements between HCCS and CcsBA . For HCCS , a minimal 16-mer peptide is required , comprised of CXXCH and adjacent alpha helix 1 , yet neither thiol is critical for recognition . For bacterial CcsBA , both thiols and histidine are required , but not alpha helix 1 . Heme attached peptide analogs are not released from the HCCS active site; thus , folding is important in the release mechanism . Peptide analogs behave as inhibitors of cytochrome c biogenesis , paving the way for targeted control . The structure of cytochrome c ( cyt c ) , as well as its key function in electron transport for aerobic respiration , have been known for over half a century ( Dickerson et al . , 1971; Ernster and Schatz , 1981 ) . Scores of newly discovered and extraordinary electron transport chains with unique cyt c proteins in bacteria are now known , such as extracellular multiheme nanowires comprised of many c-type hemes ( e . g . Deane , 2019; Wang et al . , 2019 ) . In addition to its role in respiration , cyt c is known to play other important functions , such as activation of programmed cell death in eukaryotes ( apoptosis ) ( Ow et al . , 2008; Tait and Green , 2010 ) . Regardless of its function , each c-type heme contains two thioether attachments to a conserved CysXxxXxxCysHis ( CXXCH ) motif , where the histidine acts as an axial ligand to the heme iron in the native cyt c ( Figure 1—figure supplement 1a , b; Dickerson et al . , 1971 ) . It is generally agreed that the covalently attached heme makes these energy conversion proteins particularly stable ( e . g . Allen et al . , 2005 ) . In fact , recent engineering of novel and stable heme-based catalysts has used c-heme polypeptides produced in vivo ( Kan et al . , 2017; Kan et al . , 2016; Watkins et al . , 2017 ) . To form c-heme , heme is attached stereochemically to each CXXCH motif and it appears that in the case of cyt c , folding into its native structure occurs after attachment ( Kranz et al . , 2009 ) . Cyt c biogenesis requires accessory proteins that are needed to attach the heme group and complete maturation . Three pathways have been discovered and characterized genetically , called Systems I , II , III ( Figure 1—figure supplement 1b , c ) ( reviewed in Kranz et al . , 2009; Ferguson et al . , 2008; Kranz et al . , 1998; Bowman and Bren , 2008; Simon and Hederstedt , 2011; Verissimo and Daldal , 2014; Gabilly and Hamel , 2017 ) . Systems I and II have evolved in bacteria , while System III is in most mitochondria . Each system possesses a cyt c synthase ( Figure 1—figure supplement 1c , orange ) , which attaches the two vinyl groups of heme to cysteines of CXXCH . However , the cyt c biogenesis process , starting with CXXCH recognition , to heme attachment , to release and final folding , remains largely unknown . While in vivo studies have suggested some requirements ( Babbitt et al . , 2017; Babbitt et al . , 2016; Corvest et al . , 2010; San Francisco et al . , 2013 ) , such cyt c genetic studies do not examine problems of instability , recognition , release , or folding of the cyt c variants . Direct testing of substrates without these limitations awaited the development of in vitro reconstitution . The mitochondrial System III is composed of a cyt c synthase called HCCS ( holocyt c synthase ) in the intermembrane space ( Figure 1a and Figure 1—figure supplement 1c , Pollock et al . , 1998; Dumont et al . , 1987; Babbitt et al . , 2015 ) . Bacterial systems are unrelated to HCCS and more complicated , heme attachment occurs ‘outside’ the cells; thus , these pathways export the heme and attach it to secreted , unfolded cyt c . System II is composed of a large integral membrane protein complex called CcsBA ( Beckett et al . , 2000; Dreyfuss et al . , 2003; Xie and Merchant , 1996 ) ( sometimes called ResBC [Ahuja et al . , 2009; Le Brun et al . , 2000] ) , which is proposed to both export heme and then attach it to cyt c CXXCH motifs ( Feissner et al . , 2006; Frawley and Kranz , 2009 ) . Specific factors for thiol reduction of the CXXCH motifs have also been proposed ( Bonnard et al . , 2010; Kranz et al . , 2009 ) . Large gaps in the cyt c biogenesis field remain such as CXXCH recognition requirements by each cyt c synthase and whether other general factors in the cell are needed for recognition , heme attachment , and folding . While specific proteins have been identified and functions hypothesized for each system ( reviewed in Babbitt et al . , 2015; Ferguson et al . , 2008; Gabilly and Hamel , 2017; Kranz et al . , 2009; Verissimo and Daldal , 2014 ) , there has been no in vitro reconstitution studies with purified cyt c synthases , which will be needed to address these gaps . Only recently was our group able to purify the cyt c synthases , after recombinant expression in Escherichia coli ( Frawley and Kranz , 2009; Merchant , 2009; Richard-Fogal et al . , 2009; San Francisco et al . , 2013; Sutherland et al . , 2018b ) . Here we develop and characterize the first in vitro reconstitutions of cyt c synthases , using purified human HCCS and the bacterial CcsBA . No protein factors other than the cyt c synthases are needed in vitro for attachment and folding into a native cyt c structure . In vitro reactions with a variety of peptides containing CXXCH show that the CXXCH substrates for each cyt c synthase are quite different and that post-attachment folding of cyt c is important in release from the synthase active sites . Key differences between HCCS and CcsBA include thiol ( cysteine ) requirements and the alpha helix sequence adjacent to CXXCH . Peptide analogs behave as inhibitors . Because bacteria and humans ( mitochondria ) use very different cyt c synthases , shown here to recognize distinct features of the CXXCH substrate , specific inhibitors could constitute targeted antimicrobials , facilitating chemical control of cyt c levels in selected organisms . Using purified human HCCS , we reconstituted cyt c synthase activity with equine apocyt c as substrate , initially assaying formation of a peak at 550 nm , diagnostic of cyt c’s typical UV–vis spectra ( Figure 1a ) . Recombinant human HCCS ( GST-tagged ) is functional in vivo , attaching heme to co-expressed apocyt c ( San Francisco et al . , 2013 ) in E . coli . We have previously shown that HCCS co-purifies with heme , which is liganded to His154 ( San Francisco et al . , 2013 ) . UV–vis spectra of purified HCCS shows a 423 nm and broad 560 nm absorption , typical of heme proteins , while HCCS H154A variant does not bind heme ( Figure 1b , −HL ) . We developed a ‘heme-loading ( HL ) ’ protocol to increase the levels of heme bound in HCCS ( +HL , ~30% occupied ) above the co-purified levels of endogenous heme ( −HL , ~10% occupied ) . HL was also advantageous since the HL protocol removes excess heme , thus minimizing spectral interference from free heme in reactions . HL was shown to depend on the natural His154 ligand ( Figure 1b , +HL black line ) , and loading was saturated at 2–5 µM heme ( Figure 1—figure supplement 2 ) . Initial reconstitutions were performed with wild type ( wt ) HCCS ( ±HL ) and the HCCS His154Ala variant that does not bind heme ( Figure 1b ) . Upon incubation for 1 hr in the presence of apocyt c and dithiothreitol ( DTT ) , a sharp 550 nm peak emerged , indicative of a c-type cytochrome ( Figure 1b , red with wt ) . This occurred with wt HCCS containing endogenous heme ( −HL ) and in vitro loaded heme ( +HL ) , while HCCS H154A did not produce the 550 nm peak . A second method to determine if heme has been covalently attached to the apocyt c is to separate reactions with denaturing sodium dodecyl sulphate–polyacrylamide gel electrophoresis ( SDS–PAGE ) followed by heme staining , whereby covalently attached heme electrophoreses with the polypeptide ( Figure 1c ) . Reactions with wt HCCS ( −HL and +HL ) and apocyt c confirmed that heme is covalently attached to cyt c ( 12 kDa ) in the 1 hr reaction ( Figure 1c , lanes 3 , 8 ) . As expected , no cyt c was formed with the HCCS H154A variant ( Figure 1c , lanes 5 , 10 ) . Pyridine hemochrome spectra is often used to determine if two , one , or no covalent bonds to heme are present , with two thioether bonds showing a 550 nm peak ( c-heme ) and 560 nm for none ( b-heme ) . The in vitro synthesized product has two thioether bonds , indicated by a 550 nm peak in pyridine hemochrome spectra ( Figure 1—figure supplement 3a ) . In vitro reconstitutions were studied for optimal conditions and requirements . Synthesis is optimal at 37°C ( Figure 1—figure supplement 4 ) , required DTT ( Figure 1—figure supplement 5 ) , with the cyt c product observed in 10 min ( e . g . Figure 1d , lane 5 ) . While cyt c is formed in both aerobic ( Figure 1—figure supplement 6 ) and anaerobic conditions ( Figure 1b , c ) , we decided to use anaerobic conditions for all studies since peptide substrates ( below ) under aerobic conditions required varying DTT concentrations , likely due to distinct thiol reducing requirements of individual peptides in air . To further characterize HCCS , substrates and products , we employed analytical HPLC size exclusion chromatography ( SEC ) , whereby UV–vis spectra of each separated species was recorded ( Figure 1e ) . HCCS ( brown profile ) elutes earlier than cyt c ( green profile ) , and because these are 424 nm ( heme ) profiles , it is observed in the reaction ( blue profile ) that heme in HCCS decreases while cyt c product increases . These results also demonstrate that the cyt c product is released from the HCCS active site since it elutes at the same time as purified cyt c ( holocyt c ) . We conclude that we have recapitulated in vitro the four-step process proposed previously ( Figure 1—figure supplement 7 ) for HCCS-mediated cyt c biogenesis: heme binding ( step 1 ) , apocyt c binding ( step 2 ) , thioether formation ( step 3 ) , and release ( step 4 ) ( Babbitt et al . , 2015; San Francisco et al . , 2013 ) . Next , we further characterize the released cyt c product to establish whether proper folding to the native state resulted from in vitro biogenesis . We developed a HCCS-tethered ( to glutathione beads ) release assay to isolate HCCS reaction product ( s ) , confirm that cyt c is released , and obtain high yields for product characterization ( Figure 2a ) . Spectra of the released product ( Figure 2b ) is identical to holocyt c . SDS–PAGE of stages in the bead release protocol ( Figure 2c ) showed a released product of 12 kD that heme stained and reacted with cyt c antisera ( Figure 2c , lane 2 ) . We determined spectrally that the released cyt c has folded properly , forming the Met81 ligand as well as His19 ( Figure 1—figure supplement 3b ) . Redox titrations ( Figure 2d ) showed that the redox potential of the cyt c in vitro product is the same as cyt c produced in vivo , +253 mV ( Figure 2d ) . Analyses of supernatants ( released ) , washes , and bead-retained material allowed for an estimate that at least 62% of cyt c is released from HCCS ( Figure 2c ) . Since heme in all cyt c’s is attached stereochemically ( Figure 2e ) , we performed circular dichroism ( CD ) spectra to compare the released ( in vitro ) product to cyt c made in vivo ( Figure 2f ) . CD absorption of heme ( ~420 nm ) is reduced in globins when heme binds in multiple orientations compared to a single orientation ( Aojula et al . , 1986; Nagai et al . , 2014 ) . Cyt c synthesized in vitro by HCCS shows an identical CD spectral profile as in vivo synthesized ( Figure 2f ) . We conclude that in vitro reconstitution with purified HCCS results in stereochemical heme attachment , release , and proper folding of cyt c . In vitro reconstitution of the cyt c synthases provides an opportunity to investigate chemically synthesized apocyt c peptides and analogs as substrates . For example , there are in vivo genetic results suggesting that alpha helix 1 , adjacent to the CXXCH motif ( Figure 3a ) , of native cyt c is necessary for maturation by HCCS ( San Francisco et al . , 2013; Zhang et al . , 2014; Kleingardner and Bren , 2011 ) . In fact , the bacterial cyt c has a natural deletion of Met13 in alpha helix 1 , recently shown in vivo to be the basis for the inability of HCCS to mature bacterial cyt c ( Babbitt et al . , 2016; Verissimo et al . , 2012 ) . We wanted to determine if cyt c peptides are recognized in vitro and if so the minimal sequence for recognition and heme attachment . Initially , we examined three peptides , an 11mer , 16mer , and 20mer with the 11mer lacking the sequence of alpha helix 1 ( Figure 3a ) . Heme stains of tricine SDS–PAGE were used to detect whether heme was covalently attached to peptides ( Figure 3b ) . After 1 hr , reactions showed that the 16mer ( Figure 3b , lane 6 ) and 20mer ( Figure 3b , lane 8 ) possessed an intense heme-stained peptide of 2 . 8 kDa , whereas the 11mer did not ( Figure 3b , lane 4 ) . Spectral analyses showed that the 11mer reaction looked like HCCS alone ( no peptide added ) , whereas the reactions with the 16mer and 20mer showed a 552–553 nm peak ( Figure 3c ) . We have previously shown that some recombinant HCCS is co-purified with cyt c remaining bound ( and heme attached ) ( San Francisco et al . , 2013 ) . UV/vis absorption of these HCCS/cyt c complexes exhibits a peak in the reduced state of 553–555 nm ( Figure 1—figure supplement 7 ) , whereas a purified heme attached peptide shows a 550 nm peak ( Figure 1—figure supplement 7b ) . Spectral results of HCCS reactions with 16mer and 20mer peptides ( i . e . 552–553 nm peaks , see Figure 3c ) suggest heme is covalently attached to the peptides , but that they remain in complex with HCCS , unlike full-length cyt c produced in vitro . To further test CXXCH peptide recognition , we tested a 56mer ( with alpha helix 1 and 2 of cyt c ) and a 9mer ( Figure 3—figure supplement 1 ) . While the 56mer was recognized and heme attached , the 9mer was not , consistent with the in vivo results that alpha helix 1 is required for heme attachment ( Babbitt et al . , 2016 ) . Because HCCS reaction with the 56mer yields a 555 nm absorption ( Figure 3—figure supplement 1a ) , it is likely not released . To confirm that heme-attached peptides remain bound to HCCS , we used both HPLC SEC and the bead release assay described above ( Figure 2a ) . HPLC separation ( Figure 4a ) showed that HCCS with the 20mer reaction ( blue profile ) eluted at the same time as HCCS alone ( brown profile ) , not unexpected since a small 2 . 8 kD unreleased product would not significantly alter size exclusion properties . However , the spectra of the 20mer reaction from the HPLC SEC shows the signature of a HCCS-bound cyt c product , with a peak at 553 nm . This supports the conclusion that the heme attached 20mer remains bound to HCCS upon HPLC SEC , explaining why no heme-peptide product elutes separately ( Figure 4a , compare blue and green profiles ) . Results of the bead release assay also show there is very little release of the heme-attached peptides from HCCS . Spectra of the reaction supernatant ( red ) exhibits very little heme ( Figure 4b ) , unlike with full cyt c ( Figure 2b , red ) . However , eluted HCCS from the beads show a spectra consistent with heme-attached peptide still bound , with a 555 nm peak ( Figure 4b , purple ) . Quantitation of the level of heme-attached 20mer released from HCCS was carried out using the bead release assay ( Figure 4c ) , with 14 ± 3% of the heme-attached peptide released from HCCS . We evaluated whether peptides recognized by HCCS would act as inhibitors of heme attachment to subsequent addition of apocyt c . We carried out reactions with the three peptides for 1 hr , then added apocyt c , taking samples throughout ( Figure 4d ) . The 11mer behaved as expected , as if no other substrate was present , with synthesis of cyt c occurring ( in Figure 4e , compare lanes 1–4 and 5–8 boxed bands ) . This also suggests that the 11mer is not recognized by HCCS , in that it does not prevent apocyt c from binding . However , both the 16mer ( Figure 4e , lanes 9–12 ) and 20mer ( lanes 13–16 ) showed heme attached to the peptides , but not to the apocyt c . We consider this inhibition of cyt c biogenesis ( see Discussion ) . We conclude that alpha helix 1 is necessary and sufficient for recognition and attachment to the adjacent CXXCH motif . Our findings suggest that folding of cyt c is required for optimal release from the HCCS active site ( see Discussion ) . Our previous studies with CcsBA have used recombinant GST-tagged CcsBA ( from Helicobacter ) , shown to be functional in vivo and co-purify with endogenous heme ( Feissner et al . , 2006; Frawley and Kranz , 2009; Sutherland et al . , 2018b ) . We concluded that CcsBA is both a heme exporter and a cyt c synthase with two heme binding sites ( Figure 6a ) . To increase CcsBA yields for in vitro and future structural studies , we explored various tagging and expression strategies , ultimately selecting a C-terminal hexahistidine tagged CcsBA which gave high yields ( Figure 5a ) . For unknown reasons , yields were higher when the GST ORF ( with stop codon ) , as well as a new ribosome binding site upstream of ccsBA were used ( threefold higher than GST-tagged or without the GST gene: Figure 5a , b ) . The purified hexahistidine tagged CcsBA still possessed the natural proteolysis site we have previously characterized ( Frawley and Kranz , 2009; Sutherland et al . , 2018b ) , resulting in two polypeptides ( Figure 5c , lane 4 , boxed ) . The GSTCcsBA:His construct is hereafter referred to as CcsBA:His . Using the anaerobic in vitro reconstitution conditions described above for HCCS , both the purified GST-CcsBA and metal-affinity purified CcsBA:His , both with endogenous heme , were active for heme attachment to apocyt c in vitro ( Figure 6a–d ) . For further studies here , we used the CcsBA:His due to its higher yields . We have previously shown that while wt CcsBA has heme in both the P-His/WWD and TM-His sites ( Figure 6a ) , the P-His variants possess heme only in the TM-His site ( Sutherland et al . , 2018b ) . GST:CcsBA P-His mutants are unable to attach heme in vivo to cyt c4 , yet co-purified with heme ( Sutherland et al . , 2018b ) . Since heme is proposed to attach to apocyt c from the P-His/WWD site ( Figure 6a ) , we tested whether the P-His variant functions in vitro , representing ideal negative controls for genuine in vitro attachments . Importantly , the GST:CcsBA P-His variant did not attach heme to apocyt c in vitro ( Figure 6b ) . In vitro reactions with the wt CcsBA:His shows initial spectral signatures of b-heme ( Figure 6c , black spectra ) . Within 1–3 hr , the wt CcsBA shows two peaks of reduced heme , one at 560 nm and a 550 nm peak that is characteristic of covalent heme attached in c-type cytochromes ( Figure 6c , red spectra ) . It is likely that the b-heme ( in the TM-His site ) is responsible for the absorption remaining at 560 nm . These results were confirmed by SDS–PAGE and heme stains at the different time points ( Figure 6d ) , confirming that the wt CcsBA formed cyt c . We conclude that purified wt CcsBA acts as a cyt c synthase in vitro and that heme is attached from the P-His/WWD domain , as hypothesized from in vivo results . A time course of in vitro reactions with wt CcsBA shows that the covalent attachment to apocyt c is measurable at 20 min , reaching a maximum at approximately 3 hr ( Figure 6e ) . Spectra at selected time points confirm these results ( Figure 6f , see 550 nm formation ) . To determine whether cyt c is released from CcsBA and folds into its native state , we performed HPLC SEC on CcsBA alone and from a 3 hr reaction with apocyt c ( Figure 6g ) . CcsBA in vitro synthesized cyt c is released and elutes at the same position as purified cyt c . The cyt c product ( Figure 6g , last inset ) is spectrally identical to cyt c produced by HCCS in vitro and to purified cyt c generated in vivo . We conclude that apocyt c is matured and released by CcsBA in vitro , with folding of cyt c into its native state . Similar to HCCS studies , we used the 11 , 16 , and 20mer peptides ( Figure 3a ) and heme staining of tricine SDS–PAGE , to determine whether CcsBA attaches heme to peptide substrates and if so , what sequence or structural requirements are important . In CcsBA in vitro reactions , the 20mer , 16mer , and 11mer peptides each resulted in covalent heme after 3 hr in vitro reactions ( Figure 7a ) . Spectral analyses also showed formation of 550 nm peaks ( Figure 7b ) , including reactions with the 11mer , which was not recognized by HCCS . Because the 560 nm peak also remains in reactions , likely due to heme in the TM-His site , we used second-derivative spectra to delineate and quantitate the levels of attached heme present ( Figure 7b , last panel , 550 nm ) , also confirming that all peptides possess the 550 nm absorption characteristic of c-type heme . The 56mer ( alpha helix 1 and 2 of cyt c ) and 9mer were also recognized and attached to heme by CcsBA ( Figure 7—figure supplement 1 ) . We conclude that the bacterial CcsBA cyt c synthase does not require the alpha helix 1 and that the recognition requirements are different than the mitochondrial HCCS ( see Figure 7—figure supplement 2 for parallel reactions of HCCS and CcsBA . ) The ability to biosynthesize heme-attached CXXCH peptides in vitro by HCCS and CcsBA facilitated a more detailed analysis of the cysteines and histidine in the substrates . For example , cysteine substitutions in the chemically synthesized peptides would determine whether each cysteine is required and whether non-standard thiol amino acids are recognized ( Table 1 ) . Homocysteine ( HoC ) has an additional carbon between the thiol and alpha carbon , while d-cysteine ( D-C ) rotates the thiol sidechain ( see Table 1 for structures ) . All substitutions were synthesized in the 20mer background since both HCCS and CcsBA could attach heme to it and the product is easily detected on heme stains of tricine SDS–PAGE . That is , if the peptide product has a covalent attachment , it will migrate at 2 . 8 kD and stain for heme ( Figure 3b; Figure 7a; Figure 7—figure supplement 2 ) . This method does not indicate whether a modified thiol ( HoC or D-C ) has a covalent attachment , so we also performed UV–vis and pyridine hemochrome spectroscopy to provide evidence of thioether formation . Figure 3—figure supplement 2 and Figure 7—figure supplement 3 show results of in vitro reactions of HCCS and CcsBA with the peptide analogs , as summarized in Table 1 . In the case of HCCS , all 20mer peptide variants possessed at least one covalent attachment with the exception of the SXXSH variant ( Table 1 , blue highlights ) . This indicates that HCCS does not require both cysteines for recognition , a conclusion consistent with in vivo results of engineered cyt c substrate variants ( Babbitt et al . , 2014 ) . Importantly , the HCCS/peptide complexes exhibit spectral signatures of purified HCCS/cyt c co-complex variants produced in vivo ( Babbitt et al . , 2017 ) . For example , HCCS reactions with the SXXCH peptide shows a split alpha peak at 555/560 nm ( Figure 3—figure supplement 2b ) , just as shown in vivo with the HCCS/Cys15Ser variant ( Babbitt et al . , 2017 ) . Pyridine hemochrome spectra of HCCS reaction products were used to show whether the non-natural thiols were covalently attached . Both homocysteine and the DCys18 ( Figure 3—figure supplement 2f , g ) , thiols were not attached , possessing only a single thioether , resulting in a hemochrome spectral peak of 552 nm that reflected attachment to Cys15 . However , the DCys15 variant possessed two thioether attachments , thus both thiols reacted ( Figure 3—figure supplements 2c , 550 nm pyridine hemochrome peak ) . This indicates that rotation of the first thiol ( Cys 15 ) of the CXXCH substrate is more permissive at the active site of HCCS . Lastly , we examined the role of the conserved H19 of the CXXCH motif . 20mer peptide were synthesized with H19M , H19A , and H19K substitutions . The H19A and H19K variants did not attach heme , while the H19M variant attached heme at low levels ( Figure 3—figure supplement 1 ) , suggesting methionine can act as a weak ligand in HCCS . In the case of the bacterial CcsBA , an entirely different set of rules emerge for CXXCH substrate recognition ( Table 1 , compare blue to orange highlighted variants ) . Only one 20mer cysteine variant showed any covalent attachment: the first cysteine thiol replaced with a homocysteine ( Figure 7—figure supplement 3f ) . The HoCys15 variant has two covalent linkages ( 550 nm peak ) , suggesting that the first thiol is more permissive in distance from the alpha carbon ( i . e . of the first cysteine of CXXCH ) . Because DCys15 was not attached , unlike with HCCS , rotation of the first thiol may be less permissive at the CcsBA active site . No 20mers with histidine substitutions possess covalently attached heme with CcsBA ( Figure 7—figure supplement 1 ) . It has been known for decades that the covalent , thioether attachment of heme in c-type cytochromes ( to a CXXCH motif ) , requires accessory factors , including thioredoxins and cyt c synthases . A unique feature of cyt c biogenesis is that folding into its native structure occurs after cofactor ( heme ) attachment . Many elegant in vitro studies have concerned the folding of purified cyt c , typically after denaturation and renaturation to follow the folding pathway ( e . g . Hu et al . , 2016; Pletneva et al . , 2005; Yamada et al . , 2013 ) . However , in vitro heme attachment by cyt c synthases has not been studied with purified components . Due in part to their membrane location , only recently have we been able to purify the detergent-solubilized synthases , mitochondrial HCCS ( San Francisco et al . , 2013 ) and bacterial CcsBA ( Frawley and Kranz , 2009 ) . CcsBA is an integral membrane protein that functions as a heme exporter and synthase , making its reconstitution particularly challenging . Here we have successfully reconstituted cyt c biogenesis with purified HCCS and CcsBA . Initially , we used apocyt c as substrate and endogenous heme that is co-purified with recombinant HCCS and CcsBA . For HCCS , we were also able to load heme into the active site , requiring His154 , a process proposed as step one in biogenesis ( Figure 1—figure supplement 7 ) . Besides DTT for maintaining a reducing environment , no accessory factors other than HCCS and CcsBA are necessary . In vitro reactions result in stereochemical heme attachment , release of cyt c from the synthases , and proper folding into its native cyt c conformation . The cyt c possesses His19 ( of CXXCH ) and Met81 as axial ligands and its redox potential is identical to native cyt c purified from mitochondria ( +253 mV ) . In vitro reconstitution conditions ( anaerobic , DTT ) enabled the use of CXXCH containing peptides to study biogenesis and the substrate requirements for HCCS and CcsBA . In vitro reactions with HCCS and apocyt c proceed through all four steps ( Figure 1—figure supplement 7 ) , including step 4 , release with cyt c folding . However , a 20mer peptide with CXXCH is very poorly released by HCCS , thus halting the process after step 3 . In vivo we have demonstrated that single cysteine variants of cyt c ( CXXCH motif ) are released less than the wt cyt c , since more HCCS/cyt c complex and less cyt c product is purified ( Babbitt et al . , 2014 ) . We proposed that thioether formation and consequent heme distortion contributes to release . Using cysteine peptide variants , we demonstrate in vitro that peptides with two thioethers release more than those with the single thioethers ( Figure 3—figure supplement 3 ) . Full cyt c is released at least 62 ± 5% , 20mer 14 ± 3% , and the SXXCH variant 5 ± 2% from HCCS . We conclude that folding of cyt c is necessary for optimal release from the HCCS active site ( step 4 ) . For CcsBA , we have proposed that biogenesis involves heme trafficking from an internal membrane site , liganded by two TM-His residues , to an external domain called the WWD/P-His site ( Figure 6a , Frawley and Kranz , 2009; Sutherland et al . , 2018b ) . Subsequently , it is proposed that heme from the WWD/P-His site is stereochemically attached to apocyt c ( CXXCH ) ( Sutherland et al . , 2018b ) . Preliminary data on the spectral properties of peptides with heme attached by CcsBA appear to be released , unlike HCCS . Perhaps this release is mediated by the highly conserved WWD domain in the bacterial synthase , which interfaces with the edge of heme that faces the CXXCH substrate . In vitro reconstitution with CXXCH peptides and analogs have shown that the substrate requirements for HCCS and CcsBA are quite different . There have been some in vivo studies that suggested that HCCS may require an N-terminally extended region ( from CXXCH ) , yet such approaches do not rule out , for example , folding or stability issues ( Babbitt et al . , 2016; Kleingardner and Bren , 2011; Zhang et al . , 2014 ) . A direct , in vitro approach was needed . Here we synthesized multiple CXXCH peptides ( Figure 3a ) : an 11mer lacking the N-terminal alpha helix 1 sequence , and a 16 and 20mer , which possess it . HCCS only recognizes and attaches heme to the 16 and 20mer but not the 11mer or a 9mer , while CcsBA attaches to all four peptides ( Table 1 , Figure 3—figure supplement 1 ) . Structure of this alpha helix 1 sequence is predicted by PEP-FOLD ( Shen et al . , 2014; Thévenet et al . , 2012 ) to form an alpha helix , consistent with experimental structure of cytochrome c . We conclude that the alpha helix 1 is a critical component recognized by HCCS , and that these peptides ( 16 and 20mers ) present necessary and sufficient structures for recognition ( Figure 3a ) . We used a Gremlin co-evolution/Rosetta approach ( Ovchinnikov et al . , 2017; Ovchinnikov et al . , 2015 ) to determine the structure of HCCS , facilitated by almost a billion years of HCCS evolution ( Babbitt et al . , 2015 ) . Heme was modeled into HCCS , constraining the His154 as an axial ligand , leaving the sixth ligand site open , likely bound to a weak ligand such as water ( Figure 8a ) . Figure 8b displays the minimal 16mer substrate with heme . Heme binds to HCCS via His154 in step 1 ( Figure 1—figure supplement 7 ) , before binding of the 16mer substrate ( step 2 ) . The surface at the proposed active site of HCCS is acidic ( Figure 8a ) , potentially interacting electrostatically with the basic features of alpha helix 1 ( Figure 8b ) . Moreover , during step 2 of proposed model for HCCS function ( Figure 1—figure supplement 7 ) , His19 of apocyt c forms the second axial ligand to heme at the HCCS active site . In all peptides with alpha helix 1 , spectral analysis indicated that His 19 formed this second axial ligand . We have confirmed the requirement for His19 , testing three His19 variants of the 20mer peptide , H19M , H19A , and H19K ( Figure 3—figure supplement 1 ) . Only the H19M variant showed a low amount of attached heme , with a spectrum that also implies methionine can replace the weak ligand in HCCS ( Figure 3—figure supplement 1 ) . The minimal 16mer peptide , including the His19 ligand , is modeled into HCCS in Figure 8c . These models provide an initial structural basis for HCCS function , including testable predictions . For example , to test the electrostatic hypothesis , we changed all basic lysines to aspartates , retaining a predicted alpha helix 1 ( Figure 3—figure supplement 1 , Table 1 ) . Heme was not attached to this peptide by HCCS , suggesting that the positive charge in alpha helix 1 is important . For CcsBA , a limited sequence that includes CXXCH is necessary and sufficient . Results using peptide analogs with non-standard thiol amino acids are consistent with a more stringent requirement for the CXXCH motif for CcsBA . In this respect , because bacteria often recognize hundreds of c-type cytochromes ( i . e . CXXCH motifs ) it makes evolutionary sense to recognize only the CXXCH motif , than to have a more demanding three-dimensional structure . We investigated the importance of the two thiols in CXXCH for recognition and thioether formation by synthesizing peptide analogs containing cysteine substitutions . Since the 20mer had heme attached by both HCCS and CcsBA , we used it as the base sequence for cysteine substitutions , as summarized in Table 1 . Serine , homocysteine , and d-cysteine were substituted for each cysteine ( of CXXCH ) . All substitutions were recognized by HCCS , having at least a single thioether , a result we attribute to the extended recognition requirement ( alpha helix 1 ) and the His19 axial ligand ( of CXXCH ) . We propose that this allows less dependency on the CXXCH motif . In contrast , CcsBA only recognized and attached heme to the variant with the first cysteine substituted by homocysteine . We propose that this is consistent with a more demanding recognition of the CXXCH motif at the active site of CcsBA . Clearly the serine substitutions cannot form thioethers . Consistent with this , for HCCS only the remaining cysteine had a thioether bond to heme . For HCCS , the only thiol amino acid analog with a thioether was d-cysteine substituted for the first cysteine ( 20mer DCys15 in Table 1 ) . This suggests some rotational flexibility at the first thiol , but no ‘vertical’ flexibility since the homocysteine at Cys15 did not form a thioether . In contrast , for CcsBA , since only the homocysteine at Cys15 was attached , it may possess less rotational flexibility but more ‘vertical’ flexibility at the first cysteine at its active site . It is remarkable that in spite of the commonly proposed universal CXXCH motif for all c-type cytochromes , the bacterial and mitochondrial cyt c synthases have evolved quite different recognition determinants and thus , mechanisms . As discussed above , this is likely due to the limited c-type cyts in mitochondria ( i . e . cyt c/cyt c1 ) but the large repertoire of c-type cyts in bacteria , each possessing CXXCH , and sometimes dozens of CXXCH motifs in a single bacterial protein . Multiple approaches were used to demonstrate that CXXCH peptides with alpha helix one are not released by HCCS , with single cysteine substitutions even more tightly bound ( see also Figure 3—figure supplement 3 ) . Evidence is presented that peptides recognized by HCCS inhibit heme attachment to subsequently added cyt c . Thus , peptides are inhibitors . The basis for such inhibition will require more investigations , but two possible mechanisms are noted here . First , the peptides specifically use the heme at the HCCS active site , thus precluding use by cyt c . Such a mechanism of inhibition might be considered specific dead-end use of a substrate . Second , in principle , tightly bound peptides that are not released may inhibit subsequent binding of new heme and cyt c substrates; thus , they act as substrate analog type inhibitors . Future studies will further explore these possibilities with both the mitochondrial HCCS and bacterial cyt c synthases . Escherichia coli strains were grown in Luria-Bertani ( LB; Difco ) broth with selective antibiotics and inducing reagents as required . Antibiotic/induction concentrations: carbenicillin , 50 µg/ml; chloramphenicol , 20 µg/ml , isopropyl β -D-1-thiogalactopyranoside ( IPTG; Gold Biotechnology ) , 1 . 0 mM or 0 . 1 mM; arabinose ( alfa Aesar ) , 0 . 2% ( wt/vol ) . Cloning was performed using E . coli NEB-5 α with the QuikChange II site-directed mutagenesis kit ( Agilent Technologies ) following the manufacturer’s instructions . Strains , plasmid , and primer lists are provided in Supplementary file 1 in supplemental material . GST-HCCS purifications were performed as previously described ( San Francisco et al . , 2013 ) . Briefly , starter cultures ( 100 ml ) were grown overnight at 37°C and 200 rpm . Starter cultures were used to inoculate 1 l of LB supplemented with appropriate antibiotics . One liter cultures were grown at 37°C and 120 rpm for 1 hr , and next expression of GST-HCCS was induced with 0 . 1 mM IPTG . Cells were harvested after 5 hr by centrifugation at 4500 g and cell pellets were stored at −80°C . Cell pellets were resuspended in PBS supplemented with 1 mM phenylmethansulfonul fluoride ( PMSF ) , lysed by sonication ( Branson250 sonicator ) , and cleared of cell debris by centrifugation at 24 , 000 g for 30 min at 4°C . Separation of soluble and membrane fractions was achieved by high-speed ultracentrifugation at 100 , 000 g for 45 min at 4°C . Membrane pellets were solubilized in 50 mM Tris pH 8 , 150 mM NaCl , and 1% Triton X-100 for 1 hr on ice . Solubilized membranes were added to glutathione agarose ( Pierce ) for batch pulldown . Note , GST-HCCS used for in vitro reactions were heme loaded at this step by addition of 5 µM hemin during batch pulldown ( see below ) . Columns were washed by gravity flow and eluted in 50 mM Tris pH8 , 150 mM NaCl , and 0 . 02% Triton X-100 supplemented with 20 mM glutathione . Elution was concentrated using Amicon Ultra Centrifugal Filters ( Millipore ) , and protein concentration was determined by Bradford assay ( Sigma ) . GST-CcsBA and GST-CcsBA:His purifications were performed as previously described ( Frawley and Kranz , 2009; Sutherland et al . , 2018b ) . Briefly , 5 ml starter cultures were grown for ~8 hr at 37°C with rocking . Starter cultures were diluted 1:200 into 1 l LB with selective antibiotics and grown overnight at 24°C and 240 rpm to saturation . Expression of GST-CcsBA was induced with 1 mM IPTG for 6 hr , cells were harvested at 4500 g , and cell pellets were stored at −80°C . Cell pellets were resuspended in Resin Buffer ( 20 mM Tris pH8 , 100 mM NaCl ) supplemented with 1 mM PMSF and 1 mg/ml egg white lysozyme ( Sigma-Aldrich ) . Cells were lysed , cleared of debris , and separation of membrane fraction was performed as described for GST-HCCS above . Membrane pellets were solubilized in Resin Buffer with 1% n-dodecyl-β-d-maltopyranoside ( DDM; Anatrace ) and batch purified for 2 hr with glutathione agarose ( Pierce ) . Columns were washed by gravity flow using Resin Buffer with 0 . 02% DDM and eluted in Resin Buffer with 0 . 02% DDM and 20 mM glutathione . Elution was concentrated using Amicon Ultra Centrifugal Filters ( Millipore ) , and protein concentration was determined by Bradford assay ( Sigma ) . GST-CcsBA:His , GSTCcsBA:His and *CcsBA:His were performed as described above for GST-CcsBA with the following modifications . Batch pulldowns were performed using Talon Affinity Metal Resin ( Takara ) . Gravity flow washes were performed in Resin buffer with 0 . 02% DDM supplemented with 0 mM imidazole ( wash 1 ) , 2 mM imidazole ( wash 2 ) , and 5 mM imidazole ( wash 3 ) . Protein was eluted in Resin Buffer with 0 . 02% DDM and 125 mM imidazole . Elution was concentrated using Amicon Ultra Centrifugal Filters ( Millipore ) , and protein concentration was determined by Bradford assay ( Sigma ) . To increase heme co-purification of GST-HCCS , exogenous heme was added to the affinity purification , resulting in ‘heme loaded’ HCCS . During binding of the solubilized membrane preparations to glutathione agarose , hemin ( 1 . 3 mg/ml in DMSO ) was added to a final concentration of 5 µM . Heme loading increases HCCS heme co-purification from ~10% to ~30% . To determine the optimal concentration of hemin , a range of values was tested ( see Figure 1—figure supplement 2 ) . After batch affinity purification , the column was washed ( removing unbound hemin ) and eluted as described in protein purification section . Samples were prepared in loading dye at 1:1 ( v/v ) that did not contain reducing agents and were not boiled to maintain heme signals . Samples were separated by SDS–PAGE or Tricine SDS–PAGE ( peptides ) . Heme staining was performed by transfer to nitrocellulose and detection of heme signal using the SuperSignal Femto kit ( Pierce ) ( Feissner et al . , 2003 ) , with imaging on a LI-COR odyssey Fc ( LI-Cor Biosciences ) or by in-gel heme stain with N , N , N’ , N’-tetramethylbenzidine ( TMBZ ) ( Feissner et al . , 2003; Francis and Becker , 1984; Thomas et al . , 1976 ) . Total protein was detected by staining SDS–PAGE gels with Coomassie stain or nitrocellulose blots with SYPRO Ruby Blot Stain according to the manufacturer’s instructions ( Molecular Probes ) . Immunoblots using an antibody specific to equine heart cytochrome c ( Cocalico Biologics ) were performed as previously described ( Babbitt et al . , 2016 ) . UV–vis absorption spectroscopy was obtained with a Shimadzu UV-1800 spectrophotometer . Spectra were recorded in the assay buffer and under aerobic or anaerobic conditions as indicated . Heme quantification by Soret absorbance was performed with 50 µg of protein . Pyridine hemochrome assays were performed as previously described ( Berry and Trumpower , 1987 ) in the assay buffer . If needed , sodium dithionite powder was used for protein reduction . Maturation of peptides by CcsBA was assessed by measuring the maximum or minimum of the second derivative of the final reaction spectrum . In vitro reconstitutions were performed aerobically ( HCCS ) or anaerobically ( HCCS and CcsBA ) . For anaerobic reactions , all reagents were equilibrated with N2 ( 95% ) and H2 ( 5% ) in a Coy anaerobic airlock chamber . Affinity purified synthase ( HCCS or CcsBA ) was combined with apo equine heart cytochrome c or apo peptide at indicated concentrations . Apo cyt c and peptide concentrations used were determined to be within the range for maximal heme attachment as determined by a titration . An initial spectra and sample for SDS–PAGE analysis were obtained . Five millimolar DTT was added to initiate the reaction . Reactions were placed at 37°C , and spectra and gel samples were taken at indicated time points . Gel samples were immediately placed in loading dye ( 1:1 v/v ) to stop the reaction . Apocytochrome c preparation was modified from Babul and Stellwagen , 1972 . Cytochrome c from equine ( horse ) heart was obtained from Sigma , and a 1 ml 10 mg/ml solution was prepared in water . To remove heme , 200 µl of glacial acetic acid and 1 . 5 ml of 0 . 8% silver sulfate were added and the solution was incubated at 44°C for 4 hr . Sample was dialyzed in 0 . 2 M acetic acid overnight at 4°C . To precipitate apo cytochrome c and remove silver , sample was transferred to a conical tube and 10 volumes of cold acid acetone were added . Apo cytochrome c was pelleted by spinning at 15 , 000 rpm for 20 min at 4°C . The pellet was washed with acid acetone and pelleted three times . The apoprotein was resuspended in 0 . 2M acetic acid ( ~1 ml ) , and solid urea was added until the solution turned clear . A 25-fold molar excess of 2-mercaptoethonal was added and incubated at room temperature to remove silver sulfate . Apo cytochrome c was clarified by centrifugation at 12 , 000 rpm for 10 min at room temperature . Supernatant was dialyzed in 0 . 2 M acetic acid overnight and buffer exchanged into PBS by concentration in an Amicon Concentrator with 3 kDa molecular weight cutoff . Protein concentration was determined using a BSA standard curve and Coomassie protein staining on SDS–PAGE . Affinity purified proteins or indicated in vitro reactions were resolved on an Agilent 1100 HPLC system equipped with an Agilent SEC-3 column in the purification or in vitro reaction buffer . To determine whether in vitro synthesized cytochrome c ( or peptide ) was released from the synthase , GST-HCCS bound to glutathione agarose ( 75 µl ) was combined with apocyt c ( or peptide ) under standard in vitro conditions ( 100 µl volume in addition to the 75 µl of beads ) . After a 1 hr reaction , the glutathione agarose-bound GST-HCCS were pelleted , and the supernatant was collected . Subsequently , the beads were washed to allow for analysis of protein retained on the beads . The bead fraction and supernatant were separated by SDS–PAGE and heme stained to determine which fraction contained heme attached cyt c . The amount of holo cyt c/peptide matured and released was quantitated using Image J ( Rasband , 1997 ) by determining the ratio of the supernatant derived heme band with the total peptide band signal ( beads plus supernatant ) . The supernatant was further analyzed by UV–vis spectroscopy . The supernatant from the released product assay ( above ) was extracted and pooled in an anaerobic environment and then concentrated in a 3K VivaspinTurbo cutoff filter ( Sartorius ) to obtain 300 μl of 0 . 4 mg/ml cyt c as determined by heme absorbance at 550 nm ( ext . coef . 29 . 5 mM−1 cm−1 ) . The near UV ( 500–300 nm ) signal was measured on a Jasco J-815 at room temperature in the reaction buffer ( 20 mM Tris pH 8 . 0 , 100 mM NaCl , 0 . 02% DDM , 5 mM DTT ) . The machine sensitivity was 100 mdeg , the data pitch was 0 . 5 nm , the scanning mode was continuous , the scanning speed was 50 nm/min , the response rate was 1 s , the bandwidth was 1 nm , and five accumulations were taken ( Mendez et al . , 2017 ) . A blank sample was subtracted . To compare the absorbance of the released assay product to human cyt c , each CD spectra was subtracted from a blank sample and divided by the absorbance of the protein in the CD machine . The samples were overlaid and coincide with each other . Redox potential of in vitro synthesized equine heart cytochrome c was determined by a modified Massey method as described in Sutherland et al . , 2016 , with the following modifications: The absorbance change of heme was monitored at the alpha peak at 550 nm ( negligible contribution from reference dye ) and the reduction of the reference dye , dichlorophenolindophenol , at 636 nm ( negligible contribution from heme ) . To determine whether the CXXCH containing peptides inhibited maturation of cytochrome c ( i . e . heme attachment ) , a two-step reaction was performed . Step 1: 10 μM GST-HCCS ( 30% heme occupancy ) was combined with 10 μM apo peptide for a 1 hr in vitro reaction . After 1 hr , UV–vis spectra were performed , and a sample was collected for gel analysis . Step 2: 20 μM apo cytochrome c was added to the reaction . After 1 hr , UV–vis spectra were performed and a sample was collected for gel analysis . To determine whether cytochrome c maturation was inhibited or not inhibited by the peptide , heme stain and Coomassie total protein stain were performed . Assays were performed as in Sutherland et al . , 2018b . Detailed methods are provided in Supplemental Methods . The HCCS structure was produced using Rosetta , which was informed by structural motifs ( Robetta ) and coevolutionary data ( Gremlin ) as has been described ( Ovchinnikov et al . , 2017; Ovchinnikov et al . , 2015; Sutherland et al . , 2018b; Sutherland et al . , 2018a ) , and will be detailed later .	From tiny bacteria to the tallest trees , most life on Earth carries a protein called cytochrome c , which helps to create the energy that powers up cells . Cytochrome c does so thanks to its heme , a molecule that enables the chemical reactions required for the energy-creating process . Despite both relying on cytochrome c , animals and bacteria differ in the enzyme they use to attach the heme to the cytochrome . Spotting variations in how this ‘cytochrome c synthase’ works would help to find compounds that deactivate the enzyme in bacteria , but not in humans . However , studying cytochrome c synthase in living cells is challenging . To bypass this issue , Sutherland , Mendez , Babbitt et al . successfully reconstituted cytochrome c synthases from humans and bacteria in test tubes . This allowed them to examine in detail which structures the enzymes recognize to spot where to attach the heme onto their target . The experiments revealed that human and bacterial synthases actually rely on different parts of the cytochrome c to orient themselves . Different short compounds could also block either the human or bacterial enzyme . Variations between human and bacterial cytochrome c synthase could lead to new antibiotics which deactivate the cytochrome and kill bacteria while sparing patients . The next step is to identify molecules that specifically interfere with cytochrome c synthase in bacteria , and could be tested in clinical trials .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2021	In vitro reconstitution reveals major differences between human and bacterial cytochrome c synthases
Tightly regulated Ca2+ homeostasis is a prerequisite for proper cardiac function . To dissect the regulatory network of cardiac Ca2+ handling , we performed a chemical suppressor screen on zebrafish tremblor embryos , which suffer from Ca2+ extrusion defects . Efsevin was identified based on its potent activity to restore coordinated contractions in tremblor . We show that efsevin binds to VDAC2 , potentiates mitochondrial Ca2+ uptake and accelerates the transfer of Ca2+ from intracellular stores into mitochondria . In cardiomyocytes , efsevin restricts the temporal and spatial boundaries of Ca2+ sparks and thereby inhibits Ca2+ overload-induced erratic Ca2+ waves and irregular contractions . We further show that overexpression of VDAC2 recapitulates the suppressive effect of efsevin on tremblor embryos whereas VDAC2 deficiency attenuates efsevin's rescue effect and that VDAC2 functions synergistically with MCU to suppress cardiac fibrillation in tremblor . Together , these findings demonstrate a critical modulatory role for VDAC2-dependent mitochondrial Ca2+ uptake in the regulation of cardiac rhythmicity . During development , well-orchestrated cellular processes guide cells from diverse lineages to integrate into the primitive heart tube and establish rhythmic and coordinated contractions . While many genes and pathways important for cardiac morphogenesis have been identified , molecular mechanisms governing embryonic cardiac rhythmicity are poorly understood . The findings that Ca2+ waves traveling across the heart soon after the formation of the primitive heart tube ( Chi et al . , 2008 ) and that loss of function of key Ca2+ regulatory proteins , such as the L-type Ca2+ channel , Na/K−ATPase and sodium-calcium exchanger 1 ( NCX1 ) , severely impairs normal cardiac function ( Rottbauer et al . , 2001; Shu et al . , 2003; Ebert et al . , 2005; Langenbacher et al . , 2005 ) , indicate an essential role for Ca2+ handling in the regulation of embryonic cardiac function . Ca2+ homoeostasis in cardiac muscle cells is tightly regulated at the temporal and spatial level by a subcellular network involving multiple proteins , pathways , and organelles . The release and reuptake of Ca2+ by the sarcoplasmic reticulum ( SR ) , the largest Ca2+ store in cardiomyocytes , constitutes the primary mechanism governing the contraction and relaxation of the heart . Ca2+ influx after activation of the L-type Ca2+ channel in the plasma membrane induces the release of Ca2+ from the SR via ryanodine receptor ( RyR ) channels , which leads to an increase of the intracellular Ca2+ concentration and cardiac contraction . During diastolic relaxation , Ca2+ is transferred back into the SR by the SR Ca2+ pump or extruded from the cell through NCX1 . Defects in cardiac Ca2+ handling and Ca2+ overload , for example during cardiac ischemia/reperfusion or in long QT syndrome , are well known causes of contractile dysfunction and many types of arrhythmias including early and delayed afterdepolarizations and Torsade des pointes ( Bers , 2002; Choi et al . , 2002; Yano et al . , 2008; Greiser et al . , 2011 ) . Ca2+ crosstalk between mitochondria and ER/SR has been noted in many cell types and the voltage-dependent anion channel ( VDAC ) and the mitochondrial Ca2+ uniporter ( MCU ) serve as primary routes for Ca2+ entry through the outer and inner mitochondrial membranes , respectively ( Rapizzi et al . , 2002; Bathori et al . , 2006; Shoshan-Barmatz et al . , 2010; Baughman et al . , 2011; De Stefani et al . , 2011 ) . In the heart , mitochondria are tethered to the SR and are located in close proximity to Ca2+ release sites ( García-Pérez et al . , 2008; Boncompagni et al . , 2009; Hayashi et al . , 2009 ) . This subcellular architecture exposes the mitochondria near the Ca2+ release sites to a high local Ca2+ concentration that is sufficient to overcome the low Ca2+ affinity of MCU and facilitates Ca2+ crosstalk between SR and mitochondria ( García-Pérez et al . , 2008; Dorn and Scorrano , 2010; Kohlhaas and Maack , 2013 ) . Increase of the mitochondrial Ca2+ concentration enhances energy production during higher workload and dysregulation of SR-mitochondrial Ca2+ signaling results in energetic deficits and oxidative stress in the heart and may trigger programmed cell death ( Brandes and Bers , 1997; Maack et al . , 2006; Kohlhaas and Maack , 2013 ) . However , whether SR-mitochondrial Ca2+ crosstalk also contributes significantly to cardiac Ca2+ signaling during excitation-contraction coupling requires further investigation . In zebrafish , the tremblor ( tre ) locus encodes a cardiac-specific isoform of the Na+/Ca2+ exchanger 1 , NCX1h ( also known as slc8a1a ) ( Ebert et al . , 2005; Langenbacher et al . , 2005 ) . The tre mutant hearts lack rhythmic Ca2+ transients and display chaotic Ca2+ signals in the myocardium leading to unsynchronized contractions resembling cardiac fibrillation ( Langenbacher et al . , 2005 ) . In this study , we used tre as an animal model for aberrant Ca2+ handling-induced cardiac dysfunction and took a chemical genetic approach to dissect the Ca2+ regulatory network important for maintaining cardiac rhythmicity . A synthetic compound named efsevin was identified from a suppressor screen due to its potent ability to restore coordinated contractions in tre . Using biochemical and genetic approaches we show that efsevin interacts with VDAC2 and potentiates its mitochondrial Ca2+ transporting activity and spatially and temporally modulates cytosolic Ca2+ signals in cardiomyocytes . The important role of mitochondrial Ca2+ uptake in regulating cardiac rhythmicity is further supported by the suppressive effect of VDAC2 and MCU overexpression on cardiac fibrillation in tre . Homozygous tre mutant embryos suffer from Ca2+ extrusion defects and manifest chaotic cardiac contractions resembling fibrillation ( Ebert et al . , 2005; Langenbacher et al . , 2005 ) . To dissect the regulatory network of Ca2+ handling in cardiomyocytes and to identify mechanisms controlling embryonic cardiac rhythmicity , we screened the BioMol library and a collection of synthetic compounds for chemicals that are capable of restoring heartbeat either completely or partially in tre embryos . A dihydropyrrole carboxylic ester compound named efsevin was identified based on its ability to restore persistent and rhythmic cardiac contractions in tre mutant embryos in a dose-dependent manner ( Figure 1A , E , and Videos 1–4 ) . To validate the effect of efsevin , we assessed cardiac performance of wild type , tre and efsevin-treated tre embryos ( Nguyen et al . , 2009 ) . Fractional shortening of efsevin treated tre mutant hearts was comparable to that of their wild type siblings and heart rate was restored to approximately 40% of that observed in controls ( Figure 1B–D ) . Periodic local field potentials accompanying each heartbeat were detected in wild type and efsevin-treated tre embryos using a microelectrode array ( Figure 1F–H ) . Furthermore , while only sporadic Ca2+ signals were detected in tre hearts , in vivo Ca2+ imaging revealed steady Ca2+ waves propagating through efsevin-treated tre hearts ( Figure 1I , Videos 5–7 ) , demonstrating that cardiomyocytes are functionally coupled and that efsevin treatment restores regular Ca2+ transients in tre hearts . 10 . 7554/eLife . 04801 . 003Figure 1 . Efsevin restores rhythmic cardiac contractions in zebrafish tremblor embryos . ( A ) Line scans across the atria of Tg ( myl7:GFP ) embryonic hearts at 48 hpf . Rhythmically alternating systoles and diastoles are recorded from vehicle- ( upper left ) or efsevin- treated wild type ( upper right ) and efsevin-treated tre ( lower right ) embryos , while only sporadic unsynchronized contractions are recorded from vehicle-treated tre embryos ( lower left ) . ( B , C ) Fractional shortening ( FS ) deduced from the line-scan traces . While cardiac contraction was not observed in tre , efsevin-treated wild type and tre hearts have similar levels of FS to those observed in control hearts . Ventricular FS of wild type v . s . wild type + efsevin vs tre + efsevin: 39 ± 0 . 6% , n = 8 vs 39 ± 1% , n = 10 vs 35 ± 3% , n = 6; and Atrial FS: 37 ± 1% , n = 11 vs 35 ± 2% , n = 11 vs 33 ± 2% , n = 15 . ( D ) While efsevin restored a heart rate of 46 ± 2 beats per minute ( bpm ) in tre embryos , same treatment does not affect the heart rate in wild type embryos ( 126 ± 2 bpm in vehicle-treated embryos vs 123 ± 3 bpm in efsevin-treated wild-type embryos ) . * , p < 0 . 001 by one-way ANOVA . ( E ) Dose-dependence curve for efsevin . The tre embryos were treated with various concentrations of efsevin from 24 hpf and cardiac contractions were analyzed at 48 hpf . ( F–H ) Representative time traces of local field potentials for wild type ( F ) , tre ( G ) and efsevin-treated tre ( H ) embryos clearly display periods of regular , irregular , and restored periodic electrical activity . ( I ) In vivo optical maps of Ca2+ activation represented by isochronal lines every 33 ms recorded from 36 hpf wild type ( left ) , tre ( center ) and efsevin-treated tre ( right ) embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 00310 . 7554/eLife . 04801 . 004Video 1 . The video shows a heart of a wild-type zebrafish embryo at 2 dpf . Robust rhythmic contractions can be observed in atrium and ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 00410 . 7554/eLife . 04801 . 005Video 2 . This video shows a heart of a tremblor embryo at 2 dpf . Embryos of the mutant line tremblor display only local , unsynchronized contractions , comparable to cardiac fibrillation . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 00510 . 7554/eLife . 04801 . 006Video 3 . This video shows a heart of a tremblor embryo at 2 dpf treated with efsevin . Treatment of tremblor embryos with efsevin restores rhythmic contractions with comparable atrial fractional shortening compared to wild-type embryos and approximately 40% of wild-type heart rate . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 00610 . 7554/eLife . 04801 . 007Video 4 . The video shows a heart of a wild-type zebrafish embryo at 2 dpf treated with efsevin . Treatment of wild-type embryos with efsevin did not affect cardiac performance , indicated by robust , rhythmic contractions comparable to untreated wild-type embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 00710 . 7554/eLife . 04801 . 008Video 5 . Heat map of Ca2+ transients recorded in 1 day old wild type heart . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 00810 . 7554/eLife . 04801 . 009Video 6 . Heat map of Ca2+ transients recorded in 1 day old tremblor heart . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 00910 . 7554/eLife . 04801 . 010Video 7 . Heat map of Ca2+ transients recorded in 1 day old efsevin treated tremblor heart . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 010 We next examined whether efsevin could suppress aberrant Ca2+ homeostasis-induced arrhythmic responses in mammalian cardiomyocytes . Mouse embryonic stem cell-derived cardiomyocytes ( mESC-CMs ) establish a regular contraction pattern with rhythmic Ca2+ transients ( Figure 2A , B , E , F ) . Mimicking Ca2+ overload by increasing extracellular Ca2+ levels was sufficient to disrupt normal Ca2+ cycling and induce irregular contractions in mESC-CMs ( Figure 2C , E , F ) . Remarkably , efsevin treatment restored rhythmic Ca2+ transients and cardiac contractions in these cells ( Figure 2D–F ) . Similar effect was observed in human embryonic stem cell-derived cardiomyocytes ( hESC-CMs ) ( Figure 2G ) . Together , these findings suggest that efsevin targets a conserved Ca2+ regulatory mechanism critical for maintaining rhythmic cardiac contraction in fish , mice and humans . 10 . 7554/eLife . 04801 . 011Figure 2 . Efsevin reduces arrhythmogenic events in ES cell-derived cardiomyocytes . ( A ) Line-scan analysis of Ca2+ transients in mESC-CMs after 10 days of differentiation . ( B–D ) Representative graph of Ca2+ transients detected in mESC-CMs ( B ) . After treatment with 10 mM Ca2+ for 10 min , the EB showed an irregular pattern of Ca2+ transients ( C ) . Efsevin treatment restores regular Ca2+ transients under Ca2+ overload conditions in mESC-CMs ( D ) . ( E ) Plotted intervals between peaks of Ca2+ signals detected in mESC-CMs prior to treatment ( control ) , in 10 mM Ca2+ext ( Ca2+ ) and in 10 mM Ca2+ext+10 μM efsevin ( Ca2++efsevin ) . ( F , G ) Plotted intervals of contractions detected in EBs prior to treatment ( control ) , in 10 mM Ca2+ext ( Ca2+ ) and in 10 mM Ca2+ext + 10 μM efsevin ( Ca2+ + efsevin ) for mouse ESC-CMs ( F ) and 5 mM Ca2+ext ( Ca2+ ) and in 5 mM Ca2+ext + 5 μM efsevin ( Ca2+ + efsevin ) for human ESC-CMs ( G ) . * , p < 0 . 001 by F-test . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 011 To identify the protein target of efsevin , we generated a N-Boc-protected 2-aminoethoxyethoxyethylamine linker-attached efsevin ( efsevinL ) ( Figure 3A , C ) . This modified compound retained the activity of efsevin to restore cardiac contractions in ncx1h deficient embryos ( Figure 3B , D ) and was used to create efsevin-conjugated agarose beads ( efsevinLB ) . A 32kD protein species was detected from zebrafish lysate due to its binding ability to efsevinLB and OK-C125LB , an active efsevin derivative conjugated to beads , but not to beads capped with ethanolamine alone or beads conjugated with an inactive efsevin analog ( OK-C19LB ) ( Figure 3A–E ) . Furthermore , preincubation of zebrafish lysate with excess efsevin prevented the 32kD protein from binding to efsevinLB or OK-C125LB ( Figure 3E ) . Mass spectrometry analysis revealed that this 32kD band represents a zebrafish homologue of the mitochondrial voltage-dependent anion channel 2 ( VDAC2 ) ( Figure 3F and Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 04801 . 012Figure 3 . VDAC2 is a protein target of efsevin . ( A ) Structures of efsevin and two derivatives , OK-C125 and OK-C19 . ( B ) Efsevin and OK-C125 restored rhythmic contractions in the majority of tremblor embryos , whereas OK-C19 failed to rescue the tremblor phenotype . ( C ) Structures of linker-attached compounds ( indicated by superscript L ) . ( D ) Compounds efsevinL and OK-C125L retained their ability to restore rhythmic contractions in NCX1hMO injected embryos , while the inactive derivative OK-C19L was still unable to induce rhythmic contraction . ( E ) Affinity agarose beads covalently linked with efsevin ( efsevinLB ) or OK-C125 ( OK-C125LB ) pulled down 2 protein species from zebrafish embryonic lysate , whereof one , the 32 kD upper band , was sensitive to competition with a 100-fold excess free efsevinL . The 32 kD band was not detected in proteins eluted from beads capped with ethanolamine alone ( beadsC ) or beads linked to OK-C19 ( OK-C19LB ) . Arrowheads point to the 32kD bands . ( F ) Mass Spectrometry identifies the 32kD band as VDAC2 . Peptides identified by mass spectrometry ( underlined ) account for 30% of the total sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 01210 . 7554/eLife . 04801 . 013Figure 3—figure supplement 1 . Mass Spectometry identifies VDAC2 as the target of efsevin . Image shows an example of the identification of VDAC2 peptide . Diagnostic b- and y-series ions are shown in red and blue , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 013 VDAC2 is expressed in the developing zebrafish heart ( Figure 4A ) , making it a good candidate for mediating efsevin’s effect on cardiac Ca2+ handling . To examine this possibility , we injected in vitro synthesized VDAC2 RNA into tre embryos and found that the majority of these embryos had coordinated cardiac contractions similar to those subjected to efsevin treatment ( Figure 4B , Videos 8–11 ) . In addition , we generated myl7:VDAC2 transgenic fish in which VDAC2 expression can be induced in the heart by tebufenozide ( TBF ) ( Figure 4C ) . Knocking down NCX1h in myl7:VDAC2 embryos results in chaotic cardiac movement similar to tre . Like efsevin treatment , induction of VDAC2 expression by TBF treatment restored coordinated and rhythmic contractions in myl7:VDAC2;NCX1h MO hearts ( Figure 4D , Videos 12 , 13 ) . Conversely , knocking down VDAC2 in tre hearts attenuated the suppressive effect of efsevin ( Figure 4E , Videos 14–16 ) . Furthermore , we generated VDAC2 null embryos by the Zinc Finger Nuclease gene targeting approach ( Figure 4G ) . Similar to that observed in morpholino knockdown embryos , homozygous VDAC2LA2256 embryos do not exhibit noticeable morphological defects , but the suppressive effect of efsevin was attenuated in homozygous VDAC2LA2256; NCX1MO embryos ( Figure 4F ) . These findings demonstrate that VDAC2 is a major mediator for efsevin’s effect on ncx1h deficient hearts . 10 . 7554/eLife . 04801 . 014Figure 4 . VDAC2 restores rhythmic cardiac contractions in tre . ( A ) In situ hybridization analysis showed that VDAC2 is expressed in embryonic hearts at 36 hpf ( upper image ) and 48 hpf ( lower image ) . ( B ) Injection of 25 pg in vitro synthesized VDAC2 mRNA restored cardiac contractions in 52 . 9 ± 12 . 1% ( n = 78 ) of 1-day-old tre embryos , compared to 21 . 8 ± 5 . 1% in uninjected siblings ( n = 111 ) . ( C ) Schematic diagram of myl7:VDAC2 construct ( top ) . In situ hybridization analysis showed that TBF treatment induces VDAC2 expression in the heart ( lower panel ) . ( D ) While only ∼20% of myl7:VDAC2;NCX1hMO embryos have coordinated contractions ( n = 116 ) , 52 . 3 ± 2 . 4% of these embryos established persistent , rhythmic contractions after TBF induction of VDAC2 ( n = 154 ) . ( E ) On average , 71 . 2 ± 8 . 8% efsevin treated embryos have coordinated cardiac contractions ( n = 131 ) . Morpholino antisense oligonucleotide knockdown of VDAC2 ( MOVDAC2 ) attenuates the ability of efsevin to suppress cardiac fibrillation in tre embryos ( 45 . 3 ± 7 . 4% embryos with coordinated contractions , n = 94 ) . ( F ) Efsevin treatment restores coordinated cardiac contractions in 76 . 2 ± 8 . 7% NCX1MO embryos , only 54 . 1 ± 3 . 6% VDAC2zfn/zfn;NCX1MO embryos have coordinated contractions ( n = 250 ) . ( G ) Diagram of Zinc finger target sites . VDAC2zfn/zfn carries a 34 bp deletion in exon 3 which results in a premature stop codon ( red asterisk ) . In situ hybridization analysis showing loss of VDAC2 transcripts in VDAC2zfn/zfn embryos . White arrowheads point to the developing heart . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 01410 . 7554/eLife . 04801 . 015Video 8 . This video shows a heart of a wild-type zebrafish embryo at 1 dpf . Robust rhythmic contractions can be observed in atrium and ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 01510 . 7554/eLife . 04801 . 016Video 9 . This video shows a heart of a wild-type zebrafish embryo injected with zebrafish VDAC2 mRNA at 1 dpf . Robust rhythmic contractions can be observed in atrium and ventricle . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 01610 . 7554/eLife . 04801 . 017Video 10 . This video shows a heart of a tremblor embryo at 1 dpf . Tremblor embryos display only local , unsynchronized contractions , comparable to cardiac fibrillation . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 01710 . 7554/eLife . 04801 . 018Video 11 . This video shows a heart of a tremblor embryo injected with zebrafish VDAC2 mRNA at 1 dpf . Overexpression of zebrafish VDAC2 mRNA restores rhythmic contractions in tremblor embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 01810 . 7554/eLife . 04801 . 019Video 12 . This video shows a heart of a 2 dpf Tg-VDAC2 embryo injected with a morpholino targeting NCX1h . Morpholino knock-down of NCX1h results in a fibrillating heart . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 01910 . 7554/eLife . 04801 . 020Video 13 . This video shows a heart of a 2 dpf NCX1h morphant in the Tg-VDAC2 genetic background . TBF treatment induces VDAC2 expression and restores coordinated cardiac contractions . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 02010 . 7554/eLife . 04801 . 021Video 14 . This video shows a heart of a 2 dpf wild type zebrafish embryo injected with a morpholino targeting VDAC2 . Morpholino knockdown of VDAC2 did not have obvious effects on cardiac performance . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 02110 . 7554/eLife . 04801 . 022Video 15 . This video shows a heart of a 2 dpf tremblor mutant embryo injected with a morpholino targeting VDAC2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 02210 . 7554/eLife . 04801 . 023Video 16 . This video shows a heart of a 2 dpf tremblor mutant embryo injected with a morpholino targeting VDAC2 . Efsevin treatment cannot restore coordinated cardiac contractions in the absence of VDAC2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 023 VDAC is an abundant channel located on the outer mitochondrial membrane serving as a primary passageway for metabolites and ions ( Figure 5A ) ( Rapizzi et al . , 2002; Bathori et al . , 2006; Shoshan-Barmatz et al . , 2010 ) . At its close state , VDAC favours Ca2+ flux ( Tan and Colombini , 2007 ) . To examine whether efsevin would modulate mitochondrial Ca2+ uptake via VDAC2 , we transfected HeLa cells with VDAC2 . We noted increased mitochondrial Ca2+ uptake in permeabilized VDAC2 transfected and efsevin-treated cells after the addition of Ca2+ and the combined treatment further enhanced mitochondrial Ca2+ levels ( Figure 5B ) . 10 . 7554/eLife . 04801 . 024Figure 5 . Efsevin enhances mitochondrial Ca2+ uptake . ( A ) HeLa cells were transfected with a flag-tagged zebrafish VDAC2 ( VDAC2flag ) , immunostained against the flag epitope and counterstained for mitochondria with MitoTracker Orange and for nuclei with DAPI . ( B ) Representative traces of mitochondrial matrix [Ca2+] ( [Ca2+]m ) detected by Rhod2 . Arrows denote the addition of Ca2+ . Mitochondrial Ca2+ uptake was assessed when VDAC2 was overexpressed ( left ) , cells were treated with 1 µM efsevin ( middle ) and combination of both at suboptimal doses ( right ) . Control-traces with ruthenium red ( RuRed ) show mitochondrial specificity of the signal . ( C ) Representative traces of cytosolic [Ca2+] ( [Ca2+]c ) changes upon the application of 7 . 5 µM IP3 in the presence ( + ) or absence ( − ) of RuRed . Mitochondrial Ca2+ uptake was assessed by the difference of the – and + RuRed conditions normalized to the total release ( n = 4; mean ± SE ) . ( D ) MEFs overexpressing zebrafish VDAC2 ( polycistronic with mCherry ) were stimulated with 1 μM ATP in a nominally Ca2+ free buffer . Changes in [Ca2+]c and [Ca2+]m were imaged using fura2 and mitochondria-targeted inverse pericam , respectively . Black and gray traces show the [Ca2+]c ( in nM ) and [Ca2+]m ( F0/F mtpericam ) time courses in the absence ( left ) or present ( right ) of efsevin . ( E ) Bar charts: Cell population averages for the peak [Ca2+]c ( left ) , the corresponding [Ca2+]m ( middle ) , and the coupling time ( time interval between the maximal [Ca2+]c and [Ca2+]m responses ) in the presence ( black , n = 24 ) or absence ( gray , n = 28 ) of efsevin . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 02410 . 7554/eLife . 04801 . 025Figure 5—figure supplement 1 . Local Ca2+ delivery between IP3 receptors and VDAC2 . V1V3DKO MEFs were stimulated with 100 μM ATP ( left ) or 2 μM thapsigargan ( Tg ) ( right ) . Changes in [Ca2+]c and [Ca2+]m were imaged using fura2 and mitochondria targeted inverse pericam , respectively . Representative traces obtained in three cells are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 025 Mitochondria are located in close proximity to Ca2+ release sites of the ER/SR and an extensive crosstalk between the two organelles exists ( García-Pérez et al . , 2008; Hayashi et al . , 2009; Brown and O'Rourke , 2010; Dorn and Scorrano , 2010; Kohlhaas and Maack , 2013 ) . We examined whether Ca2+ released from intracellular stores could be locally transported into mitochondria through VDAC2 in VDAC1/VDAC3 double knockout ( V1/V3DKO ) MEFs where VDAC2 is the only VDAC isoform being expressed ( Roy et al . , 2009a ) . While treatments with ATP , an IP3-linked agonist , and thapsigargin , a SERCA inhibitor , stimulated similar global cytoplasmic [Ca2+] elevation in intact cells , only ATP induced a rapid mitochondrial matrix [Ca2+] rise ( Figure 5—figure supplement 1 ) . This finding is consistent with observations obtained in other cell types ( Rizzuto et al . , 1994; Hajnóczky et al . , 1995 ) and suggests that Ca2+ was locally transferred from IP3 receptors to mitochondria through VDAC2 at the close ER-mitochondrial associations . We next investigated whether this process could be modulated by efsevin . In permeabilized V1/V3DKO MEFs , treatment with efsevin increased the amount of Ca2+ transferred into mitochondria during IP3-induced Ca2+ release ( Figure 5C ) . Also , in intact V1/V3 DKO MEFs , efsevin accelerated the transfer of Ca2+ released from intracellular stores into mitochondria during stimulation with ATP ( Figure 5D , E ) . We next examined the effect of efsevin on cytosolic Ca2+ signals in isolated adult murine cardiomyocytes . We found that efsevin treatment induced faster inactivation kinetics without affecting the amplitude or time to peak of paced Ca2+ transients ( Figure 6A ) . Similarly , efsevin treatment did not significantly alter the frequency , amplitude or Ca2+ release flux of spontaneous Ca2+ sparks , local Ca2+ release events , but accelerated the decay phase resulting in sparks with a shorter duration and a narrower width ( Figure 6B ) . These results indicate that by activating mitochondrial Ca2+ uptake , efsevin accelerates Ca2+ removal from the cytosol in cardiomyocytes and thereby restricts local cytosolic Ca2+ sparks to a narrower domain for a shorter period of time without affecting SR Ca2+ load or RyR Ca2+ release . Under conditions of Ca2+ overload , single Ca2+ sparks can trigger opening of neighbouring Ca2+ release units and thus induce the formation of erratic Ca2+ waves ( Figure 6C ) . Efsevin treatment significantly reduced the number of propagating Ca2+ waves in a dosage-dependent manner ( Figure 6C , D ) , demonstrating a potent suppressive effect of efsevin on the propagation of Ca2+ overload-induced Ca2+ waves and suggesting that efsevin could serve as a pharmacological tool to manipulate local Ca2+ signals . 10 . 7554/eLife . 04801 . 026Figure 6 . Effects of efsevin on isolated cardiomyocytes . ( A ) Electrically paced Ca2+ transients at 0 . 5 Hz ( top ) . Normalized quantification of Ca2+ transient parameters reveals no difference for transient amplitude ( efsevin-treated at 98 . 6 ± 4 . 5% of vehicle-treated ) and time to peak ( 95 ± 3 . 9% ) , but a significant decrease for the rate of decay ( 82 . 8 ± 4% of vehicle- for efsevin-treated ) ( lower panel ) . ( B ) Representation of typical Ca2+ sparks of vehicle- and efsevin treated cardiomyocytes ( top ) . No differences were observed for spark frequency ( 101 . 1 ± 7 . 7% for efsevin- compared to vehicle-treated ) , maximum spark amplitude ( 101 . 6 ± 2 . 5% ) and Ca2+ release flux ( 98 . 7 ± 2 . 8% ) . In contrast , the decay phase of the single spark was significantly faster in efsevin treated cells ( 82 . 5 ± 2 . 1% of vehicle-treated ) . Consequently , total duration of the spark was reduced to 85 . 7 ± 2% and the total width was reduced to 89 . 5 ± 1 . 4% of vehicle-treated cells . * , p < 0 . 05; *** , p < 0 . 001 . ( C ) Increasing concentrations of extracellular Ca2+ induced a higher frequency of spontaneous propagating Ca2+ waves in isolated adult murine ventricular cardiomyocytes . Efsevin treatment reduced Ca2+ waves in a dose-dependent manner . ( D ) Quantitative analysis of spontaneous Ca2+ waves spanning more than half of the entire cell . Addition of 1 µM efsevin reduced Ca2+ waves to approximately half . Increasing the concentration of efsevin to 10 µM further reduced the number of spontaneous Ca2+ waves and 25 µM efsevin almost entirely blocked the formation of Ca2+ waves . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 026 We hypothesize that efsevin treatment/VDAC2 overexpression suppresses aberrant Ca2+ handling-associated arrhythmic cardiac contractions by buffering excess Ca2+ into mitochondria . This hypothesis predicts that activating other mitochondrial Ca2+ uptake molecules would likewise restore coordinated contractions in tre . To test this model , we cloned zebrafish MCU and MICU1 , an inner mitochondrial membrane Ca2+ transporter and its regulator ( Perocchi et al . , 2010; Baughman et al . , 2011; De Stefani et al . , 2011; Mallilankaraman et al . , 2012; Csordas et al . , 2013 ) . In situ hybridization showed that MCU and MICU1 were expressed in the developing zebrafish heart ( Figure 7A ) and their expression levels were comparable between the wild type and tre hearts ( Figure 7—figure supplement 1 ) . Overexpression of MCU restored coordinated contractions in tre , akin to what was observed with VDAC2 ( Figure 7B ) . In addition , tre embryos injected with suboptimal concentrations of MCU or VDAC2 had a fibrillating heart , but embryos receiving both VDAC2 and MCU at the suboptimal concentration manifested coordinated contractions ( Figure 7C ) , demonstrating a synergistic effect of these proteins . Furthermore , overexpression of MCU failed to suppress the tre phenotype in the absence of VDAC2 activity and VDAC2 could not restore coordinated contractions in tre without functional MCU ( Figure 7B , D ) . Similar results were observed by manipulating MICU1 activity ( Figure 7E , F ) . Together , these findings indicate that mitochondrial Ca2+ uptake mechanisms on outer and inner mitochondrial membranes act cooperatively to regulate cardiac rhythmicity . 10 . 7554/eLife . 04801 . 027Figure 7 . Mitochondria regulate cardiac rhythmicity through a VDAC2-dependent mechanism . ( A ) MCU and MICU1 are expressed in the developing zebrafish hearts ( arrowhead ) . ( B ) Overexpression of MCU is sufficient to restore coordinated cardiac contractions in tre embryos ( 47 . 1 ± 1 . 6% embryos , n = 112 as opposed to 18 . 3 ± 5 . 3% of uninjected siblings , n = 64 ) while this effect is significantly attenuated when co-injected with morpholino antisense oligonucleotide targeted to VDAC2 ( 27 . 1 ± 1 . 9% embryos , n = 135 ) . ( C ) Suboptimal overexpression of MCU ( MCUS ) and VDAC2 ( VDAC2S ) in combination is able to suppress cardiac fibrillation in tre embryos ( 42 . 9 ± 2 . 6% embryos , n = 129 ) . ( D ) The ability of VDAC2 to restore rhythmic contractions in tre embryos ( 48 . 5 ± 3 . 5% embryos , n = 111 ) is significantly attenuated when MCU is knocked down by antisense oligonucleotide ( MOMCU ) ( 25 . 6 ± 2 . 4% embryos , n = 115 ) . ( E ) Overexpression of MICU1 is sufficient to restore rhythmic cardiac contractions in tre embryos ( 49 . 3 ± 3 . 4% embryos , n = 127 compared to 16 . 8 ± 1 . 4% of uninjected siblings , n = 150 ) . This effect is abrogated by VDAC2 knockdown ( MOVDAC2 , 25 . 3 ± 5 . 5% embryos , n = 97 ) . ( F ) Suboptimal overexpression of MICU1 ( MICU1S ) and VDAC2 ( VDAC2S ) in combination is able to restore rhythmic cardiac contractions in tre embryos ( 48 . 6 ± 6 . 0% , n = 106 ) . Error bars represent s . d . ; p < 0 . 05; **p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 02710 . 7554/eLife . 04801 . 028Figure 7—figure supplement 1 . Expression of MCU , MICU1 and VDAC2 . In situ hybridization analysis shows that the expression levels of MCU , MICU1 and VDAC2 are comparable between wild type and tre embryos with and without efsevin treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 04801 . 028 In summary , we conducted a chemical suppressor screen in zebrafish to dissect the regulatory network critical for maintaining rhythmic cardiac contractions and to identify mechanisms underlying aberrant Ca2+ handling-induced cardiac dysfunction . We show that activation of VDAC2 through overexpression or efsevin treatment potently restores rhythmic contractions in NCX1h deficient zebrafish hearts and effectively suppresses Ca2+ overload-induced arrhythmogenic Ca2+ events and irregular contractions in mouse and human cardiomyocytes . We provide evidence that potentiating VDAC2 activity enhances mitochondrial Ca2+ uptake , accelerates Ca2+ transfer from intracellular stores into mitochondria and spatially and temporally restricts single Ca2+ sparks in cardiomyocytes . The crucial role of mitochondria in the regulation of cardiac rhythmicity is further supported by the findings that VDAC2 functions in concert with MCU; these genes have a strong synergistic effect on suppressing cardiac fibrillation and loss of function of either gene abrogates the rescue effect of the other in tre . The regulatory roles of mitochondrial Ca2+ in cardiac metabolism , cell survival and fate have been studied extensively ( Brown and O'Rourke , 2010; Dorn and Scorrano , 2010; Doenst et al . , 2013; Kasahara et al . , 2013; Kohlhaas and Maack , 2013; Luo and Anderson , 2013 ) . Our study provides genetic and physiologic evidence supporting an additional role for mitochondria in regulating cardiac rhythmicity and reveals VDAC2 as a modulator of Ca2+ handling in cardiomyocytes . Our findings , together with recent reports of the physical interaction between VDAC2 and RyR2 ( Min et al . , 2012 ) and the close proximity of outer and inner mitochondrial membranes at the contact sites between the mitochondria and the SR ( García-Pérez et al . , 2011 ) , suggest an intriguing model . We propose that mitochondria facilitate an efficient clearance mechanism in the Ca2+ microdomain , which modulates Ca2+ handling without affecting global Ca2+ signals in cardiomyocytes . In this model , VDAC facilitates mitochondrial Ca2+ uptake via MCU complex and thereby controls the duration and the diffusion of cytosolic Ca2+ near the Ca2+ release sites to ensure rhythmic cardiac contractions . This model is consistent with our observation that efsevin treatment induces faster inactivation kinetics of cytosolic Ca2+ transients without affecting the amplitude or the time to peak in cardiomyocytes and the reports that blocking mitochondrial Ca2+ uptake has little impact on cytosolic Ca2+ transients ( Maack et al . , 2006; Kohlhaas et al . , 2010 ) . Further support for this model comes from the observation of the Ca2+ peaks on the OMM ( Drago et al . , 2012 ) and the finding that downregulating VDAC2 extends Ca2+ sparks ( Subedi et al . , 2011; Min et al . , 2012 ) and that blocking mitochondrial Ca2+ uptake by Ru360 leads to an increased number of spontaneous propagating Ca2+ waves ( Seguchi et al . , 2005 ) . Future studies on the kinetics of VDAC2-dependent mitochondrial Ca2+ uptake and exploring potential regulatory molecules for VDAC2 activity will provide insights into how the crosstalk between SR and mitochondria contributes to Ca2+ handling and cardiac rhythmicity . Aberrant Ca2+ handling is associated with many cardiac dysfunctions including arrhythmia . Establishing animal models to study molecular mechanisms and develop new therapeutic strategies are therefore major preclinical needs . Our chemical suppressor screen identified a potent effect of efsevin and its biological target VDAC2 on manipulating cardiac Ca2+ handling and restoring regular cardiac contractions in fish and mouse and human cardiomyocytes . This success indicates that fundamental mechanisms regulating cardiac function are conserved among vertebrates despite the existence of species-specific features and suggests a new paradigm of using zebrafish cardiac disease models for the dissection of critical genetic pathways and the discovery of new therapeutic approaches . Future studies examining the effects of efsevin on other arrhythmia models would further elucidate the potential for efsevin as a pharmacological tool to treat cardiac arrhythmia associated with aberrant Ca2+ handling . Zebrafish of the mutant line tremblor ( tretc318 ) were maintained and bred as described previously ( Langenbacher et al . , 2005 ) . Transgenic lines , myl7:gCaMP4 . 1LA2124 and myl7:VDAC2LA2309 were created using the Tol2kit ( Esengil et al . , 2007; Kwan et al . , 2007; Shindo et al . , 2010 ) . The VDAC2LA2256 was created using the zinc finger array OZ523 and OZ524 generated by the zebrafish Zinc Finger Consortium ( Foley et al . , 2009a , 2009b ) . Full length VDAC2 cDNA was purchased from Open Biosystems ( Huntsville , AL ) and cloned into pCS2+ or pCS2+3XFLAG . Full length cDNA fragments of zebrafish MCU ( Accession number: JX424822 ) and MICU1 ( JX42823 ) were amplified from 2 dpf embryos and cloned into pCS2+ . For mRNA synthesis , plasmids were linearized and mRNA was synthesized using the SP6 mMESSAGE mMachine kit according to the manufacturers manual ( Ambion , Austin , TX . ) . VDAC2 mRNA and morpholino antisense oligos ( 5′-GGGAACGGCCATTTTATCTGTTAAA-3′ ) ( Genetools , Philomath , OR ) were injected into one-cell stage embryos collected from crosses of tretc318 heterozygotes . Cardiac performance was analyzed by visual inspection on 1 dpf . The tre mutant embryos were identified either by observing the fibrillation phenotype at 2–3 dpf or by genotyping as previously described ( Langenbacher et al . , 2005 ) . Chemicals from a synthetic library ( Castellano et al . , 2007; Choi et al . , 2011; Cruz et al . , 2011 ) and from Biomol International LP ( Farmingdale , NY ) were screened for their ability to partially or completely restore persistent heartbeat in tre embryos . 12 embryos collected from crosses of tretc318 heterozygotes were raised in the presence of individual compounds at a concentration of 10 µM from 4 hpf ( Choi et al . , 2011 ) . Cardiac function was analyzed by visual inspection at 1 and 2 dpf . The hearts of tretc318 embryos manifest a chaotic movement resembling cardiac fibrillation with intermittent contractions in rare occasion ( Ebert et al . , 2005; Langenbacher et al . , 2005 ) . Compounds that elicit persistent coordinated cardiac contractions were validated on large number of tre mutant embryos and NCX1h morphants ( >500 embryos ) . Videos of GFP-labelled myl7:GFP hearts were taken at 30 frames per second . Line-scan analysis was performed along a line through the atria or the ventricles of these hearts ( Nguyen et al . , 2009 ) . Fraction of shortening was deduced from the ratio of diastolic and systolic width and heart rate was determined by beats per minute . Cardiac parameters were analyzed in tremblortc318 and VDAC2LA2256 at 2 dpf . 36 hpf myl7:gCaMP4 . 1 embryos were imaged at a frame rate of 30 ms/frame . Electromechanical isolation was achieved by tnnt2MO ( Milan et al . , 2006 ) . The fluorescence intensity of each pixel in a 2D map was normalize to generate heat maps and isochronal lines at 33 ms intervals were obtained by identifying the maximal spatial gradient for a given time point ( Chi et al . , 2008 ) . The mouse E14Tg2a ESC and human H9 ESC line were cultured and differentiated as previously described ( Blin et al . , 2010; Arshi et al . , 2013 ) . At day 10 of differentiation , beating mouse EBs were exposed to external solution containing 10 mM CaCl2 for 10 min before DMSO or efsevin ( 10 μM ) treatment . Human EBs were differentiated for 15 days and treated with 5 mM CaCl2 for 10 min before DMSO or efsevin ( 5 μM ) treatment . Images of beating EBs were acquired at a rate of 30 frames/s and analyzed by motion-detection software . For calcium recording , the EBs were loaded with 10 μM fluo-4 AM in culture media for 30 min at 37°C . Line-scan analysis was performed and fluorescent signals were acquired by a Zeiss LSM510 confocal microscope . 2-day-old wild type , tre , and efsevin-treated tre embryos were placed on uncoated , microelectrode arrays ( MEAs ) containing 120 integrated TiN electrodes ( 30 μm diameter , 200 μm interelectrode spacing ) . Local field potentials ( LFPs ) at each electrode were collected for three trials per embryo type over a period of three minutes at a sampling rate of 1 kHz using the MEA2100-HS120 system ( Multichannel Systems , Reutiligen , Germany ) . Raw data was low-pass filtered at a cutoff frequency of 10 Hz using a third-order Butterworth filter . Data analysis was carried out using the MC_DataTool ( Multichannel Systems ) and Matlab ( MathWorks ) . Murine ventricular cardiomyocytes were isolated as previously described ( Reuter et al . , 2004 ) . Cells were loaded with 5 µM fluo-4 AM in external solution containing: 138 . 2 mM NaCl , 4 . 6 mM KCl , 1 . 2 mM MgCl , 15 mM glucose , 20 mM HEPES for 1 hr and imaged in external solution supplemented with 2 , 5 or 10 mM CaCl2 . For the recording of Ca2+ sparks and transients , the external solution contained 2 mM CaCl2 . For Ca2+ transients , cells were field stimulated at 0 . 5 Hz with a 5 ms pulse at a voltage of 20% above contraction threshold . For all measurements , efsevin was added 2 hr prior to the actual experiment . Images were recorded on a Zeiss LSM 5 Pascal confocal microscope . Data analysis was carried out using the Zeiss LSM Image Browser and ImageJ with the SparkMaster plugin ( Picht et al . , 2007 ) . Cells were visually inspected prior to and after each recording . Only those recordings from healthy looking cells with distinct borders , uniform striations and no membrane blebs or granularity were included in the analysis . For pull down assays mono-N-Boc protected 2 , 2'- ( ethylenedioxy ) bis ( ethylamine ) was attached to the carboxylic ester of efsevin and its derivatives through the amide bond . After removal of the Boc group using TFA , the primary amine was coupled to the carboxylic acid of Affi-Gel 10 Gel ( Biorad , Hercules , CA ) . 2-day-old zebrafish embryos were deyolked by centrifugation before being lysed with Rubinfeld's lysis buffer ( Rubinfeld et al . , 1993 ) . The lysate was precleaned by incubation with Affi-Gel 10 Gel to eliminate non-specific binding . Precleaned lysate was incubated with affinity beads overnight . Proteins were eluted from the affinity beads and separated on SDS-PAGE . Protein bands of interest were excised . Gel plugs were dehydrated in acetonitrile ( ACN ) and dried completely in a Speedvac . Samples were reduced and alkylated with 10 mM dithiotreitol and 10 mM TCEP solution in 50 mM NH4HCO3 ( 30 min at 56°C ) and 100 mM iodoacetamide ( 45 min in dark ) , respectively . Gel plugs were washed with 50 mM NH4HCO3 , dehydrated with ACN , and dried down in a Speedvac . Gel pieces were then swollen in digestion buffer containing 50 mM NH4HCO3 , and 20 . 0 ng/μl of chymotrypsin ( 25°C , overnight ) . Peptides were extracted with 0 . 1% TFA in 50% ACN solution , dried down and resuspended in LC buffer A ( 0 . 1% formic acid , 2% ACN ) . Extracted peptides were analyzed by nano-flow LC/MS/MS on a Thermo Orbitrap with dedicated Eksigent nanopump using a reversed phase column ( New Objective , Woburn , MA ) . The flow rate was 200 nl/min for separation: mobile phase A contained 0 . 1% formic acid , 2% ACN in water , and mobile phase B contained 0 . 1% formic acid , 20% water in ACN . The gradient used for analyses was linear from 5% B to 50% B over 60 min , then to 95% B over 15 min , and finally keeping constant 95% B for 10 min . Spectra were acquired in data-dependent mode with dynamic exclusion where the instrument selects the top six most abundant ions in the parent spectra for fragmentation . Data were searched against the Danio rerio IPI database v3 . 45 using the SEQUEST algorithm in the BioWorks software program version 3 . 3 . 1 SP1 . All spectra used for identification had deltaCN>0 . 1 and met the following Xcorr criteria: >2 ( +1 ) , >3 ( +2 ) , >4 ( +3 ) , and >5 ( +4 ) . Searches required full cleavage with the enzyme , <4 missed cleavages and were performed with the differential modifications of carbamidomethylation on cysteine and methionine oxidation . In situ hybridization was performed as previously described ( Chen and Fishman , 1996 ) . DIG-labeled RNA probe was synthesized using the DIG RNA labeling kit ( Roche , Indianapolis , IN ) . HeLa cells were transfected with a C-terminally flag-tagged zebrafish VDAC1 or VDAC2 in plasmid pCS2+ using Lipofectamine 2000 ( Invitrogen ) . After staining with MitoTracker Orange ( Invitrogen ) cells were fixed in 3 . 7% formaldehyde and permeabilized with acetone . Immunostaining was performed using primary antibody ANTI-FLAG M2 ( Sigma Aldrich , St . Luis , MO ) at 1:100 and secondary antibody Anti-Mouse IgG1-FITC ( Southern Biotechnology Associates , Birmingham , AL ) at 1:200 . Cells were mounted and counterstained using Vectashield Hard Set with DAPI ( Vector Laboratories , UK ) . HeLa cells were transfected with zebrafish VDAC2 using Lipofectamine 2000 ( Invitrogen , Carlsbad , CA ) . 36 hrs after transfection , cells were loaded with 5 µM Rhod2-AM ( Invitrogen ) , a Ca2+ indicator preferentially localized in mitochondria , for 1 hr at 15°C followed by a 30 min de-esterification period at 37°C . Subsequently , cells were permeabilized with 100 µM digitonin for 1 min at room temperature . Fluorescence changes in Rhod2 ( ex: 544 nm , em: 590 nm ) immediately after the addition of Ca2+ ( final free Ca2+ concentration is calculated to be approximately 10 µM using WEBMAXC at http://web . stanford . edu/∼cpatton/webmaxcS . htm ) were monitored in internal buffer ( 5 mM K-EGTA , 20 mM HEPES , 100 mM K-aspartate , 40 mM KCl , 1 mM MgCl2 , 2 mM maleic acid , 2 mM glutamic acid , 5 mM pyruvic acid , 0 . 5 mM KH2PO4 , 5 mM MgATP , pH adjusted to 7 . 2 with Trizma base ) using a FLUOSTAR plate reader ( BMG Labtech , Germany ) . V1/V3 DKO MEFs were cultured as previously described ( Roy et al . , 2009a ) . Efsevin-treated ( 15 μM for 30 min ) or mock-treated MEFs were used for measurements of [Ca2+]c in suspensions of permeabilized cells or imaging of [Ca2+]m simultaneously with [Ca2+]c in intact single cells . Permeabilization of the plasma membrane was performed by digitonin ( 40 μM/ml ) . Changes in [Ca2+] in the cytoplasmic buffer upon IP3 ( 7 . 5 μM ) addition in the presence or absence of ruthenium red ( 3 μM ) was measured by fura2 in a fluorometer ( Csordás et al . , 2006; Roy et al . , 2009b ) . To avoid endoplasmic reticulum Ca2+ uptake 2 μM thapsigargin was added before IP3 . For imaging of [Ca2+]m and [Ca2+]c , MEFs were co-transfected with plasmids encoding polycistronic zebrafish VDAC2 with mCherry and mitochondria-targeted inverse pericam for 40 hr . Cells were sorted to enrich the transfected cells and attached to glass coverslips . In the final 10 min , of the efsevin or mock-treatment , the cells were also loaded with fura2AM ( 2 . 5 μM ) and subsequently transferred to the microscope stage . Stimulation with 1 μM ATP was carried out in a norminally Ca2+ free buffer . Changes in [Ca2+]c and [Ca2+]m were imaged using fura2 ( ratio of ex:340 nm–380 nm ) and mitochondria-targeted inverse pericam ( ex: 495 nm ) , respectively ( Csordas et al . , 2010 ) . All values are expressed as mean ± SEM , unless otherwise specified . Significance values are calculated by unpaired student's t-test unless noted otherwise .	The heart is a large muscle that pumps blood around the body by maintaining a regular rhythm of contraction and relaxation . If the heart loses this regular rhythm it works less efficiently , which can lead to life-threatening conditions . Regular heart rhythms are maintained by changes in the concentration of calcium ions in the cytoplasm of the heart muscle cells . These changes are synchronised so that the heart cells contract in a controlled manner . In each cell , a contraction begins when calcium ions from outside the cell enter the cytoplasm by passing through a channel protein in the membrane that surrounds the cell . This triggers the release of even more calcium ions into the cytoplasm from stores within the cell . For the cells to relax , the calcium ions must then be pumped out of the cytoplasm to lower the calcium ion concentration back to the original level . Shimizu et al . studied a zebrafish mutant—called tremblor—that has irregular heart rhythms because its heart muscle cells are unable to efficiently remove calcium ions from the cytoplasm . Embryos of the tremblor mutant were treated with a wide variety of chemical compounds with the aim of finding some that could correct the heart defect . A compound called efsevin restores regular heart rhythms in tremblor mutants . Efsevin binds to a pump protein called VDAC2 , which is found in compartments called mitochondria within the cell . Although mitochondria are best known for their role in supplying energy for the cell , they also act as internal stores for calcium . By binding to VDAC2 , efsevin increases the rate at which calcium ions are pumped from the cytoplasm into the mitochondria . This restores rhythmic calcium ion cycling in the cytoplasm and enables the heart muscle cells to develop regular rhythms of contraction and relaxation . Increasing the levels of VDAC2 or another similar calcium ion pump protein in the heart cells can also restore a regular heart rhythm . Efsevin can also correct irregular heart rhythms in human and mouse heart muscle cells , therefore the new role for mitochondria in controlling heart rhythms found by Shimizu et al . appears to be shared in other animals . The experiments have also identified the VDAC family of proteins as potential new targets for drug therapies to treat people with irregular heart rhythms .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2015	Mitochondrial Ca2+ uptake by the voltage-dependent anion channel 2 regulates cardiac rhythmicity
The yeast Dsc E3 ligase complex has long been recognized as a Golgi-specific protein ubquitination system . It shares a striking sequence similarity to the Hrd1 complex that plays critical roles in the ER-associated degradation pathway . Using biochemical purification and mass spectrometry , we identified two novel Dsc subunits , which we named as Gld1 and Vld1 . Surprisingly , Gld1 and Vld1 do not coexist in the same complex . Instead , they compete with each other to form two functionally independent Dsc subcomplexes . The Vld1 subcomplex takes the AP3 pathway to reach the vacuole membrane , whereas the Gld1 subcomplex travels through the VPS pathway and is cycled between Golgi and endosomes by the retromer . Thus , instead of being Golgi-specific , the Dsc complex can regulate protein levels at three distinct organelles , namely Golgi , endosome , and vacuole . Our study provides a novel model of achieving multi-tasking for transmembrane ubiquitin ligases with interchangeable trafficking adaptors . Ubiquitin-dependent protein down-regulation and quality control are important for maintaining the integrity of all organelles ( Arvan et al . , 2002; Wang et al . , 2011 ) . Besides the well-characterized ER protein quality control system ( Ellgaard and Helenius , 2003; Ruggiano et al . , 2014 ) , other organelles such as the plasma membrane ( Babst , 2014; Okiyoneda et al . , 2010; Zhao et al . , 2013 ) , Golgi ( Dobzinski et al . , 2015; Reggiori and Pelham , 2002 ) , endosomes ( Fukuda et al . , 2006; Léon et al . , 2008; Nakamura et al . , 2005 ) , peroxisome ( Platta et al . , 2009; Williams et al . , 2008 ) , and mitochondria ( Narendra et al . , 2008 ) , either contain integral membrane E3 ligases , or have adaptor proteins to recruit cytosolic E3 ligases for the removal of damaged organelle proteins or the down-regulation of unnecessary proteins ( Hettema et al . , 2004; Lin et al . , 2008 ) . Focusing on the down-regulation of lysosomal membrane proteins , we recently identified two independent E3 ligase systems on the budding yeast vacuole ( yeast lysosome ) membrane , each governing a different subset of vacuolar membrane proteins . The Ssh4-Rsp5 system selectively ubiquitinates a vacuolar lysine transporter Ypq1 to down-regulate the lysine import activity after lysine withdrawal ( Li et al . , 2015b; Sekito et al . , 2014 ) , whereas Zn2+ withdrawal results in the selective ubiquitination of a Zn2+ importer Cot1 by the Dsc E3 ligase complex ( Li et al . , 2015a ) . Interestingly , Rsp5 can work together with the Dsc complex to down-regulate a Zn2+ exporter Zrt3 when excessive Zn2+ is present in the cytoplasm ( Li et al . , 2015a ) . These ubiquitinated membrane proteins will be directly internalized into the vacuole lumen by the ESCRT machinery for degradation ( Zhu et al . , 2017 ) . Originally discovered in fission yeast , the S . pombe Dsc complex contains six components , including Tul1 , Dsc2 , Dsc3 , Dsc4 , Ubx3 , and the AAA+ ATPase Cdc48 ( Stewart et al . , 2012 , 2011 ) . These components , with the exception of Dsc4 , also exist in budding yeast ( Dobzinski et al . , 2015; Li et al . , 2015a; Tong et al . , 2014 ) . Strikingly , most Dsc components share sequence similarity to the Hrd1 E3 ligase complex , a key player in ER protein quality control . Tul1 is a multi-spanning membrane RING domain E3 ligase that is related to Hrd1 . Other components , including Dsc2 , Dsc3 , Ubx3 , are homologous to Der1 , Usa1 , and Ubx2 of the Hrd1 complex , respectively ( Stewart et al . , 2012 , 2011 ) . Furthermore , both complexes contain the same AAA+ ATPase Cdc48 . The striking similarity suggests the Dsc complex might play a role in protein quality control at the downstream organelles of the secretory pathway . Probably the biggest controversy about the Dsc complex is its subcellular localization . In S . pombe , the Dsc complex has been shown to be critical to the proteolytic activation of the sterol regulatory element binding protein ( SREBP ) transcription factor , which is a Golgi membrane protein . Consistently , fission yeast Dsc complex has been shown to localize to the Golgi ( Burr et al . , 2017; Stewart et al . , 2011 ) . In S . cerevisiae , it is also generally accepted that the Dsc complex is a Golgi-specific E3 ligase complex . Tul1 was initially identified as a Golgi protein quality control E3 ligase through its ability to recognize and ubiquitinate an artificial folding mutant of Pep12 , which is cycled between Golgi and endosomes ( Reggiori and Pelham , 2002 ) . Recently , it has been shown that the Dsc complex is also responsible for the ubiquitination and degradation of another Golgi membrane protein Yif1 , after either amino acid starvation or rapamycin treatment ( Dobzinski et al . , 2015 ) . However , the Arabidopsis homologue of Tul1 , FLY1 , is predominantly localized to the late endosome , instead of the Golgi ( Voiniciuc et al . , 2013 ) . Furthermore , Graham and colleagues recently demonstrated that the budding yeast Tul1 participates in the ubiquitination and recycling of an exocytic v-SNARE Snc1 at the early endosome ( Xu et al . , 2017 ) . Lastly , we observed that Dsc complex is responsible for the down-regulation of some vacuole membrane proteins in budding yeast ( Li et al . , 2015a ) . How can a Golgi E3 ligase complex ubiquitinate a vacuole membrane protein such as Cot1 ? In this study , we resolved this controversy by identifying two new components of the Dsc complex , which we named as Gld1 ( Golgi/endosome Localized Dsc protein 1 ) and Vld1 ( Vacuole Localized Dsc protein 1 ) . Gld1 and Vld1 are two similar tetra-spanning membrane proteins that compete with each other to form functionally independent Dsc subcomplexes at the ER . The Gld1-containing subcomplex takes the VPS ( vacuolar protein sorting ) pathway for its localization and is cycled between endosomes and Golgi by the retromer complex . In contrast , the Vld1-containing subcomplex travels through the AP3 pathway to the vacuole membrane . Together , this novel mechanism enables the cell to achieve protein regulation and probably quality control at three distinct organelles , namely Golgi , endosomes , and the vacuole , using just one RING domain E3 ligase Tul1 . We propose that plant and mammalian cells might use a similar strategy to target their membrane-residing E3 ligases to different organelles in order to expand their substrate repertoire . Two independent pathways have been identified to deliver proteins from the Golgi to vacuole ( Figure 1A ) . The VPS pathway transports vacuolar proteases such as CPY , Pep4 and Prb1 to the vacuole lumen via the intermediate endosomal compartments marked by the Pep12 SNARE protein ( Bowers and Stevens , 2005 ) . It is also used to deliver ubiquitinated membrane cargoes to the vacuole lumen for degradation . Furthermore , some vacuolar membrane proteins , such as Vph1 and Ssh4 , also utilize the VPS pathway to reach the vacuole surface ( Zhu et al . , 2017 ) . The VPS pathway can be blocked by deletion of PEP12 , or the cargo can be trapped in an aberrant endosome ( i . e . the class E compartment ) by the deletion of genes encoding ESCRT machinery ( Bowers and Stevens , 2005; Odorizzi et al . , 1998a; Zhu et al . , 2017 ) . As an independent targeting mechanism , the AP3 pathway transports a subset of the vacuolar membrane proteins such as alkaline phosphatase ( ALP ) from Golgi to the vacuole ( Figure 1A ) ( Cowles et al . , 1997 ) . These membrane proteins typically contain an acidic di-leucine targeting motif ( D/EXXXLL , where X can be any amino acid ) , which can be recognized at the late Golgi by the AP3 adaptor complex for sorting into carrier vesicles that then directly target and fuse with the vacuole ( Cowles et al . , 1997 ) . Deletion of the AP3 complex leads to an accumulation of AP3 cargoes at the Golgi and forces them to traffick to the vacuole membrane via the VPS pathway ( Cowles et al . , 1997; Li et al . , 2015b; Llinares et al . , 2015; Odorizzi et al . , 1998b ) . Previously , we reported that the Dsc complex has three distinct subcellular localizations , including Golgi , endosomes , and vacuole . At the steady state , about 60% of the Dsc complex localizes to the vacuole membrane , whereas the rest appears to be on punctae that co-localize with the Golgi and endosomes ( Li et al . , 2015a ) . This multi-localization pattern is very interesting among membrane proteins in the endomembrane trafficking pathway . At steady state , most endomembrane proteins are either localized to the vacuole membrane , or to punctae that include Golgi and endosomes . For example , Vph1 , a VO subunit of the vATPase complex localizes to the vacuole membrane ( Manolson et al . , 1992 ) , while its isoform , Stv1 , localizes to Golgi and endosomes ( Manolson et al . , 1994 ) . As another example , Ssh4 and Ear1 , two homologous adaptor proteins for the E3 ligase Rsp5 , localize to the vacuole and Golgi/endosomes , respectively ( Léon et al . , 2008; Li et al . , 2015b ) . Furthermore , to maintain their punctae localization , some Golgi and endosomal membrane proteins are constantly recycled by the retromer complex ( Burd and Cullen , 2014; Hettema et al . , 2003; Seaman et al . , 1997 ) . Only in retromer mutants do these Golgi/endosome proteins mislocalize to the vacuole membrane ( Burd and Cullen , 2014; Strochlic et al . , 2007 ) . Then , how does the Dsc complex achieve three distinct subcellular localizations in wild type cells ? Intrigued by its localization pattern , we set out to determine the targeting pathway utilized by the Dsc complex . We chose Ubx3 to represent the Dsc complex because: ( 1 ) Ubx3 forms a stable complex with the rest of the Dsc components . Antibody-mediated depletion of Ubx3 almost completely depletes the entire Dsc complex from the cell lysate; ( 2 ) Ubx3 is the only Dsc component that can be chromosomally tagged with fluorescent proteins without abolishing Dsc function in budding yeast ( Li et al . , 2015b ) . To begin the investigation , we tagged Ubx3 with mNeonGreen , a green fluorescent protein that is similar to GFP in size , but about two times brighter than GFP ( Shaner et al . , 2013 ) . In the wild type strain , Ubx3-mNeonGreen ( hereafter referred to as Ubx3-nG ) localized to the vacuole membrane and intracellular punctae , consistent with our previous observation ( Figure 1B ) ( Li et al . , 2015a ) . In contrast , deleting VPS27 , which encodes an essential component of the ESCRT machinery , led to a partial accumulation of the Ubx3-nG in the class E compartment . A significant amount of Ubx3-nG , however , was still able to reach the vacuole membrane ( Figure 1B ) . As a control , Vph1-mCherry ( hereafter referred to as Vph1-mCh ) , a cargo of the VPS pathway , was almost entirely trapped in the class E compartment ( Zhu et al . , 2017 ) . Furthermore , deleting the gene encoding the endosomal t-SNARE Pep12 blocked the trafficking of Vph1-mCh and led to a ‘cytosolic’ accumulation of Vph1-mCh as numerous small vesicles ( Figure 1B ) . The same PEP12 deletion , however , only resulted in a partial accumulation of Ubx3-nG in the cytosol , with a significant amount of the protein reaching the vacuole membrane ( Figure 1B ) . Together , these results suggest that the Dsc complex might use both VPS and AP3 pathways for its trafficking . However , VPS and AP3 cargoes use different signals for their targeting . How can one protein complex contain both targeting signals ? A possible explanation might be that the Dsc complex has two distinct sub-populations . One uses the AP3 pathway , the other uses the VPS pathway . To test whether a fraction of the Dsc complex uses the AP3 pathway for vacuolar delivery , we deleted APL6 , which encodes a key component of the AP3 adaptor complex ( Cowles et al . , 1997; Odorizzi et al . , 1998b ) . Interestingly , although APL6 deletion did not change the punctae localization of Ubx3-nG , it led to an accumulation of the Ubx3-nG in the vacuole lumen . The result is different from the reported AP3 pathway cargoes such as ALP and Ypq1 , which partially accumulate at the Golgi and partially reach the vacuole membrane via the VPS pathway ( Cowles et al . , 1997; Li et al . , 2015b ) , in the absence of the AP3 adaptor complex . This implied that the AP3 Dsc subcomplex , when forced into the VPS pathway , was recognized by an unidentified endosomal protein quality control machinery as an ‘abnormal’ protein complex and degraded in the vacuole lumen . However , this hypothesis cannot explain the fact that Dsc complex normally exists on the Golgi and endosomes at the steady state , unless the AP3 Dsc sub-population carries a distinct unknown subunit that does not exist in the Golgi/endosome subcomplex . To investigate if there are unknown components in the Dsc complex that determine its trafficking pathways , we chromosomally tagged Ubx3 with the Flag tag at the C-terminus and performed immunoprecipitation ( IP ) experiments to isolate the Dsc complex . Our previous research has demonstrated that the Dsc complex containing Ubx3-Flag is still able to ubiquitinate a vacuole membrane substrate Cot1 after Zn2+ withdrawal ( Li et al . , 2015a ) , indicating that Ubx3-Flag is still functional . As shown in Figure 2A–B , all known components of the budding yeast Dsc complex , including Dsc2 , Dsc3 , Tul1 , and Cdc48 , can be co-purified with Ubx3-Flag . Importantly , two new bands at ~27 and ~31 KDa , respectively , co-immunoprecipitated with Ubx3-Flag . Mass spectrometry analysis identified the 27 KDa band as Yir014W ( Figure 2—figure supplement 1A , hereafter referred to as 014W ) , and the 31 KDa band as Ypr109W ( Figure 2—figure supplement 1B , hereafter referred to as 109W ) , both of which are proteins of unknown function . To confirm that these two proteins are genuine components of the Dsc complex , we chromosomally labelled them with the Flag tag at the C-termini and performed reciprocal IP experiments . As shown in Figure 2C and D , both 014W-Flag and 109W-Flag can pull-down other tested Dsc complex components , including Ubx3 , Dsc2 , Dsc3 , and Tul1 , whereas Vph1 , an abundant vacuolar membrane protein , cannot be co-immunoprecipitated . Bioinformatics analysis indicated that both proteins have four transmembrane helices , and they share a significant sequence similarity ( 53% , Figure 2E and Figure 2—figure supplement 1C ) ( Sievers et al . , 2011; Tsirigos et al . , 2015 ) . We also performed position-specific iterated BLAST ( PSI-BLAST ) search of sequenced protein databases . Intriguingly , both 014W and 109W showed a low sequence similarity to Dsc4 from several sequenced fungus species , including Stemphylium lycopersici , Escovopsis weberi , Tolypocladium ophioglossoides CBS 100239 , and Ceratocystis fimbriata CBS 114723 ( data not shown ) , suggesting they might be the ‘missing’ Dsc4 in budding yeast . We were puzzled by the finding that two similar proteins coexist in the same protein complex . One possible explanation was that 014W and 109W exist in different sub-populations of the Dsc complex , considering that the complex has three distinct sub-cellular locations . Indeed , as shown in Figure 2F , in a yeast strain that was co-expressing 014W-HA and 109W-Flag from their chromosomal loci , these two proteins did not co-immunoprecipitate with each other , although each was capable of pulling down all other tested Dsc components , including Ubx3 , Dsc2 , Dsc3 , and Tul1 . Together , we have identified two novel components of the Dsc complex ( 014W and 109W ) , which are similar to each other . Each protein is capable of forming a stable complex with the known Dsc components . However , 014W and 109W do not coexist in the same sub-population . These results support the existence of two distinct Dsc subcomplexes within the cell . To further investigate if 014W and 109W form distinct Dsc subcomplexes , we chromosomally tagged them with mNeonGreen ( nG ) to directly visualize their sub-cellular localizations . Strikingly , 014W-nG localized exclusively to the FM4-64 labelled vacuole membrane , while 109W-nG only localized to intracellular punctae ( ~3 . 7 per cell , n = 110 cells , Figure 3A–B and Figure 3—source data 1 ) . Further analysis revealed that ~ 37 . 4% of the 109W punctae co-localized with the Mars-Sec7 labelled trans-Golgi compartment , while ~56 . 9% of the 109W punctae co-localized with the FM4-64 labelled endosomes ( Figure 3C–D and Figure 3—source data 2 ) . The distinct localizations of 109W and 014W are consistent with a model in which these two proteins form distinct subcomplexes that travel separately to either the Golgi/endosome or the vacuole membrane . A prediction of our model is that 109W should co-localize with the Ubx3 punctae . However , our attempts to show the co-localization by simultaneously tagging these two proteins with red and green fluorescent proteins proved to be difficult . Mysteriously , tagging either 109W or Ubx3 with different red fluorescent proteins , including mCherry , DsRed , and tagRFP , always resulted in the degradation of the fusion proteins in the vacuole lumen ( data not shown ) . Inspired by the ‘knock sideways’ technology ( Haruki et al . , 2008; Robinson et al . , 2010 ) , we developed a new assay , which we named as Rapamycin Induced Co-localization ( RICo ) assay , to show that 109W co-localizes with the Ubx3-nG punctae . As shown in Figure 3E , we fused the FRB moiety to mCherry and expressed the fusion protein under a weak SSH4 promoter ( 75 copies per cell ) . Meanwhile , 109W was chromosomally tagged with FKBP . In the absence of rapamycin , FRB-mCherry appeared to be cytosolic . However , 45 min after the addition of rapamycin , FRB-mCherry was recruited to 109W-FKBP and appeared as punctae that co-localized with Ubx3-nG ( Figure 3E ) . Together , our data suggest that 014W and 109W form distinct Dsc subcomplexes at the vacuole membrane and Golgi/endosomes , respectively . Since both proteins have not been characterized before , we named 014W as Vacuole Localized Dsc protein1 ( Vld1 ) and 109W as Golgi/endosome Localized Dsc protein1 ( Gld1 ) . Because Vld1 and Gld1 are the only unique components within the vacuole and Golgi/endosome subcomplexes and they share a significant protein sequence similarity , we hypothesized that they may compete with each other to determine the subcellular locations of the Dsc complex . As an initial step , we deleted genes encoding either Vld1 or Gld1 to test if these mutants affect the corresponding Dsc subcellular localizations . As shown in Figure 4A , deleting VLD1 eliminated the vacuole membrane localization of Ubx3-nG , whereas deleting GLD1 abolished its Golgi and endosome localizations . Intriguingly , deletion of GLD1 gene also resulted in a partial accumulation of smaller punctae in the endoplasmic reticulum ( ER ) ( Figure 4A and B ) , as indicated by its co-localization with the ER marker DsRed-HDEL ( Figure 4B ) , probably because not all Ubx3-nG were assembled into the Dsc complex and the excessive Ubx3-nG were trapped at the ER . This result also suggested that the assembly of the Dsc complex might happen at the ER . As a direct test of the competition hypothesis , we asked if gradually elevating the expression levels of either Vld1 or Gld1 can recruit more and more Dsc complex to their corresponding subcellular locations . As shown in Figure 4C , in a vld1∆ strain transformed with an empty vector , Ubx3-nG localized exclusively to the intracellular punctae . However , after expressing Vld1 using its native promoter , a significant amount of Ubx3-nG was recruited to the vacuole membrane from the punctae . Further increasement of the Vld1 expression level with a GPD promoter resulted in an exclusive vacuole membrane localization of Ubx3-nG ( Figure 4C ) . Conversely , elevating the Gld1 expression level in a gld1∆ strain gradually recruited Ubx3-nG from the vacuole membrane to intracellular punctae ( Figure 4—figure supplement 1A ) . We also used immunoprecipitation experiments to verify the imaging results . As shown in Figure 4—figure supplement 1B , overexpression of Vld1-HA under the GPD promoter can displace Gld1-GFP from the Dsc complex , and vice versa . All together , these experiments provided a direct evidence that Vld1 and Gld1 compete with each other for the Dsc complex . One prediction of the competition hypothesis is that both vacuole and Golgi/endosome subcomplexes should function independently . As stated above , Zn2+ withdrawal leads to the degradation of the vacuolar Zn2+ transporter Cot1 ( Li et al . , 2015a ) , whereas amino acid starvation ( or rapamycin treatment ) triggers the degradation of a Golgi resident protein Yif1 ( Dobzinski et al . , 2015 ) . Both processes are initiated by the Dsc complex-mediated protein ubiquitination . As shown in Figure 5 , deletion of TUL1 , the RING domain E3 ligase that exists in all Dsc subcomplexes , blocked the degradation of both GFP-Yif1 and Cot1-GFP . GFP-Yif1 was stabilized on the intracellular punctae after 4 hr of amino acid starvation ( Figure 5A ) , whereas Cot1-GFP was stabilized on the vacuole membrane ( Figure 5C ) . However , deletion of the Golgi/endosome-specific component , Gld1 , significantly reduced GFP-Yif1 degradation , whereas Cot1-GFP degradation was unaffected ( Figure 5 ) . Conversely , deletion of the VLD1 gene only drastically delayed the degradation of Cot1-GFP without affecting the turnover of GFP-Yif1 ( Figure 5 ) . Taken together , we conclude that Vld1 competes with Gld1 to form two functionally independent Dsc subcomplexes that localize to distinct subcellular locations . The Vld1 subcomplex localizes to the vacuole membrane , whereas the Gld1 subcomplex localizes to the Golgi and endosomes . At these locations , they may govern the ubiquitination of distinct organelle membrane proteins . To determine the minimum subunit requirement for the proper assembly and trafficking of the complex , we took a reductive approach by deleting Dsc subunits and testing if any mutant affects the trafficking of the complex , as indicated by the localizations of Vld1-nG , Gld1-nG , and Ubx3-nG . All three tagged proteins are still functional because they can support the degradation of both Cot1-GFP and GFP-Yif1 ( Figure 6—figure supplement 1 ) . As shown in Figure 6A and B , deletion of either TUL1 or DSC3 did not affect the locations of remaining complex components , as demonstrated by Vld1-nG , Gld1-nG , and Ubx3-nG . However , deletion of DSC2 dramatically changed their localization patterns . After DSC2 deletion , Vld1-nG , Gld1-nG , and Ubx3-nG were all trapped in smaller punctae that co-localized with the ER marker , DsRed-HDEL ( Figure 6A–C and Figure 6—figure supplement 2A–B ) . Similarly , deletion of UBX3 caused the accumulation of Vld1-nG and Gld1-nG as smaller punctae that co-localized with DsRed-HDEL ( Figures 6 and Figure 6—figure supplement 2A–B ) , indicating the importance of Ubx3 in the complex assembly . Lastly , double deletion of VLD1 and GLD1 caused the accumulation of Ubx3-nG punctae at the ER ( Figure 6B and C ) . Together , our results suggest a model in which the complex assembly may occur at the ER , where Dsc2 and Ubx3 form a core complex with either Vld1 or Gld1 to determine the subcellular localizations . Tul1 and Dsc3 are not part of the core complexes . Consistent with this model , double deletion of TUL1 and DSC3 did not affect the proper targeting of Gld1-nG , Vld1-nG , or Ubx3-nG ( Figure 6D ) . To further verify the core complexes model , we performed a series of IP experiments using various Dsc deletion mutants . As shown in Figure 6E , in the absence of Tul1 and Dsc3 , Ubx3-Flag was still able to interact with Dsc2 , Gld1-GFP , and Vld1-HA . In contrast , either deletion of DSC2 or the double deletion of both GLD1 and VLD1 abolished all the interactions between Ubx3-Flag and the rest of the Dsc subunits . Single deletions of either GLD1 or VLD1 did not affect the protein-protein interactions within the counterpart subcomplex ( Figure 6E ) . A similar result has been observed by the Espenshade group ( Tong et al . , 2014 ) . Using immunoprecipitation analysis of the Dsc deletion mutants , they concluded that Tul1 and Dsc3 are not essential for the assembly of the complex , whereas Dsc2 and Ubx3 forms a core complex ( Tong et al . , 2014 ) . With our analysis , we now know the core complex should also include either Vld1 or Gld1 , and more importantly , the two core complexes determine the subcellular localizations of the Dsc complex . The observation that Gld1 and Vld1 double deletion leads to the ER accumulation of Ubx3-nG enabled us to test if spDsc4 can support the assembly and trafficking of Dsc complex in budding yeast . As shown in Figure 6—figure supplement 2C , overexpression of spDsc4 under the ADH1 promoter allows Ubx3-nG to traffic out of ER in ~60% of yeast cells . However , the chimeric Dsc complex is unstable and Ubx3-nG appears to accumulate in the vacuolar lumen , presumably due to the low sequence similarity . Nevertheless , this result is consistent with the hypothesis that Gld1 and Vld1 might be the counterpart of Dsc4 in budding yeast . As shown above in Figure 1B , our initial analysis of the Dsc trafficking pathways suggested a confusing model that the complex utilizes both AP3 and VPS pathways for its subcellular localizations . Since the Vld1 subcomplex localizes exclusively to the vacuole membrane , it is reasonable to hypothesize that this subcomplex travels through the AP3 pathway . Indeed , deleting genes encoding either ESCRT machinery , as represented by vps27∆ , or the endosomal t-SNARE Pep12 , did not affect the vacuole localization of Vld1-nG , although the vacuolar trafficking of Vph1-mCh , a VPS pathway cargo , was abolished in both mutants ( Figure 7A ) . In contrast , deleting APL6 resulted in the accumulation of Vld1-nG on cytosolic punctae , presumably the Golgi and endosome compartment , and the vacuolar degradation of Vld1-nG ( Figure 7A ) . The vacuolar degradation of Vld1-nG in the apl6∆ strain was consistent with the above mentioned response of Ubx3-nG in the APL6 deletion strain ( Figure 1B ) , suggesting that the entire Vld1 subcomplex , when forced to travel through the VPS pathway , was recognized by the endosomal protein quality control system and turned over in the vacuole lumen . It is very likely that Vld1 was the ‘culprit’ that caused the ubiquitination and degradation since the remaining subunits are identical between the two subcomplexes . As stated above , AP3 cargoes normally contain an acidic di-leucine motif for their vacuolar sorting ( Odorizzi et al . , 1998b ) . Examining the Vld1 protein sequence revealed a putative acidic di-leucine motif ( 233EITPLL238 ) close to the C-terminus of the protein . Fungal sequence alignment indicated that this motif is conserved among orthologues from different fungal species ( Figure 7B ) ( Cliften et al . , 2003; Kellis et al . , 2003 ) . In order to test if this motif is important for the AP3 pathway trafficking , we mutated the sequence to 233AITPAA238 . However , these changes caused an instability of Vld1-nG , and nearly all of it was degraded by an unknown mechanism ( data not shown ) . As an alternative , we deleted the last six amino acids of Vld1 ( 237LLNIAE242 ) , including the di-leucine ( vld1∆6AA-nG ) . This deletion caused a partial punctate ( Golgi and endosomes ) accumulation and partial vacuolar degradation of vld1∆6AA-nG in wildtype cells ( Figure 7C ) , a phenotype similar to that of Vld1-nG in the apl6∆ strain ( Figure 7A ) . These results indicated the importance of the di-leucine motif . Further deletion of the PEP12 gene , which eliminates the VPS pathway , completely blocked the vld1∆6AA-nG trafficking and the protein appeared as numerous tiny ‘cytosolic’ punctae outside the vacuole ( Figure 7C ) . In contrast , the trafficking of an AP3 pathway cargo , mCh-ALP , was not affected by the PEP12 deletion . We also tested whether the acidic E233 residue is important for the Vld1 trafficking . As shown in Figure 7D , a single E233 to A233 mutation ( vld1E233A-nG ) was sufficient to cause trafficking defects similar to vld1∆6AA–nG . All together , these results strongly indicate that the putative acidic di-leucine motif is important for the trafficking of Vld1 through the AP3 pathway . The observation that Gld1 co-localizes with FM4-64 labelled endosomes ( Figure 3C and D ) supports the hypothesis that the protein travels through the VPS pathway . To confirm this , we tested if the ESCRT or PEP12 deletion mutants can affect the trafficking of Gld1 . As shown in Figure 7E , a fraction of the Gld1-nG punctae co-localized with the Vph1-mCherry labelled class E compartment in vps27∆ strain . The Golgi population of Gld1-nG , however , did not accumulate in the class E compartment . In the pep12∆ strain , Gld1-nG trafficking was also disrupted and the protein appeared as ‘cytosolic’ , with very few punctae left . This phenotype was also similar to that of the VPS pathway cargo Vph1-mCherry ( Figure 7E ) . To further verify that Gld1 trafficks through the VPS pathway , we tested if Gld1 is cycled between the Golgi and endosomes by the retromer , a sorting apparatus that functions on the endosomes to prevent important endosomal proteins from mis-targeting to the vacuole membrane ( Burd and Cullen , 2014; Seaman et al . , 1997 ) . As shown in Figure 7F , deleting VPS35 , an essential subunit in the retromer complex , resulted in the mis-localization of Gld1-nG to the vacuole membrane . Taken together , we conclude that the Dsc subcomplexes use two independent trafficking pathways for their subcellular localizations . The Vld1-containing subcomplex trafficks through the AP3 pathway to reach the vacuole membrane , whereas the Gld1-containing subcomplex travels through the VPS pathway and is cycled between Golgi and endosomes by the retromer machinery ( Figure 7G ) . Protein ubiquitination and degradation is an essential process to rapidly down-regulate protein levels and remove unfolded/damaged proteins in eukaryotic cells . How to recognize so many different substrates in a regulated fashion using a limited number of E3 ligases is a fundamental challenge faced by all eukaryotic cells . For example , the human proteome contains ~19 , 000–20 , 000 different proteins ( Kim et al . , 2014; Wilhelm et al . , 2014 ) , but only has ~640 E3 ligases ( Morreale and Walden , 2016 ) . This challenge is further complicated by the presence of different organelles that divide the cell into numerous compartments , which requires the proper targeting of E3 ligases to different organelle surfaces . The human proteome has 6 , 000–7 , 000 predicted transmembrane proteins . However , there are only ~50 predicted transmembrane E3 ligases ( Lussier et al . , 2012; Nakamura , 2011 ) . With so few transmembrane E3 ligases , how does the cell achieve the regulated ubiquitination for so many membrane proteins ? During evolution , eukaryotic cells have developed several ingenious ways to expand the substrate repertoire of existing E3 ligases . At the protein level , one strategy is through the expansion of the interchangeable F-box proteins in the SCF E3 ligase complex . F-box is an approximately 50 amino acids motif that mediates protein-protein interaction ( Kipreos and Pagano , 2000 ) . As part of the SCF E3 ligase complex , F-box proteins are responsible for recognizing specific substrates . Many SCF complexes only differ in their F-box protein subunits . Therefore , by expanding the F-box protein family , the cell can recognize different substrates using a similar SCF complex . For example , although budding yeast has only 11 F-box proteins , C . elegans has more than 300 F-box proteins ( Kipreos and Pagano , 2000 ) , and A . thaliana has over 1 , 300 F-box proteins ( Hua et al . , 2011 ) . At the organelle level , another strategy is to use adaptor proteins to recruit cytosolic E3 ligases to different organelle surface . For example , in budding yeast , the NEDD4 family E3 ligase Rsp5 can be recruited to the plasma membrane ( Lin et al . , 2008; MacGurn et al . , 2012 ) , Golgi ( Hettema et al . , 2004 ) , endosomes ( Léon et al . , 2008 ) , and vacuole ( Li et al . , 2015b ) by different PY motif containing proteins ( Wang et al . , 2001 ) . The PY motif interacts with the WW motifs of Rsp5 to recruit the ligase . A similar recruitment mechanism has been observed for the human NEDD4 family members . The human NEDD4-2 can be recruited to the plasma membrane by the PY motif within the surface epithelial sodium channel ( ENaC ) protein in order to down-regulate ENaC ( Fotia et al . , 2003 ) . In this study , we report a novel mechanism to directly target a membrane-residing E3 ligase complex to different organelles . Through the characterization of targeting pathways of the budding yeast Dsc complex , we unexpectedly uncovered that the ER can assemble two distinct Dsc subcomplexes using Gld1 and Vld1 . Gld1 and Vld1 then guide the Dsc complex through two independent trafficking pathways ( VPS and AP3 ) to reach Golgi/endosomes and lysosomes , respectively . This mechanism allows the cell to achieve protein regulation and probably quality control at three distinct organelles , namely Golgi , endosomes , and vacuole , using just one RING domain E3 ligase Tul1 ( Figure 7G ) . Because both VPS and AP3 pathways are conserved from yeast to human ( Bonifacino and Traub , 2003 ) , it is reasonable to speculate that human cells might use a similar strategy to expand the substrate repertoire of their membrane E3 ligases . Consistent with this hypothesis , several human RING domain E3 ligases have been identified in the endomembrane trafficking pathways , such as March2 , March3 , March11 , RNF152 , and RNF167 ( Deng et al . , 2015; Lussier et al . , 2012; Nakamura , 2011; Nakamura et al . , 2005 ) . It remains to be verified whether there is a similar multi-localization mechanism for these E3 ligases . Endosomes are intermediate transporting organelles between the Golgi ( or plasma membrane for endocytosis ) and lysosomes . On the endosome , ubiquitinated membrane cargoes are sorted and internalized into the lumen as intraluminal vesicles by the ESCRT machinery , whereas cargo receptors such as LDL receptor in human and Vps10 in yeast will be recycled back to either plasma membrane or Golgi ( Goldstein and Brown , 2009; Seaman et al . , 1997 ) . As stated above , many studies have reported that endosomes also contain E3 ligase systems in yeast , plants , and metazoans ( Léon et al . , 2008; Nakamura , 2011; Tian et al . , 2015; Voiniciuc et al . , 2013 ) . However , it remains to be addressed as to the functions of these endosomal E3 ligases . One reasonable hypothesis is to counteract the activity of endosomal deubiquitinases ( Dubs ) ( Kee et al . , 2005; Kee et al . , 2006; Léon et al . , 2008 ) . During the endomembrane trafficking , ubiquitinated cargoes can be deubiquitinated by the Dubs and later recycled by the retromer . The presence of endosomal E3 ligases enables the cell to have the capability of determining whether a cargo needs to be degraded or recycled at the endosome stage ( Léon et al . , 2008 ) . In this study , we made a surprising observation that the Vld1-containing Dsc subcomplex , when forced into the VPS pathway , was recognized by an unidentified endosomal protein quality control ( EQC ) system and degraded inside the vacuole lumen ( Figures 1B and 7A ) . This observation indicated that endosomal E3 ligases may indeed have the capability of recognizing mislocalized proteins and removing them , although the identity of the E3 ligase still needs to be addressed . Strikingly , the only difference between the two subcomplexes were the Vld1 and Gld1 subunits , which are homologous to each other . Yet , the EQC system was able to recognize Vld1 and selectively ubiquitinate the protein . Considering endosomes are constantly receiving exogenous proteins delivered by the transporting vesicles , it is stunning to realize that endosomes can tell the difference between transporting cargoes and mislocalized membrane proteins . Apparently , future investigations , such as identifying the E3 ligase and expanding the substrate list of the EQC system , are needed to address this striking quality control mechanism . All yeast strains and plasmids used in this study are listed in Supplementary file 1 . Both Difco YPD broth and Difco Yeast Nitrogen Base ( YNB ) w/o amino acids were purchased from Thermo Fisher Scientific . All yeast strains were grown at 26°C in either YPD or YNB media before further analysis . The MS analysis was performed by the Mass Spectrometry and Metabolomics Core at the Michigan State University . Essentially , eluted samples from IP experiments were separated by SDS-PAGE and stained with SYPRO Ruby protein gel stain ( S12000 , Invitrogen ) . Two bands at 27 KDa and 31 KDa were excised and in-gel digested with trypsin . Then , digested peptides were extracted and eluted peptides were sprayed into a ThermoFisher Q-Exactive mass spectrometer using a FlexSpray spray ion source . Survey scans were taken in the Orbi trap ( 35000 resolution , determined at m/z 200 ) and the top ten ions in each survey scan are then subjected to automatic higher energy collision induced dissociation ( HCD ) with fragment spectra acquired at 17 , 500 resolution . The resulting MS/MS spectra were converted to peak lists using Mascot Distiller , v2 . 6 . 1 and searched against a database of all protein sequences available from SwissProt using the Mascot searching algorithm , v 2 . 6 . 0 . The Mascot output was then analyzed using Scaffold , v4 . 7 . 5 to probabilistically validate protein identifications . Assignments validated using the Scaffold 1%FDR confidence filter were considered true . The Zn2+ minus YNB media was prepared according to the methods of Li et al ( Li et al . , 2015a ) . For the Cot1-GFP degradation assay , yeast cells were grown in YNB media to mid-log phase ( OD600: 0 . 4∼0 . 8 ) , after 20 min of pre-incubation with 2 µg/ml doxycycline , the cells were collected at 2500 g for 5 min . After two times of washing with water , the cells were resuspended in Zn2+ minus YNB media that contained 2 ug/ml doxycycline and incubated at 26°C for an appropriate amount of time ( typically , 6–8 hr ) before further analysis . In our previous experiments at Cornell University , normally 4–6 hr was enough for a complete degradation of Cot1-GFP in WT . After moving to University of Michigan , we found that the degradation of Cot1-GFP was relatively slower , probably due to a higher level of residual Zn2+ in the MilliQ water system , so we have to extend the Zn2+ withdrawal treatment to 6–8 hr to get a complete degradation of Cot1-GFP . For GFP-Yif1 degradation assay , yeast cells were grown in YNB media to mid-log phase ( OD600: 0 . 4∼0 . 8 ) , then collected at 2500 g for 5 min . After two times of washing with water , the cells were resuspended in nitrogen starvation medium ( YNB lacking amino acids and ammonium sulfate , with 2% glucose ) and incubated at 26°C for an appropriate amount of time ( typically , 2–4 hr ) before being collected for further analysis . Microscopy was performed with a DeltaVision Elite system ( GE Healthcare Life Sciences ) , equipped with an Olympus IX-71 inverted microscope , a sCMOS camera , a 100X/1 . 4 Oil Super-Plan APO objective , and a DeltaVision Elite Standard Filter Set with the FITC filter ( Excitation:475/28 , Emission: 525/48 ) for mNeonGreen and the TRITC filter ( Excitation:542/27 , Emission: 594/45 ) for mCherry , DsRed , and FM4-64 . Yeast cells , except for those transformed with the pRS416-DsRed-HDEL plasmid , were grown in YPD at 26°C overnight . Yeast cells transformed with the pRS416-DsRed-HDEL plasmid were grown in YNB minus uracil at 26°C overnight . Before imaging , yeast cells were briefly washed with water and immediately imaged in milliQ water at room temperature . Image acquisition , deconvolution , and maximum projection analysis were performed with the program softWoRx . The image cropping and adjustment were performed using the ImageJ software ( National Institutes of Health ) . A rapamycin resistant strain ( SEY6210 . 1 , tor1-1 , fpr1∆::NAT ) ( Zhu et al . , 2017 ) was used to develop the RICo assay . First , Ubx3 and Gld1 were chromosomally tagged with mNeonGreen and 2xFKBP , respectively . Then , pRS305-pSSH4-FRB-mCherry plasmid was integrated into the yeast genome for stably expressing FRB-mCherry under the SSH4 promoter . Yeast cells were grown in YPD media to OD600 ∼3 , before being incubated with 1 µg/ml rapamycin at 26°C for 45 min , then collected for imaging . Yeast cells were grown overnight in YPD to late log phase . 1–1 . 5 ml of culture was collected , washed once with 1 ml YNB complete media , and resuspended with 100 ul of YNB complete media . Yeast cells were then labeled with FM4-64 ( T3166 , Invitrogen , 10 ug/ml final concentration ) for 10 min at room temperature in the dark ( Vida and Emr , 1995 ) . For the endosome staining , cells were immediately washed with 1 ml ice cold YNB complete media to remove the FM4-64 and kept on ice to stop the membrane trafficking . For vacuole membrane staining , 1 ml room temperature YNB complete media was added to the 10 min FM4-64 incubating cells and the incubation was continued for another 50 min in the dark to ensure the vacuole membrane staining . The cells were then collected , and washed with 1 ml ice cold YNB complete media , and kept on ice . Cells were resuspended in milliQ water and imaged by fluorescence microscopy . The immunoprecipitation assay was adapted from Li et al ( Li et al . , 2015a ) , with some modifications . Essentially , 1 liter of yeast culture ( OD600 ∼1 . 5 ) was harvested by spinning at 4 , 000 g for 10 min , resuspended in 50 ml weakening buffer ( 100 mM Tris-HCl , pH 8 . 8 , and 10 mM DTT ) and incubated at room temperature for 10 min to weaken the cell wall . The cells were then resuspended with 25 ml spheroplasting media ( 2% glucose , 1 × amino acids , 1M Sorbitol , 20 mM Tris-HCl , pH 7 . 5 , in YNB ) containing 100 µl of 10 mg/ml Zymolyase 100T ( 120493–1 , Amsbio ) , and incubated at 30°C for 30 min with gentle rocking . After washing once with 20 ml spheroplasting media , the cells were resuspended with 20 ml lysis buffer ( 20 mM HEPES , pH 7 . 2 , 50 mM KOAc , 10 mM EDTA , 200 mM Sorbitol , with Protease Inhibitor Cocktail ( 21169500 , Roche , Switzerland ) and ruptured on ice by 20 strokes in a Dounce homogenizer . The membrane fraction was collected by a 10 min 13 , 000 g spin at 4°C before being resuspended in 1 ml IP buffer ( 50 mM Hepes-KOH , pH 6 . 8 , 150 mM KOAc , 2 mM MgOAc , 1 mM CaCl2 , 15% glycerol ) supplemented with protease inhibitors . Then , the resuspended membrane was dissolved in 10 ml IP buffer containing 1% Triton X-100 at 4°C for 30 min , with gentle rocking . Insoluble material was removed by spinning at 14000 rpm ( Sorval SS-34 rotor ) for 10 min . The resulting lysate was incubated with 70 µl either M2 anti-FLAG resin ( A2426 , Sigma-Aldrich , St Louis , MO ) or anti-HA resin ( E6779 , Sigma-Aldrich ) for 3 hr at 4°C , with gentle rocking . The resin was then washed six times with 0 . 1% Triton X-100 in IP buffer . For anti-FLAG resin , bound proteins were eluted with 200 µl 3xFLAG peptide ( F3290 , Sigma-Aldrich , 100 µg/ml , dissolved in IP buffer containing 0 . 1% Triton X-100 ) . For anti-HA , the resin was incubated with 2 × SDS PAGE sample buffer ( 150 mM Tris , pH 6 . 8 , 2% SDS , 100 mM DTT and bromophenol blue ) at 42°C for 5 min to dissociate bound proteins . The resulting eluates were then analyzed by either Western blotting or silver staining . Total cell lysates were prepared from 7 OD600 cultures by incubating on ice for 1 hr in 10% TCA . After washing once with 0 . 1% TCA , samples were bead-beated for 5 min in 2 × urea buffer ( 150 mM Tris , pH 6 . 8 , 6 M urea , 6% SDS ) and incubated 5 min at 65˚C . After addition of 2 × sample buffer ( 150 mM Tris , pH 6 . 8 , 2% SDS , 100 mM DTT and bromophenol blue ) , samples were bead-beated again for 5 min and heated for another 5 min at 65˚C . Samples were then centrifuged at 21 , 000 g for 5 min and the supernatant was collected . The samples were separated by 10% polyacrylamide gels and transferred to nitrocellulose membranes for western blotting analysis . Antibodies used in this study were G6PDH ( A9521; Sigma-Aldrich ) , mouse anti-GFP ( SC9996; Santa Cruz Biotechnology , Inc . Santa Cruz , CA ) , rabbit anti-GFP ( TP401; Torrey Pines Biolabs , Secaucus , NJ ) , Flag ( F7425; Sigma-Aldrich ) , rabbit anti-HA ( 715500 , Life technologies , Camarillo , CA ) , mouse anti-HA ( 12CA5; Sigma-Aldrich ) and Vph1 ( 10D7 , Invitrogen , Carlsbad , CA ) . Antibodies against Dsc2 , Dsc3 , Ubx3 , and Tul1 were generous gifts from P . Espenshade ( Johns Hopkins University , Baltimore , MD ) .	Proteins perform many tasks and , to remain healthy , each cell must ensure that its proteins are in good condition and present at the right levels . Plants , animals and fungi all largely deal with damaged , or otherwise unneeded , proteins by tagging them with a small marker called ubiquitin . The tagged proteins are then rapidly destroyed , which prevents them from harming the cells . Enzymes known as E3 ligases attach ubiquitin to proteins . Yet , the number of E3 ligases is dwarfed by the number of proteins modified with ubiquitin . For instance , humans have approximately 20 , 000 different proteins , about one third of which are found in or on cell membranes . However , there are only around 600 E3 ligases , and only about 50 of them are associated with cell membranes . This is further complicated by the fact that proteins are also present in distinct compartments within the cell . The Dsc complex , for example , is an E3 ligase from yeast that is found within a compartment of the cell known as the Golgi . It was thus expected to only attach ubiquitin to Golgi proteins . Yet some recent studies showed that the Dsc complex could also tag proteins present in two other compartments of yeast cells: the endosome and vacuole . How can the Dsc complex act on proteins in three distinct compartments ? The Dsc complex is actually made from multiple proteins , and Yang et al . now report two new protein components . Biochemical and genetic tools showed that these two proteins do not co-exist in the same Dsc complex . Instead , they compete with each other to form two different kinds of Dsc complexes , which Yang et al . refer to as subcomplexes . Further work showed that the two new proteins determine the route taken by the Dsc complex along the cell’s protein transport pathway . One subcomplex is transported to the vacuole and the other cycles between the Golgi and endosomes . Thus , by changing just one component , the Dsc complex can be sent to different locations within the cell . These findings describe a new mechanism that enables E3 ligases to multi-task on a wide range of proteins , even across distinct compartments of the cell . Future work will determine whether plant and animal cells also use a similar strategy . Since defects in protein quality control contribute to many human diseases , such as Alzheimer's and Parkinson's disease , working out how E3 ligases work is important for the field of biomedicine .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2018	Sorting of a multi-subunit ubiquitin ligase complex in the endolysosome system
Epithelial tissues are primed to respond to insults by activating epithelial cell motility and rapid inflammation . Such responses are also elicited upon overexpression of the membrane-bound protease , Matriptase , or mutation of its inhibitor , Hai1 . Unrestricted Matriptase activity also predisposes to carcinoma . How Matriptase leads to these cellular outcomes is unknown . We demonstrate that zebrafish hai1a mutants show increased H2O2 , NfκB signalling , and IP3R -mediated calcium flashes , and that these promote inflammation , but do not generate epithelial cell motility . In contrast , inhibition of the Gq subunit in hai1a mutants rescues both the inflammation and epithelial phenotypes , with the latter recapitulated by the DAG analogue , PMA . We demonstrate that hai1a has elevated MAPK pathway activity , inhibition of which rescues the epidermal defects . Finally , we identify RSK kinases as MAPK targets disrupting adherens junctions in hai1a mutants . Our work maps novel signalling cascades mediating the potent effects of Matriptase on epithelia , with implications for tissue damage response and carcinoma progression . The transmembrane serine protease , Matriptase , encoded by the ST14 gene , has potent oncogenic properties and is consistently dysregulated in human carcinomas . Overexpression of Matriptase in the mouse epidermis leads to epidermal papillomas , ulcerative and invasive carcinomas , and inflammation ( List et al . , 2005; Martin and List , 2019 ) . These effects of Matriptase are mitigated by a cognate serine protease inhibitor , HAI-1 . Clinically , an increase in the Matriptase:HAI-1 ratio has been found in a range of tumours and is predictive of poor outcome ( Martin and List , 2019 ) . Loss of mouse Hai1 leads to epidermal and intestinal barrier defects , epithelial inflammation , and failure of placental labyrinth formation , which are all due to unrestricted Matriptase activity ( Kawaguchi et al . , 2011; Nagaike et al . , 2008; Szabo et al . , 2007 ) . The response of epithelia to unregulated Matriptase activity appears conserved across vertebrates . Mutation of the zebrafish orthologue , Hai1a , also results in epidermal defects , including loss of membrane E-cadherin , aberrant mesenchymal behaviour of keratinocytes , which form cell aggregations over the body and loss of fin fold integrity . The epidermis also displays sterile inflammation and is invaded by highly active neutrophils . Genetic ablation of the myeloid lineage demonstrated that the keratinocyte phenotypes are not a consequence of the inflammation ( Carney et al . , 2007 ) . The strong hai1afr26 allele is embryonic lethal , dying within the first week , whilst the more mild allele , hai1ahi2217 , is semi-viable , with epithelial defects resolved through sphingosine-1-phosphate-mediated entosis and cell extrusion ( Armistead et al . , 2020 ) . All hai1a mutant phenotypes can be ameliorated by reduction of Matriptase levels ( Carney et al . , 2007; Mathias et al . , 2007 ) . Due to the clinical implications of its dysregulation , the signalling pathways activated pathologically by Matriptase are of interest . The G-protein-coupled receptor , proteinase-activated receptor-2 ( Par2 ) , is essential for the oncogenic and inflammatory effects of uninhibited Matriptase in zebrafish and mouse ( Sales et al . , 2015; Schepis et al . , 2018 ) . Par2 is directly activated by Matriptase proteolysis and signals through a number of heterotrimeric Gα protein subunits . Early studies in keratinocytes linked Par2 activation with intracellular Ca++ mobilisation via phospholipase C , thus implicating Gq subunit as the canonical target ( Schechter et al . , 1998 ) . Alternate Gα subunits , including Gi , Gs , and G12/13 , are now known to also be activated by Par2 ( Zhao et al . , 2014 ) . Par2 displays biased agonism , and the logic of the pathway utilised depends on cell context and the activating protease . In vitro experiments using HEK293 cells implicated both Par2 and Gi in Matriptase-mediated Nfκb pathway activation ( Sales et al . , 2015 ) . Whilst this explains the inflammatory phenotype of uninhibited Matriptase , it does not address whether Par2 promotes carcinoma phenotypes directly in keratinocytes in vivo . In zebrafish , as the keratinocyte defects are not dependent on inflammation , but are dependent on Par2 , it is likely that there is a direct effect of Par2 on promoting keratinocyte motility . Par2 can also transactivate EGFR through an unknown mechanism , and inhibition of EGFR alleviates certain basal keratinocyte phenotypes of zebrafish hai1a mutants ( Schepis et al . , 2018 ) . Thus , the identity , contribution , and interactions of the pathways downstream of Matriptase and Par2 remain unclear . Here through use of the zebrafish hai1a mutant , we comprehensively map the essential pathways downstream of zebrafish Matriptase and Par2 , leading to inflammation and epithelial disruption . Neutrophils in hai1a embryos invade the epidermis , are highly motile , but move randomly ( Carney et al . , 2007; Mathias et al . , 2007; Figure 1A–E , Video 1 ) . To establish the nature of their stimulus , we tested if neutrophils in hai1a altered their behaviour in the presence of a large fin wound . In wild-type larvae , neutrophils were recruited from a great distance and tracked to the wound with high directionality . However , neutrophils in the hai1a mutant appeared largely apathetic to the wound and remained migrating randomly . There was a mild increase in neutrophil speed in hai1a larvae following wounding , indicating that they retain capacity to respond to additive stimuli ( Figure 1—figure supplement 1A–D , Video 2 ) . Co-labelling of basal keratinocyte nuclei ( using TP63 immunostaining ) , neutrophils ( Tg ( fli1:EGFP ) y1 transgenic ) , and TUNEL labelling of apoptotic cells highlighted that whilst the epidermis of hai1a mutants , unlike WT , had regions of apoptosis at 24hpf ( arrowhead , Figure 1—figure supplement 1E , F ) , neutrophils were not associated , but rather found at nascent TUNEL-negative aggregates of basal keratinocytes ( arrow ) . We conclude that epidermal cell death does not directly contribute to inflammation and that the effector stimulating neutrophils in hai1a mutants is as , or more , potent as that of wounds . To identify the neutrophil activator in hai1a , we employed an unbiased approach using 2D gel proteomics to compare the wild-type proteome with that of strong hai1a alleles . The dandruff ( ddf ) mutant has many phenotypic similarities to the strong hai1afr26 allele ( van Eeden et al . , 1996 ) . Crosses between ddfti251 or ddft419 and hai1ahi2217 failed to complement , and sequencing of hai1a cDNA from both ddf alleles identified a nonsense mutation in the ddft419 allele ( c . 771T>G; p . Tyr257Ter ) and a missense mutation of a highly conserved amino acid in the ddfti251 allele ( c . 749G>A; p . Gly250Asp ) ( Figure 1F , Figure 1—figure supplement 1G–I ) . We used both alleles for comparative 2D protein gel analysis at 24hpf and 48hpf . Rather than finding proteins with altered molecular weight , Peroxiredoxin4 ( Prdx4 ) was identified as having a higher pI in both hai1at419 and hai1ati251 mutants at 24hpf and 48hpf , indicative of a change in oxidation state ( Figure 1G , Figure 1—figure supplement 1J , K ) . Peroxiredoxins are hydrogen peroxide scavengers , and its altered oxidation state suggested that hai1a has higher H2O2 levels , a known activator of inflammation in larval zebrafish ( Niethammer et al . , 2009 ) . Pentafluorobenzenesulfonyl fluorescein ( PFBSF ) staining Maeda et al . , 2004 demonstrated significantly higher levels of H2O2 in the trunk and tails of hai1a mutants at 24hpf and 48hpf ( Figure 1H–J , Figure 1—figure supplement 1L , M ) . This increase in H2O2 in hai1a was observed as early as 16hpf , and thus preceded presentation of hai1a phenotypes ( Figure 1K , L ) . To demonstrate that , as with other phenotypes , the H2O2 increase in hai1a was due to unrestrained activity of Matriptase1a , we used a matriptase1a mutant allele , st14asq10 , which prematurely terminates the protein at 156 amino acids ( Figure 2A , Figure 2—figure supplement 1A–C; Lin et al . , 2019 ) . Zygotic st14a mutants showed no overt phenotype; however , maternal zygotic mutants lacked ear otoliths ( Figure 2B , C ) . As expected , when crossed into the hai1a background , embryos lacking otoliths ( st14asq10; hai1ahi2217 double mutants ) never displayed the hai1a epidermal and neutrophil phenotypes ( Figure 2D–F; Table 1 ) . Double mutants also had significantly reduced H2O2 levels ( Figure 2F , Figure 2—figure supplement 1D ) . To determine if reduced H2O2 could account for the rescue of hai1a phenotypes by st14a mutation , we used genetic and pharmacological inhibition of the main enzyme responsible for generating H2O2 in zebrafish , Duox . A morpholino directed against duox successfully reduced H2O2 levels ( Figure 2 , Figure 2—figure supplement 1D ) and neutrophil inflammation in hai1a mutants but did not rescue the epithelial defects ( Figure 2F , G ) . Treatment with a known Duox inhibitor , diphenyleneiodonium ( DPI ) , also resulted in amelioration of neutrophil inflammation , but not epithelial aggregates , in hai1a mutants ( Figure 2G , Figure 2—figure supplement 1E ) . We conclude that Matriptase1 activity leads to excess H2O2 in hai1a mutants , which partially accounts for the neutrophil inflammation , but not epidermal defects . Duox is regulated by calcium through its EF-Hand domains , and calcium flashes are known to generate H2O2 in epidermal wounds in Drosophila ( Razzell et al . , 2013 ) . We injected hai1afr26 with RNA encoding the calcium reporter GCaMP6s . Timelapse imaging at 24hpf indicated that hai1a mutants had significantly more calcium flashes in both the trunk and tail ( Figure 3A , B , E , Figure 3—figure supplement 1A , B , Video 3 ) . Increased intracellular calcium dynamics was observable as early as 16hpf , concomitant with increased H2O2 , but prior to onset of hai1a phenotypes ( Figure 3G , H , Video 4 ) . Release of calcium from intracellular stores is regulated by IP3 receptors located on the endoplasmic reticulum . The frequency and number of calcium flashes in the trunk and tail of hai1a mutants are reduced by treatment with the IP3R inhibitor , 2-APB compared to control ( Figure 3C , D , F , Figure 3—figure supplement 1C , D , Video 5 ) . Reducing calcium flashes in hai1a mutant embryos with 2-APB also significantly reduced H2O2 levels ( Figure 3I , J , Figure 3—figure supplement 1E ) and partially reduced inflammation; however , the epidermal defects were not noticeably rescued ( imaged by DIC ( Differential Interference Contrast ) or labelled with the TP63 antibody ) ( Figure 3I–K ) . We observed similar reduction in neutrophil inflammation , but not rescue of epidermal defects , in hai1a mutants treated with thapsigargin , which inhibits the replenishment of ER calcium stores by SERCA ( Figure 3K , Figure 3—figure supplement 1F , G ) . This suggests , in hai1a mutants , that IP3R-dependent calcium flashes activate Duox , flooding the epidermis with H2O2 and leading to inflammation . Increased Matriptase , Par2 activity , or hydrogen peroxide levels are known to activate NfkB signalling ( Kanke et al . , 2001; Sales et al . , 2015; Schreck et al . , 1991 ) . We crossed the hai1afr26 allele to the NfkB sensor transgenic line Tg ( 6xHsa . NFKB:EGFP ) nc1 . In WT embryos , NfkB signalling was mostly restricted to neuromasts at 48hpf , whilst in hai1a mutants we observed an increase in fluorescence at 24hpf and a strong increase at 48hpf . Fluorescence at both timepoints was noted in epidermal aggregates and fin folds , locations of strong inflammation ( Figure 4A , B , Figure 4—figure supplement 1A , B ) . This increase in signalling in 48hpf hai1a mutant embryos was confirmed by qRT-PCR of the NfkB target gene , nfkbiaa ( Figure 4C ) . Unlike calcium and H2O2 , NfkB signalling is not present at early stages prior to phenotype ( Figure 4—figure supplement 1C , D ) . To determine the extent that NfkB signalling accounts for the hai1a phenotypes , we generated a mutant in the ikbkg ( =ikkg or nemo ) gene , which encodes a scaffold protein required for activating the NfkB pathway ( Rothwarf et al . , 1998 ) ( ikbkgsq304 Gly80ValfsTer11; Figure 4—figure supplement 1E ) . Crossing this mutant to hai1ahi2217 on the Tg ( mpx:eGFP ) i114 background realised a very strong rescue of neutrophil inflammation at 48hpf , but no improvement of hai1a epidermal defects ( Figure 4D–I ) . To demonstrate that this increase in NfkB signalling was dependent on H2O2 , we injected hai1ahi2217; Tg ( 6xHsa . NFKB:EGFP ) nc1 embryos with duox MO . We noted a strong reduction in NfkB pathway activation compared to uninjected hai1ahi2217 mutant controls ( Figure 4J , K ) . Conversely , genetic ablation of NfkB signalling , through use of the ikbkg mutant , did not reduce H2O2 levels in hai1a mutants ( Figure 4—figure supplement 1F , G ) . Similarly , we tested if reduction of calcium flashes could also reduce NfkB signalling in hai1a mutants using 2-APB but noticed only a slight reduction ( Figure 4—figure supplement 1H , I ) . We propose that upon loss of Hai1a , IP3R-mediated release of calcium activates Duox to increase H2O2 . This acts upstream of NfkB pathway activation , which occurs at later stages , and is necessary for the inflammation phenotype . IP3 is generated from cleavage of PIP2 by Phospholipase C . The sensitivity of the hai1a mutants to 2-APB implies that IP3 levels are increased and therefore there may be an increase in Phospholipase C activity . Numerous attempts to inhibit PLC in hai1a mutants failed , and we were unable to find a dosage window that rescued without gross embryo deformity . Hence , we tested rescue of hai1a mutants with YM-254890 , an inhibitor of the heterotrimeric G protein alpha subunit , Gq , which directly activates PLC isoforms . We found that not only did this significantly reduce neutrophil inflammation ( Figure 5D , F ) , but surprisingly , it also significantly rescued the epidermal defects in hai1a mutants , with a significant reduction in TP63-positive epidermal aggregates in the trunk and improved tail fin fold integrity at 48hpf ( Figure 5A–E ) . As IP3R inhibition only blocks inflammation in hai1a mutants , but an inhibitor of a PLC activator ( Gq ) additionally reduces the epidermal defects , we considered that diacyl glycerol ( DAG ) might contribute to the epidermal defects as the second product of PIP2 cleavage ( along with IP3 ) . Indeed , treating WT embryos from 15hpf to 24hpf with 125 ng/ml phorbol 12-myristate 13-acetate ( PMA ) , a DAG analogue , resulted in embryos with striking similarities to strong hai1a mutants , including a thin or absent yolk sac extension , lack of head straightening , lack of lifting the head off the yolk , and multiple epidermal aggregates on the skin ( Figure 6A–C ) . These aggregates were due , at least partially , to displacement of basal keratinocytes as shown by TP63 staining where the basal keratinocyte nuclei lost their uniform distribution ( Figure 6D , E ) . Treatment from 24hpf to 48hpf with 125 ng/ml PMA led to a fin defect similar to the dysmorphic hai1a mutant fin ( Figure 6F , G ) . It has been shown that the basal keratinocytes in hai1a lose their epithelial nature and adopt a partially migratory phenotype ( Carney et al . , 2007; Video 6 ) . We treated Tg ( krtt1c19e:lyn-tdtomato ) sq16 larvae ( Lee et al . , 2014 ) with 37 . 5 ng/ml PMA for 12 hr and imaged the basal epidermis at 3dpf by light-sheet timelapse . Whilst the DMSO-treated transgenic larvae had very stable keratinocyte membranes and shape , PMA treatment led to a less stable cell membrane topology and dynamic cell shape , similar to hai1a mutants ( Figure 6H , Videos 7 and 8 ) . Kymographs taken from Video 7 highlighted both the more labile and weaker cell membrane staining following PMA treatment ( Figure 6I ) . The potency of PMA was dependant on region and reduced with age . Most PMA-treated Tg ( mpx . eGFP ) i114 larvae at 48hpf also had more neutrophils in the epidermis than untreated controls , which were highly migratory ( Figure 6F–G , J–K′ , Video 8 ) . We determined H2O2 levels in PMA-treated embryos using PFBSF staining and found that it was significantly increased in both trunk and tail at 24hpf ( Figure 6L–O , R ) . In contrast , when we treated GCaMP6s RNA-injected embryos with PMA , we failed to see an increase in calcium flashes , as seen in hai1a ( Figure 6P , Q , S ) . To see if the heightened H2O2 and inflammation was also correlated with increased NfkB signalling , we treated Tg ( 6xHsa . NFKB:EGFP ) nc1 embryos with 125 ng/ml PMA . There was a robust increase in fluorescence , indicating that PMA activates the NfkB pathway ( Figure 6T , U ) . The phenocopy and the rescue of hai1a by PMA and Gq inhibition respectively imply that DAG is elevated in hai1a mutants . Elevated cellular DAG leads to relocalisation of Protein Kinase C isoforms to the plasma and nuclear lipid membranes where they bind DAG and become activated . Using a GFP-tagged PKCδ fusion protein ( Sivak et al . , 2005 ) , we showed that in the WT embryo there was largely diffuse cytoplasmic PKCδ-GFP signal , however , it translocated to plasma and nuclear membranes in hai1a mutants , indicating increased levels of DAG ( Figure 7A , B , Figure 7—figure supplement 1A , B ) . This is indeed relevant to the epidermal defects , as treatment of hai1ahi2217 embryos with the PKC inhibitor , GFX109203 , reduced the epidermal aggregates and disruption of fin morphology as imaged by DIC or immunostaining for TP63 ( Figure 7C–H ) . Neutrophil inflammation in the epidermis was somewhat reduced , but not to a significant degree ( Figure 7E–I ) . Thus , these experiments strongly suggest that epithelial defects of hai1a are due to DAG generation and PKC activation . We next sought to determine which pathways downstream of PKC are responsible for the epidermal defects . The MAPK pathway is a major target pathway of multiple PKC isoforms , and activation of this pathway in zebrafish epidermis has previously been shown to induce papilloma formation which have very similar attributes to hai1a mutant aggregates ( Chou et al . , 2015 ) . Although whole embryo western analysis of hai1a mutants failed to show an overall increase in pERK ( Armistead et al . , 2020 ) , we performed wholemount immunofluorescent analysis in case there was only a localised effect . Indeed , we observed a significant and localised increase in cytoplasmic pERK immunoreactivity ( phospho-p44/42 MAPK ( Erk1/2 ) ( Thr202/Tyr204 ) ) in the regions of epidermal aggregate formation in hai1a mutants and in PMA-treated embryos , including under the yolk at 24hpf and in the fins at 24hpf and 48hpf ( Figure 8A–K , Figure 8—figure supplement 1A–F ) . There was no increase in total ERK levels in the mutant ( Figure 8—figure supplement 1M , N ) . Increased pERK was seen in both the cytoplasm and nucleus of TP63-positive cells but was only increased in the nucleus of periderm cells ( Figure 8E–E′ , Figure 8—figure supplement 1D ) . To establish that this is an early marker of aggregate formation , and not a sequela , we stained hai1a mutant embryos at earlier timepoints . We found that at 16hpf regions of the epidermis have pERK staining before overt aggregation formation ( Figure 8G–H ) , whilst nascent aggregates also contain pERK staining which increases in number over time ( Figure 8—figure supplement 1G–L ) . To determine if elevated pERK is causative of epidermal defects , we attempted to rescue using pERK inhibitors . Initially we used PD0325901; however , this appeared to give fin fold deformities , even in WT embryos ( Anastasaki et al . , 2012 ) , precluding ability to assess rescue in hai1a , although there was a noticeable reduction in epidermal aggregates forming under the yolk-sac extension ( data not shown ) . Instead , we tried U0126 and CI-1040 , other well-known pERK inhibitors ( Allen et al . , 2003; Favata et al . , 1998 ) . Both inhibitors showed a significant reduction in hai1a mutant epidermal aggregates under the yolk , and restoration of the overall and tail epithelial morphology , with embryos showing a hai1a phenotype class significantly reduced ( Figure 9A–G , Figure 9—figure supplement 1A–F ) . Similarly , the epidermal defects of the trunk , yolk , and tail following PMA treatment were also ameliorated by concomitant U0126 treatment ( Figure 9H , I , Figure 9—figure supplement 1G , H ) . Rescue of aggregates and tail morphology following PMA treatment or in hai1a mutants could be visualised by immunolabelling TP63 in basal keratinocyte nuclei ( Figure 9J–O , Figure 9—figure supplement 1I , J ) . Initiating U0126 treatment later at 26hpf led to only a partial rescue , indicating that the epidermal phenotypes were likely due to sustained pERK activation ( Figure 9—figure supplement 1K–M′ ) . Treatment with U0126 did not significantly reduce neutrophil inflammation of hai1a mutants or PMA treatment ( Figure 9L–P ) . This suggests that the inflammation phenotype is not simply a consequence of the epidermal defects . Furthermore , dye penetration assays showed that the epithelial barrier was not globally and overtly compromised in hai1a , underscoring that inflammation is not simply a consequence of epithelial defects ( Figure 9—figure supplement 2A–H ) . It has been shown that the epidermal defects in hai1a are associated with loss of E-cadherin from adherens junctions ( Carney et al . , 2007 ) . As there was a rescue of the epithelial phenotype following pERK inhibition , we looked at the status of the adherens junction marker β-catenin . Whilst the WT basal epidermal cells of the 48hpf tail showed strong staining at the membrane , hai1a mutants and PMA-treated embryos showed a significant loss of β-catenin at the membrane and increase in the cytoplasm ( Figure 9Q–V , Y , Z ) . Treatment of hai1a mutants with U0126 restored the membrane localisation of β-catenin ( Figure 9W , X , AA ) . As increased pERK appeared to contribute strongly to loss of adherens junctions and removal of E-cadherin/β-catenin from the membrane , we sought to determine how pERK signalling might affect adherens junctions . We predicted that this would occur through a cytoplasmic target of pERK as we have previously shown that there is no transcriptional downregulation of E-cadherin levels in hai1a , making a nuclear transcription factor target less likely to be relevant ( Carney et al . , 2007 ) . The p90RSK family of kinases represents direct cytoplasmic targets of Erk1/2 phosphorylation which regulate cell motility , and thus were good candidates for mediators disrupting cell-cell adhesion ( Čáslavský et al . , 2013; Tanimura and Takeda , 2017 ) . We determined that at least RSK2a ( =p90RSK2a , encoded by rps6ka3a ) is expressed in basal keratinocytes at 24hpf ( Figure 10A , B ) . To gauge if there was an alteration in phosphorylation of RSK family members in the epidermis of hai1a mutants , we used an antibody which detects a phosphorylated site of mouse p90RSK ( Phospho-Thr348 ) . This site is phosphorylated in an ERK1/2-dependent manner ( Romeo et al . , 2012 ) . We noticed a substantial increase in cytoplasmic signal in both hai1a mutants and PMA-treated embryos . Where p90RSK-pT348 signal was largely nuclear in both basal and periderm cells in WT , it was more broadly observed in hai1a mutant fins , with an increase in the cytoplasm leading to a more uniform staining ( Figure 10C–D′ ) . This increase in cytoplasmic levels of p90RSK-pT348 was observable at 17hpf prior to epithelial defects ( Figure 10—figure supplement 1A–C ) . p90RSK cytoplasmic signal was lost upon U0126 and GFX109203 treatments , showing that it was pERK and PKC dependant ( Figure 10E , E′ , Figure 10—figure supplement 1D , E ) . Similarly , increased cytoplasmic p90RSK-pT348 was observed following PMA treatment which was reduced by co-treatment with U0126 ( Figure 10F–H′ ) . The increase in cytoplasmic p90RSK-pT348 signal , and its reduction by U0126 , was significant in both hai1a mutants and PMA-treated embryos ( Figure 10I , J ) . If phosphorylation of an RSK protein is required for mediating the pERK epidermal defects in hai1a mutants , then inhibition of RSK should rescue the epidermal defects . As morpholino-targeted inhibition of rps6ka3a was unsuccessful , we employed established pan-RSK inhibitors BI-D1870 and dimethyl fumarate ( Andersen et al . , 2018; Sapkota et al . , 2007 ) . Dimethyl fumarate treatment reduced the extent of cytoplasmic p90RSK-pT348 in hai1a ( Figure 10—figure supplement 1F , G ) . We noted that both inhibitors were able to reduce epidermal aggregates in hai1a mutants and restore fin morphology when visualised by DIC or TP63 immunofluorescence ( Figure 10K–N , Figure 10—figure supplement 1H , I , K , L ) . Reduction of mutant phenotype classes was significant at both 24hpf and 48hpf ( Figure 10—figure supplement 1J ) . We then assayed if RSK inhibition can reduce the aberrant cytoplasmic E-cadherin staining in hai1a mutant basal keratinocytes and observed that dimethyl fumarate treatment restored membrane localisation of E-cadherin in the mutants ( Figure 10O–Q′ ) . Thus , phosphorylation of RSK proteins is altered in hai1a mutants , and their inhibition appears to restore E-cadherin to the membrane and reduce epidermal aggregate formation . There are a number of similarities between loss of Hai1a in zebrafish and overexpression of Matriptase in the mouse epidermis , including inflammation , hyperproliferation , and enhanced keratinocyte motility , suggesting conservation of downstream pathways . What the conserved ancestral role of the Matriptase-Hai1 might have been is unclear . Matriptase dysregulation in the mouse is associated with cancer progression ( Martin and List , 2019 ) . Tumours have long been considered to represent non-healing wounds , and the cellular- and tissue-level phenotypes of hai1a have similarities to tumours . Epidermal cells in zebrafish transformed by MAPK activation both promote and respond to inflammation through similar mechanisms to wound responses ( Feng et al . , 2010; Schäfer and Werner , 2008 ) . Further , tissue damage of the zebrafish epidermis perturbs osmolarity and releases nucleotides , leading to inflammation and epithelial cell motility , with the resulting phenotypes strikingly similar to hai1a mutants ( de Oliveira et al . , 2014; Enyedi and Niethammer , 2015; Gault et al . , 2014; Hatzold et al . , 2016 ) . Indeed , the tissue responses initiated by loss of zebrafish Hai1a have been previously suggested to represent an early injury response ( Schepis et al . , 2018 ) , whilst PAR2 synergises with P2Y purinergic and EGF receptors to promote cell migration in scratch assays ( Shi et al . , 2013 ) . Thus our analysis supports the previous hypothesis of the Hai1-Matriptase system as a component of tissue injury responses ( Schepis et al . , 2018 ) , which , if inappropriately activated , promotes carcinoma . The various molecular pathways known to be activated by Matriptase have not been fully delineated or integrated . Par2 has previously been shown to be required for the hai1a phenotype in zebrafish and contributes to the phenotypes of Matriptase overexpression in the mouse . Exactly which heterotrimeric G-protein Par2 is activating in vivo and how this links to phenotypes has not been identified . Our analyses allow us to propose a pathway downstream of Par2 which accounts for both the inflammatory and the epidermal phenotypes ( Figure 11 ) . Firstly , inhibition of Gq rescued both the inflammation and epithelial defects . PAR2 activation of Gq has been documented to occur in many cell types including keratinocytes , where inhibition of Gq and PKC reduces PAR2-mediated Nfκb signalling ( Böhm et al . , 1996; Goon Goh et al . , 2008; Macfarlane et al . , 2005 ) . Although we were unable to rescue hai1a phenotypes with a PLC inhibitor due to toxicity , genetic sensors demonstrated increased levels of Ca++ and DAG in hai1a epidermis . Our analysis demonstrated that the different products of PIP2 hydrolysis appear to invoke the two main hai1a phenotypes to different extents . IP3R-dependent calcium release in hai1a epidermis was required for Duox activity , high hydrogen peroxide levels , and , later , increased NfkB signalling . Reduction of these attenuated the inflammatory , but not epithelial , defects . Conversely , inhibiting the DAG receptor , PKC , rescued the epithelial phenotypes , and the inflammation slightly . The DAG analogue , PMA , phenocopied the epidermal defects of hai1a mutants but also increased H2O2 , NfkB , and neutrophil inflammation , indicating that PKC activation may be sufficient , but not necessary , for inflammation . This is in line with known activation of Duox and IKK by PKC ( Rigutto et al . , 2009; Turvey et al . , 2014 ) . In addition , expression of activated Ras in zebrafish keratinocytes has been shown to lead to H2O2 release and neutrophil attraction ( Feng et al . , 2010 ) . Thus , there is likely to be dual contribution to the inflammatory phenotype from IP3 and DAG . It is important to stress however that the inflammation is not simply a result of epithelial defects or an overt loss of barrier . Firstly , we see increase in Ca++ and H2O2 very early in the epidermis prior to skin defects . Secondly , barrier assays failed to conclusively show a broad increase in permeability . Finally , rescue of epithelial defects by PKC and pERK inhibition did not fully rescue the inflammation . We conclude in our model that DAG contributes to both aspects of the phenotype , but IP3 promotes only the inflammation . Seminal experiments in transgenic mice overexpressing Matriptase in the epidermis and treated with a DMBA/PMA regime concluded that Matriptase and PMA activate functionally similar carcinoma promoting pathways ( List et al . , 2005 ) . Our subsequent analysis suggests that this would include the MAPK pathway as we see increased phosphorylated-ERK in the epidermis of both hai1a mutants and also PMA-treated embryos . That we can rescue the epithelial defects using a MEK inhibitor indicated that this increase in epidermal pERK is likely critical to the phenotype . The MAPK pathway is known to regulate cell motility ( Tanimura and Takeda , 2017 ) . In the zebrafish epidermis , misexpression of activated MEK2 generated papillomas with remarkable resemblance to the epidermal aggregates in hai1a mutants ( Chou et al . , 2015 ) , and which are not overtly proliferative . In astrocytes and oesophageal or breast tumour cell lines , PAR2 stimulates migration and invasiveness through MAPK/ERK , activation of which required Gq and PIP2 hydrolysis ( Jiang et al . , 2004; McCoy et al . , 2010; Morris et al . , 2006; Sheng et al . , 2019 ) . One of the main molecular defects defined for zebrafish hai1a is the removal of adherens junction proteins from the membrane ( Carney et al . , 2007 ) . MAPK signalling has been shown to reduce E-cadherin expression at adherens junctions and promote cytoplasmic accumulation through phosphorylation of the effector , RSK ( Čáslavský et al . , 2013 ) . Like Matriptase , activation of RSK2 is associated with tumour progression , promoting invasiveness and metastasis of glioblastomas and head and neck squamous cell carcinomas ( Kang et al . , 2010; Sulzmaier et al . , 2016 ) . Promotion of invasiveness has also been noted for activated RSK1 , which promotes invasion of melanoma clinically as well as in vitro and zebrafish melanoma models ( Salhi et al . , 2015 ) . Intriguingly , proximity protein labelling has identified p120-catenin as a target of RSK phosphorylation . This catenin promotes cell-cell adhesion by stabilising cadherins at junctions , a function inhibited by RSK phosphorylation ( Méant et al . , 2020 ) . More broadly , RSK2 activity promotes cell motility through other mechanisms , including inactivation of Integrins and activation of the RhoGEF , LARG ( Gawecka et al . , 2012; Shi et al . , 2018 ) . Thus , we propose that pERK signalling , through RSK members , significantly contributes to dissolution of adherens junctions and the hai1a epidermal phenotype . We observed increased pERK in the cytoplasm and also the nucleus of keratinocytes , with comparatively more nuclear levels in periderm cells . Thus , whilst RSKs are phosphorylated by pERK , it is also likely that other cytoplasmic and also nuclear targets , such as cFos and Ets transcription factors , may also be activated , and that there are underlying transcriptional changes in hai1a mutants . It is not clear why pERK shows slightly different subcellular localisation patterns between the two different epidermal layers , but the two layers do respond differently to ErbB2 inhibition ( Schepis et al . , 2018 ) , whilst calcium is recently described to alter nuclear shuttling of pERK ( Chuderland et al . , 2020 ) . Our model for how Matriptase invokes cellular responses is highly likely to be incomplete . Indeed , others have indicated MMPs , HB-EGF , EGFR , and AKT and are downstream of Matriptase and PAR2 function ( List et al . , 2005; Schepis et al . , 2018; Darmoul et al . , 2004; Chung et al . , 2013; Rattenholl et al . , 2007 ) . Furthermore , Matriptase promotes HGF–cMet signalling in mouse ( Szabo et al . , 2011 ) . We do not think that these conflict with our model but will interface with it . A number of reports have demonstrated that PI3K/AKT and MEK/ERK function in parallel downstream of PAR2 ( Sheng et al . , 2019; Tanaka et al . , 2008; van der Merwe et al . , 2009 ) . Furthermore , there is evidence that PKC activates both MEK/ERK and EGFR independently following PAR2 stimulation , and that PI3K is activated by PAR2 via Gq ( Wang and DeFea , 2006; Al-Ani et al . , 2010 ) . Cell identity , subcellular localisation , β-arrestin scaffolding , and biased agonism/antagonism are known to generate alternative downstream outputs from PAR2 ( Zhao et al . , 2014 ) . To understand fully the roles of Matriptase and PAR2 in epithelial homeostasis and carcinoma , it will be critical to map how , when , and where they activate different downstream pathways . Fish were housed at the IMCB and the NTU zebrafish facilities under IACUC numbers #140924 and #A18002 , respectively , and according to the guidelines of the National Advisory Committee for Laboratory Animal Research . Embryos were derived by natural crosses and staged as per Kimmel et al . , 1995 and raised in 0 . 5× E2 medium ( 7 . 5 mM NaCl , 0 . 25 mM KCl , 0 . 5 mM MgSO4 , 75 μM KH2PO4 , 25 μM Na2HPO4 , 0 . 5 M CaCl2 , 0 . 35 mM NaHCO3 ) . Anaesthesia was administered in E2 medium ( embryos ) or fish tank water ( adults ) using 0 . 02% pH 7 . 0 buffered Tricaine MS-222 ( Sigma ) . The hai1a/ddf alleles used were hai1ahi2217 , hai1afr26 , ddfti251 , and ddft419 . The st14asq10 allele was generated previously ( Lin et al . , 2019 ) . For imaging neutrophils and keratinocytes , the transgenic lines Tg ( mpx:EGFP ) i114 ( Renshaw et al . , 2006 ) and Tg ( krtt1c19e:lyn-tdtomato ) sq16 ( Lee et al . , 2014 ) were used , whilst early leukocytes were imaged with Tg ( fli1:EGFP ) y1 ( Redd et al . , 2006 ) . To image NfkB pathway activity , the Tg ( 6xHsa . NFKB:EGFP ) nc1 sensor line was used ( Kanther et al . , 2011 ) . Calcium imaging was performed by injection of GCaMP6s RNA ( see below ) or using a Tg ( actb2:GCaMP6s , myl7:mCherry ) lkc2 stable transgenic line , generated via plasmid ( Chen et al . , 2017 ) and Tol2 RNA co-injection . Adult fin clips or embryos were isolated following anaesthesia , and genomic DNA extracted by incubation at 55°C for 4 hr in Lysis buffer ( 10 mM Tris pH 8 . 3 , 50 mM KCl , 0 . 3% Tween20 , 0 . 3% Nonidet P-40 , 0 . 5 µg/µl Proteinase K ) . PCRs were performed using GoTaq ( Promega ) on a Veriti thermal cycler ( Applied Biosystems ) and purified with a PCR purification kit ( Qiagen ) . TRIzol ( Invitrogen ) was used for RNA extraction following provided protocol , and cDNA generated from 1 µg total RNA using SuperScript III Reverse Transcriptase ( Invitrogen ) with Oligo ( dT ) 12-18 primer . For qPCR , iTaq SYBR green ( Bio-Rad ) was used to amplify , with reaction dynamics measured on a Bio-Rad CFX96 Real-Time PCR Detection System . For measuring nfkbiaa mRNA by qPCR , the following primers ( 5′ to 3′ ) were used to amplify a region encoded on exons 4 and 5: F-AGACGCAAAGGAGCAGTGTAG , R-TGTGTGTCTGCCGAAGGTC . Reference gene was eef1a1l1 and the primers used amplified between exon 3 to 4: F-CTGGAGGCCAGCTCAAACAT , R- ATCAAGAAGAGTAGTACCGCTAGCATTAC . RNAs for GCaMP6s and PKCδ-GFP were synthesised from pCS2-based plasmids containing the respective coding sequences ( Sivak et al . , 2005; Chen et al . , 2017 ) . These were linearised with NotI ( NEB ) , and RNA in vitro transcribed with mMESSAGE mMACHINE SP6 Transcription Kit ( Ambion ) . RNA for Tol2 was generated from the pT3Ts-Tol2 plasmid , linearised with SmaI ( NEB ) , and transcribed with the mMESSAGE mMACHINE T3 Transcription Kit ( Ambion ) . RNA for injection was purified by lithium chloride precipitation . Embryos were aligned on an agarose plate and injected at the one-cell stage with RNA or morpholino diluted in Phenol Red and Danieau’s buffer using a PLI-100 microinjector ( Harvard Apparatus ) . Injection needles were pulled from borosilicate glass capillaries ( 0 . 5 mm inner diameter , Sutter ) on a Sutter P-97 micropipette puller . The Duox morpholino ( AGTGAATTAGAGAAATGCACCTTTT ) was purchased from GeneTools and injected at 0 . 4 mM with 0 . 2 mM of the tp53 morpholino ( GCGCCATTGCTTTGCAAGAATTG ) . To generate the ikbkg mutant , TALEN vectors targeting the sequence ATGGAGGGCTGG in second exon were designed and constructed by ToolGen ( http://toolgen . com ) . TALEN vectors were linearised with PvuII ( NEB ) and purified using a PCR purification kit ( Qiagen ) , and then used for in vitro transcription with the MEGAshortscript T7 kit ( Ambion ) . About 170–300 pg of supplied ZFN RNAs or purified TALEN RNAs were then injected into one-cell stage WT zebrafish embryos , which were raised to 24 hr , then genomic DNA extracted . For detection of fish with edited loci , PCR was performed on genomic DNA of injected fish with primers flanking the target site , cloned by TA cloning into pGEMT-Easy ( Promega ) or pCR2 . 1-TOPO-TA ( Invitrogen ) and individual clones sequenced to establish efficiency . Other embryos were raised to adulthood and their offspring were similarly genotyped to identify founder mutants . All compounds for treating embryos were dissolved in DMSO , diluted in 0 . 5× E2 Embryo Medium and embryos treated by immersion . The compounds , and concentrations used , with catalogue numbers were diphenyleneiodonium chloride ( DPI ) , 40 µM ( D2926 , Sigma ) ; thapsigargin , 6 . 25 µM ( T9033 , Sigma ) ; bisindolylmaleimide I ( GF109203X ) , 85 µM ( S7208 , Selleckchem ) ; YM-254890 , 32 µM ( 10-1590-0100 , Focus Biomolecules ) ; 2-aminoethyl diphenylborinate ( 2-APB ) , 2 . 5 µM ( D9754 , Sigma ) , BI-D1870 , 1 . 2 µM ( Axon-1528 , Axon Medchem ) ; dimethyl fumarate , 9 µM ( 242926 , Sigma ) ; phorbol 12-myristate 13-acetate ( PMA ) , 37 . 5 or 125 ng/ml ( P8139 , Sigma ) ; U0126 , 100 µM ( 9903 , Cell Signaling Technology ) ; PD184352 ( CI-1040 ) , 1 . 3 µM ( S1020 , Selleckchem ) . Unless otherwise stated , controls for all experiments were exposed to 0 . 5% DMSO carrier in 0 . 5× E2 Embryo Medium . Batches of 100 WT , ddft419 , and ddfti251 embryos were collected at 24 hr and 48 hr , dechorionated , deyolked , and protein extracted as per Alli Shaik et al . , 2014 . Protein was precipitated in 100% methanol at 4°C , then resuspended in 2-D cell lysis buffer ( 30 mM Tris-HCl , pH 8 . 8 , containing 7 M urea , 2 M thiourea , and 4% CHAPS ) . 2-D DIGE and mass spectrometry protein identification was performed by Applied Biomics ( Hayward , CA ) . Protein samples were labelled with either Cy2 , Cy3 , or Cy5 , mixed , and then subjected to 2-D DIGE to separate individual proteins . Gels were scanned using Typhoon TRIO ( Amersham BioSciences ) and analysed by Image QuantTL and DeCyder ( ver . 6 . 5 ) software ( GE-Healthcare ) . Spots with more than 1 . 5-fold change were picked , in-gel trypsin digested , and protein identification performed by MALDI-TOF mass spectrometry and MASCOT search engine in the GPS Explorer software ( Matrix Science ) . A probe corresponding to the final 1078 bp of rps6ka3a ( RSK2a; NM_212786 . 1 ) was generated by cloning a PCR-derived cDNA fragment into in pGEMT-Easy ( Promega ) , linearising with ApaI ( NEB ) and transcribing a DIG probe with SP6 RNA polymerase ( Roche ) . Whole-mount in situ hybridisation developed with NBT/BCIP ( Roche ) was performed as described ( Thisse and Thisse , 2008 ) . For antibody staining , embryos were fixed in 4% paraformaldehyde overnight at 4°C and then washed in PBT ( 0 . 1% Triton in PBS ) , permeabilised in −20°C acetone for 7 min , washed in PBT , blocked for 3 hr in Block solution ( PBT supplemented with 4% BSA and 1% DMSO ) , then incubated overnight at 4°C with primary antibody diluted in Block solution , washed extensively in PBT , re-blocked in Block solution , then incubated overnight at 4°C with fluorescent secondary antibody diluted in Block solution . Following extensive PBT washing , embryos were cleared in 80% glycerol/PBS before imaging . Primary antibodies used and their dilutions are as follows: Chicken anti-eGFP antibody , 1:500 ( ab13970 , Abcam ) , Rabbit anti-eGFP , 1:500 ( Tp401 , Torrey Pines Biolabs ) , Rabbit anti-FITC , 1:200 ( #71-1900 , Thermo Fisher ) , Rabbit anti-beta catenin , 1:200 ( ab6302 , Abcam ) , Mouse anti-E-cadherin , 1:200 ( #610181 , BD Biosciences ) , Mouse anti-Tp63 , 1:200 ( CM163 , Biocare Medical ) , Rabbit anti-phospho-p44/42 MAPK ( Erk1/2 ) ( Thr202/Tyr204 ) , 1:100 ( #4370 , Cell Signaling Technology ) , Rabbit anti-p44/42 MAPK ( Erk1/2 ) , 1:100 ( #9102 , Cell Signaling Technology ) , and Rabbit anti-p90RSK ( Phospho-Thr348 ) , 1:100 ( A00487 , GenScript ) . All secondary antibodies were purchased from Invitrogen and used at 1:700 and were Alexa Fluor-488 Donkey anti-rabbit ( A21206 ) , Alexa Fluor-647 Donkey anti-rabbit ( A31573 ) , Alexa Fluor-546 Donkey anti-mouse ( A10036 ) , and Alexa Fluor-488 Goat anti-chicken ( A-11039 ) . Nuclei were counterstained using 5 µg/ml of DAPI ( 4' , 6-diamidino-2-phenylindole , dihydrochloride; D1306 , Invitrogen ) added during secondary antibody incubation . To stain hydrogen peroxide , embryos were incubated for 60 min at room temperature with 12 . 5 µM PFBSF ( #10005983 , Cayman Chemicals ) , then rinsed in Embryo Medium , anaesthetised , and imaged . Fluorescent TUNEL staining was performed using the Fluorescein In Situ Cell Death Detection Kit ( 11684795910 , Roche ) , with the fluorescein detected by antibody staining using rabbit anti-FITC , and co-immunostained for TP63 and eGFP . Epidermal permeability assays were conducted by immersing 36hpf embryos in 2 . 5 mg/ml fluorescein isothiocyanate-dextran 3–5 kDa ( Sigma ) or 0 . 075% methylene blue for 30 min and then destained in E2 medium . Still and timelapse imaging was performed on upright Zeiss AxioImager M2 , Zeiss Light-sheet Z . 1 , upright Zeiss LSM800 Confocal Microscope or Zeiss AxioZoom V16 microscopes . Embryos were mounted in 1 . 2% Low Melting Point Agarose ( Mo Bio Laboratories ) in 0 . 5× E2 medium in 35 mm glass-bottom imaging dishes ( MatTek ) or in a 1 mm inner diameter capillary for light-sheet timelapse . When imaging was performed on live embryos , the embryo media were supplemented with buffered 0 . 02% Tricaine and imaging conducted at 25°C . Image processing was done using Zen 3 . 1 software ( Zeiss ) , Fiji ( ImageJ , ver . 1 . 52p ) , or Imaris ( Bitplane ) and compiled using Photoshop 2020 ( Adobe ) . Neutrophils were tracked with TrackMate in Fiji or using the Spot function in Imaris . Kymographs were generated using the Reslice function in Fiji following generation of a line of interest across image . Fluorescence intensities were calculated using the Average Intensity function in Fiji following generation of a Region of Interest and masking of the DAPI channel to exclude the nucleus when required . In statistical analyses , n = number of embryos or cells measured , and as defined in the figure legend . GraphPad Prism was used for statistical analyses and graph generation . In all statistical tests , p<0 . 05 , p<0 . 01 , **p<0 . 001 . Tests used are indicated in the associated figure legend and were Student’s t-test , Chi-squared test , Mann–Whitney test , or ANOVA with Bonferroni or Dunn’s post-tests .	Cancer occurs when normal processes in the cell become corrupted or unregulated . Many proteins can contribute , including one enzyme called Matriptase that cuts other proteins at specific sites . Matriptase activity is tightly controlled by a protein called Hai1 . In mice and zebrafish , when Hai1 cannot adequately control Matriptase activity , invasive cancers with severe inflammation develop . However , it is unclear how unregulated Matriptase leads to both inflammation and cancer invasion . One outcome of Matriptase activity is removal of proteins called Cadherins from the cell surface . These proteins have a role in cell adhesion: they act like glue to stick cells together . Without them , cells can dissociate from a tissue and move away , a critical step in cancer cells invading other organs . However , it is unknown exactly how Matriptase triggers the removal of Cadherins from the cell surface to promote invasion . Previous work has shown that Matriptase switches on a receptor called Proteinase-activated receptor 2 , or Par2 for short , which is known to activate many enzymes , including one called phospholipase C . When activated , this enzyme releases two signals into the cell: a sugar called inositol triphosphate , IP3; and a lipid or fat called diacylglycerol , DAG . It is possible that these two signals have a role to play in how Matriptase removes Cadherins from the cell surface . To find out , Ma et al . mapped the effects of Matriptase in zebrafish lacking the Hai1 protein . This revealed that Matriptase increases IP3 and DAG levels , which initiate both inflammation and invasion . IP3 promotes inflammation by switching on pro-inflammatory signals inside the cell such as the chemical hydrogen peroxide . At the same time , DAG promotes cell invasion by activating a well-known cancer signalling pathway called MAPK . This pathway activates a protein called RSK . Ma et al . show that this protein is required to remove Cadherins from the surface of cells , thus connecting Matriptase’s activation of phospholipase C with its role in disrupting cell adhesion . An increase in the ratio of Matriptase to HAI-1 ( the human equivalent of Hai1 ) is present in many cancers . For this reason , the signal cascades described by Ma et al . may be of interest in developing treatments for these cancers . Understanding how these signals work together could lead to more direct targeted anti-cancer approaches in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "genetics", "and", "genomics" ]	2021	Matriptase activation of Gq drives epithelial disruption and inflammation via RSK and DUOX
During organogenesis , inductive signals cause cell differentiation and morphogenesis . However , how these phenomena are coordinated to form functional organs is poorly understood . Here , we show that cell differentiation of the Drosophila trachea is sequentially determined in two steps and that the second step is synchronous with the invagination of the epithelial sheet . The master gene trachealess is dispensable for the initiation of invagination , while it is essential for maintaining the invaginated structure , suggesting that tracheal morphogenesis and differentiation are separately induced . trachealess expression starts in bipotential tracheal/epidermal placode cells . After invagination , its expression is maintained in the invaginated cells but is extinguished in the remaining sheet cells . A trachealess cis-regulatory module that shows both tracheal enhancer activity and silencer activity in the surface epidermal sheet was identified . We propose that the coupling of trachealess expression with the invaginated structure ensures that only invaginated cells canalize robustly into the tracheal fate . A fundamental question in biology is how cells coordinately shape functional organs with complex architecture during embryogenesis . Extensive studies have uncovered how inductive signals , such as morphogens , prime cell differentiation and morphogenesis ( Heisenberg and Bellaïche , 2013; Perrimon et al . , 2012 ) , leading to segregated organs with uniquely specified cells . Due to the graded nature of the inductive signals , the initial territories of an organ primordial placode are occupied by cells with various degrees of commitment . Furthermore , cells modulate their own physical properties by changing gene expression to drive morphogenesis , but each cell behavior is dynamic and fluctuating . Therefore , mechanisms to coordinate these phenomena are of critical importance . Without a coordination mechanism , tissues would be mixed with improperly specified cells that would interfere with organ functions . The sequence of signaling , gene expression and morphogenesis is not unidirectional , and the feedback input from morphogenesis to gene expression is proposed to be crucial ( Chan et al . , 2017; Gilmour et al . , 2017 ) . However , the generality of the proposed feedback mechanisms from morphogenesis to gene expression and cell differentiation in a wide range of developmental systems remains to be determined . Epithelial invagination is an important morphogenetic process in which three-dimensional tubular organs are formed from a two-dimensional flat sheet ( Andrew and Ewald , 2010; Kondo and Hayashi , 2015; Sawyer et al . , 2010 ) , and the Drosophila trachea is a useful model system for analyzing three-dimensional epithelial morphogenesis ( Hayashi and Kondo , 2018; Loganathan et al . , 2016 ) . Tracheal morphogenesis is initiated by placode specification; ten pairs of tracheal placodes form in the dorsal anterior part of the epidermis in each segment by stage 10 , followed by invagination , branching and fusion ( Figure 1A ) . In this process , the tracheal placodes first appear as a group of cells expressing trachealess ( trh ) , which is considered to be a master regulator of tracheal morphogenesis ( Chung et al . , 2011; Isaac and Andrew , 1996; Wilk et al . , 1996 ) , and then EGF signaling and mitosis synergistically drive invagination by generating centripetal pressure and inducing epithelial sheet buckling , respectively ( Kondo and Hayashi , 2013; Nishimura et al . , 2007; Ogura et al . , 2018 ) . Finally , FGF signaling triggers tracheal branching ( Figure 1A ) ( Glazer and Shilo , 1991; Sutherland et al . , 1996 ) . trh encodes a bHLH-PAS transcription factor that is critical for tracheal morphogenesis . Its expression is primarily induced under the combinatorial control of activation through JAK-STAT signaling and inhibition through Wg and Dpp signaling before invagination ( Brown et al . , 2001; Wilk et al . , 1996 ) , and STAT-responsive enhancers for trh have been identified ( Sotillos et al . , 2010 ) . After invagination , all of the tracheal cells continue expressing trh , while no other surrounding epithelial cells , such as epidermal cells , express this factor . However , it is not well understood how trh expression is strictly restricted only to invaginated tracheal cells . Although trh is proposed to maintain its own expression through an auto-regulatory mechanism ( Wilk et al . , 1996; Zelzer and Shilo , 2000 ) , it is still unclear whether all the cells that start expressing Trh expression take part in the invagination and generation of trachea , or if some of these cells fail to invaginate , and if so , how they shut off the auto-regulatory control of trh . In this article , we first show that trh plays a critical role in maintaining the invaginated structure but not in initiating invagination . Second , we reveal that the tracheal placode cells initiating trh expression later become either tracheal or epidermal cells , and the maintenance of trh expression is tightly associated with the change from the epithelial sheet to the invaginated structure . On the basis of our findings , we propose that the transcriptional coordination of trh expression , tracheal cell fate specification and invaginated structures during epithelial invagination ensures that only the invaginated cells are canalized robustly into the tracheal fate . We previously reported that mitosis can drive tracheal invagination , alone or in combination with EGFR signaling ( Kondo and Hayashi , 2013 ) . Although all embryonic cells undergo multiple cell divisions , mitosis-induced invagination occurs only in the tracheal placode . Non-tracheal epidermal cells quickly recover their flat epithelial architecture after mitosis , suggesting that the tracheal placode cells possess a special ability to couple mitosis with invagination and tubule formation ( Kondo and Hayashi , 2013 ) . Since trh is considered a master regulator of tracheal morphogenesis ( Isaac and Andrew , 1996; Wilk et al . , 1996 ) , we reasoned that trh is involved in this mitosis-induced invagination . Previous studies showed that in trh mutants , the tracheal tissue is completely missing in late-stage embryos , and no invagination occurs ( Isaac and Andrew , 1996; Wilk et al . , 1996; Younossi-Hartenstein and Hartenstein , 1993 ) . However , in the stage-10 tracheal placode , di-phosphorylated ERK , a hallmark of EGFR activation , was detected even in trh mutants ( Figure 1—figure supplement 1 ) ( Ogura et al . , 2018 ) , suggesting that some early tracheal development processes were taking place . Live imaging of trh1 ( an EMS-induced missense allele ) mutants at single-cell resolution revealed an unexpected finding: apical constriction forming a tracheal pit appeared in the center of the placode region , followed by mitosis in the pit cells and rapid , deep invagination as seen in the control , although the onset of invagination was delayed ( Figure 1B , C ) . Over the next 90 min , the invaginated structure gradually returned to the surface epidermal layer and merged with these cells to form a segmental furrow , leaving no trace of the tracheal structure ( Figure 1C ) . Consistent with this live imaging analysis of trh1 , in fixed samples with a heteroallelic combination of TALEN-induced trh null alleles ( trhA14-3/trhB16-11 , Figure 1—figure supplement 2 ) ( Kondo et al . , 2014 ) , would-be tracheal placode cells labeled with the R14E10 trh primary enhancer ( containing the trh66 STAT-responsive element that mediates stage-10 trh expression in the placode ( Sotillos et al . , 2010 ) , Figure 1D ) also formed a tracheal pit ( Figure 1E ) and formed invaginated structures during stages 11–13 ( Figure 1F ) . These R14E10-positive cells returned to the surface epidermal layer at stage 15 ( Figure 1G ) . The appearance of invaginated structures at stage 12 and the disappearance of these structures at stages 15–16 were observed , as shown in Figure 1F and G , respectively , with 100% penetrance ( stage 12: eight embryos , stage 15–16: ten embryos ) . In addition , the overexpression of trh ( trh-OE ) by R14E10-GAL4 in the trh mutants rescued the phenotype ( Figure 1—figure supplement 3A ) . One of the Trh target genes is breathless ( btl ) , which encodes an FGF receptor ( Ohshiro and Saigo , 1997 ) , and we reported that FGF signaling through Btl is able to trigger invagination independent of EGF signaling and mitotic rounding ( Kondo and Hayashi , 2013 ) . However , the btl-OE in the trh mutants using R14E10-GAL4 did not rescue tracheal formation from the transiently invaginated tracheal placodes ( Figure 1—figure supplement 3B ) , indicating that other trh target genes are required to support FGF signaling-triggered tracheal morphogenesis . In addition , inhibiting apoptosis by p35 did not prevent the transiently invaginated cells from returning to the epidermis , indicating that apoptotic cell removal is not the major cause of this anomaly ( Figure 1—figure supplement 3C ) . These results demonstrated that trh is essential for maintaining the invaginated structure , whereas it is dispensable for initiating invagination . Thus , tracheal formation proceeds by two successive and genetically separable steps: ( 1 ) invagination triggered by the mechanical forces generated through the combined activities of mitosis , EGFR , and FGFR signaling; and ( 2 ) maintenance of the invaginated structure controlled by trh . Since mitosis-triggered invagination was maintained in rho bnl mutants ( Kondo and Hayashi , 2013 ) , EGFR and FGFR signaling are dispensable for the maintenance of the invaginated structure . For Trh to function as a determinant of the invaginated structures , its expression must be tightly sustained only in the invaginated cells but not in the surface epidermal cells . To reveal the relationship between trh expression and epithelial geometry , we attempted to analyze the impact of reducing the number of invaginated tracheal cells on Trh expression . In rho bnl double mutants that lose both EGF and FGF signaling in tracheal cells , tracheal invagination is impaired , and the trachea is composed of a smaller number of cells than that of the control . If a similar number of cells initiates trh expression in control and rho bnl mutants , some trh-positive placode cells are expected to remain in the surface epidermis , and these cells may face a conflict between their fate and tissue geometry . JAK-STAT signaling induces trh expression through a STAT-responsive trh enhancer in the tracheal placodes at stage 10 of embryogenesis before invagination ( Figure 2A ) ( Brown et al . , 2001; Sotillos et al . , 2010 ) . After invagination , Trh is detected only in all invaginated cells , including the most proximal spiracular branch ( Figure 2C , F ) . The number of initial Trh+ cells in the tracheal placode before invagination ( stage 10 ) was 58 . 2 ± 5 . 1 ( mean ±S . D . ) in controls and 67 . 2 ± 8 . 5 in the rho bnl mutants , respectively , indicating that tracheal fate specification was not compromised in the rho bnl mutants ( Figure 2A , B , G ) . The increase in the number of initial Trh+ cells in the rho bnl placodes reflects the expansion of the tracheal placode due to the earlier role of rho in restricting the size of tracheal placode ( Raz and Shilo , 1993 ) . The R14E10 fragment contains the trh66 STAT-responsive element that mediates stage9-10 trh expression ( Sotillos et al . , 2010 ) , and the number of initial R14E10 + cells ( 57 . 9 ± 2 . 7 cells , monitored by using the R14E10-lacZ transgene at stage 10 ) is almost the same as the number of initial Trh+ cells . In contrast , after cycle-16 mitosis and invagination , the resultant tracheae of the rho bnl mutants were composed of a smaller number of Trh+ cells ( trh-on cells , 31 . 2 ± 6 . 2 cells ) than those of the controls ( 87 . 6 ± 6 . 3 cells ) ( Figure 2C , D , G ) . These findings demonstrate that in the rho bnl mutants , the number of initial Trh+ cells at stage 10 was reduced at stages 13–14 . This reduction is due to either the disappearance of Trh+ cells from the epithelium or the loss of Trh expression in cells that failed to invaginate . To discriminate these possibilities , we traced the fate of cells initiating trh expression by labeling them with nls-lacZ driven by R14E10 . The R14E10-GAL4-induced nls-lacZ product ( β-galactosidase , β-gal ) persisted in the initial trh+ cells after termination of R14E10-GAL4 transcription , allowing us to distinguish trh-off ( Trh– , LacZ+ ) and trh-on ( Trh+ ) cells derived from the initial trh-on cell population after invagination ( from stage 13 onward ) . Even in control embryos , there were trh-off cells ( 31 . 9 ± 6 . 7 cells ) in the epidermis ( Figure 2C , G ) , while all of the trh-on cells were found in the invaginated tubule region ( 87 . 6 ± 6 . 3 cells ) . The sum of the trh-on and trh-off cells ( 119 . 5 ± 10 . 5 cells ) agreed well with the prediction from the number of initial Trh+ cells ( 58 . 2 ± 5 . 1 cells ) and the number of initial R14E10 + cells ( 57 . 9 ± 2 . 7 cells , Figure 3—figure supplement 2B , C ) after one round of cycle-16 mitosis during invagination . The results showed that 27% of the initial trh+ cells lost their Trh expression , all of which remained in the epidermis . Many of the trh-off cells remained during the rest of embryogenesis and formed trichomes on their apical surface ( data not shown ) , suggesting that they adopted the epidermal fate . We next asked if the loss of trh expression in trh-off cells was due to their failure to become part of the tube by tracing the fate of trh-expressing cells in the rho bnl mutants . After invagination , the resultant tracheae were composed of a smaller number of trh-on cells ( 31 . 2 ± 6 . 2 cells ) , as mentioned above , and surrounded by an increased number of epidermal trh-off cells ( 50 . 4 ± 7 . 2 cells ) than those of the controls , with a total of 81 . 6 ± 9 . 8 cells ( Figure 2D , G ) . Blocking apoptosis by p35 in the rho bnl mutants increased the number of both trh-on and trh-off cells ( Figure 2E , G ) . In the rho bnl mutants with or without p35 , 62% of the surviving initial trh+ cells remained in the epidermis and lost their trh expression . These findings indicated that the reduction in trh-on cells in the rho bnl mutants was not simply due to their disappearance from the epithelium . Losses of EGF , FGF , and mitosis in the rho CyclinA ( CycA ) bnl triple mutant caused a more severe invagination defect ( Kondo and Hayashi , 2013 ) . The initial trh expression at stage 10 was nearly normal even in the triple mutant ( 53 . 9 ± 7 . 1 cells ( Figure 2—figure supplement 1A , B ) . If the cells maintaining Trh expression and forming tubes are predetermined before invagination , the triple mutants are supposed to possess half the number of trh-on cells observed in the rho bnl double mutants because the CycA mutation eliminates cycle-16 mitosis . However , although all the invaginated cells were Trh-positive , the number of Trh-on tracheal cells in these triple mutants was much smaller than expected ( 6 . 0 ± 2 . 5 cells , Figure 2—figure supplement 1A , B ) , which strongly argues against the model in which the tube-forming trh-on cells are predetermined before invagination . These observations support the possibility that the trh expression in the stage 10 tracheal placodes is maintained only in the successfully invaginated tubule cells , independent of the depth of invagination . The placode cells that failed to invaginate and remained in the epidermis lost their trh expression . These results imply that a mechanism exists to maintain trh expression only in the invaginated tubule cells and extinguish it in the superficial epidermal cells ( Figure 2H ) . A transcriptional reporter of trh [1-eve-1 , a lacZ enhancer trap of trh ( Perrimon et al . , 1991; Wilk et al . , 1996 ) elicited reporter β-gal expression that was limited to the invaginated tracheal cells ( Figure 3B ) , suggesting that the epidermal expression of trh is repressed at the level of transcription . We then tested the properties of several previously identified trh enhancers ( Sotillos et al . , 2010 ) . Among the eight trh upstream regions with phylogenetically conserved STAT-binding sites , trh47 and trh66 drive reporter expression from the early stage of tracheal development , and trh67 is proposed to be a trh-dependent auto-regulatory element ( Sotillos et al . , 2010 ) . We found that the two primary enhancers , trh47 ( Figure 3A , Figure 3—figure supplement 1A ) and trh66 ( covered by R14E10 , Figure 3A , Figures 1D and 2C ) , were active in both tubule cells and the surrounding epidermal cells , suggesting that they did not reproduce the epidermal extinction of trh . In contrast , trh67 did not drive reporter expression in all of the Trh-positive invaginated cells , suggesting that additional cis-elements control the tube-specific maintenance of trh expression ( Figure 3—figure supplement 1B ) . We then searched for additional trh enhancers from a systematic enhancer mapping resource ( the FlyLight project: https://www . janelia . org/project-team/flylight ) ( Jenett et al . , 2012; Jory et al . , 2012; Manning et al . , 2012 ) and identified another enhancer immediately upstream of the proximal trh promoter ( R15F01 , Figure 3A ) . R15F01 showed strict tube-specific activity after invagination , and its activity was maintained in the invaginated tubule cells throughout embryogenesis ( Figure 3B’ , G ) and in postembryonic stages ( not shown ) , in contrast to trh47 and trh66 ( R14E10 ) , which show transient activity only from embryonic stage 10 to 11 ( Sotillos et al . , 2010 ) . We note that a few epidermal cells sporadically showed leaked R15F01 activity , especially when we monitored the activity using R15F01-GAL4 reporter transgenes ( Figure 3B’ ) . In addition , when we used a direct lacZ reporter ( R15F01-lacZ ) , R15F01’s activity became detectable in part of the tracheal placode before invagination ( Figure 3F , Figure 3—figure supplement 2G ) , slightly later than that of the other early enhancers ( trh47 or R14E10 ) . In addition , we noticed that R15F01 repressed the function of an adjacent mini-yellow gene ( mini-y included in attP40 or attP2 , a transgene landing site on chromosome 2L or chromosome 3L , respectively [Groth et al . , 2004] ) in the adult epidermis ( Figure 3E and Figure 3—figure supplement 2A ) . The function of the endogenous yellow gene on chromosome X was not affected ( data not shown ) , suggesting that R15F01 represses mini-y expression in cis . These results suggested that R15F01 is a cis-regulatory module ( CRM ) that simultaneously functions as an enhancer in tracheal tube cells and a silencer in epidermal sheet cells . Using the mutant combinations that prevented invagination to various degrees ( combinations of rho , bnl and CycA ) , no or only a few epidermal cells showed R15F01 activity , while the small tracheal tubes were R15F01-positive irrespective of the depth of invagination ( Figure 3C , D , H–K ) . In the rho bnl mutants , R15F01 activation was detected at stage 10 before invagination . The number of initial R15F01 + cells in the rho bnl mutants before cycle-16 mitosis and invagination ( 39 . 5 ± 4 . 7 cells , Figure 3—figure supplement 2D , G ) is smaller than that of control embryos ( 47 . 3 ± 6 . 1 , Figure 3—figure supplement 2G ) , indicating that EGF signaling is involved in the initial activation but is not essential . In addition , the number is larger than the number of invaginated cells with Trh expression after cycle-16 mitosis ( Trh-on cells ) in the rho bnl mutants ( two times 37 . 7 ± 6 . 3 cells VS trh-on cells 31 . 2 ± 6 . 2 cells ( Figure 2G ) ; note that the number of cells is doubled after cycle-16 mitosis ) , suggesting that R15F01 activity is controlled by a multistep mechanism . Even in unpaired 1 ( upd1 ) , upd2 , and upd3 triple mutants ( Df ( 1 ) BSC352 ) , which are deficient for all Upd cytokines and are unable to activate JAK-STAT signaling , invaginated trachea appeared in some segments , and only these invaginated cells became R15F01-active ( Figure 3L ) . These results indicate that the tubule-specific maintenance of R15F01 activity after invagination is independent of JAK-STAT , EGFR , and FGFR signaling . These results also support the idea that the tube-forming trh-on cells are not predetermined before invagination and that only the successfully invaginated cells secondarily sustain the R15F01 activity for Trh expression independent of the initial induction of its activation . To analyze the contribution of trh to tube-specific R15F01 maintenance , we monitored R15F01 activity in trh mutants . R15F01 activation was detected in the trh mutants before invagination , but the number of β-gal-positive cells ( 29 . 4 ± 6 . 7 ) was smaller than that of the control ( 47 . 3 ± 6 . 1 ) ( Figure 3—figure supplement 2E , G ) , indicating that trh is involved in the initial activation of R15F01 but is not essential . After mitosis cycle 16 , transient invagination and disappearance of invaginated architecture in the trh mutants , 19 . 1 ± 6 . 8 cells still maintained R15F01 reporter expression at stage 15 ( Figure 3—figure supplement 2F , H ) . Because this number is smaller than the number of initially activated cells , R15F01 activity could be maintained in a trh-independent manner . As mentioned above , in addition , trh-OE in the trh mutants by R14E10-GAL4 ( early transient activation of trh in placodes , and no secondary regulation of trh expression ) could rescue tracheal morphogenesis , and more cells took on a tubular architecture than in the trh mutants ( Figure 1—figure supplement 3A ) . We found that all these invaginated cells showed R15F01 activity ( Figure 3M , β-gal ) , while some trh-OE cells did not take part in invagination , and these surface-remaining cells did not show R15F01 reporter expression even when we overexpressed a phosphomimetic active form of Trh ( Figure 3M , RFP + epidermal cells ) . These results strongly suggest that Trh is not sufficient to maintain longer R15F01 activity and that secondary regulatory mechanisms that are potentially associated with invagination are required . trh-OE by R15F01-GAL4 in the trh mutants could also partially rescue tracheal morphogenesis ( Figure 3—figure supplement 3 ) . Because of the later onset of R15F01 and/or the smaller number of R15F01-active cells than in R14E10 ( Figure 3—figure supplement 2 ) , the invagination defect of the trh mutants was not rescued at stage 12 . On the other hand , in later stage 15 , R15F01-positive cells were able to maintain invaginated structures , although they showed an incomplete branching pattern , possibly due to a limited number of R15F01-positive cells . To identify functional elements within the 3963 bp fragment of R15F01 , we divided R15F01 into eight fragments ( D1-D8 ) , constructed four deletions ( del1-del4 ) and assayed their regulatory activity before invagination ( stage-10 tracheal placode ) , their activity in the trachea and epidermis after invagination ( from stage 13 onward ) , and their cis-inhibition effect on mini-yellow in adult flies ( Figure 4A ) . The results showed that the enhancer activity for tracheal expression was mapped to two sub-fragments: D7 , which was sufficient to drive expression in the tracheal placode and invaginated tracheal tubules ( Figure 4A , D , H , I ) , and D1 , which drove tracheal expression after invagination in a somewhat sporadic manner ( Figure 4A , C , F , G ) . No other sub-fragment showed tracheal enhancer activity ( Figure 4—figure supplement 1B–G ) . In addition , fragment D7 also drove expression in the epidermal region near the tracheal pit ( Figure 4D , I ) , indicating that D7 drives both tracheal and epidermal expression and is not sufficient to reproduce the tube-restricted pattern after invagination . Second , the mini-y silencer activity was mapped to sub-fragment D1 , and its removal from R15F01 ( i . e . , the del4 construct ) abrogated this mini-yellow silencing ( Figure 4A , Figure 4—figure supplement 1A ) . The epidermal enhancer activity of D7 in embryos was not detected in full-length R15F01 , suggesting that an epidermal silencer that represses D7’s epidermal activity must reside in another part of R15F01 . D1 is likely to contain this activity , since it possesses the silencer activity for mini-yellow , and the combination of D1 with D7 in constructs del2 and del3 reproduced the expression pattern of R15F01 ( Figure 4A , Figure 4—figure supplement 1I , J ) . In addition , the del4 construct , which did not contain the D1 or D2 fragment , also showed epidermal suppression of D7 activity ( Figure 4A , E , Figure 4—figure supplement 1K ) . These data indicated that silencer elements reside in at least two regions , one in the D1-D2 fragment with mini-y silencer activity and the other in the region included in del4 ( in the D3 to D6 and/or D8 fragment ) . To examine whether these silencers could act on a heterologous epidermal enhancer , we constructed chimeric reporters consisting of the shavenbaby ( svb ) epidermal enhancer and sub-fragments of R15F01 ( Figure 5A ) . The svb E6B element enhances reporter gene expression in the dorsal epidermal cells that form trichomes ( Figure 5B , D ) ( Frankel et al . , 2011 ) . We found that the fragments containing D1 and D2 ( D1-D2 ) , from D3 to D6 ( D3-D6 ) , and from D1 to D6 ( in the del1 construct ) were able to silence the svb E6B activity ( Figure 5C , E , F ) . Since the D1-D2 and D3-D6 inserts have no overlap , these data confirmed that R15F01 contains multiple and redundant silencer elements that act dominantly over the svb E6B epidermal enhancers . To further characterize the difference in the silencing activity of R15F01 between epidermal and internal epithelial tissues , we constructed another chimeric reporter in which the UAS element was fused to R15F01 . When we crossed the 3×UAS-GFP reporter with arm-GAL4 , which is an epithelial ubiquitous GAL4 driver ( Sanson et al . , 1996 ) , GFP expression was detected ubiquitously , with an intense signal in epidermal and hindgut cells ( Figure 5G ) . Then , we crossed the 3×UAS-R15F01-fused GFP reporter with arm-GAL4 , and epidermal GFP expression was nearly undetectable , while hindgut GFP expression was still detectable ( Figure 5H ) . The tracheal GFP signal was enhanced , possibly due to the tracheal enhancers in R15F01 . Embryos harboring only the 3×UAS-R15F01 reporter without arm-GAL4 showed the tracheal GFP signal but not hindgut GFP expression ( Figure 5I ) , indicating that the hindgut activity was driven by arm-GAL4 . These results are also consistent with the notion that R15F01 silences enhancer activities in the surface epidermis but not in internal tubular organs . We found that in trh mutants , a subset of the would-be tracheal placode cells undergo invagination but fail to maintain the invaginated structure . This observation contradicts previous reports claiming that trh is required for invagination ( Isaac and Andrew , 1996; Wilk et al . , 1996 ) . We consider that our results based on live imaging of early tracheal invagination processes identified crucial tracheal cell behavior that was missed in previous works that mainly focused on late embryonic phenotypes . The phenotype of trh mutants indicates that the conversion of the epithelial sheet of the tracheal placode into a tube through invagination and the stabilization of the invaginated structures are genetically separable steps . In addition , inductive signals , such as JAK/STAT signaling , are considered to prime both tracheal differentiation ( i . e . , trh expression ) and invagination separably . This finding is also consistent with the idea that morphogenetic movement and cell differentiation can be uncoupled ( Ip et al . , 1994 ) . We suggest that epithelial tissue can assume two alternative stable structures , sheet or tube , and according to cell fate , each epithelial tissue assumes one of these structures . In the tracheal system , invagination forces include the contraction of myosin cables regulated by EGF signaling and a cell migratory force stimulated by FGF signaling ( Kondo and Hayashi , 2013; Nishimura et al . , 2007; Ogura et al . , 2018 ) . If both signaling pathways are absent , transient tissue instability caused by clustered mitosis allows invagination ( Kondo and Hayashi , 2013 ) . This mitotic cue is sufficient for the conversion of the Trh + placodes from the sheet state to the more stable tube state . Once invaginated by any driving forces , the tracheal cells robustly maintain the invaginated structure under the control of Trh . In trh mutants , although the placode cells are able to initiate invagination , the degree of invagination is much smaller than that of the control . This is consistent with our recent report that trh controls the propagation of EGFR activation but not the initial activation of EGFR in the placodes ( Ogura et al . , 2018 ) , indicating that trh contributes to tracheal invagination through EGFR signaling propagation in part . However , even when both EGFR and FGFR signaling are lost , Trh+ cells are able to maintain the invaginated structure . In addition , although btl , which encodes an FGFR , is one of the important downstream genes of trh and FGFR signaling can trigger invagination when EGFR signaling and mitosis 16 are eliminated in the placodes , the btl-OE in the trh mutants was not sufficient to rescue invagination and tubule maintenance . These results indicate that the maintenance of the invaginated structure is largely dependent on trh in an EGFR and FGFR signaling-independent manner . While canonical tissue-folding processes are driven by apically concentrated myosin through active apical constriction ( Martin and Goldstein , 2014 ) , tracheal invagination has unique properties , including passive apical constriction under centripetal pressure from neighboring cells , the acceleration of the invagination through mitotic cell rounding ( Kondo and Hayashi , 2013 ) , and a lack of apically concentrated myosin localization in the invaginating cells ( Kondo and Hayashi , 2013 ) . This is consistent with a recent report that myosin regulatory light chain depletion using the deGradFP system does not cause significant abnormalities in tracheal morphogenesis ( Ochoa-Espinosa et al . , 2017 ) . Candidates for Trh-downstream effectors that maintain invaginated structures are Crossveinless-c ( cv-c ) , a Rho family GTPase-activating protein ( RhoGAP ) , and Crumbs ( Crb ) , both of which are expressed in the tracheal cells under the control of trh ( Brodu and Casanova , 2006; Letizia et al . , 2011; Röper , 2012 ) . However , both cv-c mutants and crb mutants are able to maintain their tubular tracheal geometry ( Cela and Llimargas , 2006; Letizia et al . , 2011 ) , suggesting that trh controls the epithelial tissue geometry through the activation of multiple genes . It has recently been reported that in the salivary gland , the overexpression of a constitutive active form of Arp2/3 activator causes reversal of invaginated structures into epidermis ( Chung et al . , 2017 ) , suggesting that the difference in F-actin organization is also important to stabilize epithelial structure ( sheet or tube ) . In addition , trh is also known to re-organize microtubule structures ( Brodu et al . , 2010 ) . During dorsal fold formation in the gastrulating fly embryo , the invaginating cells do not show apical myosin enrichment , whereas the apical microtubule network plays an important role in cell shortening through a polarity-dependent basal shift of AJs ( Takeda et al . , 2018; Wang et al . , 2012 ) . Therefore , re-organization of both F-actin and microtubule architectures , but not apically concentrated myosin , might synergistically support the maintenance of invaginated structures . Further investigation of the trh-downstream transcriptome profile and tubule stabilization mechanisms will be important for understanding the diversity of cellular mechanisms of epithelial morphogenesis . We found that trh expression is strictly maintained only in invaginated cells , whereas it is extinguished in cells remaining in the adjacent surface epidermis , irrespective of the depth of invagination . Because some of the initial Trh-expressing cells do not invaginate and lose their Trh expression , the initial priming of Trh expression is not sufficient for cells to take part in invagination and does not result in autocatalytic maintenance of its expression . Therefore , there should be a mechanism that determines cells that maintain Trh expression after the initial primed state . One possibility is that these cells are predetermined independent of initial priming of trh expression before invagination and then form the tubular architecture precisely . The other possibility is that maintenance of Trh expression is tightly associated with invagination . Our data using various invagination mutants ( especially the comparison between rho bnl mutants and rho CycA bnl mutants , Figure 2 and Figure 2—figure supplement 1 , as mentioned in the results section above ) strongly suggest that the cells that retain trh expression and invaginated structures are not predetermined , and tissue-geometry-dependent mechanisms are involved in tight coupling of invaginated structures and Trh expression . Since Trh is essential for stabilizing invaginated structures , this coupling may ensure that only invaginated cells canalize robustly into the tracheal fate and further supports the formation of a tubular tracheal system after invagination . Our findings strongly suggest that the R15F01 CRM is crucial for invagination-restricted Trh expression and is composed of multiple enhancers and redundant epidermal silencers . We also note that because R15F01 is activated in part of the placode cells before invagination , there should be a mechanism for this transient activation , regardless of tissue architecture . After invagination , R15F01 activity is maintained only in invaginated cells independent of the degree of invagination , suggesting that R15F01 senses both initial placode activation cues and morphogenetic invagination cues for trh expression . In addition , R15F01 is also reported to be a conserved Polycomb response element ( PRE ) ( Hauenschild et al . , 2008; Schuettengruber et al . , 2014 ) . ChIP signals for several Polycomb factors , such as Polycomb , Pleiohomeotic , Polyhomeotic distal , and Dorsal switch protein 1 , are highly concentrated along the R15F01 region , especially in the D1 region , which shows strong silencer activity for mini-y , and this pattern is conserved across Drosophila species ( Schuettengruber et al . , 2014 ) ( ChIP atlas; http://chip-atlas . org and GSE60428 ) . Since these analyzes were performed using whole embryos , it is still unclear in which cells these Polycomb factors associate with the R15F01 locus . It was also recently reported that some developmental enhancers can also function as PREs ( Erceg et al . , 2017 ) . These findings suggest that R15F01 functions as a developmental enhancer in tracheal cells , while it operates as a PRE in other cells , including epidermal cells . If so , counteracting this PRE activity at the trh locus only in invaginated cells may be the critical step for coupling Trh expression and tubular architecture . The terminal differentiation of tracheal cells is likely to be a consequence of the relocation of tracheal primordial cells from the surface epidermis to the inside of the trachea and the suppression of the epidermal silencer activity of trh . An essential remaining question is how the tracheal cells couple Trh expression with invaginated structures during morphogenesis . This study showed that known signaling pathways involved in early tracheal morphogenesis , such as JAK-STAT , EGFR and FGFR signaling , are dispensable for maintaining R15F01 activity only in invaginated cells . One possible mechanism is sensing the change in epithelial geometry from sheet to tube through mechano-transduction pathways . Cells are known to sense rigidity in their environment , mechanical stress , and their own morphology and cytoskeletal architecture and to control gene expression and chromatin organization in response to these factors ( Chan et al . , 2017; Kirby and Lammerding , 2018; Kumar et al . , 2017; Panciera et al . , 2017 ) . Another possibility is that the cell can detect geometrical conversion through a change in the local concentration of secreted molecules ( Gilmour et al . , 2017 ) . Buckling and bending of the intestinal epithelium affect the local concentration of Sonic hedgehog ( Shh ) to help define the positions of stem cells in chicks ( Shyer et al . , 2015 ) . In addition , lumen formation and the luminal accumulation of FGF promote differentiation at the zebrafish lateral line ( Durdu et al . , 2014 ) . At present , we do not have evidence that any signaling pathways affected by secreted ligands or known mechano-transduction pathways , such as Hippo , Src-Arm , or Ca2+ , are involved in invagination-responsive Trh expression . In addition , it is still unclear whether cells directory sense the invaginated structure to terminally differentiate into tracheal cells or whether the relocation from the surface epidermis to the inside of the embryos just allows cells to acquire a non-epidermal fate . It remains a future challenge to discriminate among these possibilities . Because all tissues and organs must match their architecture with their cellular phenotype to function properly , similar canalization mechanisms that couple gene expression , cell fate , and tissue geometry may play fundamental roles in shaping functional organs . Although morphogenetic feedback is proposed to be important for organogenesis , the cellular mechanisms for sensing tissue geometry have only begun to be elucidated . Further study into the cellular and genetic mechanisms by which tracheal cells monitor the process of morphogenesis and adjust their cell fate would help us understand the robustness of animal morphogenesis . The fly strains used in this study are listed in the Key Resources Table . Plasmids were constructed using PrimeSTAR max or PrimeSTAR HS ( Takara Bio ) and an in-fusion PCR cloning kit ( Clontech ) unless otherwise noted . For pBPGUw-R15F01 and pBPGUw-R14E10 , each corresponding region was amplified from genomic DNA , subcloned into pENTR-TOPO , and recombined into pBPGUw [pBPGUw was a gift from Gerald Rubin ( Addgene plasmid #17575 ) ( Pfeiffer et al . , 2008 ) using LR recombinase . For pBPGUw-lacZ , the lacZ coding sequence ( CDS ) was amplified from pCaSpeR-hs-lacZ and subcloned into HindIII-digested pBPGUw . For pBPGUw-R15F01-lacZ , the R15F01 fragment was recombined into pBPGUw-lacZ from pENTR-TOPO-R15F01 using LR recombinase . For pBPGUw- ( D1 ~D8 or del1 ~del4 ) -lacZ , each truncated fragment of R15F01 was amplified from pBPGUw-R15F01 and subcloned into AatII/NaeI-digested pBPGUw-R15F01-lacZ . For pBPGUw- ( D1 , D7 , or del4 ) -GAL4 , each truncated fragment of R15F01 was amplified and subcloned into AatII/NaeI-digested pBPGUw-R15F01 . For pBPGUw-trh47 or trh67 , each fragment ( Sotillos et al . , 2010 ) was amplified from genomic DNA and subcloned into AatII/NaeI-digested pBPGUw-R14E10 . For pBPGUw-svbE6B- ( D1-2 , D3-6 or D1-6 ) -lacZ , each chimeric fragment was amplified from genomic DNA and plasmids containing a R15F01 fragment and subcloned into AatII/NaeI-digested pBPGUw-R15F01-lacZ . For pBPGUw-3×UAS-GAP43GFP and pBPGUw-3×UAS-R15F01-GAP43GFP , the GFP CDS with the GAP43 palmitoylation sequence was amplified from pUbi-GAP-CAAX ( Kondo and Hayashi , 2013 ) by PCR and subcloned into HindIII-digested pBPGUw ( pBPGUw-GAP43GFP ) . A 3×UAS fragment were generated by annealing two oligo DNAs and subcloned into AatII/NeaI-digested pBPGUw-GAP43GFP . A R15F01 fragment fused to 3×UAS was amplified by PCR and subcloned into AatII/NeaI-digested pBPGUw-GAP43GFP . The primer sequences used in this study are listed in Supplementary file 1 . Transgenic strains were generated by φC31-mediated transgene integration into the attP target sites of attP2 , attP40 , or ZH-51C ( Bischof et al . , 2007; Groth et al . , 2004 ) using plasmid DNAs constructed as described above . Plasmid DNA injections were performed in our laboratory or by BestGene . Information about DNA constructs using the attP landing site is included in the table of fly strains . Most of the lacZ reporters were integrated into the attP40 site , except for del1-lacZ , svbE6B-D1-2-lacZ , and svbE6B-D1-6-lacZ . Although one strain of del1-lacZ integrated at the attP40 site was obtained , it had a frame-shift mutation in the lacZ CDS . Therefore , we used this attP40 line with the frameshift to observe the adult body color ( Figure 4—figure supplement 1A ) and the ZH-51C line to analyze the embryonic β-gal expression ( Figure 4—figure supplement 1H ) . svbE6B-D1-2-lacZ and svbE6B-D1-6-lacZ were also integrated into the ZH-51C site because we could not obtain transformants in which these transgenes were integrated into the attP40 site . Embryos were prepared for live imaging as previously reported ( Kondo and Hayashi , 2013 ) . Imaging was performed using an Olympus FV-1000 with a 60x oil immersion objective ( PLAPON 60XO , numerical aperture 1 . 42 , Olympus ) at 25°C ( Figure 1C ) or a Zeiss LSM800 with a 63x oil immersion objective ( Objective Plan-Apochromat 63x/1 . 4 Oil DIC , Zeiss ) ( Figure 1B ) at 25°C with a setting below saturated signal intensity . Images were processed using FIJI software ( https://fiji . sc/ ) , and all projection views were generated using the custom FIJI plugin CoordinateShift ( written by Housei Wada , https://signaling . riken . jp/en/en-tools/imagej/ ) . ‘XY’ showed the Z-projection view . ‘YZ’ and ‘XZ’ showed the X-projection and Y-projection views of a boxed area in the ‘XY’ panels . The range of intensity was adjusted using FIJI software , avoiding saturation of the signal . Embryos were dechorionated in 50% bleach for 2 min and fixed in 1:1 4% PFA containing 1 mM CaCl2 and heptane for 20 min at room temperature . The vitelline membrane was removed by shaking in 1:1 methanol and heptane . Embryos were washed in PBSTwx ( PBS with 0 . 2% Tween 20% and 0 . 2% Triton X-100 ) 3 times for 15 min each and blocked in PBSTwx with 1% BSA for 60 min at room temperature . Samples were stained with the primary antibody at 4 °C overnight and washed in PBSTwx 3 times for 15 min each . Secondary antibody or phalloidin staining was performed at room temperature for 3 hr . The antibodies used in this study are listed in the Key Resources Table . For dpERK antibody staining , the fluorescent signal was amplified using the Tyramide Signal Amplification system with anti-mouse IgG-biotin , Reagents A and B in the ABC kit , and Cy3 Tyramide . After staining , the embryos were washed in PBSTwx 3 times for 15 min each and mounted in Vectashield Mounting Medium with DAPI ( Vector Laboratories ) or SlowFade Diamond Antifade Mountant with DAPI ( Molecular Probes ) . Images of fixed embryos were taken using a Zeiss Apotome . 2 equipped with ORCA-Flash V2 ( Hamamatsu Photonics ) and a 20x dry objective ( Objective Plan-Apochromat 20x/0 . 8 , Zeiss ) or a 63x water immersion objective ( Objective C-Apochromat 63x/1 . 20 W , Zeiss ) with a setting below saturated signal intensity unless otherwise noted . For Figure 1—figure supplement 1 , images of fixed embryos were taken using an Olympus FV-1000 with a 20x objective lens ( UPLSAPO 20X numerical aperture 0 . 75 , Olympus ) . Images were processed using FIJI software , and all projection views were generated using a custom FIJI plugin CoordinateShift ( https://signaling . riken . jp/en/en-tools/imagej/ ) . The Z-projection , X-projection , and Y-projection regions were manually determined for each image . ‘XY ( epi ) ’ showed the Z-projection view at the surface-epidermis level , and ‘XY ( tra ) ’ showed the Z-projection view at the level inside the trachea . ‘YZ’ and ‘XZ’ showed the X-projection and Y-projection views of a boxed area in the ‘XY’ panels . All images were acquired with a 63x water immersion objective ( Objective C-Apochromat 63x/1 . 20 W , Zeiss ) were smoothed with a 1-Sigma ( radius ) Gaussian Blur filter . The dynamic range of intensity was adjusted while avoiding saturation of the signal . For the DE-cad signal , the intensity was adjusted to show the epithelial tissue geometry more clearly; therefore , some dense signals were oversaturated . For the svbE6B chimeric reporter analyzes in Figure 5B–F , the β-gal signal was collected and adjusted using the same parameters among strains using the same transgene landing site ( attP40 or ZH-51c ) . For the 3 × UAS chimeric reporter analyzes in Figure 5G–I , the GFP signal was collected and adjusted using the same parameters . All images were converted into 8-bit images and assembled using Adobe Illustrator for figures . In all images , the anterior side is to the left , and the dorsal side is up . Since tracheal morphogenesis is left-right symmetric , right-side images were flipped to adjust the directions of the anterior-posterior and dorsal-ventral axes . The numbers of Trh-positive cells and β-gal-positive cells were counted manually using Z-stack images taken by a Zeiss ApoTome . 2 and FIJI software with the Cell Counter plug-in . The 4th , 5th , and 6th tracheal metameres of each embryo were used for this quantification . Boxplots and beeswarms were drawn , and statistical analyzes were performed using R software ( https://www . r-project . org/ ) . Exact Wilcoxon-Mann-Whitney Tests were performed using the Wilcox_test function from the coin package . Steel-Dwass tests were performed using pSDCFlig from the NSM3 package with the Asymptotic method . Images of 1-day-old adult flies were taken using a Leica S8APO stereomicroscope equipped with an Olympus AIR01 digital camera . Images were adjusted using Adobe Photoshop and assembled using Adobe Illustrator for figures .	Cells in developing organs have two important decisions to make: where to be and what cell type to become . If cells end up in the wrong places , they can stop an organ from working , so it is vital that one decision depends upon the other . The so-called progenitor cells responsible for forming the trachea , for example , can either become part of a flat sheet or part of a tube . The cells on the sheet need to become epidermal cells , while the cells in the tube need to become tracheal cells . Work on fruit flies found that a gene called 'trachealess' plays an important role in this process . Without it , developing flies cannot make a trachea at all . At the start of trachea development , some of the cells form thickened structures called placodes . The progenitor cells in the placodes start to divide , and the structures buckle inwards to form pockets . These pockets then lengthen into tubes . The trachealess gene codes for a protein that works as a genetic switch . It turns other genes on or off , helping the progenitor cells inside the pockets to become tracheal cells . But , it is not clear whether trachealess drives the formation of the pockets: the progenitor cells first decide what to be; or whether pocket formation tells the cells to use trachealess: the progenitor cells first decide where to be . To find out , Kondo and Hayashi imaged developing fly embryos and saw that the trachealess gene does not start pocket formation , but that it is essential to maintain the pockets . Flies without the gene managed to form pockets , but they did not last long . Looking at embryos with defects in other genes involved in pocket formation revealed why . In these flies , some of the progenitor cells using trachealess got left behind when the pockets started to form . But rather than forming pockets of their own ( as they might if trachealess were driving pocket formation ) , they turned their trachealess gene off . Progenitor cells in the fly trachea seem to decide where to be before they decide what cell type to become . This helps to make sure that trachea cells do not form in the wrong places . A question that still remains is how do the cells know when they are inside a pocket ? It is possible that the cells are sensing different mechanical forces or different chemical signals . Further research could help scientists to understand how organs form in living animals , and how they might better recreate that process in the laboratory .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2019	Two-step regulation of trachealess ensures tight coupling of cell fate with morphogenesis in the Drosophila trachea
Vemurafenib and dabrafenib selectively inhibit the v-Raf murine sarcoma viral oncogene homolog B1 ( BRAF ) kinase , resulting in high response rates and increased survival in melanoma . Approximately 22% of individuals treated with vemurafenib develop cutaneous squamous cell carcinoma ( cSCC ) during therapy . The prevailing explanation for this is drug-induced paradoxical ERK activation , resulting in hyperproliferation . Here we show an unexpected and novel effect of vemurafenib/PLX4720 in suppressing apoptosis through the inhibition of multiple off-target kinases upstream of c-Jun N-terminal kinase ( JNK ) , principally ZAK . JNK signaling is suppressed in multiple contexts , including in cSCC of vemurafenib-treated patients , as well as in mice . Expression of a mutant ZAK that cannot be inhibited reverses the suppression of JNK activation and apoptosis . Our results implicate suppression of JNK-dependent apoptosis as a significant , independent mechanism that cooperates with paradoxical ERK activation to induce cSCC , suggesting broad implications for understanding toxicities associated with BRAF inhibitors and for their use in combination therapies . BRAF inhibitors ( BRAFi ) have revolutionized the treatment of melanoma ( Flaherty et al . , 2010; Chapman et al . , 2011; Sosman et al . , 2012; Falchook et al . , 2012; Hauschild et al . , 2012; Long et al . , 2012 ) . Their clinical use is associated with the development of keratinocytic tumors including cSCC ( Flaherty et al . , 2010; Chapman et al . , 2011; Sosman et al . , 2012; Hauschild et al . , 2012; Falchook et al . , 2012; Long et al . , 2012 ) . Mechanistic studies of this have centered on paradoxical ERK activation , which is most evident in BRAF-wild-type , RAS-mutant cells , as the primary mechanism ( Karreth et al . , 2009; Halaban et al . , 2010; Hatzivassiliou et al . , 2010; Heidorn et al . , 2010; Poulikakos et al . , 2010 ) . This is supported by the findings that RAS mutations are significantly enriched in cSCC arising in patients treated with vemurafenib relative to sporadic cSCC ( Oberholzer et al . , 2011; Su et al . , 2012 ) , and by the low rate of cSCC in patients treated with combined BRAFi and MEK inhibitor ( MEKi ) ( Flaherty et al . , 2012 ) . In one model , drug binding relieves the autoinhibition of BRAF whereupon it is recruited to the membrane by activated RAS and dimerizes with CRAF , driving MEK-dependent ERK activation ( Heidorn et al . , 2010 ) . Other studies show ERK hyperactivation resulting from drug-induced CRAF transactivation ( Hatzivassiliou et al . , 2010; Poulikakos et al . , 2010 ) and modulation of RAS spatiotemporal dynamics ( Cho et al . , 2012 ) . Inhibitor-induced KSR1-BRAF dimers modulate the activity of ERK ( McKay et al . , 2011 ) and also affect MEK signaling by activating KSR1 kinase activity ( Brennan et al . , 2011; Hu et al . , 2011 ) . These models all highlight the importance of CRAF in driving MEK-dependent hyperactivation of ERK . Because of the rapid development of these cSCC on BRAFi therapy and the enrichment for RAS mutations , pre-existing genetic lesions are likely present prior to therapy , which are then ‘unmasked’ following initiation of BRAFi therapy . The fact that many arise in sun-damaged skin suggests that prior chronic UV exposure is an important predisposing event ( Su et al . , 2012 ) . We instead hypothesized that vemurafenib and PLX4720 could also affect the susceptibility of cells to apoptosis and in so doing , contribute to the acceleration of tumor development . We studied the acute ultraviolet radiation ( UVR ) response because this is the most important environmental risk factor in the development of skin cancer and because many BRAFi-induced cSCC arise in sun-damaged areas ( Su et al . , 2012 ) . PLX4720 and vemurafenib share structural features ( Tsai et al . , 2008; Bollag et al . , 2010 ) and have similar activities , as is the case in our studies . We performed our initial studies using cSCC ( SRB1 , SRB12 , COLO16 ) and keratinocyte ( HaCaT ) cell lines . Cells treated with 1 kJ/m2 of UVB ( FS40 lamp ) undergo apoptosis within 24 hr ( Figure 1A–D ) . Surprisingly , this apoptosis was suppressed by at least 70% in cells concomitantly treated with 1 μM PLX4720 ( Figure 1A–D ) compared to control DMSO-treated cells as measured by FACS for Annexin V+; TMRE ( tetramethylrhodamine ) -low cells ( Figure 1E , Figure 1—figure supplement 1A–C ) . Similar results were obtained using doxorubicin as the inducer of apoptosis , and similar suppression of apoptosis was obtained using 1 μM PLX4720 in all cells ( Figure 1—figure supplement 2A , B ) . Importantly , these cells have no oncogenic RAS or BRAF mutations ( Table 1 ) , and PLX4720 conferred no significant proliferative advantage to the tested cells ( Figure 1—figure supplement 3 ) even when used at concentrations that inhibit the proliferation of BRAFV600E melanoma cell lines ( Tsai et al . , 2008 ) . 10 . 7554/eLife . 00969 . 003Figure 1 . PLX4720 suppresses UV-induced apoptosis . The cSCC and HaCaT cell lines were either unirradiated or irradiated with 1 kJ/m2 of UVB in the absence ( ‘o’ , 1:2000 DMSO ) or presence ( ‘+’ ) of 1 μM PLX4720 and isolated for FACS analysis and protein extracts 24 hr later . ( A ) SRB1 , ( B ) SRB12 , ( C ) COLO16 , and ( D ) HaCaT cells show at least 70% suppression of apoptosis in the presence of PLX4720 as measured by FACS for Annexin V+ , TMRE-low cells ( n = 6 for each cell line , ‘’ denotes statistical significance at p<0 . 05 ) . ( E ) A representative FACS plot for COLO16 is shown . Annexin V+ , TMRE-low cells are contained in the upper left quadrant ( boxed ) , which was significantly populated in UV-irradiated cells , but not in the absence of UV , or in the presence of PLX4720 . ( F ) Western blots probed for the MAP kinases demonstrated strong phospho-JNK and phospho-p38 induction following irradiation and significant suppression by PLX4720 . Phospho-ERK was slightly induced following irradiation , and at 24 hr , paradoxical hyperactivation in the presence of PLX4720 was observed , particularly in SRB1 and HaCaT cells . ( G ) Western blots showed that BIM was not upregulated in these BRAF-wild-type cells , consistent with intact ERK signaling . MCL1 was downregulated by irradiation and not modulated by PLX4720 , whereas NOXA expression was strongly induced in irradiated cells and suppressed by PLX4720 . ( H ) Western blots of BRAFV600E melanoma cell lines , A375 and WM35 , demonstrated suppression of UV-mediated induction of phospho-JNK and phospho-p38 by PLX4720 at 24 hr . As expected , phospho-ERK is shut down in PLX4720-treated cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 00310 . 7554/eLife . 00969 . 004Figure 1—figure supplement 1 . PLX4720 potently suppresses apoptosis in cSCC , HaCaT cell lines , and NHEK cells . Representative FACS plots of Annexin V vs TMRE in SRB1 ( A ) , SRB12 ( B ) , HaCaT ( C ) , and NHEK ( D ) cells demonstrated low levels of apoptosis ( Annexin V+ , TMRE-low in quadrant 1 ) in unirradiated cells in the presence and absence of 1 μM PLX4720 . Significant levels of apoptosis were seen in all control-treated irradiated cells , which were significantly suppressed in the presence of PLX4720 , by at least 70% in all cells tested . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 00410 . 7554/eLife . 00969 . 005Figure 1—figure supplement 2 . PLX4720 suppresses doxorubicin-induced JNK activation and apoptosis in cSCC and HaCaT cell lines . COLO16 and HaCaT cell lines were either treated with doxorubicin or PBS and lysed 24 hr later in the absence ( ‘o’ , 1:2000 DMSO ) or presence ( “+” ) of 1 μM PLX4720 . ( A ) COLO16 and ( B ) HaCaT cells showed significant decrease in apoptosis measured by FACS for Annexin V+ , TMRE-low cells ( n = 3 for each cell line , ‘’ denotes statistical significance at p<0 . 05 ) . ( C ) Western blots were probed for phospho-JNK and total JNK , showing a potent activation of JNK by doxorubicin that is significantly suppressed by PLX4720 . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 00510 . 7554/eLife . 00969 . 006Figure 1—figure supplement 3 . PLX4720 does not confer a proliferative advantage to cSCC and HaCaT cell lines . ( A ) SRB1 , ( B ) SRB12 , ( C ) COLO16 , and ( D ) HaCaT cells were treated with DMSO ( 1:2000 ) or the indicated concentrations of PLX4720 for at least 28 days during which cells were serially passaged and counted . Over that time frame there was a slight decrement in the proliferation of SRB12 and HaCaT cells in the presence of 1 μM PLX4720 . All cells treated at 5 μM PLX4720 exhibited decreased proliferation . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 00610 . 7554/eLife . 00969 . 007Table 1 . Lack of BRAF and RAS mutations in cSCC and HaCaT cell linesDOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 007ALK_F1174LIV_T3520CAGALK_F1245C_T3734GALK_F1245VI_T3733GABRAF_G464EVA_G1391ATCBRAF_G466R_G1396CABRAF_K601E_A1801GBRAF_V600EAG_T1799ACG_FCTNNB1_S45APT_T133GCACTNNB1_T41APS_A121GCSEGFR_Y813C_A2438GGNAS_R201SC_C601ATKRAS_G12SRC_G34ACTKRAS_Q61EKX_C181GATMET_H1112_A3335GTMET_Y1248HD_T3742CGPIK3CA_A1046V_C3137TPIK3CA_C420R_T1258CPIK3CA_E110K_G328APIK3CA_E418K_G1252APIK3CA_F909L_C2727GPIK3CA_H1047RL_A3140 GTPIK3CA_H701P_A2102CPIK3CA_N345K_T1035APIK3CA_Q060K_C178APIK3CA_R088Q_G263APIK3CA_S405F_C1214TTNK2_R99Q_G296ABRAF_G466EVA_G1397ATCBRAF_V600LM_G1798 TACTNNB1_S37APT_T109GCACTNNB1_S45CFY_C134GTAEGFR_G719_G2155TAEGFR_L858R_T2573GEGFR_T790M_C2369TEPHA3_K761N_G2283FGFR2_S252W_C755GFOXL2_C134W_C402GKIT_K642E_A1924GKIT_R634W_C1900TKIT_V560D_T1679AKIT_V825A_T2474CKIT_Y553N_T1657AKRAS_G12DAV_G35ACTMET_N375S_A1124GNRAS_G12SRC_G34ACTPIK3CA_E453K_G1357APIK3CA_E545AGV_A1634CGTPIK3CA_H1047RL_A3140 GT . . 1 . PIK3CA_K111N_G333CPIK3CA_M1043V_A3127GPIK3CA_P539R_C1616GBRAF_E586K_G1756ABRAF_G469EVA_G1406ATCCTNNB1_S33APT_T97GCACTNNB1_S37CFY_C110GTAEGFR_L861_T2582AGEGFR_T854I_C2561TFGFR2_N549KK_T1647GAFRAP_R2505P_G7514CFRAP_S2215Y_C6644TIDH2_R172MK_G515 TAJAK2_V617F_G1849TKIT_L576P_T1727CKIT_N566D_A1696 GKRAS_A146PT_G436CANRAS_G12DAV_G35ACTKRAS_Q61HHE_A183CTGKRAS_G13SRC_G37ACTNRAS_G13DAV_G38ACTNRAS_Q61HHQ_A183TCGPDGFRA_N659Y_A1975TPDGFRA_V561D_T1682APIK3CA_E545KQ_G1633ACPIK3CA_H1047Y_C3139TPIK3CA_Q546EK_C1636 GAPIK3CA_Y1021HN_T3061CARET_M918T_T2753CAKT1_G173R_G517CAKT2_E17K_G49ABRAF_G469R_G1405CABRAF_L597R_T1790GBRAF_V600_G1800CTNNB1_G34EVA_G101ATCEGFR_S720P_T2158CGNA11_Q209LP_A626 TCIDH1_R132CGS_C394TGAIDH2_R140LQ_G419 TAIDH2_R140W_C418TIDH2_R172S_G516TKIT_D816HNY_G2446CATKIT_V559ADG_T1676CAGKRAS_G10R_G28AKRAS_Q61LPR_A182TCGMET_H1112Y_C3334TMET_M1268T_T3803CMET_T1010I_C3029TNRAS_A146T_G436ANRAS_Q61EKX_C181GATPDGFRA_D842V_A2525TPDGFRA_D842_G2524TAPDGFRA_N659K_C1977APIK3CA_E542KQ_G1624ACPIK3CA_G1049R_G3145CPIK3CA_M1043I_G3129ATCPIK3R1_D560Y_G1678TPRKAG2_N488I_A1463TAKT2_G175R_G523CAKT3_G171R_G511AALK_F1174L_C3522AGALK_I1171N_T3512AALK_R1275QL_G3824ATBRAF_D594GV_A1781 GTCTNNB1_D32HNY_G94CATFBWX7_R465C_C1393TFBWX7_R479QL_G1436ATFBWX7_R505HLP_G1514ATCFGFR3_G370C_G1108TGNAQ_Q209H_A627TIDH2_R140W_C419TIDH2_R172GW_A514 GTKIT_N822KNK_T2466GCAKRAS_G13DAV_G38ACTPDPK1_D527E_C1581GPIK3CA_E542VG_A1625TGPIK3CA_E545D_G1635CTPIK3CA_T1025SA_A3073TGPIK3CA_Y1021C_A3062GPIK3R1_N564K_C1693AGPRKAG1_R70Q_G209AAKT1_E17K_G49AAKT1_K179M_A536TBRAF_V600EAG_T1799ACG_RCDK4_R24C_C70TCDK4_R24H_G71ACTNNB1_D32AGT_A95CGVFBWX7_R465HL_G1394ATFGFR3_G697C_G2089TFGFR3_K650MT_A1949 TCFGFR3_R248C_C742TFGFR3_S371C_A1111TFGFR3_Y373C_A1118GGNAS_R201H_G602AIDH1_R132HL_G395ATKIT_N822YHD_A2464TCGMET_R988C_C2962TMET_Y1253D_T3757 GNRAS_G13SRC_G37ACTNRAS_Q61RPL_A182GCTPIK3CA_Q546LPR_A1637TCGTNK2_E346K_G1036APIK3CA_H1047RL_A3140 GTALK_F1245C_T3734GThe listed gene mutations were screened by Sequenom INT16/20 panel ( Characterized Cell Line Core , MD Anderson Cancer Center ) and HRAS was sequenced by Sanger sequencing . All examined loci were wild-type in the cSCC cell lines SRB1 , SRB12 , COLO16 , and keratinocyte cell line HaCaT . The PIK3R1_M326I_G978 polymorphism was found in the SRB12 cell line . Because the p38 and JNK stress-activated MAP kinases are well-established critical mediators of UV-induced apoptosis ( Derijard et al . , 1994; Chen et al . , 1996; Tournier et al . , 2000; Hildesheim et al . , 2004 ) , we explored the status of JNK and p38 activation by assessing phospho-JNK and phospho-p38 levels by Western blot ( Figure 1F ) . Phospho-JNK levels in particular were highly upregulated upon UV irradiation and were significantly suppressed by treatment post-radiation with 1 μM PLX4720 in cSCC and HaCaT cell lines ( Figure 1F ) . Similar effects were seen with 1 μM vemurafenib ( data not shown ) and in cells stressed with doxorubicin ( Figure 1—figure supplement 2C ) . Importantly , ERK signaling remained intact , as evidenced both by the paradoxical activation of ERK ( upregulation of phospho-ERK ) and by the failure to upregulate BIM levels ( Figure 1F , G ) . This pro-apoptotic BCL2 family member is upregulated by inhibition of ERK signaling ( Collins et al . , 2005 ) and in BRAFV600E melanoma cells treated with vemurafenib ( Paraiso et al . , 2011 ) . Since NOXA is a downstream effector of UV-induced apoptosis ( Naik et al . , 2007 ) , we examined its expression and found that NOXA expression is induced by UV irradiation and suppressed by PLX4720 in all cell lines ( Figure 1G ) , suggesting that inhibition of NOXA expression may be a mechanism of PLX4720-induced suppression of apoptosis . Finally , we examined the expression of the antiapoptotic BCL2 family member MCL1 because it is downregulated by UV exposure ( Figure 1G ) , but as previously reported ( Paraiso et al . , 2011 ) , unaffected by PLX4720 ( Figure 1G ) . To test the generality of these effects in cells in which ERK activity is suppressed by BRAFi , we extended our analysis to the BRAFV600E melanoma cells A375 and WM35 . As expected , phospho-ERK expression was strongly suppressed by PLX4720 ( Figure 1H ) . Phospho-JNK and phospho-p38 were significantly upregulated following UV-irradiation ( Figure 1H ) , showing that signaling to JNK and p38 is intact in BRAFV600E melanoma cells . Here again , there was significant suppression of both phospho-p38 and phospho-JNK induction by PLX4720 ( Figure 1H ) , and similar effects were seen with vemurafenib ( data not shown ) . We next examined the responses of primary normal human epidermal keratinocytes ( NHEKs ) to vemurafenib . UV-induced apoptosis was significantly suppressed ( approximately 70% ) by vemurafenib in these cells ( Figure 2A , Figure 1—figure supplement 1D ) , and the UV-induced upregulation of phospho-JNK and phospho-p38 was likewise suppressed most significantly at 6 and 24 hr ( Figure 2B ) . As in the cSCC and HaCaT cell lines , activation of ERK was observed following exposure to vemurafenib ( Figure 2B ) . The presence of cleaved caspase-3 correlated with high levels of apoptosis in the UV-treated cells and its absence with rescue by vemurafenib at 24 hr post-irradiation ( Figure 2C ) . In probing members of the BCL2 family , we found similar results to those in the cSCC and HaCaT cell lines . BIM and MCL1 were unaffected by vemurafenib but NOXA induction at 24 hr post-UV irradiation was diminished by vemurafenib ( Figure 2C ) . The advantage of using primary cells is that p53 is intact . In NHEKs , p53 is stabilized by 24 hr post-UV irradiation and this is unaffected by vemurafenib ( Figure 2C ) . However , since BCL2 family members can be modulated by JNK ( Tournier et al . , 2000; Haeusgen et al . , 2011 ) and p53 ( Oda et al . , 2000 ) in apoptosis , the inhibition of NOXA expression by PLX4720 and vemurafenib ( Figures 1G and 2C ) likely reflects p53-independent regulation of NOXA given that p53 is mutant in HaCaT ( Lehman et al . , 1993 ) cells , p53 is undetectable in SRB12 cells , and p53 levels do not change with radiation in SRB1 , COLO16 , or HaCaT cells , ( Figure 2—figure supplement 1 ) . PUMA , BAX , BCL2 , BCL-XL , and BCL2A1 expression were unchanged following irradiation and were unchanged by PLX4720 or vemurafenib exposure ( data not shown , Figure 2—figure supplement 2 ) . We conclude from our results that vemurafenib and PLX4720 suppress UV-induced apoptosis by inhibiting JNK signaling and NOXA induction in BRAF and RAS WT cells . 10 . 7554/eLife . 00969 . 008Figure 2 . Vemurafenib and PLX4720 suppress apoptosis and JNK signaling in primary human keratinocytes and cSCC cells independently of MEK/ERK signaling . Normal human epidermal keratinocytes ( NHEKs ) were irradiated with 1 kJ/m2 of UVB in the absence ( ‘o’ , 1:2000 DMSO ) or presence ( ‘+’ ) of 1 μM vemurafenib and isolated for FACS analysis and protein extracts 24 hr later . ( A ) Apoptosis was significantly suppressed ( 70% ) in the presence of vemurafenib as measured by FACS for Annexin V+ , TMRE-low cells ( n = 6 , ‘’ denotes statistical significance at p<0 . 05 ) . ( B ) Western blot analysis showed induction of phospho-JNK and phospho-p38 within 1 hr following irradiation , which persisted for at least 24 hr and which was suppressed by vemurafenib at all time points . ( C ) MCL1 and BIM expression was not significantly modulated by vemurafenib; however , NOXA induction , which occurred at 24 hr , was reduced by vemurafenib . In these primary cells , p53 protein was stabilized by 24 hr and vemurafenib did not affect overall levels . Suppression of apoptosis , as measured by cleaved caspase-3 levels , was observed in the presence of vemurafenib-treated irradiated cells , consistent with the FACS results . To test the relevance of MEK signaling , cSCC ( SRB1 ) and NHEK cells were irradiated with 1 kJ/m2 of UVB in the absence ( ‘o’ , 1:2000 DMSO ) or presence of 1 μM PLX4720 singly or in combination with 0 . 6 μM ( NHEK ) or 1 . 2 μM ( SRB1 ) AZD6244 ( MEKi ) and isolated for FACS analysis and protein extracts 24 hr later . ( D ) SRB1 and ( E ) NHEK cells showed induction of phospho-JNK at 24 hr following irradiation , by Western in the presence ( lane 7 ) and absence ( lane 5 ) of MEKi . The addition of MEKi to PLX4720 did not affect the suppression of JNK activation ( compare lanes 6 , 8 ) despite potent suppression of phospho-ERK . ( F ) SRB1 and ( G ) NHEK cells exhibited a strong suppression of UV-induced apoptosis by PLX4720 ( Annexin V+ , TMRE-low cells; n = 6 , ‘’ denotes statistical significance at p<0 . 05 ) that was likewise unaffected by the addition of MEKi . To test whether upstream kinases in the JNK pathway were inhibited , MKK4 and MKK7 activation was probed in cells . ( H ) Both phospho-MKK4 and phospho-MKK7 were induced in HaCaT and NHEK cells following irradiation , and this was suppressed in the presence of 1 μM PLX4720 and vemurafenib , respectively . ( I ) In all cSCC cell lines , SRB1 , SRB12 , COLO16 , phospho-MKK4 and phospho-MKK7 are strongly induced following irradiation , and this is suppressed in all lines by 1 μM PLX4720 . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 00810 . 7554/eLife . 00969 . 009Figure 2—figure supplement 1 . p53 does not respond to stress in cSCC and HaCaT cell lines . cSCC cell lines SRB1 , SRB12 , and COLO16 were either unirradiated or irradiated with 1 kJ/m2 of UVB in the absence ( ‘o’ , 1:2000 DMSO ) or presence ( ‘+’ ) of 1 μM PLX4720 and isolated for protein extracts 24 hr later . ( A ) Western blots of total p53 reveal that none of the cell lines upregulate p53 in response to UV irradiation . SRB12 cells do not express p53 . ( B ) HaCaT cells are known to be mutant for p53 and the presence of p53 in unstressed cells , combined with the failure to upregulate levels following UV radiation , is a hallmark of functionally inactive p53 in cell lines . Loading controls are the same as those in Figure 1F . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 00910 . 7554/eLife . 00969 . 010Figure 2—figure supplement 2 . BCL2 family members BCL2 , BCL-XL , and BCL2A1 are not modulated by acute UV exposure or PLX4720 . HaCaT cells were either unirradiated or irradiated with 1 kJ/m2 of UVB in the absence ( ‘o’ , 1:2000 DMSO ) or presence ( ‘+’ ) of 1 μM PLX4720 and isolated for protein extracts 24 hr later . Western blots for BCL2 ( 2875P , Cell Signaling ) and BCL-XL ( 2764P/clone 54H6 , Cell Signaling ) expression show that expression of neither is changed by acute UV exposure and or by PLX4720 . c , qRT-PCR for BCL2A1 mRNA expression , referenced to GAPDH ( Taqman ) shows that BCL2A1 expression is likewise unchanged by acute UV exposure or by PLX4720 . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 010 Although BRAFi-induced JNK inhibition is observed in BRAF-WT as well as BRAFV600E cells ( Figure 1F , H , 2B ) , with opposite effects on ERK signaling , we sought to further demonstrate that JNK inhibition and paradoxical ERK activation are independent and separable . We treated SRB1 and NHEK cells with the MEK inhibitor ( MEKi ) AZD6244 and PLX4720 singly and in combination with and without UV irradiation . While MEKi effectively abrogated ERK phosphorylation and activation ( Figure 2D , E ) , this left PLX4720-mediated suppression of UV-induced JNK activation ( Figure 2D , E ) and apoptosis ( Figure 2F , G ) unaffected in both SRB1 and NHEK cells . Because JNK and p38 isoforms are not significantly inhibited by PLX4720 or vemurafenib directly , ( Tsai et al . , 2008; Bollag et al . , 2010 ) we probed the phosphorylation status of MKK4 and MKK7 ( MAP2K7 ) , the two proximal kinases that synergistically phosphorylate JNK and that are required for JNK activation ( Tournier et al . , 2001; Haeusgen et al . , 2011 ) . The phosphorylation of both MKK4 and MKK7 , corresponding to their activation , was significantly upregulated in control UV-irradiated cells and inhibited by PLX4720 in all cSCC cell lines , HaCaT , and NHEK cells ( Figure 2H , I ) . We then performed a kinome screen of PLX4720 and vemurafenib against a panel of 38 kinases reported to be upstream of JNK ( Keshet and Seger , 2010; Haeusgen et al . , 2011 ) and other kinases previously tested against PLX4720 and vemurafenib ( Tsai et al . , 2008; Bollag et al . , 2010 ) using a quantitative competitive binding assay ( Davis et al . , 2011 ) at four concentrations ( 50 nM , 200 nM , 1 μM , 10 μM ) . We extended previously reported results obtained on this platform ( Davis et al . , 2011 ) by testing a wider concentration range and by additionally testing vemurafenib . Reported biochemical IC50s for vemurafenib ( Bollag et al . , 2010 ) and PLX4720 ( Tsai et al . , 2008 ) against multiple kinases including BRAFV600E , MAP4K5 , SRMS , and BRK were quantitatively similar to the estimated Kd , confirming the validity of this assay ( Tables 2 and 3 ) . We confirmed that ZAK and MKK4 ( MAP2K4 ) have high binding affinities comparable to that of the intended target , BRAF ( estimated Kd below 50 nM ) for both PLX4720 ( Davis et al . , 2011 ) and vemurafenib , and confirmed activity against MAP4K5 ( Bollag et al . , 2010 ) ( Tables 2 and 3 ) . To demonstrate an effect on activity , in vitro kinase assays were performed ( Figure 3A–C ) and revealed biochemical IC50s of 187 ± 5 nM , 460 ± 41 nM , and 354 ± 26 nM for ZAK , MKK4 , and MAP4K5 , respectively . All of these values are within the range of reported correspondences between binding assays and activity-based assays and with reported data ( Anastassiadis et al . , 2011; Davis et al . , 2011 ) . Importantly , at 1 μM vemurafenib used in our experiments , the residual activity of ZAK , MKK4 , and MAP4K5 kinases , was 18 . 9 ± 0 . 5% , 29 . 6 ± 1 . 1% , and 25 . 7 ± 0 . 6% , respectively . 10 . 7554/eLife . 00969 . 011Table 2 . Quantitative competitive binding assays reveal additional kinase targets of PLX4720DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 011Gene NameEntrez gene SymbolPercent control ( 50 nM ) Percent control ( 200 nM ) Percent control ( 1000 nM ) Percent control ( 10 μM ) Calculated estimate of IC50 ( nM ) Published biochemical IC50 ( nM ) ASK1MAP3K589989710014 , 179 . 29ASK2MAP3K694100100100BLKBLK9178321446 . 56BRAF ( V600E ) BRAF38193 . 90 . 132 . 0413BRKPTK647142 . 40 . 230 . 38130DLKMAP3K12959810092FGRFGR6938112 . 5153 . 47HPK1MAP4K110010010047LZKMAP3K13941009675MAP3K1MAP3K1961009284MAP3K15MAP3K1594979159MAP3K2MAP3K2100938741MAP3K3MAP3K394979875MAP3K4MAP3K410010010065MAP4K2MAP4K2981009967MAP4K3MAP4K3100959056MAP4K4MAP4K4929910046MAP4K5MAP4K5961006381257 . 42MEK3MAP2K310010010064MEK4MAP2K448272 . 60 . 0537 . 96MEK6MAP2K68210010047MINKMINK1891009855MKK7MAP2K710010010084MLK1MAP3K9100100100100>10 , 000>5000MLK2MAP3K101008210076MLK3MAP3K11100100100100MST1STK41009384556709 . 79>5000OSR1OXSR1100949542PAK1PAK193978322RIPK1RIPK199878550SRMSSRMS1 . 90 . 550 . 0500 . 64STK39STK3910010010059TAK1MAP3K790888549TAOK1TAOK1879489657532 . 57>5000TAOK2TAOK2921009351TAOK3TAOK3100989658TNIKTNIK97897924ZAKZAK2040 . 70 . 19 . 47Quantitative competitive binding assays were performed for a group of kinases previously tested against PLX4720 as well as a group of MAP kinases upstream of JNK . Published biochemical IC50s for PLX4720 are listed ( see main text ) for comparison and demonstrate good quantitative correspondence between estimated Kd from binding assays and biochemical IC50s . ZAK and MKK4 ( MAP2K4 ) were very tightly bound by PLX4720 with estimated Kd below 50 nM . Bold text indicates the kinases tested for inhibition by PLX4720 with in-vitro kinase assays . 10 . 7554/eLife . 00969 . 012Table 3 . Quantitative competitive binding assays reveal additional kinase targets of vemurafenibDOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 012Percent control ( 50 nM ) Percent control ( 200 nM ) Percent control ( 1000 nM ) Percent control ( 10 μM ) Calculated estimate of IC50 ( nM ) Published biochemical IC50 ( nM ) ASK1MAP3K590949710011 , 972 . 22>1000ASK2MAP3K6949810074BLKBLK9666300 . 55518 . 03547BRAF ( V600E ) BRAF63255 . 40 . 564 . 7831BRKPTK663286 . 90 . 3568 . 04213DLKMAP3K1298976692FGRFGR6549131 . 6149 . 2663HPK1MAP4K195886715LZKMAP3K13100999374MAP3K1MAP3K198848981MAP3K15MAP3K15841008491MAP3K2MAP3K291918983MAP3K3MAP3K3879710094MAP3K4MAP3K495928746MAP4K2MAP4K299829546MAP4K3MAP4K380908224MAP4K4MAP4K4969283232842 . 34>1000MAP4K5MAP4K562334 . 10 . 158 . 2151MEK3MAP2K3100969854MEK4MAP2K4194 . 10 . 20 . 056 . 82MEK6MAP2K6919787214080 . 69>1000MINKMINK1100100916614 , 761 . 44>1000MKK7MAP2K797959485MLK1MAP3K910093974113 , 979 . 88>1000MLK2MAP3K1092968778MLK3MAP3K119810010077MST1STK499835112OSR1OXSR11001008998PAK1PAK199989146RIPK1RIPK1921009973SRMSSRMS249 . 60 . 75011 . 1518STK39STK391001009766TAK1MAP3K793888688TAOK1TAOK1911009779TAOK2TAOK29892957011 , 770 . 83>1000TAOK3TAOK39298928015 , 468 . 75>1000TNIKTNIK95946611ZAKZAK91 . 80 . 250 . 054 . 03Quantitative competitive binding assays were performed for a group of kinases previously tested against vemurafenib as well as a group of MAP kinases upstream of JNK . Published biochemical IC50s for vemurafenib are listed ( see main text ) for comparison and demonstrate good quantitative correspondence between estimated Kd from binding assays and biochemical IC50s . ZAK and MKK4 ( MAP2K4 ) were very tightly bound by vemurafenib with estimated Kd below 50 nM . Bold text indicates the kinases tested for inhibition by vemurafenib with in-vitro kinase assays . 10 . 7554/eLife . 00969 . 013Figure 3 . PLX4720 and vemurafenib suppress apoptosis and JNK signaling through inhibition of off-target kinases . ( A–C ) In-vitro kinase assays for ZAK , MKK4 , and MAP4K5 were performed across a 10-point concentration range from 0 . 05 to 1000 nM in triplicate , revealing significant inhibition of kinase activity within the nM range for vemurafenib . ( D ) Lentiviral shRNA knockdown of ZAK singly or in combination with MKK4 and MAP4K5 ( triple knockdown , ‘TKD’ ) was performed revealing potent suppression of apoptosis as measured by FACS for Annexin V+ , TMRE-low cells ( n = 5 , ‘’ denotes statistical significance at p<0 . 05 , ‘’ at p<0 . 01 , ‘NS’ is not significant ) at 24 hr following single dose UVB irradiation at 720 J/m2 . ZAK knockdown and triple knockdown cells exhibit 70% and 94% suppression of apoptosis , respectively , relative to PLX4720-treated cells expressing a non-suppressing shRNA control ( scramble , ‘SCR’ ) . ( E ) Western blots of lysates obtained at 1 and 6 hr post-UV irradiation show potent induction of phospho-MKK4 , phospho-MKK7 , and phospho-JNK which are all suppressed with progressively increasing effect in ZAK single knockdown ( ‘shZAK2’ ) and triple knockdown ( ‘TKD’ ) HaCaT cells . ( F ) Western blots of HaCaT cells electroporated with pcDNA3-wild-type ( WT ) ZAK and the gatekeeper mutant pcDNA3- ( T82Q ) ZAK show equivalent expression . ( G ) HaCaT cells overexpressing ZAK ( WT ) and ZAK ( T82Q ) were irradiated with a single dose of UVB irradiation at 720 J/m2 in the absence ( ‘o’ ) and presence ( ‘+’ ) of 1 μM PLX4720 and apoptosis measured by FACS for Annexin V+ , TMRE-low cells ( n = 4 , ‘’ at p<0 . 01 , ‘NS’ is not significant ) at 24 hr . ZAK ( WT ) cells are sensitive to PLX4720-mediated suppression of apoptosis ( bar 3 vs 4 ) , but drug-treated ZAK ( T82Q ) -expressing cells undergo significantly more apoptosis than drug-treated ZAK ( WT ) cells ( bar 4 vs 8 ) , with bypass of PLX4720-induced suppression as compared to drug-treated ZAK ( WT ) cells ( paired t-test , p=0 . 005 ) . ( H ) Western blots of ZAK ( WT ) and ZAK ( T82Q ) -expressing HaCaT cells at 1 hr and 6 hr post-irradiation show that phospho-JNK activation is intact in both cell lines in the absence of drug ( lanes 3 , 7 ) , but that drug-treated ZAK ( T82Q ) -expressing HaCaT cells have significantly more phospho-JNK activation at both 1 and 6 hr post-irradiation , as compared to drug-treated ZAK ( WT ) -expressing cells ( lane 4 vs 8 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 01310 . 7554/eLife . 00969 . 014Figure 3—figure supplement 1 . Knockdown of ZAK potently inhibits JNK activation and UV-induced apoptosis . ( A ) Western blot of HaCaT cells expressing two shRNA clones and HaCaT TKD cells ( containing shZAK2 ) all show significant knockdown of ZAK protein , though shZAK2 produces slightly less knockdown . Approximately 54 . 6% knockdown ( ImageJ ) of MAP4K5 is observed in TKD cells . ( B ) HaCaT cells , expressing non-silencing scramble-shRNA ( ‘SCR’ ) , shZAK1 , shZAK2 , or TKD , were either unirradiated ( black bars ) or irradiated ( open bars ) with 1 kJ/m2 of UVB in the absence ( ‘o’ , 1:2000 DMSO ) or presence ( ‘+’ ) of 1 μM PLX4720 and analyzed by FACS for apoptosis ( Annexin V+ , TMRE- ) at 24 hr . UV-induced apoptosis is significantly suppressed by both ZAK shRNA clones in HaCaT cells and in TKD cells . The shZAK2 clone , which results in less knockdown than shZAK1 , produces correspondingly less suppression of UV-induced apoptosis ( 93 . 7% for shZAK1 vs 67 . 8% for shZAK2 ) . shZAK1-expressing HaCaT cells , TKD cells , and PLX4720-treated HaCaT scrambled-shRNA-expressing cells show similar degrees of suppression , again consistent with the fact that ZAK can account for the majority of the effect of BRAFi-induced suppression of JNK signaling . ( C ) HaCaT cells , treated as above , were processed for Western blots at 1 and 6 hr following UV exposure to assess JNK activation . Significant suppression of phospho-JNK is observed at 1 hr and 6 hr post-irradiation in all cell lines where ZAK is knocked down , as well as TKD cells and SCR cells treated with PLX4720 . In comparing the shZAK1 and shZAK2-expressing HaCaT cells , the degree of phospho-JNK inhibition correlates exactly with the degree of knockdown of ZAK particularly at 1 hr: less phospho-JNK inhibition is observed with less ZAK knockdown . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 01410 . 7554/eLife . 00969 . 015Figure 3—figure supplement 2 . Single knockdown of MKK4 or MAP4K5 partially inhibits JNK activation and UV-induced apoptosis . ( A ) Western blot of HaCaT cells expressing two shRNA clones against MKK4 and MAP4K5 show significant knockdown of targets proteins ( shMKK4-1: 89 . 3% , shMKK4-2: 71 . 9% , shMAP4K5-1: 86 . 4% , shMAP4K5-2: 84 . 1%; ImageJ ) . ( B ) HaCaT cells , expressing non-silencing scramble-shRNA ( ‘SCR’ ) , shMKK4-1 , shMKK4-2 , shMAP4K5-1 , or shMAP4K5-2 were either unirradiated ( black bars ) or irradiated ( open bars ) with 720 J/m2 of UVB in the absence ( ‘o’ , 1:2000 DMSO ) or presence ( ‘+’ ) of 1 μM PLX4720 and analyzed by FACS for apoptosis ( Annexin V+ , TMRE- ) at 24 hr . UV-induced apoptosis is suppressed most substantially by MKK4 ( up to 27 . 3% ) , but not substantially by MAP4K5 ( up to 11 . 6% ) in HaCaT cells . These results are consistent with the fact that ZAK can account for the majority of the effect of BRAFi-induced suppression of JNK signaling . Importantly , since MKK4 is important for JNK activation , and ZAK activates MKK4 , the partial suppression of phospho-JNK activation and apoptosis is expected . ( C ) HaCaT cells , treated as above , were processed for Western blots at 1 hr following UV exposure to assess JNK activation . Significant activation of phospho-JNK is still observed at 1 hr post-irradiation in all cell lines , as compared to SCR cells treated with PLX4720 . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 01510 . 7554/eLife . 00969 . 016Figure 3—figure supplement 3 . Knockdown of ZAK potently inhibits JNK activation and UV-induced apoptosis in SRB1 cells . ( A ) Western blot of SRB1 cells expressing two shRNA clones ( shZAK1 , shZAK2 ) all show significant knockdown of ZAK protein . ( B ) SRB1 cells , expressing non-silencing scramble-shRNA ( ‘SCR’ ) , shZAK1 , or shZAK2 , were either unirradiated ( black bars ) or irradiated ( open bars ) with 720 J/m2 of UVB in the absence ( ‘o’ , 1:2000 DMSO ) or presence ( ‘+’ ) of 1 μM PLX4720 and analyzed by FACS for apoptosis ( Annexin V+ , TMRE- ) at 24 hr . UV-induced apoptosis is significantly suppressed by both ZAK shRNA clones in SRB1 cells . shZAK1/2-expressing SRB1 cells and PLX4720-treated SRB1 scrambled-shRNA-expressing cells show similar degrees of suppression ( 90% , 92 . 5% of drug-treated cells ) , again consistent with the fact that ZAK can account for the majority of the effect of BRAFi-induced suppression of JNK-dependent apoptosis . ( C ) SRB1 cells , treated as above , were processed for Western blots at 1 hr following UV exposure to assess JNK activation . Significant suppression of phospho-JNK is observed at 1 hr post-irradiation in all cell lines where ZAK is knocked down , as well as in SCR cells treated with PLX4720 . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 01610 . 7554/eLife . 00969 . 017Figure 3—figure supplement 4 . Vemurafenib and PLX4720 inhibit multiple kinases upstream of JNK and p38 . The schematic shows MAP kinases upstream of JNK and p38 that are inhibited by these BRAF inhibitors ( gray-shaded ) . Vemurafenib and PLX4720 inhibit ZAK ( principally ) and MKK4 ( MEK4/MAP2K4 ) , resulting in inhibition of MKK7 and MKK4 and , ultimately , JNK . p38 activation was diminished by drug exposure in some contexts , but not to the degree that JNK activation was . Vemurafenib and PLX4720 also inhibit MAP4K5 , which has been shown to be upstream of MKK4 and JNK . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 017 To examine the requirements for ZAK , MAP4K5 , and MKK4 in activating JNK activation and apoptosis more directly , we performed lentiviral shRNA knockdown experiments in HaCaT cells . HaCaT cells with knockdown of ZAK ( ‘shZAK2’ ) showed a strong suppression of ZAK protein expression ( Figure 3—figure supplement 1A ) and of UV-induced apoptosis , showing 70% suppression of apoptosis relative to that achieved by PLX4720 in control ( ‘SCR’ ) cells ( Figure 3D , Figure 3—figure supplement 1A ) . An additional clone of shRNA against ZAK ( ‘shZAK1’ ) showed similar results ( Figure 3—figure supplement 1B , C ) , demonstrating that even greater knockdown of ZAK can account for nearly the entire effect of PLX4720 on JNK activation and apoptosis . Western blots show significant suppression of phospho-MKK4/MKK7 in shZAK2 knockdown cells ( Figure 3E ) . Triple knockdown cells ( ‘TKD’ ) with combined shRNA knockdown of ZAK , MKK4 , and MAP4K5 kinases , as confirmed by Western ( Figure 3E , Figure 3—figure supplement 1A ) , showed comparable suppression of apoptosis to that of drug-treated control cells ( Figure 3D , Figure 3—figure supplement 1B ) and substantial suppression of phospho-MKK4/MKK7 induction ( Figure 3E ) . Furthermore , single knockdown of MKK4 and MAP4K5 ( Figure 3—figure supplement 2A ) , only partially suppresses UV-induced apoptosis or phospho-JNK induction in HaCaT cells ( Figure 3—figure supplement 2B , C ) . Knockdown of ZAK alone was able to account for 91 . 3% of the suppression of UV-induced apoptosis in a distinct cell line , SRB1 ( Figure 3—figure supplement 3A , B ) , with corresponding suppression of phospho-JNK induction ( Figure 3—figure supplement 3C ) . As knockdown of ZAK alone can account for up to 93 . 7% of the effect of PLX4720 treatment , we conclude that the potent inhibition of JNK activation and resultant apoptosis by PLX4720 and vemurafenib is due to the off-target inhibition of ZAK principally , with smaller additional contributions from inhibition of MKK4 and MAP4K5 , which abrogates the activation of the two kinases essential for JNK phosphorylation and activation: MKK4 and MKK7 ( Figures 2H–I , 3D–E , Figure 3—figure supplement 4 ) . Consistent with our findings , ZAK has been shown to be critically important for JNK activation upstream of MKK4 and MKK7 ( Wang et al . , 2005 ) and doxorubicin-induced apoptosis ( Sauter et al . , 2010; Wong et al . , 2013 ) . Our biochemical and shRNA data showed that knockdown of ZAK suppressed phospho-JNK activation and apoptosis ( Figure 3D–E , Figure 3—figure supplements 1 , 3 ) and that the degree of knockdown correlated with the degree of JNK and apoptosis suppression . To show that PLX4720 suppresses apoptosis primarily through direct action on ZAK in cells , we employed a chemical-genetic approach by engineering a gatekeeper mutant ZAK ( T82Q ) . Gatekeeper mutant kinases , in which the threonine ( T ) is replaced by a larger amino acid , in our case glutamine ( Q ) , are often rendered insensitive to small molecule inhibitors and are an important mechanism of drug resistance ( Daub et al . , 2004; Whittaker et al . , 2010 ) . We overexpressed equivalent amounts of ZAK ( T82Q ) wild-type ZAK ( WT ) in HaCaT cells ( Figure 3F ) , and compared their UV responses . Whereas ZAK ( WT ) cells were sensitive to PLX4720-mediated suppression of apoptosis ( Figures 1D and 3G ) , drug-treated ZAK ( T82Q ) -expressing cells underwent 2 . 13-fold more apoptosis than drug-treated ZAK ( WT ) cells ( bar 4 vs 8; p=0 . 005 ) , corresponding to 76 . 9% of the levels of apoptosis in untreated cells ( bars 3 , 7 vs 8; p=0 . 08 ) ( Figure 3G ) . The effects on apoptosis corresponded to higher levels of phospho-JNK , even in drug-treated cells expressing the ZAK ( T82Q ) mutant as compared to drug-treated ZAK ( WT ) -expressing cells at both 1 hr and 6 hr post-irradiation ( lane 4 vs 8; Figure 3H ) . Sustained activation of JNK is necessary for apoptosis ( Tobiume et al . , 2001; Kamata et al . , 2005; Ventura et al . , 2006 ) , and our results show that PLX4720-treated ZAK ( T82Q ) -expressing cells retain higher activation across 1–6 hr as compared to PLX4720-treated ZAK ( WT ) cells . We then explored whether vemurafenib or PLX4720-mediated suppression of JNK and apoptosis is relevant in vivo . We first examined cSCC arising in patients treated with vemurafenib and compared them to sporadic cSCC that were histologically similar , arising in individuals never treated with vemurafenib ( Figure 4A–E ) . Phospho-JNK and cleaved caspase-3 expression were assessed by immunohistochemistry and then quantified following normalization by unit area ( mm2 ) of tumor tissue ( malignant keratinocytes ) only ( Figure 4A–D , Figure 4—figure supplement 1 ) . Sporadic cSCC arising in patients never treated with vemurafenib ( n = 15 ) contained substantially greater expression of phospho-JNK ( p=0 . 013; Figure 4A , E ) and cleaved caspase-3 ( p=0 . 042; Figure 4C , E ) as compared to lesions arising in vemurafenib-treated patients ( n = 16; Figure 4B , D , E ) . Therefore , we found significant reductions in phospho-JNK and cleaved caspase-3 expression in human cSCC suggesting that suppression of JNK activity and apoptosis occur in vivo in patients treated with vemurafenib . 10 . 7554/eLife . 00969 . 018Figure 4 . Vemurafenib and PLX4720 suppress apoptosis and JNK signaling in vivo . ( A–D ) cSCC samples from vemurafenib-treated patients and non-treated patients were analyzed by immunohistochemistry for phospho-JNK and cleaved caspase-3 expression . cSCC arising in vemurafenib-treated patients show decreased expression of phospho-JNK ( B ) and cleaved caspase-3 ( D ) as compared to sporadic cSCC in patient never treated with vemurafenib ( A and C ) . Scale bar is 100 μm . ( E ) Comparisons of stained cells normalized to mm2 of tumor area revealed significant suppression of both phospho-JNK and cleaved caspase 3 expression in vemurafenib-treated cSCC ( ‘’ , p<0 . 05 ) . ( F and G ) Hematoxylin-stained cryosections of skin harvested at 24 hr post-irradiation showed extensive apoptosis ( arrowheads ) with vacuolated blebbed cells and clumped pyknotic nuclei in control-treated mice ( F ) and significantly fewer apoptotic cells in PLX4720-treated mice ( G ) . Scale bar is 50 μm . ( H–I ) Vehicle-treated ( ‘o’ ) and PLX4720-treated ( ‘+’ ) mice were unirradiated or irradiated once , and epidermis was harvested at 1 hr , 6 hr , and 24 hr post-irradiation . ( H ) Significant UV-induced upregulation of both phospho-JNK and phospho-p38 were observed within 1 hr , with significant suppression of phospho-JNK in PLX4720-treated mice by 6 hr and minimal suppression of phospho-p38 . Phospho-ERK levels remained constant . The upstream regulators of JNK , MKK4 and MKK7 , were both significantly activated within 1 hr of irradiation , and potently suppressed in PLX4720-treated mice . Cleaved caspase-3 levels increased within 6 hr and were suppressed in PLX4720-treated mice . ( I ) Noxa was induced most significantly at 6 hr and was potently suppressed by PLX4720 at all time points ( ‘*’ , p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 01810 . 7554/eLife . 00969 . 019Figure 4—figure supplement 1 . Double staining of sporadic cSCC confirms phospho-JNK and cleaved caspase-3 expression within keratinocytes of tumors . Sections were processed for standard immunohistochemistry and stained with primary antibodies against phospho-JNK , cleaved caspase-3 ( Cell Signaling; peroxidase–DAB ) as before , together with antibodies against cytokeratins 5/6 ( clone D5/16 B4—Thermo; peroxidase–AEC ) . Results of the double staining show that in all cases , phospho-JNK staining and cleaved caspase 3 staining was observed exclusively in keratinocytes within tumors . Keratinocytes ( CK5/6 + ) are significantly larger and have more cytoplasm than macrophages or lymphocytes . Scale bar is 50 μM . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 019 We then probed the acute , in vivo , short-term UV response in skin by pre-treating C57BL/6 mice with PLX4720 administered by oral gavage 40–80 mg/kg twice a day for 2–4 days ( Tsai et al . , 2008 ) . Following depilation , mice were irradiated once using a solar simulator ( Oriel ) with 10 kJ/m2 UVB . The skin was harvested at 1 hr , 6 hr , and 24 hr post-irradiation . Consistent with our other results , we found significant apoptosis of epidermal keratinocytes in irradiated mouse skin that was suppressed by PLX4720 treatment ( 12 . 7 ± 0 . 4 apoptotic cells/mm vs 4 . 9 ± 0 . 3 apoptotic cells/mm skin , [n = 3 pairs] , p<10−5; Figure 4F , G ) , a finding corroborated by cleaved caspase-3 levels , which were induced within 6 hr of irradiation and suppressed in PLX4720-treated mice ( Figure 4H ) . As expected , phospho-JNK and phospho-p38 were significantly upregulated following UV irradiation and phospho-JNK was significantly suppressed by PLX4720 ( Figure 4H ) . The upstream kinases MKK4 and MKK7 were likewise activated by UV radiation and suppressed by PLX4720 at 1 and 6 hr post-irradiation , confirming the importance of this mechanism of PLX4720-induced JNK signaling suppression in vivo . Finally , Noxa mRNA expression , as measured by qPCR , was strongly induced by UV exposure , a response significantly dampened by PLX4720 treatment ( Figure 4I ) . We also used the Hairless mouse model of squamous cell carcinoma to assess whether PLX4720 would affect UV-driven tumor development . This is particularly relevant since it appears that UV exposure is an important initiating event in BRAFi-accelerated cSCC ( Su et al . , 2012 ) . Unlike the DMBA/TPA model , in which lesions almost universally harbor Hras mutations ( Brown et al . , 1990 ) , the Hairless model has a very low frequency of Ras mutation in papillomas and carcinomas ( van Kranen et al . , 1995 ) , more similar to sporadic human cSCC . The cohorts ( n = 5 each ) were identically irradiated thrice weekly ( 12 . 5 kJ/m2 per week UVB ) for 72 days before starting on PLX4720 treatment vs vehicle control . Within 20 days of administration of drug , hyperkeratotic papules were visible on the backs of PLX4720-treated animals ( Figure 5A , B ) , which steadily grew into cSCC over the following several weeks ( Figure 5C , D ) . Within this period of 150 days ( 78 days of drug treatment ) , control-treated mice had not yet developed any visible lesions ( Figure 5E ) . When we quantified the effects of each of these drug treatments , we found significant decreases in both phospho-JNK expression ( p=0 . 046; Figure 5F , G , J ) and cleaved caspase 3 expression ( p=0 . 019; Figure 5H–J ) in PLX4720-treated mice as compared to control-treated mice . Importantly , we sequenced the entire coding regions for Ras ( Hras , Kras , Nras ) and found no mutations in any of the tumors in PLX4720-treated mice , as compared to one of 14 papillomas and carcinomas in a cohort of control-treated chronically-irradiated Hairless mice ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 00969 . 020Figure 5 . PLX4720 and JNK inhibition dramatically accelerate cSCC development in the UV-driven Hairless mouse model . ( A–E ) Chronically-irradiated Hairless mice were treated with PLX4720 ( n = 5 ) , or vehicle ( n = 5 ) starting at day 72 ( arrow , E ) . Tumors were induced within 20 days of PLX-4720 treatment ( B ) , whereas only erythema was seen in control animals ( A ) . The tumors in PLX4720-treated mice progressed to well-differentiated cSCC ( C , scale bar 75 μm ) , steadily increasing in size and number ( D , day 132 ) . ( E ) Even at 150 days ( 78 days of drug treatment ) , only PLX4720-treated mice had tumors and the differences in tumor number persisted throughout ( ‘’ , p=0 . 0026 ) . ( F–J ) cSCC from mice were harvested and assessed for phospho-JNK and cleaved caspase 3 expression by immunohistochemistry . Tumors from PLX4720-treated animals showed significantly lower levels of phospho-JNK ( G ) and cleaved caspase 3 ( I ) as compared to control-treated animals ( F and H ) . Differences in these parameters were significant across all comparisons ( J , ‘’ , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 02010 . 7554/eLife . 00969 . 021Figure 5—figure supplement 1 . cSCC and papillomas arising in Hairless mice treated with PLX4720 do not have Ras mutations . ( A and B ) cDNA was reverse-transcribed from total RNA , PCR-amplified with the above primers ( B ) and analyzed by Sanger sequencing for mutations in both directions . No mutations in Hras , Kras , or Nras were detected in any of the papillomas ( n = 5 ) or carcinomas ( n = 3 ) isolated from PLX4720-treated mice . One of the papillomas from untreated mice had a heterozygous point mutation ( A ) in Hras ( G35A , G12E ) among 14 samples ( 12 papillomas , 2 cSCC ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 021 While the effects of these BRAFi on JNK-dependent apoptosis is clear and independent of ERK activity , the relative contribution of paradoxical ERK activation vs JNK pathway inhibition to tumorigenesis has not been precisely addressed ( Figure 6A ) . To accomplish this , we took advantage of the fact that paradoxical ERK activation requires intact CRAF ( Hatzivassiliou et al . , 2010; Heidorn et al . , 2010; Poulikakos et al . , 2010 ) ( Figure 6A ) . We used isogenic , matched WT and Craf-deficient ( Craf−/− ) mouse embryonic fibroblasts ( MEFs ) and transformed them with adenovirus E1A and human HRASG12V to enable anchorage-independent growth ( Figure 6—figure supplement 1 ) . These Craf−/− cells do not exhibit strong paradoxical MEK or ERK activation , consistent with previous reports ( Poulikakos et al . , 2010 ) ( Figure 6—figure supplement 1A ) . Wild-type and matched Craf-deficient MEFs were plated in soft agar assays ( Su et al . , 2012 ) and treated with PLX4720 . Both WT and Craf-deficient MEFs exhibited a significant colony formation advantage in the presence of drug ( Figure 6B–D ) . Based upon this analysis , we estimated that the effect of paradoxical ERK activation to be 60% and other effects , including inhibition of JNK activity , to account for the rest ( 40% ) of the total colony growth advantage ( Figure 6D ) . To assess the role of JNK signaling directly , we used HRASG12V-transformed HaCaT cells with ( ‘TKD’ ) and without ( ‘SCR’ ) triple lentiviral knockdown of ZAK , MAP4K5 , and MKK4 ( Figure 3D , E ) , to perform similar colony formation assays to assess responses to PLX4720 treatment ( Figure 6E ) . Drug treatment conferred a significant colony formation advantage in both sets of cells , which exhibit equivalent paradoxical ERK activation ( Figure 6—figure supplement 1B ) . Yet , untreated TKD HaCaT cells produced more colonies than SCR HaCaT cells suggesting that JNK pathway suppression results in an advantage in the absence of drug and paradoxical ERK activation ( Figure 6E ) . Drug-treated SCR and TKD HaCaT cells , had elevated colony counts to similar levels , as expected , because both lines would experience similar degrees of both paradoxical ERK activation and JNK inhibition , and TKD cells ( knocked down for ZAK , MAP4K5 , MKK4 ) are unlikely to experience any further suppression of JNK signaling ( Figure 3D ) . Based on this , we estimated the effect of JNK pathway inhibition to be 17 . 6% ( Figure 6E ) . When combined with the MEF experiment , we estimate that the effect of JNK inhibition contributes approximately 17 . 6–40% of the total effect of PLX4720-accelerated colony formation ( Figure 6D , E ) . Importantly , although we can quantify these individual contributions , it is clear in many contexts in cancer that hyperproliferation and inhibition of apoptosis are highly cooperative ( Hanahan and Weinberg , 2011 ) , and our data do not preclude the possibility that one or both are individually required . 10 . 7554/eLife . 00969 . 022Figure 6 . Paradoxical ERK activation and JNK pathway inhibition make significant and separable contributions to BRAFi-induced growth . ( A ) We envision two separable , parallel mechanisms by which PLX4720 and vemurafenib contribute to cSCC development . Drug-induced paradoxical ERK activation and inhibition of JNK signaling occur in parallel , but the former depends on intact CRAF . ( B and C ) Representative soft agar colonies of E1A and HRASG12V-transformed wild-type ( WT ) ( B ) and Craf−/− ( C ) MEFs , following exposure to 0 . 2 μM and 1 . 0 μM PLX4720 over 4–6 weeks show significant colony-forming advantages conferred by BRAFi . ( D ) The fold-change in colony counts of transformed wild-type ( WT ) ( n = 22 replicates ) and Craf−/− ( n = 14 replicates ) MEFs demonstrate a dose-dependent increase in colonies , particularly for WT MEFs . The difference between colony formation advantages conferred by 1 . 0 μM PLX4720 in WT vs Craf−/− MEFs was interpreted to reflect the contribution of paradoxical ERK signaling ( red arrow ) , which depends upon Craf , and is 60% of the total effect ( black arrow ) , with the remainder composed of other effects including JNK inhibition ( blue arrow ) . All differences between each MEF population were significant ( ‘’ , p<0 . 001 ) ( E ) The fold-change in colony counts of transformed HaCaT cells with ( ‘TKD’ ) and without ( ‘SCR’ ) triple lentiviral shRNA knockdown of ZAK , MAP4K5 , and MAP2K4 , show significant differences between 1 . 0 μM PLX4720-treated and control-treated conditions ( ‘’ , p<10−10 ) . Importantly , untreated TKD cells had a significant advantage over untreated SCR HaCaT cells ( ‘’ , p<0 . 01 ) , which we interpreted to be the contribution of JNK signaling inhibition , of 17 . 6% ( blue arrow ) . Drug-treated SCR and TKD cells both had a similar degree of total colony formation advantage ( averaged as black arrow ) , as expected , since the TKD cells are not expected to have any additional suppression of JNK signaling in the presence of drug ( ‘NS’ , p=0 . 17 , Figure 3D ) . Therefore , the colony counts for these two distinct systems ( D and E ) , when taken together , show that JNK pathway inhibition accounts for approximately 17 . 6–40% and paradoxical ERK activation accounts for approximately 60–82 . 4% of the total effects of PLX4720 on tumor growth . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 02210 . 7554/eLife . 00969 . 023Figure 6—figure supplement 1 . Paradoxical MEK and ERK activation require intact Craf . Wild-type ( WT ) and isogenic Craf−/− MEFs were retrovirally transduced with HRASG12V and adenovirus E1A thereby enabling anchorage-independent growth for soft agar assays . ( A ) WT MEFs exhibit paradoxical MEK and ERK activation , effects that are significantly reduced in Craf−/− MEFs , particularly for MEK activation . ( B ) HRASG12V–transformed HaCaT cells with ( ‘TKD’ ) and without ( ‘SCR’ ) triple knockdown of ZAK , MAP4K5 , and MAP2K4 show equivalent paradoxical ERK activation . ( C ) Transformed WT and Craf−/− MEFs show equivalent expression of E1A ( sc-25 , Santa Cruz ) and RAS ( sc-32 , Santa Cruz ) . ( D ) Transformed HaCaT cells show equivalent expression of RAS . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 02310 . 7554/eLife . 00969 . 024Figure 6—figure supplement 2 . Dabrafenib fails to suppress apoptosis and phospho-JNK upregulation following UV irradiation at bioequivalent doses as compared to PLX4720 . Based upon human pharmacokinetic data and in vitro experiments , dabrafenib and PLX4720 were compared in multiple settings at bioequivalent doses ( 0 . 05 μM and 1 . 0 μM , respectively ) . ( A ) Both BRAFi suppress the growth of A375 and WM35 BRAFV600E melanoma cell lines to the same degree at these doses . ( B and C ) At these doses , dabrafenib fails to suppress UV-induced apoptosis significantly in HaCaT and SRB1 cells . ( D and E ) Likewise , dabrafenib fails to suppress phospho-JNK induction , whereas PLX4720 potently suppresses phospho-JNK induction as shown earlier . ( F ) Dabrafenib inhibits ZAK kinase with an estimated IC50 of 28 . 92 ± 2 . 23 nM , with no significant inhibition of MAP4K5 or MKK4 up to 1 μM . At 0 . 01 μM of dabrafenib , the retained activity of ZAK kinase is over 64% . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 02410 . 7554/eLife . 00969 . 025Figure 6—figure supplement 3 . Dabrafenib produces a colony formation advantage only in WT MEFs . At 0 . 05 μM , dabrafenib produce a significant growth advantage in E1A-HRASG12V- transformed WT MEFs . In E1A-HRASG12V-transformed Craf−/− MEFs , dabrafenib fails to confer a significant growth advantage , suggesting that in the absence of significant paradoxical ERK activation , dabrafenib does not have a relevant off-target effect that results in a growth advantage . DOI: http://dx . doi . org/10 . 7554/eLife . 00969 . 025 The recently updated combination trial of the BRAFi dabrafenib and MEKi trametinib shows a low 7% cSCC rate in 54 patients , ( Flaherty et al . , 2012 ) suggesting that combined MEK inhibition can reduce , but not eliminate cSCC formation , nevertheless reinforcing a role for paradoxical ERK activation ( Su et al . , 2012 ) . However , clinical trial data on dabrafenib alone at 150 mg PO BID , shows an overall aggregated cSCC rate of 6 . 1% ( Falchook et al . , 2012; Hauschild et al . , 2012; Long et al . , 2012 ) vs 22% for vemurafenib at 960 mg PO BID in several hundred patients ( Flaherty et al . , 2010; Chapman et al . , 2011; Sosman et al . , 2012; Menzies et al . , 2013 ) . We interpreted this as being reflective of differences between vemurafenib and dabrafenib , as opposed to unequivocal proof that paradoxical ERK activation is the only mechanism involved . To explore this further , we used similar assays to assess the effects of dabrafenib on apoptosis , JNK signaling , and colony formation . In stark contrast to vemurafenib , dabrafenib has little effect on apoptosis and JNK signaling at doses that are biologically equivalent based upon growth inhibition of BRAFV600E melanoma cells and human pharmacokinetic data ( Falchook et al . , 2012; Gowrishankar et al . , 2012 ) ( Figure 6—figure supplement 2A ) . Peak serum concentrations of dabrafenib at 150 mg PO BID in humans ( Falchook et al . , 2012 ) are over 50-fold lower ( 1 . 55 μM ) than mean sustained serum levels of vemurafenib ( 86 μM ) at 960 mg PO BID ( Flaherty et al . , 2010 ) , and the GI50 for the A375 melanoma cell line is less than 0 . 01 μM for dabrafenib ( Greger et al . , 2012 ) vs 0 . 50 μM for PLX4720 ( Tsai et al . , 2008 ) . Even at 0 . 05 μM , dabrafenib did not significantly impact UV-induced phospho-JNK upregulation or apoptosis in HaCaT and SRB1 cells ( Figure 6—figure supplement 2B–E ) . We profiled dabrafenib activity against ZAK , MKK4 , and MAP4K5 , and found that ZAK is a significant off-target kinase for dabrafenib as well , but at 0 . 01 μM , over 64% of activity is retained ( Figure 6—figure supplement 2F ) . Neither MKK4 nor MAP4K5 is substantially inhibited by dabrafenib up to 1 μM ( Figure 6—figure supplement 2F ) . Using transformed WT and Craf-deficient MEFs in soft agar assays , we also showed that dabrafenib enhanced colony formation in WT MEFs , but not in Craf-deficient MEFs ( Figure 6—figure supplement 3 ) . Our results suggest that while both dabrafenib and vemurafenib cause equivalent paradoxical ERK activation in BRAF-wild-type cells ( Figure 6—figure supplement 1A–B ) , only vemurafenib confers a significant colony formation advantage in Craf-deficient cells that have no significant paradoxical MEK/ERK activation , implicating off-target effects as a key difference between the two drugs with respect to cSCC development ( Menzies et al . , 2013 ) . We have discovered an unexpected and novel effect of the BRAFi PLX4720 and vemurafenib in inhibiting apoptosis in vitro and in vivo through the ERK-independent suppression of JNK signaling ( Tournier et al . , 2000 ) . Our studies implicate the off-target binding and inhibition of these compounds to ZAK primarily ( Figure 3—figure supplements 1–3 ) , with additional contributions of MKK4 ( MAP2K4 ) and MAP4K5 inhibition , thus implicating inhibition of JNK signaling at all three upstream tiers of MAP kinase signaling ( Figure 3—figure supplement 4 ) . Although MKK4 knockdown alone could suppress UV-induced apoptosis and phospho-JNK induction by up to 27 . 3% ( Figure 3—figure supplement 2 ) , this is expected , given that ZAK signals through MKK4 and MKK7 ( Gross et al . , 2002 ) ( Figure 2H , I , 3E , 4H ) and MKK4 is important ( with MKK7 ) for full JNK activation ( Tournier et al . , 2001; Haeusgen et al . , 2011 ) . Additionally , UV-mediated induction of NOXA is suppressed in cell lines , primary NHEKs , and in vivo , indicating that this BCL2 family member may be a critical effector of apoptosis in this context ( Figures 1G and 4I ) . In chronically-irradiated Hairless mice , development of well-differentiated papillomas and cSCC is substantially accelerated by PLX4720 treatment without the need for Ras mutation and with a dramatic reduction in latency by at least 10 weeks ( Figure 5E , Figure 5—figure supplement 1 ) . While there is enrichment for RAS mutations in human cSCC arising in vemurafenib-treated patients vs controls ( Oberholzer et al . , 2011; Su et al . , 2012 ) , up to 30–40% of these lesions do not have RAS mutations . Our novel mechanism of BRAFi-mediated apoptosis suppression is the off-target inhibition of several kinases in the JNK pathway , which is independent from , and compatible with , paradoxical ERK–dependent mechanisms ( Su et al . , 2012 ) ( Figure 6A ) . Importantly , our approach ( Figure 6B–E ) has not only allowed us to quantify the contribution of the effect on apoptosis ( 17 . 6–40% ) vs paradoxical ERK activation ( 60–82 . 4% ) , but also shows that the growth advantage conferred by BRAFi in BRAF-WT cells is not accounted for entirely by paradoxical ERK activation . Our results have also shown that there are significant differences between the BRAFi vemurafenib and dabrafenib ( Figure 6—figure supplements 2 , 3 ) with respect to these off-target effects in cells ( even though their relative selectivities for BRAF over ZAK are similar ) and this may , in part , explain why they differ in rates of cSCC ( Flaherty et al . , 2010; Chapman et al . , 2011; Falchook et al . , 2012; Hauschild et al . , 2012; Long et al . , 2012; Sosman et al . , 2012 ) . At present it is unclear why ZAK appears to be a common off-target kinase and whether structural similarities with other kinases may explain this ( Sauter et al . , 2010; Wong et al . , 2013 ) . ZAK has been previously studied in the context of bacterial toxin and doxorubicin-mediated cytokine signaling ( Jandhyala et al . , 2008; Sauter et al . , 2010; Stone et al . , 2012; Wong et al . , 2013 ) , cardiac ( Huang et al . , 2004 ) and ischemic stress responses ( Su et al . , 2012 ) , and in cellular responses to ionizing radiation ( Gross et al . , 2002; Tosti et al . , 2004; Vanan et al . , 2012 ) . It is widely expressed across tissues including epidermis , but most prominently in heart , liver , and muscle ( Abe et al . , 1995; Miyata et al . , 1999; Liu et al . , 2000; Bloem et al . , 2001; Gross et al . , 2002; Su et al . , 2004 ) , and has purported tumor suppressive roles in lung cancer ( Yang et al . , 2010 ) and tumor promoting ones in partially transformed mouse skin epidermal cells ( Cho et al . , 2004 ) . ZAK is a MAP3K that is upstream of both JNK and p38 signaling ( Gross et al . , 2002; Tosti et al . , 2004; Jandhyala et al . , 2008; Cheng et al . , 2009; Stone et al . , 2012; Wong et al . , 2013 ) and signals to JNK through MKK4 and MKK7 ( Gross et al . , 2002 ) ( Figures 2H , I , 3E , 4H ) . Accordingly , macrophages derived from ZAK-deficient mice have profound defects in activation of both JNK and p38 signaling following doxorubicin exposure ( Wong et al . , 2013 ) . In the setting of UV-induced apoptosis as we have examined here , JNK activity is the major driver of apoptosis ( Derijard et al . , 1994; Chen et al . , 1996; Tournier et al . , 2000 ) , also by virtue of the fact that phospho-p38 induction by UV is inconstant ( Figures 1F , H , 2B , 4H ) ; although where it is induced , PLX4720/vemurafenib treatment suppresses it ( Figure 1F ( SRB1 , HaCaT cells ) , Figures 1H , 2B ) . These results are consistent with the model ( Figure 3—figure supplement 4 ) that ZAK signals to both JNK and p38 , but is principally necessary for activating JNK in stress-induced apoptosis . Because off-target kinases in the JNK pathway are affected by vemurafenib/PLX4720 , one expects that these kinases would be affected in all cells regardless of BRAF status . Indeed , melanoma cells expressing BRAFV600E also exhibit suppression of JNK activity following irradiation ( Figure 1H ) . However , in BRAFV600E-expressing melanoma cells , the effect of blocking BRAF activity alone clearly dominates , because these cells are exquisitely dependent upon BRAF activity ( Tsai et al . , 2008 ) . Therefore , although off-target kinases are inhibited , the cellular context of dependence on particular kinases is still highly relevant and likely dictates the outcome . Our findings suggest a tumor suppressive role for JNK signaling in the context of drug-induced cSCC , though the role of JNK in cancer is highly context-dependent and is partly related to differing functions of the individual isoforms and partial redundancy ( Tournier et al . , 2000 ) . Nonetheless , there is ample in vivo evidence showing that JNK can function in a tumor suppressive role . Genetically-engineered mice lacking Jnk1 and Jnk2 have increased ( She et al . , 2002 ) and decreased ( Chen et al . , 2001 ) susceptibility , respectively , to chemical carcinogenesis in skin , though these mice also have opposite defects in epidermal differentiation ( Weston et al . , 2004 ) . In mouse models , lack of Jnk1/2 activity suppresses Ras-driven tumorigenesis in lung ( Cellurale et al . , 2011 ) and promotes it in Ras-driven and Trp53-deficient breast cancer models ( Cellurale et al . , 2010 , 2012 ) . In the context of Pten-deficiency , loss of Jnk1/2 or Mkk4/Mkk7 promotes aggressive prostate adenocarcinoma ( Hubner et al . , 2012 ) . Importantly , the effects of JNK on cancer are not always tumor cell autonomous , as JNK activity supports a pro-tumorigenic inflammatory microenvironment in hepatocellular carcinoma ( Das et al . , 2011 ) . Our results have important clinical implications and suggest careful consideration of combining certain BRAFi with therapeutic modalities that induce apoptosis such as radiation or chemotherapy , particularly with respect to off-target tissues ( keratinocytes in skin ) . We have shown that off-target inhibition of kinases , even at higher IC50s , can contribute biologically significant effects , particularly if they are in the same pathway . Finally , our results show that kinase inhibitors must be considered in terms of their entire spectrum of activity , which can dramatically affect pathways distinct from those affected by inhibition of the intended target . All studies were conducted under institutionally-approved IRB ( LAB08-0750 ) and ACUF ( 06-09-06332 ) protocols for the protection of human and animal subjects , respectively . Cutaneous SCC cell lines ( SRB1 , SRB12 , COLO16 ) were obtained from Jeffrey N Myers ( MD Anderson ) , HaCaT cells from Norbert Fusenig ( German Cancer Research Center ) , and WM35 and A375 melanoma cell lines from Michael Davies ( MD Anderson ) . The cell lines were validated by STR DNA fingerprinting using the AmpFℓSTR Identifiler kit according to manufacturer instructions ( Applied Biosystems , Grand Island , NY ) . The STR profiles were compared to known ATCC fingerprints ( ATCC . org ) , to the Cell Line Integrated Molecular Authentication database ( CLIMA ) version 0 . 1 . 200808 ( http://bioinformatics . istge . it/clima/ ) and to the MD Anderson fingerprint database . The STR profiles matched known DNA fingerprints ( HaCaT ) or were unique ( SRB1 , SRB12 , COLO16 ) . The cells were cultured in DMEM/Ham’s F12 50/50 ( Cellgro ) supplemented with 10% Fetal Bovine Serum ( FBS ) ( Sigma ) , glutamine , and Primocin ( Invivogen ) . NHEKs ( Lonza ) were cultured in media according to manufacturer’s instructions . Irradiation was performed using an FS40 sunlamp dosed by an IL1700 radiometer . Following irradiation , cells were treated with PLX4720 ( Plexxikon ) , vemurafenib ( Selleck Chemicals ) or DMSO ( 1:2000 ) . Primary antibodies ( Cell Signaling ) used for Western blot analysis included p53 ( 2527P , clone 7F5 ) , phospho-/total p44/42 MAPK ( 4370S , cloneD13 . 14 . 4E/9102S ) , phospho-/total p38 MAPK ( 4511S , clone D3F9/9212S ) , phospho-/total JNK ( 4668S , clone 81E11/9252S ) , BIM ( 2933 , clone C34C5 ) , MCL1 ( 5453P , clone D35A5 ) , cleaved caspase-3 ( 9661L , clone D175 ) , phospho-/total MKK7 ( 4171S/4172S ) , phospho-/total MKK4 ( 9156S/9152S ) , phospho-/total MEK ( 9121S/9122 ) , MAP4K5 ( ab56848; Abcam ) and NOXA ( mA1-41000; Thermo Scientific ) . GAPDH ( 21 , 182 , clone 14C10; Cell Signaling ) and beta-actin ( A5060; Sigma ) were probed to ensure even loading of protein samples . Immunohistochemistry was performed for phospho-JNK ( V7931; Promega ) and cleaved caspase-3 ( Cell Signaling as above ) . Antibody against ZAK was generously provided by R Ruggieri ( Feinstein Institute for Medical Research ) . TMRE ( Invitrogen ) was used as a measure of mitochondrial membrane potential , Annexin V-FITC or Annexin V-APC ( Invitrogen ) as a probe for apoptosis , and Sytox Blue ( Invitrogen ) as an indicator for dead cells . At 24 hr post-irradiation , floating and adherent cells were collected and stained with TMRE , Annexin V and Sytox Blue . Data was collected and analyzed using a flow cytometer ( Fortessa , Becton Dickinson ) and FlowJo Software ( Tree Star ) . Data were calculated and charts were plotted using GraphPad Prism 5 software . Cell were lysed in standard buffers with protease inhibitors ( Roche ) and phosphatase inhibitors ( Santa Cruz ) with extracts run on SDS/polyacrylamide gels and transferred to Immobilon-P transfer membrane ( Millipore ) . Blots were blocked in TBST ( 10 mM Tris-HCL pH8 , 150 mM NaCl , 0 . 5% Tween ) with milk or BSA , probed with primary antibodies , corresponding HRP-conjugated secondary antibodies , and signals detected using ECL kit ( Amersham ) . Cutaneous squamous cell carcinomas biopsied from patients treated with or without BRAF-inhibitor were obtained either under clinical trials ( Roche ) or separate IRB approval ( LAB08-0750 ) . Staining levels were quantified by counting positively labeled cells and dividing by the total area of the tumor tissue within each sample . To measure tumor areas , all samples were photographed , tumor cells outlined , and total pixel numbers calculated using included image analysis tools in Adobe Photoshop and standardized to a hemacytometer to convert to mm2 . To measure apoptosis in irradiated skin , pyknotic or dyskeratotic epidermal keratinocytes were counted and normalized to length ( mm ) of epidermis . PLX4720 and vemurafenib were prepared in DMSO and tested in duplicate at four concentrations ( 50 nM , 200 nM , 1000 nM , 10 μM ) against a panel of 38 kinases using a quantitative competitive binding assay ( KINOMEscan , San Diego , CA ) . Average percent inhibition was reported . Estimated Kd values were derived by averaging pointwise estimates calculated using a transformed Hill equation at each concentration of drug . In vitro kinase assays were performed using human full-length ZAK ( MBP substrate , ATP 2 . 5 μM ) , amino acids 33-end MKK4 ( JNK1 substrate , ATP 0 . 1 μM ) , and full length MAP4K5 ( MBP substrate , ATP 10 μM ) ( Reaction Biology ) . Assays for BRAFV600E and ASK1 against vemurafenib were run in parallel revealing IC50s of 31 . 6 ± 2 . 9 nM for BRAFV600E and no significant inhibition of ASK1 , as previously reported ( Bollag et al . , 2010 ) . Lentiviral shRNA knockdown was accomplished using standard lentiviral methods using 293T cells and psPAX2/pVSV . G packaging plasmids . shRNA clones against ZAK ( clones V2LHS_239842 , V3LHS_336769 ) , MKK4 ( clones V3LHS_646205 , V3LHS_386825A ) , and MAP4K5 ( clones 196277A , 334084 ) , as well as a non-silencing shRNA were obtained from Open Biosystems in the GIPZ vector . Following transduction , cells were puromycin-selected and FACS sorted to obtain cells with high-level suppression . Degree of mRNA suppression was quantified by qPCR using Taqman probes using internally controlled ( 2-color , same well ) GAPDH probes to ensure proper normalization . ZAK ( T82Q ) mutant was generated in the pcDNA3 mammalian expression vector . HaCaT cells were electroporated using the Neon transfection system 24 hr prior to irradiation . Transfection efficiencies were estimated to be 70–80% by GFP fluorescence . Wild-type C57BL/6 mice , 5–8 weeks old , were pretreated with PLX4720 in 5% DMSO in 1% methylcellulose for 2–4 days at 40–80 mg/kg twice a day or control 5% DMSO in 1% methylcellulose by oral gavage . The mice were shaved and depilated ( Nair ) 24 hr prior to irradiation with a solar simulator ( Oriel ) dosed at 10 kJ/m2 of UVB . Epidermis was harvested and protein extracts run on Western blots and probed as above . For chronically-irradiated Hairless mice , 3–4 week old males were irradiated thrice weekly for a total weekly dose of 12 . 5 kJ/m2 UVB ( solar simulator , Oriel ) . At 72 days , PLX4720 treatment was started using drug-impregnated chow ( Plexxikon ) with vehicle chow in the control cohort . Following plating of bottom agar ( 0 . 6% Bacto Agar ) with media and appropriate amounts of drug , 2500 to 10 , 000 cells per well ( transformed WT , Craf−/− MEFs; transformed HaCaT SCR and TKD cells ) were embedded in top agar ( 0 . 3% ) and plated in 24-well plates . Control or drug-treated media was replaced every 48 hr for 4–6 weeks . The plates were stained with 1% crystal violet and colonies counted by bright-field microscopy . All data are represented as means ± SEM . All experiments were performed in triplicate at least . Student’s t-test was used for comparison between two groups . p≤0 . 05 was considered significant .	Over 50% of melanomas , a highly lethal form of skin cancer , carry mutations in a gene called BRAF . The BRAF gene encodes an enzyme that helps to regulate the proliferation of cells , but mutations in this gene lead to the excessive proliferation that is seen in cancer . Clinical trials have shown that a drug called vemurafenib can be used to treat patients who carry the mutated BRAF genes and go on to develop melanoma , but around one fifth of these patients developed another type of skin cancer called cSCC ( cutaneous squamous cell carcinoma ) . The cSCC tumors often develop in areas where the sun has damaged the patient’s skin , and it is thought that their growth is then accelerated by vemurafenib activating another enzyme , ERK , which causes the excessive proliferation of skin cells . Vin et al . have now found that vemurafenib might also cause cSCC tumors by blocking another signaling pathway . The experiments were performed in human cells and also in mice , and the results were then verified in human cSCC samples . Cells that are exposed to UV radiation usually die , but when treated with vemurafenib , some 70% of the cells that would have died instead survived . The stress from the UV radiation activates the JNK signaling pathway , which causes the irradiated cells to die . However , Vin et al . found that cSCC cells had very low levels of JNK signaling because treatment with vemurafenib had the unintended effect of inhibiting three enzymes that are needed to fully activate the JNK signaling pathway . Vin et al . estimate that suppression of JNK signaling and cell death is responsible for about 17 . 6 to 40% of the effect on cSCC growth seen in melanoma patients , with activation of the ERK pathway accounting for the rest . These unexpected findings suggest that combining vemurafenib treatment with radiation or chemotherapy should be done with caution as these effects could affect their efficacy . It also suggests that future drugs should be designed in a way that avoids these types of effects by making sure they do not inhibit important ‘off-target’ enzymes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2013	BRAF inhibitors suppress apoptosis through off-target inhibition of JNK signaling
Cyanogenic glucosides are among the most widespread defense chemicals of plants . Upon plant tissue disruption , these glucosides are hydrolyzed to a reactive hydroxynitrile that releases toxic hydrogen cyanide ( HCN ) . Yet many mite and lepidopteran species can thrive on plants defended by cyanogenic glucosides . The nature of the enzyme known to detoxify HCN to β-cyanoalanine in arthropods has remained enigmatic . Here we identify this enzyme by transcriptome analysis and functional expression . Phylogenetic analysis showed that the gene is a member of the cysteine synthase family horizontally transferred from bacteria to phytophagous mites and Lepidoptera . The recombinant mite enzyme had both β-cyanoalanine synthase and cysteine synthase activity but enzyme kinetics showed that cyanide detoxification activity was strongly favored . Our results therefore suggest that an ancient horizontal transfer of a gene originally involved in sulfur amino acid biosynthesis in bacteria was co-opted by herbivorous arthropods to detoxify plant produced cyanide . Plants have developed a remarkable diversity of chemical defenses to deter herbivores from feeding . Cyanogenesis is one of the most ancient and widespread of these defenses , and more than 2500 plant species are known to synthesize cyanogenic glucosides and cyanolipids as phytoanticipins ( constitutive defense compounds present in the plant prior to herbivore attack ) . Upon tissue disruption by herbivore feeding , cyanogenic glucosides are degraded by the plant β-glucosidases and α-hydroxynitrile lyases , which results in the release of toxic hydrogen cyanide ( HCN ) and other toxic products such as their aglycones ( Gleadow and Woodrow , 2002; Poulton , 1990; Spencer , 1988; Zagrobelny et al . , 2004; Figure 1 ) . The released cyanide is a potent inhibitor of the mitochondrial respiratory chain and has a disruptive effect on various metabolic pathways ( Solomonson , 1981 ) , thus providing a broad defense against generalist herbivores . 10 . 7554/eLife . 02365 . 003Figure 1 . Schematic overview of the cysteine biosynthesis pathway in Metazoa ( purple ) and Plants/Bacteria ( blue ) , the release of HCN during plant cyanogenesis ( green ) and the main HCN detoxification pathway in arthropods ( red ) . The two reactions catalyzed by the gene product of the Tu-CAS gene are marked by an orange background and are indicated by CYS and CAS . CAS detoxifies cyanide by incorporation into cysteine forming β-cyanoalanine , which can be further metabolized by nitrilases . CYS catalyzes the second step of the cysteine synthesis pathway in bacteria and plants , after serine is converted to O-acetylserine . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 003 In their arms race with plants , arthropods have evolved several mechanisms to overcome plant cyanogenesis . A well-documented case of co-evolution is found in insect lepidopteran specialists that sequester the ingested cyanogenic glucosides in their own defense against predators . Remarkably , cyanogenic compounds have become so crucial for some species such as burnet moths ( Zygaenidae ) that they not only sequester , but also synthesize these compounds de novo by convergent evolution of the biosynthetic pathways ( Jensen et al . , 2011 ) . Next to sequestration , other mechanisms have evolved to cope with the toxic effects of HCN , such as avoiding the ingestion of cyanogenic compounds or detoxifying HCN upon plant release ( Despres et al . , 2007 ) . In animals , HCN is thought to be detoxified by two main pathways . The enzyme rhodanese converts cyanide into thiocyanate , but this biochemical reaction is not very common and thought to be inefficient ( Beesley et al . , 1985; Davis and Nahrstedt , 1985; Long and Brattsten , 1982 ) . Alternatively , the conversion of HCN and cysteine into β-cyanoalanine and sulfide has been suggested as the main detoxification pathway in arthropods ( Figure 1 ) . This is supported by several biochemical surveys showing a correlation between β-cyanoalanine synthesis and HCN exposure in lepidopteran species tolerant to HCN ( Meyers and Ahmad , 1991; Stauber et al . , 2012 ) . However , the enzyme that catalyzes this crucial reaction in arthropods has not been identified to date . The conversion of HCN into β-cyanoalanine by an enzyme called β-cyanoalanine synthase ( CAS ) has been best studied in bacteria and plants that need to protect themselves from HCN during the synthesis of cyanogenic glucosides or ethylene . The enzymes responsible for CAS activity also have cysteine synthase activity ( CYS ) and are referred to as CYS or CAS depending on their substrate specificity . CYS catalyzes the conversion of O-acetylserine into cysteine , an essential final step in the cysteine biosynthesis pathway unique for plants and bacteria ( Bonner et al . , 2005 ) . Animals synthesize cysteine by a different pathway and use related enzymes such as cystathionine-β-synthase ( CBS ) and cystathionine-γ-lyase ( CGL ) , which form together with CAS and CYS a group of pyridoxal-5′-phosphate dependent enzymes ( Finkelstein et al . , 1988; Figure 1 ) . Here we identify the enzyme responsible for detoxification of cyanide to β-cyanoalanine in a spider mite and show that this enzyme is horizontally acquired from bacteria and is also widely distributed in Lepidoptera . In order to gain a better insight into arthropod defenses against plant cyanogenesis , we used the two-spotted spider mite T . urticae as a generalist herbivore model in a host plant adaptation experiment . This species is one of the most polyphagous arthropod pests known to date , and feeds on more than 1100 plant species from more than 140 plant families , including many cyanogenic plants ( Grbic et al . , 2011 ) . We transferred a spider mite strain reared on acyanogenic bean plants ( Phaseolus vulgaris ) to a cultivar of Phaseolus lunatus containing high levels of well characterized cyanogenic glucosides such as linamarin and lotaustralin ( Ballhorn et al . , 2006; Jones , 1998; Wybouw et al . , 2012 ) and allowed the strain to adapt to this host plant for more than 30 generations . Gene expression differences between mites feeding on P . vulgaris and on P . lunatus were then determined using genome-wide microarrays ( Dermauw et al . , 2013 ) . In contrast to the broad response previously detected after host plant changes from Fabaceae to Solanaceae and Brassicaceae ( Grbic et al . , 2011; Dermauw et al . , 2013 ) , only a limited set of 28 genes ( absolute fold change ≥2 , Benjamini-Hochberg corrected p-value <0 . 05 ) was found differentially expressed between the parental and adapted lines ( Supplementary file 1 ) . Within this small set of genes , 18 had an increased expression in the adapted line , while 10 exhibited a lower expression . The most differentially expressed genes after transfer to cyanogenic lima bean encode closely related , small cytoplasmic proteins of unknown function . They were also seen in the transcriptional response after transfer to tomato suggesting that these genes could be part of a broad general stress response of T . urticae ( Dermauw et al . , 2013 ) . Among the genes with an increased expression were three cytochrome P450 genes , known to respond readily to host plant changes ( Grbic et al . , 2011; Dermauw et al . , 2013 ) . Moreover , we identified a gene ( tetur10g01570 , Tu-CAS ) encoding a predicted cytosolic protein with high similarity to bacterial cysteine synthases ( Conserved Domain Database ( CDD ) : COG0031 ) . The reported microarray data have been deposited at the Gene Expression Omnibus ( Wybouw et al . , 2013 ) . Within chelicerates , we only detected close homologues of Tu-CAS in two closely related tetranychid mites . By sequencing PCR amplicons ( see below , this section ) and a tBLASTn-search in a published transcriptome ( Liu et al . , 2011 ) , CAS genes were identified in Tetranychus evansi and Panonychus citri , respectively . A tBLASTn-search in published genomes of mesostigmatid mites and of ticks ( Metastigmata ) did not reveal homologues ( Supplementary file 2 ) . Broadening the search to arthropods by tBLASTn-searches in NCBI databases and additional arthropod genome portals ( Supplementary file 2 ) , we only detected close homologues of Tu-CAS in lepidopteran genomes ( Bombyx mori , Danaus plexippus , Heliconius melpomene , Manduca sexta and Plutella xylostella ) ( Figure 2 ) . On average , Tu-CAS showed 75% similarity with lepidopteran protein sequences , and genes encoding these proteins were intronless in both mites and insects . By searching additional NCBI EST-databases and published lepidopteran transcriptomes , a total of 20 ( complete and partial ) homologous sequences were identified in arthropods . 10 . 7554/eLife . 02365 . 004Figure 2 . Panel A: Phylogenetic analysis of β-substituted alanine synthases , showing arthropod sequences nested within bacterial cysteine synthases . The fungal CYS , metazoan and fungal CBS as well as the plant , oomycete and nematode CYS and CAS sequences are marked with a different color . The two branches of nematode sequences , marked as cysl-1 and cysl-2 , include the sequences coded by the two genes previously characterized in C . elegans ( Budde and Roth , 2011 ) . The CYS and CAS groups within Plantae represent plant protein sequences with CYS and CAS activity , respectively ( Yamaguchi et al . , 2000 ) . The asterisk represents a CYS sequence of the mealybug P . citri acquired by horizontal gene transfer from its endosymbiont ( Husnik et al . , 2013 ) . Panel B: Detailed view of the bacterial CYS sequences showing the embedded sequences of tetranychid mites and Lepidoptera . In both panels support values of only important nodes are shown . The scale bar represents 0 . 5 and 0 . 2 substitutions per site in panel A and panel B , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 00410 . 7554/eLife . 02365 . 005Figure 2—figure supplement 1 . MUSCLE alignment of cysteine synthases and β-cyanoalanine synthases discussed in this study . Residue numbering is shown for the G . max CAS sequence ( Yi et al . , 2012 ) . The five residues forming an active site loop are marked by asterisks . The amino acid and pyridoxal-5′-phosphate binding sites are indicated by a green and a blue background respectively , while the Schiff base Lys is indicated by a red background ( Bonner et al . , 2005; Yi et al . , 2012 ) . The unique 9 amino acid insertion seen in arthropod enzymes and their closest bacterial homologues is highlighted in orange . The three residues that convert G . max CYS into CAS by creating a triple mutant ( Yi et al . , 2012 ) are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 005 A phylogenetic reconstruction of these arthropod proteins with CYS , CAS and CBS enzymes of plants , fungi , oomycetes , bacteria and Metazoa indicated that these homologous arthropod sequences might be monophyletic . They were embedded with high node support within bacterial cysteine synthase sequences , indicative of a horizontal gene transfer ( Figure 2 ) . The most closely related sequences are from bacteria that belong to the α- and β-Proteobacteria , two of which are Methylobacterium species , free-living epiphytic bacteria known to establish endophytic colonies . These species are reported to be transferred from the plant to phytophagous insects and to survive inside arthropod hosts ( Kutschera , 2007; Rampelotti-Ferreira et al . , 2010 ) , facilitating a potential horizontal gene transfer . Alternatively , Proteobacteria are known endosymbionts of arthropods and often reside in the reproductive organs for vertical transmission to following generations ( Wernegreen , 2002 ) . Because of this intimate relationship , successful horizontal gene transfer is more likely to occur ( Hotopp , 2011 ) . The mite and insect sequences formed a branch in bacterial cysteine synthase enzymes , some of which have documented dual CYS and CAS activities ( Omura et al . , 2003 ) ( Figure 2 ) . The arthropod protein sequences and their closest bacterial homologues shared a unique 9 amino acid insertion not present in cysteine synthases of other organisms , but the residues known to be crucial for substrate and cofactor binding in plants and bacteria showed conservation ( Bonner et al . , 2005; Yi et al . , 2012 ) . The lysine residue ( Lys95 , Glycine max CAS numbering , Yi et al . , 2012 ) that forms a Schiff base linkage to the cofactor pyridoxal-5′-phosphate was also conserved in arthropods ( Figure 2—figure supplement 1 ) . To exclude the possibility that the Tu-CAS sequence was derived from contaminating bacterial DNA , we examined its position in the T . urticae genome into more detail . Tu-CAS is located on a 3 Mb large scaffold ( scaffold 10 ) and is flanked by typical eukaryotic genes , tetur10g01550 and tetur10g01580 . Both genes contain introns with splice sites confirmed by EST or RNA-seq data ( Grbic et al . , 2011 , GenBank: LIBEST_025606 , Figure 3 ) . Tetur10g01580 encodes a nudix hydrolase ( CDD: cl00447 ) highly similar to other arthropod nudix hydrolases ( BLASTp hits with E-value <1e−45 ) . To rule out that the linkage of Tu-CAS to these spliced genes is a genome assembly artefact , we took several independent genomic approaches . First , we remapped the T . urticae Sanger reads used for the genome assembly of the London strain of T . urticae ( Grbic et al . , 2011 ) and examined the Illumina read coverage of the Tu-CAS region in two re-sequenced strains ( EtoxR and Montpellier ) that are geographically distinct from the London strain ( Grbic et al . , 2011; Van Leeuwen et al . , 2012 ) . Neither Sanger nor Illumina-reads revealed any inconsistencies in the Tu-CAS region in any of these strains ( Figure 3 ) . Second , using a PCR approach , we amplified a 6 kb genomic region bracketing Tu-CAS with the surrounding spliced genes ( tetur10g01550 and tetur10g01580 , Figure 4 , Figure 4—figure supplement 1 ) . Using a similar strategy we also amplified a 6 kb region in the closely related spider mite T . evansi ( Figure 4 , Figure 4—figure supplement 1 ) and a nucleotide dot plot between the amplified region of these two species showed clear synteny and the absence of discontinuity around the CAS gene ( Figure 4 ) . This would not be expected by bacterial contamination of either genome . Last , gene compositions of T . urticae and bacteria were analyzed by determining both the GC-content at the synonymous third codon position ( GC3 ) and the overall GC-content ( GC ) of the genes to look if amelioration occurred . Amelioration is the process by which the DNA composition of the newly acquired gene becomes homogenized to match the composition of the recipient genome ( all genes in the recipient genome are subject to the same mutational processes ) ( Lawrence and Ochman , 1997 ) . Indeed , the GC/GC3 content of Tu-CAS was most similar to the GC/GC3 content of genes from the T . urticae genome , and quite distinct from the GC/GC3 content of genes ( including the CYS/CAS genes ) from the three annotated bacterial genomes in the closest sister clade to the apparent monophyletic arthropod clade ( Figure 5 ) . Taken together , these data provide strong evidence that Tu-CAS is a sequence integrated in the genome of T . urticae and does not represent bacterial contamination . Sequence data is available at Genbank ( accession numbers: KF981736 and KF981737 ) . 10 . 7554/eLife . 02365 . 006Figure 3 . Coverage plot of Tu-CAS ( tetur10g01570 ) and its surrounding region in the genome of T . urticae . Gene models of Tu-CAS and its neighboring genes are depicted as follows: blue and red rectangles represent coding sequences and untranslated regions , respectively , while introns are shown as dashed lines . ( + ) and ( − ) represent the forward and reverse strand , respectively . Underneath the gene models , indicated in green , are the length and position of amplicons obtained by PCR ( Figure 4 ) . Next , an alignment of paired-end Sanger reads ( and corresponding coverage plot ) with the T . urticae genome of the London strain is displayed . Paired-end Sanger reads for which both reads are mapped in or extend nearby the indicated region are denoted by thin lines to show pair connections ( shown are all paired-end Sanger reads that were produced from 2 . 5 , 8 . 5 , and 35 . 5 kb libraries used for assembly of the T . urticae genome [Grbic et al . , 2011] ) . The Sanger reads coverage plot is followed by coverage plots of Illumina-reads from genomic DNA sequencing of the Montpellier and EtoxR strain of T . urticae ( Grbic et al . , 2011; Van Leeuwen et al . , 2012 ) . The coverage plot at the bottom shows Illumina RNA-seq read coverage produced from adult T . urticae polyA selected RNA ( Grbic et al . , 2011 ) . Numbers between brackets represent the sequence depth . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 00610 . 7554/eLife . 02365 . 007Figure 4 . Nucleotide dot plot of the PCR amplified genomic region bracketing T . urticae and T . evansi CAS with adjacent intron-containing eukaryotic genes . The dot-plot was constructed with 95% identity in a 21 bp window , with the T . evansi and the T . urticae amplified region on the y- and x-axis , respectively . From the T . urticae region , the gene models and their genomic positions on the 10th scaffold are specified below the x-axis . The ( + ) and ( − ) signs represent the forward and reverse strand , respectively . Blue and black bars indicate exons and introns respectively , while the untranslated regions are depicted as red bars . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 00710 . 7554/eLife . 02365 . 008Figure 4—figure supplement 1 . Agarose gel of PCR products , bracketing CAS with adjacent eukaryotic genes in T . urticae and T . evansi . λ: lambda DNA , digested with Pstl; A: tetur10g01550—tetur10g01570 fragment in T . urticae; B: tetur10g01570—tetur10g01580 fragment in T . urticae; C; tetur10g01550—tetur10g01570 fragment in T . evansi; D: tetur10g01570—tetur10g01580 fragment in T . evansi . Primers used for the amplification of the fragments are listed in Supplementary file 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 00810 . 7554/eLife . 02365 . 009Figure 5 . Graph showing the GC/GC3 gene contents of T . urticae and putative bacterial donor species . The GC/GC3 contents of the T . urticae genome and the three annotated bacterial genomes of bacteria residing in the sister clade closest to the apparent monophyletic arthropod clade ( Figure 2 ) are shown . The GC/GC3 content of the specific CYS/CAS gene of each organism ( T . urticae in red , bacteria in blue ) is highlighted . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 009 In order to functionally test whether arthropod enzymes are active and still able to catalyze both reactions after horizontal gene transfer , we recombinantly expressed Tu-CAS in Escherichia coli and obtained mg quantities after affinity purification . Subsequent biochemical assays confirmed that Tu-CAS catalyzes both reactions ( Figure 6 ) . In order to determine which of the two reactions ( cysteine synthesis , CYS and β-cyanoalanine synthesis , CAS ) was favored by the Tu-CAS enzyme , we calculated the ratio of the specificity constants for each reaction . The specificity constant kcat/Km defines , at any concentration , the specificity of an enzyme for a particular substrate . The CAS/CYS ratio of the specificity constants was 33 . 7 , showing that CAS activity was strongly favored over CYS activity ( Table 1 ) . ( The ratio was calculated from the respective values of Vmax/Km , as kcat = Vmax/[E] and as the reactions were measured with the same enzyme preparation . ) This preferred CAS activity is typical of known plant CAS enzymes and clearly different from CYS enzymes ( Table 1; Yamaguchi et al . , 2000; Wada et al . , 2004; Bogicevic et al . , 2012; Yi et al . , 2012 ) . Enzymatic activity was dependent on pyridoxal-5′-phosphate as a cofactor , and the substrate-dependent formation of β-cyanoalanine was further confirmed by thin layer chromatography ( TLC ) and LC-MS ( Figure 7 ) . 10 . 7554/eLife . 02365 . 010Figure 6 . The two reactions catalyzed by recombinant Tu-CAS ( cysteine synthase , CYS and β-cyanoalanine synthase , CAS ) , showing the kinetic plots and calculated Vmax and Km values . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 01010 . 7554/eLife . 02365 . 011Table 1 . Specificity constants for the two activities of CYS-like enzymes ( cysteine synthase , CYS and β-cyanoalanine synthase , CAS ) DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 011CAS reactionCYS reactionCAS/CYSActivity ( s−1 ) Km ( mM ) Specificity constant ( mM−1 . s−1 ) Activity ( s−1 ) Km ( mM ) Specificity constant ( mM−1 . s−1 ) Ratio of specificity constantsTetranychus urticae CAS2 . 1350 . 3126 . 84†0 . 6463 . 170 . 203†33 . 7Arabidopsis thaliana CAS2 . 660 . 14192 . 08 . 030 . 25076Glycine max CAS38 . 90 . 81481 . 828 . 870 . 205234Glycine max CYS0 . 210 . 300 . 757 . 53 . 615 . 970 . 044Corynebacterium glutamaticum CYSn . dn . d–435762†–Lactobacillus casei CYSn . dn . d–890 . 6148†–n . d . : not determined . *data as Vmax in µmol . min−1mg−1 . †data as Vmax/Km in µmol . min−1mg−1mM−1 . CAS activity was measured with cysteine as substrate while CYS activity was measured with O-acetylserine as substrate . Data for the plants A . thaliana and G . max were obtained in Yamaguchi et al . ( 2000 ) and Yi et al . ( 2012 ) , respectively , while data for the bacteria C . glutamaticum and L . casei were retrieved from Wada et al . ( 2004 ) and Bogicevic et al . ( 2012 ) , respectively . 10 . 7554/eLife . 02365 . 012Figure 7 . Panel A: Formation and accumulation of β-cyanoalanine by recombinant Tu-CAS as visualized by TLC analysis . Controls; C1: no cysteine control , C2: no cyanide control , C3: no enzyme control . Time course 0–60 min after adding 3 . 75 µg of recombinant Tu-CAS . Standards; BCA: 5 µg β-cyanoalanine , CST: 5 µg cysteine . Panel B: LC-MS identification of β-cyanoalanine as a reaction product of Tu-CAS . The enzymatically produced β-cyanoalanine was scraped from silica plates after TLC separation of reaction mixtures , and was analyzed by LC-MS . β-cyanoalanine was identified in the reaction mixture on the basis of a similar elution time on LC and the characteristic ion of m/z = 113 which is [M-H]- as compared to the standard . ( The base peak in panel B2 at m/z = 141 is a contaminant from the silica gel [2SiO +2H2O + OH−] . ) B1: total ion current ( TIC ) chromatogram and mass spectrum of the β-cyanoalanine standard , B2: TIC chromatogram and mass spectrum of Tu-CAS reaction mixture after separation on TLC . DOI: http://dx . doi . org/10 . 7554/eLife . 02365 . 012 The very high sequence similarity between Tu-CAS and the lepidopteran proteins strongly suggests that the lepidopteran proteins can catalyze the same two reactions ( cyanide detoxification and cysteine synthesis ) and that these proteins are responsible for the known wide occurrence of CAS activity in lepidopteran species ( Witthohn and Naumann , 1987; Meyers and Ahmad , 1991; Stauber et al . , 2012 ) . In arthropods , the ability to detoxify HCN plays a crucial ecological role and is thought to have allowed the exploitation of cyanogenic plants by circumventing the toxic effects of HCN . We have shown here that the two-spotted spider mite increases transcript levels of a horizontally transferred β-cyanoalanine synthase upon adaptation on cyanogenic bean . Phylogenetic evidence alone does not constitute strong evidence for horizontal gene transfer ( HGT ) , because in the absence of introns in the sequence , a contamination of the mite genome sequence with a sequence from a bacterial symbiont or commensal cannot be excluded . However , the results of mite genomic analysis , codon amelioration , synteny and genomic PCR , combined with the phylogenetic evidence , unambiguously prove that T . urticae CAS has a bacterial origin and was laterally transferred and incorporated into the mite genome prior to the divergence in the Tetranychidae . A homologous lateral gene transfer has also occurred in Lepidoptera ( Figure 2 ) raising the question of the time and number of HGT events needed to explain the phylogenetic pattern of distribution of the CYS/CAS genes . Several hypotheses can be discussed . ( a ) A single HGT to a common ancestor of mites and insects , followed by selective losses resulting in the present phylogenetic distribution . The broad sampling of arthropod species that revealed the absence of any CYS/CAS-like sequence ( even as a distant recognizable trace ) argues against the origin of the gene in the common ancestor of tetranychid mites and Lepidoptera , followed by multiple independent losses . These losses would have to be numerous: seven in the hexapods ( Trichoptera , Diptera , Hymenoptera , Coleoptera/Strepsiptera , Hemiptera/Heteroptera/Thysanoptera/Phthiraptera , Orthoptera , Odonata/Ephemeroptera ) and six further in other arthropods ( Copepoda/Branchiopoda , Myriapoda , Aranea , Scorpiones , Metastigmata and Mesostigmata ) . This figure of 13 losses is an absolute minimum that implies that the loss occurred each time at the origin of the lineage , that is before any speciation beyond the point of coalescence . A loss later in the history of each lineage would rapidly increase the total number of losses . We strongly believe that this hypothesis is not parsimonious , and therefore that there was no CYS/CAS gene in the common ancestor of mites and Lepidoptera . ( b ) Alternatively , two HGT events might have occurred ( one to a mite ancestor , one to a lepidopteran ancestor ) or ( c ) a single HGT from a donor bacterial species , followed by a transfer between a mite and a lepidopteran . These two hypotheses are impossible to distinguish based on the topology of the tree ( Figure 2 ) . The sampling of bacterial species phylogenetically close to the presumed donor species is too shallow at present . The apparent monophyly of the arthropod CAS gene may be due to the fact that both the mite and the insect gene came from very closely related bacteria , but these bacteria are not represented in the tree . A future survey of Proteobacteria likely to be associated with arthropods ( directly or through a plant host ) may resolve this question by finding several potential bacterial donor species that would invalidate the apparent monophyly ( i . e . , split the tree at the point of arthropod coalescence ) . The transfer from a lepidopteran to mite is improbable , because the mite sequence would branch with the closest relative to the lepidopteran donor , rather than being basal . Conversely , we cannot exclude the transfer from a mite to a lepidopteran , as sampling within the Prostigmata is extremely limited . We therefore favor hypothesis ( b ) , with a very old transfer to an ancestral lepidopteran , and a second transfer to an ancestral mite . Current sampling of mite species is insufficient to give a good estimate of time , but we note the absence of a CAS enzyme in Mesostigmata ( typically parasitic and predatory mites ) and Metastigmata ( ticks ) , and transfer is likely to have occurred after the split of Prostigmata in the Lower Devonian about 400 MYA ( Dabert et al . , 2010 ) . The hypothesis of multiple transfers is also compatible with the presence of a CYS-like sequence in the mealybug Planococcus citri , apparently derived from its endosymbiont ( Husnik et al . , 2013 ) . In that case , the phylogenetic tree clearly distinguishes the HGT event from that under study here . In Lepidoptera , the CAS gene was subsequently duplicated in H . melpomene and D . plexippus that have 3 and 2 copies of the gene , respectively ( Sun et al . , 2013 ) . The importance of the laterally acquired gene for HCN detoxification in those species was not previously apparent . These species of Papilionoidea , like burnet moths , not only thrive on cyanogenic plants , but have themselves evolved the ability to synthesize cyanogenic compounds de novo ( Jensen et al . , 2011 ) that now have crucial functions in their life history . Cyanogenesis serves in predator defense by releasing HCN , but also stores reduced nitrogen that can be mobilized for chitin synthesis , and plays a role in mate choice by determining the attractiveness of nuptial gifts from male to female partners ( Zagrobelny et al . , 2007 ) . Our results support the early idea that CAS activity is needed for the exploitation of cyanogenic glucosides in insects ( Meyers and Ahmad , 1991; Witthohn and Naumann , 1987; Zagrobelny et al . , 2008 ) whether they are sequestered from the plant , or synthesized de novo . Higher β-cyanoalanine synthase activity in Spodoptera eridania than in Trichoplusia ni is related to higher cyanide tolerance ( Meyers and Ahmad , 1991 ) , and this enzyme activity is widespread in Lepidoptera ( Witthohn and Naumann , 1987 ) . Moreover , it was recently shown that specialist pierid butterflies that feed on Brassicales , release equimolar concentrations of HCN upon metabolism of benzylglucosinolates , turning the ‘mustard oil bomb’ into a ‘cyanide bomb’ ( Stauber et al . , 2012 ) . When Pieris rapae feeds on a cyanogenic ( dhurrin-containing ) plant that this species does not normally consume , an increased production of β-cyanoalanine and thiocyanate is observed , thus implicating both a CAS and a rhodanese activity ( Stauber et al . , 2012 ) . It was therefore proposed that the ability of P . rapae to metabolize HCN allowed the primary host transfer from Fabales to Brassicales ( Stauber et al . , 2012 ) . The gene for either CAS or rhodanese has not been identified in arthropods before , and their respective role in detoxification of HCN is not formally demonstrated . There is no close homologue of the known rhodanese ( thiosulfate sulfurtransferase ) gene in T . urticae or in Lepidoptera . However , we identified Tu-CAS and functionally demonstrated that the enzyme it encodes converts HCN to β-cyanoalanine in vitro . Such evidence is difficult to obtain in vivo with mites , as it would be difficult to exclude the possibility that a plant or a bacterial enzyme rather than the mite enzyme catalyzes the reaction in vivo . We argue that the presence of the same gene in lepidopteran species that display this activity in vivo ( Meyers and Ahmad , 1991; Stauber et al . , 2012 ) is strong evidence for the function of the laterally transferred CAS gene . It will be of great interest to confirm that the homologous CAS genes of Lepidoptera that we identified indeed encode a β-cyanoalanine synthase , and to provide evidence of its protective role against HCN poisoning . Next to the detoxification function , the CYS activity acquired after horizontal gene transfer may also have enhanced the sulfur amino acid economy of mites and lepidopterans ( Figure 1 ) . To date , nematodes were the only animal species thought to synthesize cysteine independently from methionine by CYS activity ( Budde and Roth , 2011 ) . However , nematode CYS sequences grouped with plants and oomycetes , clearly outside the arthropod-bacterial clade , suggestive of a different origin of CYS between metazoan subgroups ( Figure 2 ) . Duplications of cys genes were observed in nematodes , and indeed genetic evidence suggests that in Caenorhabditis elegans cyanide resistance is conferred by the cysl-2 gene , probably encoding an enzyme with CAS activity while cysl-1 is a classical cysteine synthase gene ( Budde and Roth , 2011 ) . The acquisition of an alternative cysteine biosynthesis route fits into previously documented horizontal gene transfers in T . urticae that include a cobalamin-independent methionine synthase gene , genes for carotenoid biosynthesis , as well as laterally acquired genes that likewise respond to host plant change such as intradiol ring-cleavage dioxygenases and a cyanase gene ( Grbic et al . , 2011; Wybouw et al . , 2012; Dermauw et al . , 2013 ) . The latter encodes an enzyme that decomposes cyanate ( CNO− ) , a bacterial or photochemical decomposition product of cyanide , to carbon dioxide and ammonia ( Wybouw et al . , 2012 ) . This enzyme may serve as a second line of spider mite defense against cyanogenic plants , or alternatively may have a regulatory function in the amino acid and pyrimidine metabolism as previously suggested ( Wybouw et al . , 2012 ) . For a polyphagous herbivore , horizontal gene transfer might play an important role in gaining independence from the varying plant nutrients and defense compounds . It remains unclear which reaction ( CAS/CYS ) of Tu-CAS provides the strongest adaptive advantage , but the CYS activity might be one of the reasons why these horizontally transferred genes have been retained in organisms that are at present not living on cyanogenic plants . In conclusion , a horizontal gene transfer from a bacterial ancestor underlies the exploitation of cyanogenic host plants in some arthropod lineages and made the subsequent evolution of a convergent pathway for synthesis of cyanogenic glucosides possible , as shown in burnet moths ( Jensen et al . , 2011 ) . The London strain of T . urticae ( Grbic et al . , 2011 ) was maintained on acyanogenic P . vulgaris L . cv ‘Prelude’ . The strain London-CYANO originated from this population and was transferred to cyanogenic P . lunatus as previously described ( Wybouw et al . , 2012 ) . Before the start of experiments the cyanogenic potential of both plant species was determined , confirming the acyanogenic nature of P . vulgaris and revealing high levels of cyanogenic precursors produced in the P . lunatus cultivar ( Wybouw et al . , 2012 ) . For this study , young adult female mites were collected for gene-expression analysis 35 generations after the initial host shift . The T . evansi strain was maintained in the laboratory on Solanum lycopersicum L . cv ‘Moneymaker’ as previously described ( Wybouw et al . , 2012 ) . All strains were maintained in climatically controlled rooms at 26°C , 60% RH and 16:8 hr light:dark photoperiod . Total RNA samples were isolated with the RNeasy minikit ( Qiagen , Belgium ) and were subsequently treated with DNase ( Turbo DNA-free kit , Ambion , Belgium ) . RNA was extracted from 100–120 young adult female mites in four replicates . Cy5- or Cy3-labeled cRNA was produced using the Low Input Quick Amp Labeling Kit ( Agilent Technologies , Belgium ) as previously described ( Dermauw et al . , 2013 ) . Microarray hybridization and scanning procedures were performed as previously described ( Dermauw et al . , 2013 ) , using the GE2_107_Sep09 protocol . The data was transferred to GeneSpring GX 11 . 0 software ( Agilent Technologies ) for statistical analyis . Probes were flag filtered ( only probes that had flag-value ‘present’ in 50% of all replicates were retained ) and linked to T . urticae genes using the ‘Create New Gene-Level Experiment’ option . Differentially expressed genes were identified by a Student's t test with the cutoff for Fold Change ( FC ) and corrected p-value ( Benjamini-Hochberg correction ) set at 2 and 0 . 05 , respectively . The array design is accessible under the GEO-platform format GPL16890 ( Wybouw et al . , 2013 ) . A full-length Tu-CAS ( tetur10g01570 ) homologue was retrieved in T . evansi by sequencing PCR products bracketing the CAS gene with neighboring genes ( Supplementary file 4 ) . Additional Tu-CAS homologues were identified by conducting BLASTp and/or tBLASTn searches in NCBI , UniProt , P . citri transcriptome ( Liu et al . , 2011 ) and diverse arthropod genome portals ( Supplementary file 2 ) using Tu-CAS as query . As several best BLAST-hits ( E-value ≤1e−90 ) included members of the order Lepidoptera , transcriptome databases from Lepidoptera not included in the NCBI database were also mined for Tu-CAS homologues ( Supplementary file 3 ) . This approach yielded 35 arthropod and bacterial protein sequences , which according to the Conserved Domain Database all contained a motif typical for cysteine synthases ( COG0031 ) ( Marchler-Bauer et al . , 2011 ) . This dataset was further complemented with cysteine synthase M ( cysM ) protein sequences from bacteria and cysteine synthase protein sequences from fungi , Chromalveolata , plants , nematodes , and Planococcus citri and its three best BLASTp hits , harboring a cysteine synthase CDD motif ( COG0031 or PLN2565 ) . Finally , a diverse set of cystathionine-β-synthase protein sequences , related to cysteine synthases and also belonging to the group of pyrodixal-5′-phosphate dependent β-substituted alanine synthases were added as an outgoup . The final dataset contained 90 protein sequences . Accession numbers of protein sequences , their trivial name , CDD classification ( Marchler-Bauer et al . , 2011 ) and cellular localization ( Horton et al . , 2007 ) are listed in Supplementary file 3 . Protein sequences were aligned with MUSCLE ( Edgar , 2004 ) using default settings . Model selection was done with ProtTest 2 . 4 and according to the Akaike information criterion the model LG+I+G was optimal for phylogenetic analysis ( Abascal et al . , 2005 ) . A maximum-likelihood analysis was performed using Treefinder ( Jobb et al . , 2004 ) with edge-support calculated by 1000 pseudoreplicates ( LR-ELW ) . Resulting trees were midpoint rooted prior to further analysis ( Hess and De Moraes Russo , 2007 ) . Phylogenetic trees were visualized and edited using MEGA5 ( Tamura et al . , 2011 ) and CorelDraw X6 ( Corel inc . , UK ) , respectively . Paired-end T . urticae Sanger reads ( available in the Trace Archive at the NCBI website , http://www . ncbi . nlm . nih . gov/Traces/home/ ) were remapped to the T . urticae genome ( Grbic et al . , 2011 ) using Bowtie 2 . 1 . 0 ( Langmead et al . , 2009 ) and the preset parameter option ‘–very-sensitive’ . Resulting SAM files were converted into BAM files using SAMtools ( Li et al . , 2009 ) . Illumina-reads from genomic DNA sequencing of the London , Montpellier and EtoxR strains of T . urticae and Illumina RNA-seq reads from adult T . urticae polyA selected RNA were mapped as previously described ( Grbic et al . , 2011; Van Leeuwen et al . , 2012 ) . Read alignments and coverage were visualized with IGV 2 . 3 ( Thorvaldsdottir et al . , 2013 ) using the most recent T . urticae genome annotation ( ‘Tetur_gff3_20130708’ , accessible at http://bioinformatics . psb . ugent . be/orcae/-overview/Tetur ) ; for display , Sanger reads were arranged in Adobe Illustrator CS5 while maintaining alignment coordinates . Genomic DNA was collected from T . evansi and T . urticae by phenol-chloroform extraction ( Van Leeuwen et al . , 2008 ) . Primer pairs were designed to amplify a genomic region of Tu-CAS and adjacent genes on either the 5′ or the 3′ end on the 10th scaffold of the T . urticae genome ( Supplementary file 4 ) . The Expand Long Range PCR kit ( Roche , Belgium ) was used to conduct PCR , and fragments were sequenced with primers listed in Supplementary file 4 . Some primer pairs designed on the T . urticae genome sequence also successfully amplified genomic fragments of T . evansi ( Supplementary file 4 ) . The resulting fragments were sequenced by primer walking ( Supplementary file 4 ) . A nucleotide dot-blot between the two spider mite species was constructed using the MEGALIGN program of DNASTAR software , allowing 5% mismatch in a 21 bp window . We analyzed overall GC contents and at the third codon position ( GC3 ) of whole coding nucleotide sequences using UGENE ( Okonechnikov et al . , 2012 ) of all coding sequences of T . urticae and Achromobacter xylosoxidans , Methylobacterium radiotolerans and Methylobacterium sp . GXF4 . These three bacterial genomes were selected based on our phylogenetic analysis as the closest fully annotated bacterial genomes to the arthropod clade ( Figure 2 ) . Recombinant Tu-CAS was produced by the GenScript Corporation ( Piscataway , NJ , USA ) . After codon optimization of the Tu-CAS coding sequence ( Supplementary file 5 ) , an E3 expression vector was used to transform E . coli cells . The transformed cells were cultured in 3 l LB . Using a Ni2+-column , the N-His-tagged Tu-CAS protein was purified from the supernatant . After sample sterilization via a 0 . 22 µm filter , the recombinant protein was stored in a buffer containing: 50 mM Tris , 150 mM NaCl , 2 mM DTT , 10% glycerol at a pH of 8 . 0 and finally kept at −80°C . The concentration and purity of the recombinant protein sample was determined respectively by a Bradford protein assay ( Bradford , 1976 ) and a densitometric analysis of a Coomassie Blue-stained SDS-PAGE gel . Chemicals for the activity assays were purchased from Sigma–Aldrich ( Belgium ) , except β-cyanoalanine , which was acquired from VWR , Cayman Chemical . All reactions were carried out in gastight 7 ml vials with a screw cap having a PTFE/rubber septum ( Supelco–Sigma–Aldrich , Belgium ) . Reagents were added to the reaction volumes using gastight syringes ( Hamilton , series 1700 , Gastight , 1750RN , VWR Belgium ) . Prior to measuring enzyme activity , recombinant Tu-CAS was incubated at 30°C for 10 min in the appropriate reaction buffer containing 500 µM pyridoxal-5′-phosphate . For measuring CAS activity , reactions were executed in a 0 . 2 M Tris buffer at pH 8 . 5 . The CAS activity assay was a modification of the method of Hendrickson ( Hendrickson and Conn , 1969 ) . The standard reaction was started with 0 . 5 ml of 0 . 01 M cysteine , 0 . 5 ml of 0 . 01 M sodium cyanide and 1 . 5 µg recombinant Tu-CAS . All CAS assays were performed at 37°C on a mechanical shaker . CAS activity was quantified by spectrophotometrically measuring the H2S formed at 650 nm ( PowerWavex340 , BioTek Instruments Inc . , Winooski , VT , USA ) by the method of Siegel ( Siegel , 1965 ) . The CYS activity assay was based on Lunn et al . , 1990 using 0 . 5 µg recombinant Tu-CAS per reaction . Standard substrate concentrations were 10 mM and 2 mM for O-acetylserine and sodium sulfide , respectively . The reaction product cysteine was quantified by measuring the absorbance at 560 nm by the method of Gaitonde ( Gaitonde , 1967 ) . For both activity assays , the spontaneous formation of the measured reaction product and its potential presence in protein preparations was corrected by respectively non-enzyme controls and zero time point controls . Each experimental assay condition was analyzed using three independent and three technical replicates . Kinetic data was fitted to the Hill equation , from which the Km and Vmax values for O-acetylserine and cysteine were calculated for respectively the CYS and CAS reaction . Recombinant Tu-CAS was incubated at 30°C for 10 min in a 1 mM phosphate buffer containing 500 µM pyridoxal-5′-phosphate . The standard reaction was executed in a 320 µl reaction volume at 37°C on a mechanical shaker in 1 mM phosphate buffer pH 8 . 5 , with 80 µl of 7 . 5 mM cysteine and 80 µl of 15 mM sodium cyanide . Each reaction was started by adding 3 . 75 µg of recombinant protein and was terminated by snap freezing at different time points . No-substrate and no-enzyme controls were included in the analysis . Reaction mixtures were defrosted at 4°C and a 20 µl aliquot was spotted on a thin-layer chromatograph ( HPTLC Silica gel 60 F254 , Merck , Darmstadt Germany ) and run with a mobile phase of ( ethanol/28% ammonium hydroxide/water ) with a ( 18/1/4 ) ratio . Five µg of β-cyanoalanine and cysteine were spotted as standards . After drying , the TLC plate was treated with a ninhydrin solution ( 20 g ninhydrin in 600 ml ethanol ) for amino-acid visualization . In this validated TLC set-up ( Yoshikawa et al . , 2000 ) , cysteine and β-cyanoalanine can be identified both by color and relative mobility ( Rf-value ) . β-cyanoalanine consistently displayed a blue-green color with a Rf value of around 0 . 8 . In contrast , cysteine colored red and exhibited a consistent different Rf value . The identification of β-cyanoalanine as a blue spot at 0 . 8 Rf was further confirmed by LC-MS analysis . After TLC separation , the zone around 0 . 8 Rf was scrapped from the TLC plate . After scrapping , the plate was colored as described above to confirm that the correct blue/green zone was collected . The collected silica was mixed with 200 µl ddH2O , vortexed and centrifuged at 21000×g for 5 min . The supernatant was collected and directly analyzed using an Agilent 1100 Series LC-MSD with a diode array detector operating at 220 nm . The column oven was programmed at 35°C using a Phenomenex Luna 5u column ( particle size ) C18 ( 2 ) , 100A ( pore size ) with a column size of 250 × 3 . 0 mm . A gradient elution program driven by a quarternary pump was used at a flow rate of 0 . 5 ml/min ( injection volume: 20 µl ) . The acetonitrile/water gradient used was 0–2 min ( 5% acetonitrile ) ; 2–17 min ( 5–100% acetonitrile ) ; 17–22 min ( 100% acetonitrile ) ; 22–24 . 5 min ( 100–5% acetonitrile ) ; 24 . 5–27 min ( 5% acetonitrile ) . The mass spectrometer was operated using the SCAN mode in the electrospray ionization mode . The analyzing sector contained a quadrupole analyzer and an electron multiplier detector .	Hydrogen cyanide is a poison that is deadly for most forms of life . Also known as prussic acid , it has killed countless humans throughout history in accidents and during the Holocaust . Hydrogen cyanide is also used by plants to defend themselves against insects and other herbivorous animals . Many plants produce chemicals called cyanogenic glycosides that can be converted into hydrogen cyanide when the plant is eaten . This is an ancient and efficient defense against all sorts of herbivores , including humans . For instance , cassava is a key source of food in sub-Saharan Africa and South America , but it contains cyanogenic glucosides and is highly toxic if eaten in unprocessed form . However , some insects and mites can thrive on cyanogenic plants , often to the extent of becoming pests on these plants . Certain moths , such as burnet moths , have gone further and now depend on cyanogenic glucosides for their own defenses against predators such as birds . How these mites and insects are capable of fending off cyanide toxicity has long remained a mystery . Now Wybouw et al . have identified a mite enzyme that detoxifies hydrogen cyanide to produce a compound called beta-cyanoalanine . Remarkably , the DNA that encodes this enzyme did not evolve in animals but originally belonged to a bacterium . Wybouw et al . show that the gene was transferred to the genome of the spider mite Tetranychus urticae perhaps a few hundred million years ago . An equivalent gene was also found in moths and butterflies , which explains why these insects can thrive on plants that produce hydrogen cyanide . This lateral gene transfer from bacteria to animals is a remarkable coalition of two kingdoms against another , and illustrates a new aspect of the chemical warfare between plants and animals . This study also increases our awareness of the importance of laterally transferred genes in the genomes of higher organisms .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "plant", "biology", "genetics", "and", "genomics" ]	2014	A gene horizontally transferred from bacteria protects arthropods from host plant cyanide poisoning
Sensory hair cells in the ear utilize specialized ribbon synapses . These synapses are defined by electron-dense presynaptic structures called ribbons , composed primarily of the structural protein Ribeye . Previous work has shown that voltage-gated influx of Ca2+ through CaV1 . 3 channels is critical for hair-cell synapse function and can impede ribbon formation . We show that in mature zebrafish hair cells , evoked presynaptic-Ca2+ influx through CaV1 . 3 channels initiates mitochondrial-Ca2+ ( mito-Ca2+ ) uptake adjacent to ribbons . Block of mito-Ca2+ uptake in mature cells depresses presynaptic-Ca2+ influx and impacts synapse integrity . In developing zebrafish hair cells , mito-Ca2+ uptake coincides with spontaneous rises in presynaptic-Ca2+ influx . Spontaneous mito-Ca2+ loading lowers cellular NAD+/NADH redox and downregulates ribbon size . Direct application of NAD+ or NADH increases or decreases ribbon size respectively , possibly acting through the NAD ( H ) -binding domain on Ribeye . Our results present a mechanism where presynaptic- and mito-Ca2+ couple to confer proper presynaptic function and formation . Neurotransmission is an energy demanding process that relies heavily on mitochondria . In neurons , mitochondrial dysfunction has been implicated in synaptopathies that impact neurodevelopment , learning and memory , and can contribute to neurodegeneration ( Flippo and Strack , 2017; Lepeta et al . , 2016; Todorova and Blokland , 2016 ) . In hair cells , sensory neurotransmission relies on specialized ribbon synapses to facilitate rapid and sustained vesicle release that is particularly energy demanding ( reviewed in: Johnson et al . , 2019; Lagnado and Schmitz , 2015; Matthews and Fuchs , 2010; Safieddine et al . , 2012 ) . Although mitochondrial dysfunction has been implicated in hearing loss ( Böttger and Schacht , 2013; Fischel-Ghodsian et al . , 2004; Kokotas et al . , 2007 ) , the precise role mitochondria play at hair-cell synapses remains unclear . Ribbon synapses are characterized by a unique presynaptic structure called a ‘ribbon’ that tethers and stabilizes synaptic vesicles at the active zone ( reviewed in: Matthews and Fuchs , 2010 ) . In hair cells , neurotransmission at ribbon synapses requires the presynaptic-Ca2+ channel CaV1 . 3 ( Brandt et al . , 2003; Kollmar et al . , 1997; Sidi et al . , 2004 ) . Hair-cell depolarization opens CaV1 . 3 channels , resulting in a spatially restricted increase of Ca2+ at presynaptic ribbons that triggers vesicle fusion . Tight spatial regulation of presynaptic Ca2+ is important for ribbon-synapse function and requires efficient Ca2+ clearance through a combination of Ca2+ pumps , Ca2+ buffers and intracellular Ca2+ stores ( Carafoli , 2011; Mulkey and Malenka , 1992; Tucker and Fettiplace , 1995; Yamoah et al . , 1998; Zenisek and Matthews , 2000 ) . While ER-Ca2+ stores have been implicated in hair-cell neurotransmission , whether mitochondrial-Ca2+ ( mito-Ca2+ ) stores play a role in this process remains unclear ( Castellano-Muñoz and Ricci , 2014; Kennedy , 2002; Lioudyno et al . , 2004; Tucker and Fettiplace , 1995 ) . In addition to a role in hair-cell neurotransmission , presynaptic Ca2+ and CaV1 . 3 channels also play an important role during inner-ear development . In mammals , prior to hearing onset , auditory hair cells fire spontaneous Ca2+ action potentials ( Eckrich et al . , 2018; Marcotti et al . , 2003; Tritsch et al . , 2007; Tritsch et al . , 2010 ) . In mammalian hair cells , these Ca2+ action potentials are CaV1 . 3-dependent and are thought to be important for synapse and circuit formation . In support of this idea , in vivo work in zebrafish hair cells found that increasing or decreasing voltage-gated Ca2+ influx through CaV1 . 3 channels during development led to the formation of smaller or larger ribbons respectively ( Sheets et al . , 2012 ) . Furthermore , in mouse knockouts of CaV1 . 3 , auditory outer hair cells have reduced afferent innervation and synapse number ( Ceriani et al . , 2019 ) . Mechanistically , how CaV1 . 3-channel activity regulates ribbon size and innervation , and whether hair-cell Ca2+ stores play a role in this process is not known . Cumulative work has shown that ribbon size varies between species and sensory epithelia ( reviewed in Moser et al . , 2006 ) ; these variations are thought to reflect important encoding requirements of a given sensory cell ( Matthews and Fuchs , 2010 ) . In auditory hair cells , excitotoxic noise damage can also alter ribbon size and lead to hearing deficits ( Jensen et al . , 2015; Liberman et al . , 2015 ) . Excitotoxic damage is thought to be initiated by mito-Ca2+ overload and subsequent ROS production ( Böttger and Schacht , 2013; Wang et al . , 2018 ) . Mechanistically , precisely how ribbon size is established during development or altered under pathological conditions is not fully understood . One known way to regulate ribbon size is through its main structural component Ribeye ( Schmitz et al . , 2000 ) . Perhaps unsurprisingly , previous work has shown that overexpression or depletion of Ribeye in hair cells can increase or decrease ribbon size respectively ( Becker et al . , 2018; Jean et al . , 2018; Lv et al . , 2016; Sheets , 2017; Sheets et al . , 2011 ) . Ribeye is a splice variant of the transcriptional co-repressor Carboxyl-terminal binding protein 2 ( CtBP2 ) – a splice variant that is unique to vertebrate evolution ( Schmitz et al . , 2000 ) . Ribeye contains a unique A-domain and a B-domain that is nearly identical to full-length CtBP2 . The B-domain contains a nicotinamide adenine dinucleotide ( NAD+ , NADH or NAD ( H ) ) binding site ( Magupalli et al . , 2008; Schmitz et al . , 2000 ) . NAD ( H ) redox is linked to mitochondrial metabolism ( Srivastava , 2016 ) . Because CtBPs are able to bind and detect NAD+ and NADH levels , they are thought to function as metabolic biosensors ( Stankiewicz et al . , 2014 ) . For example , previous work has demonstrated that changes in NAD ( H ) redox can impact CtBP oligomerization and its transcriptional activity ( Fjeld et al . , 2003; Thio et al . , 2004 ) . Interestingly , in vitro work has shown that both NAD+ and NADH can also promote interactions between Ribeye domains ( Magupalli et al . , 2008 ) . Whether NAD+ or NADH can impact Ribeye interactions and ribbon formation or stability has not been confirmed in vivo . In neurons , it is well established that during presynaptic activity , mitochondria clear and store Ca2+ at the presynapse ( Devine and Kittler , 2018 ) . Additionally , presynaptic activity and mito-Ca2+ can couple together to influence cellular bioenergetics , including NAD ( H ) redox homeostasis ( reviewed in: Kann and Kovács , 2007; Llorente-Folch et al . , 2015 ) . Based on these studies , we hypothesized that Ca2+ influx through CaV1 . 3 channels may regulate mito-Ca2+ , which in turn could regulate NAD ( H ) redox . Changes to cellular bioenergetics and NAD ( H ) redox could function to control Ribeye interactions and ribbon formation or impact ribbon-synapse function and stability . To study the impact of mito-Ca2+ and NAD ( H ) redox on ribbon synapses , we examined hair cells in the lateral-line system of larval zebrafish . This system is advantageous because it contains hair cells with easy access for in vivo pharmacology , mechanical stimulation and imaging cellular morphology and function . Within the lateral-line , hair cells are arranged in clusters called neuromasts . The hair cells and ribbon synapses in each cluster form rapidly between 2 to 3 days post-fertilization ( dpf ) but by 5–6 dpf , the majority of hair cells are mature , and the system is functional ( Kindt et al . , 2012; McHenry et al . , 2009; Metcalfe , 1985; Murakami et al . , 2003; Santos et al . , 2006 ) . Thus , these two ages ( 2–3 dpf and 5–6 dpf ) can be used to study mito-Ca2+ and NAD ( H ) redox in developing and mature hair cells respectively . Using this sensory system , we find that presynaptic-Ca2+ influx drives mito-Ca2+ uptake . In mature hair cells , mito-Ca2+ uptake occurs during evoked stimulation and is required to sustain presynaptic function and ultimately synapse integrity . In developing hair cells , mito-Ca2+ uptake coincides with spontaneous rises in presynaptic Ca2+ . Blocking these spontaneous changes in Ca2+ leads to the formation of larger ribbons . Using a redox biosensor , we demonstrate that specifically in developing hair cells , decreasing mito-Ca2+ levels increases the NAD+/NADH redox ratio . Furthermore , we show that application of NAD+ or NADH can promote the formation of larger or smaller ribbons respectively . Overall , our results suggest that in hair cells presynaptic-Ca2+ influx and mito-Ca2+ uptake couple in hair cells to impact ribbon formation and function . In neurons , synaptic mitochondria have been shown to influence synapse size , plasticity and function ( Flippo and Strack , 2017; Todorova and Blokland , 2016 ) . Based on this work , we hypothesized that mitochondria may impact synapses in hair cells . Therefore , we examined the proximity of mitochondria relative to presynaptic ribbons in zebrafish lateral-line hair cells . We visualized mitochondria and ribbons using transmission electron microscopy ( TEM ) and in vivo using Airyscan confocal microscopy . Using TEM , we examined sections that clearly captured ribbons ( Example , Figure 1C ) . Near the majority of ribbons ( 81% ) we observed a mitochondrion in close proximity ( <1 µm ) ( Figure 1D , median ribbon-to-mitochondria distance = 174 nm , n = 17 out of 21 ribbons ) . To obtain a more comprehensive understanding of the 3D morphology and location of mitochondria relative to ribbons in live cells , we used Airyscan confocal microscopy . To visualize these structures in living cells , we used transgenic zebrafish expressing MitoGCaMP3 ( Esterberg et al . , 2014 ) and Ribeye a-tagRFP ( Sheets et al . , 2014 ) in hair cells to visualize mitochondria and ribbons respectively . Using this approach , we observed tubular networks of mitochondria extending from apex to base ( Figure 1A–B , E–E’ , Figure 1—figure supplement 1A , Video 1 ) . At the base of the hair cell , we observed ribbons nestled between branches of mitochondria . Overall our TEM and Airyscan imaging suggests that in lateral-line hair cells , mitochondria are present near ribbons . In zebrafish hair cells , robust rises in mito-Ca2+ have been reported during mechanical stimulation ( Pickett et al . , 2018 ) . Due to the proximity of the mitochondria to the ribbon , we predicted that rises in mito-Ca2+ levels during mechanical stimulation could be related to presynapse-associated rises in Ca2+ . To test this prediction , we used a fluid-jet to mechanically stimulate hair cells and evoke presynaptic activity . During stimulation , we used MitoGCaMP3 to monitor mito-Ca2+ in lateral-line hair cells . As previously reported , we observed robust mito-Ca2+ uptake during stimulation ( Figure 1E–F , Figure 1—figure supplement 2 ) . We examined the subcellular distribution of MitoGCaMP3 signals over time and observed an increase in MitoGCaMP3 ( ∆F ) signals that initiated near ribbons ( Figure 1E , ∆F ) . During the latter part of the stimulus , and even after the stimulus terminated , the MitoGCaMP3 signals propagated apically , away from the ribbons ( Example , Figure 1E’–E’’ , regions 1–3 , ∆F/F0 ) . We characterized the time-course of MitoGCaMP3 signals with regards to onset kinetics and return to baseline . During a 2 s stimulus , we detected a significant rise in MitoGCaMP3 signals 0 . 6 s after stimulus onset ( Figure 1—figure supplement 1B , ∆F/F0 ) . Interestingly , after the stimulus terminated , MitoGCaMP3 levels took approximately 5 min to return to baseline ( Figure 1—figure supplement 1C–C’ , ∆F/F0 ) . Despite this long time-course of recovery to baseline , we were still able to evoke additional rises in MitoGCaMP3 signal 10 s after stimulation ( Figure 1—figure supplement 1D , ∆F/F0 ) . As previously reported , the kinetics of MitoGCaMP3 signals in hair-cell mitochondria were quite different from signals observed using cytosolic GCaMP3 ( CytoGCaMP3 ) in hair cells ( Pickett et al . , 2018 ) . Compared to MitoGCaMP3 signals , CytoGCaMP3 signals had faster onset kinetics and a faster return to baseline ( Figure 1—figure supplement 1B–C , time to rise: 0 . 06 s , post-stimulus return to baseline: 12 s ) . These differences in kinetics indicate that mito-Ca2+ loading operates over slower timescales compared to the cytosolic compartment . It also confirms that hair-cell stimulation can initiate long lasting increases in mito-Ca2+ . To verify that MitoGCaMP3 signals reflect Ca2+ entry into mitochondria , we applied Ru360 , an antagonist of the mito-Ca2+ uniporter ( MCU ) . The MCU is the main pathway for rapid Ca2+ entry into the mitochondrial matrix ( Matlib et al . , 1998 ) . We found that stimulus-evoked MitoGCaMP3 signals were blocked in a dose-dependent manner after a 30 min treatment with Ru360 ( Figure 1F , Figure 1—figure supplement 1F; IC50 = 1 . 37 µM ) . We confirmed these results by applying TRO 19622 , an antagonist of the voltage-dependent anion channel ( VDAC ) . VDAC enables transport of ions including Ca2+ across the outer mitochondrial membrane ( Schein et al . , 1976; Shoshan-Barmatz and Gincel , 2003 ) . We observed that similar to the MCU antagonist Ru360 , a 30 min treatment with the VDAC antagonist TRO 19622 also impaired stimulus-evoked MitoGCaMP3 signals ( 10 µM TRO 19622 , Figure 1—figure supplement 1E ) . Due to the initiation of mito-Ca2+ near ribbons , we examined whether presynaptic-Ca2+ influx through CaV1 . 3 channels was the main source of Ca2+ entering the mitochondria . To examine CaV1 . 3 channel contribution to mito-Ca2+ uptake , we applied isradipine , a CaV1 . 3 channel antagonist . Similar to blocking the MCU and VDAC , blocking CaV1 . 3 channels eliminated all stimulus-evoked MitoGCaMP3 signals at the base of the cell ( Figure 1F ) . Previous work in zebrafish-hair cells demonstrated that isradipine specifically blocks CaV1 . 3 channels without impairing mechanotransduction ( Zhang et al . , 2018 ) . For our current study we confirmed whether Ru360 and TRO 19622 specifically block synaptic mito-Ca2+ uptake without impairing mechanotransduction . We measured apical , mechanically evoked Ca2+ signals in hair bundles before and after a 30 min treatment with 10 µM Ru360 or TRO 19622 . Neither compound blocked mechanotransduction ( Figure 2—figure supplement 1A–B’ ) . Overall our MitoGCaMP3 functional imaging indicates that in hair cells , evoked mito-Ca2+ uptake initiates near ribbons and this uptake is dependent on MCU , VDAC and CaV1 . 3 channel function . Interestingly , we observed that mito-Ca2+ uptake was only present in ~40% of cells ( Examples , Figure 2A’ and Figure 1—figure supplement 2; n = 10 neuromasts , 146 cells ) . This observation is consistent with previous work demonstrating that only ~30% of hair cells within each neuromast cluster have presynaptic-Ca2+ signals and are synaptically active ( Zhang et al . , 2018 ) . Because presynaptic-Ca2+ signals initiate near mitochondria , it is probable that mito-Ca2+ uptake may occur specifically in hair cells with synaptic activity . To test whether evoked mito-Ca2+ uptake occurred exclusively in cells with presynaptic-Ca2+ influx , we performed two-color functional imaging . We used a double transgenic approach that utilized a membrane-localized GCaMP6s ( GCaMP6sCAAX; green ) to measure presynaptic-Ca2+ signals at the base of hair cells ( Jiang et al . , 2017; Sheets et al . , 2017 ) , and we concurrently used MitoRGECO1 ( red ) to examine mito-Ca2+ signals ( Figure 2A–B’ ) . Our two-color imaging approach revealed a strong correlation between the magnitude of evoked GCaMP6sCAAX and MitoRGECO1 signals ( Figure 2B , R2 = 0 . 77 , p<0 . 0001; n = 136 cells ) . We found that the median MitoRGECO1 signals were 100% larger in presynaptically active hair cells compared to presynaptically silent hair cells ( Figure 2B’ ) . Together these results suggest that mito-Ca2+ uptake occurs specifically in hair cells with evoked presynaptic-Ca2+ influx . Although we observed mito-Ca2+ uptake specifically in hair cells with active Ca2+ channels , the impact of mito-Ca2+ uptake on the function of hair-cell synapses was unclear . Based on previous studies in neurons and bipolar-cell ribbon synapses ( Billups and Forsythe , 2002; Chouhan et al . , 2010; Kwon et al . , 2016; Levy et al . , 2003; Zenisek and Matthews , 2000 ) , we reasoned that mitochondria may be important to remove excess Ca2+ or provide ATP for hair-cell neurotransmission . To determine if mito-Ca2+ uptake impacted presynaptic function , we assayed evoked presynaptic-Ca2+ signals by monitoring GCaMP6sCAAX signals adjacent to ribbons as described previously ( Figure 2—figure supplement 1C–C’; Sheets et al . , 2017; Zhang et al . , 2018 ) . We examined GCaMP6sCAAX signals in mature hair cells at 5–6 dpf when neuromast organs are largely mature ( Kindt et al . , 2012; McHenry et al . , 2009; Metcalfe , 1985; Murakami et al . , 2003; Santos et al . , 2006 ) . Using this approach , we assayed presynaptic GCaMP6sCAAX signals before and after a 30 min application of the MCU antagonist Ru360 ( Figure 2C–D’ ) . We found that during short , 200 ms stimuli , GCaMP6sCAAX signals at ribbons were reduced after complete MCU block ( 10 µM Ru360 , Figure 2C–C’ ) . Reduction of GCaMP6sCAAX signals were further exacerbated during sustained 10 s stimuli , even when the MCU was only partially blocked ( 2 µM Ru360 , Figure 2D–D’ ) . A similar reduction in GCaMP6sCAAX signals were observed after a 30 min application of the VDAC inhibitor TRO 19622 ( Figure 2—figure supplement 1D-E’ , 10 µM TRO 19622 ) . These results suggest that in mature hair cells , evoked mito-Ca2+ uptake is critical for presynaptic-Ca2+ influx , especially during sustained stimulation . MCU block could impair presynaptic-Ca2+ influx through several mechanisms . It could impair the biophysical properties of CaV1 . 3 channels , for example , through Ca2+-dependent inactivation ( Platzer et al . , 2000; Schnee and Ricci , 2003 ) . MCU block could also impact CaV1 . 3 channel localization . In addition , mito-Ca2+ has been implicated in synapse dysfunction and cell death ( Esterberg et al . , 2014; Vos et al . , 2010; Wang et al . , 2018 ) , and MCU block could be pathological . To distinguish between these possibilities , we assessed whether synaptic components or hair-cell numbers were altered after MCU block with Ru360 . To quantify ribbon-synapse morphology after MCU block , we immunostained mature-hair cells ( 5 dpf ) with CaV1 . 3 , Ribeye b and MAGUK antibodies to label CaV1 . 3 channels , presynaptic ribbons and postsynaptic densities ( MAGUK ) respectively . We first applied 2 μM Ru360 for 1 hr , a concentration that partially reduces evoked mito-Ca2+ uptake ( See Figure 1F ) , yet is effective at reducing sustained presynaptic-Ca2+ influx ( See Figure 2D–D’ ) . At this dose , Ru360 had no impact on hair cell or synapse number ( Figure 3E ) . We also observed no morphological change in ribbon or postsynapse size ( Figure 3F , Figure 3—figure supplement 1C , Figure 3—figure supplement 2 ) . After the 1 hr 2 µM Ru360 treatment , CaV1 . 3 clusters were still present at synapses , but the channels were at a significantly higher density compared to controls ( Figure 2E–H ) . These findings indicate that in mature hair cells , partial MCU block may impair presynaptic function by altering CaV1 . 3 channel density . We also tested a higher dose of Ru360 ( 10 µM ) that completely blocks evoked mito-Ca2+ uptake ( See Figure 1F ) . Interestingly , a 30 min or 1 hr 10 µM Ru360 treatment had a progressive impact on synapse and cellular integrity . After a 30 min treatment with 10 µM Ru360 we did not observe fewer complete synapses per hair cell or fewer hair cells compared to controls ( Figure 3E; Hair cells per neuromast , control: 16 . 3 , 30 min 10 µM Ru360: 15 . 5; p=0 . 5 ) . But after the 30 min treatment , ribbons were significantly larger ( Figure 3F ) . The effects of MCU block became more pathological after a 1 hr , 10 µM Ru360 treatment . After 1 hr , there were both fewer hair cells per neuromast ( Hair cells per neuromast , control: 18 . 1 , 1 hr 10 µM Ru360: 12 . 0; p>0 . 0001 ) and fewer synapses per hair cell ( Figure 3E ) . Similar to 30 min treatments with Ru360 , after 1 hr , ribbons were also significantly larger ( Figure 3F ) . Neither 30 min nor 1 hr 10 µM Ru360 treatment altered postsynapse size ( Figure 3—figure supplement 2 ) . Overall , our results indicate that in mature hair cells , partial block of mito-Ca2+ uptake may impair presynaptic function by altering CaV1 . 3 channel clustering , without seemingly altering other gross pre- or post-synaptic morphology . Complete block of mito-Ca2+ uptake is pathological; it impairs presynaptic function , alters presynaptic morphology , and results in a loss of synapses and hair-cells . In addition to evoked presynaptic- and mito-Ca2+ signals in hair cells , we also observed instances of spontaneous presynaptic- and mito-Ca2+ signals ( Example , Figure 4A–A’’’ , Video 2 ) . Numerous studies have demonstrated that mammalian hair cells have spontaneous presynaptic-Ca2+ influx during development ( Eckrich et al . , 2018; Holman et al . , 2019; Marcotti et al . , 2003; Tritsch et al . , 2007; Tritsch et al . , 2010 ) . Therefore , we predicted that similar to mammals , spontaneous presynaptic-Ca2+ uptake may be a feature of development . Furthermore , we predicted that spontaneous mito-Ca2+ uptake may correlate with instances of spontaneous presynaptic-Ca2+ influx . First , we tested whether spontaneous presynaptic-Ca2+ signals were a feature of development . In zebrafish neuromasts , hair cells are rapidly added between 2–3 dpf , but by 5–6 dpf relatively fewer cells are added and the hair cells and the organs are largely mature ( Kindt et al . , 2012; McHenry et al . , 2009; Metcalfe , 1985; Murakami et al . , 2003; Santos et al . , 2006 ) . Therefore , we examined the magnitude and frequency of spontaneous , presynaptic GCaMP6sCAAX signals in developing ( 3 dpf ) and mature ( 5 dpf ) hair cells . We found that in developing hair cells , spontaneous GCaMP6sCAAX signals occurred with larger magnitudes and at a higher frequency compared to those in mature hair cells ( Figure 4B–C ) . Our spontaneous GCaMP6sCAAX imaging demonstrates that similar to mammals , spontaneous presynaptic-Ca2+ activity is a feature of developing zebrafish hair cells . Next we tested whether spontaneous mito-Ca2+ uptake and presynaptic-Ca2+ influx were correlated . For this analysis we concurrently imaged GCaMP6sCAAX and MitoRGECO1 signals in the same cells for 15 mins to measure presynaptic- and mito-Ca2+ responses respectively . We found that spontaneous presynaptic-Ca2+ influx was often associated with spontaneous mito-Ca2+ uptake ( Example , Figure 4A–A’’’ ) . Overall , we observed a high correlation between the rise and fall of these two signals within individual cells ( Figure 4A’’–A’’’ ) . Both of these signals and their correlation were abolished by application of the CaV1 . 3-channel antagonist isradipine ( Figure 4—figure supplement 1 ) . Together these experiments indicate that , similar to our evoked experiments , spontaneous presynaptic- and mito-Ca2+ signals are correlated . Previous work in zebrafish demonstrated that CaV1 . 3 channel activity plays a role in regulating ribbon size specifically during development ( Sheets et al . , 2012 ) . This work found that a transient , 1 hr pharmacological block of CaV1 . 3 channels increased ribbon size , while CaV1 . 3 channel agonists decreased ribbon size ( Figure 5E; Sheets et al . , 2012 ) . Therefore , we reasoned that spontaneous CaV1 . 3 and mito-Ca2+ activities could function together to control ribbon size in developing hair cells . To characterize the role of spontaneous mito-Ca2+ uptake on ribbon size , we applied the MCU antagonist Ru360 to developing hair cells ( 3 dpf ) . After this treatment , we quantified ribbon-synapse morphology by immunostaining hair cells to label presynaptic ribbons and postsynaptic densities . After a 1 hr application of 2 μM Ru360 to block the MCU , we observed a significant increase in ribbon size in developing hair cells ( Figure 5A–B , E , Figure 5—figure supplement 1C ) . In contrast , this same treatment did not impact ribbon size in mature hair cells ( Figure 3F , Figure 3—figure supplement 1C ) . We also applied a higher concentration of Ru360 ( 10 µM ) to developing hair cells for 1 hr . In developing hair cells , after a 1 hr 10 µM Ru360 treatment , we also observed a significant increase in ribbon size ( Figure 5A , C , E ) . Unlike in mature hair cells ( Figure 3 ) , in developing hair cells , these concentrations of the MCU antagonist did not alter the number of hair cells or the number of synapses per hair cell ( Figure 5D; Hair cells per neuromast , control: 9 . 0 , 1 hr 10 µM Ru360: 8 . 8 , p=0 . 3 ) . All morphological changes were restricted to the ribbons , as MCU block did not alter the size of the postsynapse ( Figure 5—figure supplement 2 ) . In addition to larger ribbons , at higher concentrations of Ru360 ( 10 µM ) we also observed an increase in cytoplasmic , non-synaptic Ribeye aggregates ( Figure 5F , G ) . Previous work in zebrafish reported both larger ribbons and cytoplasmic aggregates of Ribeye in CaV1 . 3a-deficient hair cells ( Sheets et al . , 2011 ) . These parallel phenotypes indicate that spontaneous presynaptic-Ca2+ influx and mito-Ca2+ uptake may couple to shape Ribeye aggregation and ribbon size . Our results suggest that during development , spontaneous Ca2+ entry through both CaV1 . 3 and MCU channels continuously regulate ribbon formation; blocking either channel increases Ribeye aggregation and ribbon size . Our results indicate that spontaneous Ca2+ influx through CaV1 . 3 channels and subsequent loading of Ca2+ into mitochondria regulates ribbon size in developing hair cells . But how do these two Ca2+ signals converge to regulate ribbon size ? It is possible that mitochondria could buffer Ca2+ during spontaneous presynaptic activity and function to decrease resting levels of cytosolic Ca2+ ( cyto-Ca2+ ) ; cyto-Ca2+ levels could be a signal that regulates ribbon size . To examine resting cyto-Ca2+ levels in hair cells , we examined the fluorescence signal change of the cytosolic Ca2+ indicator RGECO1 ( CytoRGECO1 ) before and after a 30 min pharmacological manipulation of CaV1 . 3 or MCU channels ( Figure 6A–C ) . We observed that treatment with the CaV1 . 3 channel antagonist isradipine and agonist Bay K8644 decreased and increased resting CytoRGECO1 fluorescence respectively ( Figure 6B ) . However , treatment with MCU blocker Ru360 did not significantly shift resting CytoRGECO1 fluorescence levels ( Figure 6B ) . Similar results with Ru360 were observed in developing and mature hair cells ( Figure 6B–C ) . These data suggest that , unlike CaV1 . 3 channel function , MCU function and associated mito-Ca2+ uptake does not play a critical role in buffering steady state cyto-Ca2+ levels . Alternatively , it is possible that rather than impacting cyto-Ca2+ levels , both CaV1 . 3 and MCU activity are required to load and maintain Ca2+ levels within the mitochondria . In this scenario , mito-Ca2+ levels could be a signal that regulates ribbon size . To test this possibility , we used MitoGCaMP3 to examine resting mito-Ca2+ levels before and after modulating CaV1 . 3 or MCU channel function ( Figure 6D–F ) . We observed that blocking CaV1 . 3 channels with isradipine or the MCU with Ru360 decreased resting MitoGCaMP3 fluorescence ( Figure 6E–F ) . Conversely , CaV1 . 3 channel agonist Bay K8644 increased resting MitoGCaMP3 fluorescence ( Figure 6E ) . These results were consistent in developing and mature hair cells ( Figure 6E–F ) . Our resting MitoGCaMP3 measurements indicate that the effects of CaV1 . 3 channel and MCU activity converge to regulate mito-Ca2+ levels . When either of these channels are blocked , the resting levels of mito-Ca2+ decrease . Therefore , if presynaptic-Ca2+ influx and mito-Ca2+ regulate ribbon size through a similar mechanism , they may act through mito- rather than cyto-Ca2+ homeostasis . If mito-Ca2+ levels signal to regulate ribbon size , how is this signal transmitted from the mitochondria to the ribbon ? An ideal candidate is NAD ( H ) homeostasis . Ribeye protein , the main component of ribbons contains a putative NAD ( H ) binding site . Because mitochondria regulate NAD ( H ) redox homeostasis ( Jensen-Smith et al . , 2012 ) , we reasoned that there may be a relationship between mito-Ca2+ levels , NAD ( H ) redox , and ribbon size . To examine NAD ( H ) redox , we created a stable transgenic line expressing Rex-YFP , a fluorescent NAD+/NADH ratio biosensor in hair cells ( Figure 6G ) . We verified the function of the Rex-YFP biosensor in our in vivo system by exogenously applying NAD+ or NADH for 30 min . We found that incubation with 100 µM NAD+ increased while 5 mM NADH decreased Rex-YFP fluorescence; these intensity changes are consistent with an increase and decrease in the NAD+/NADH ratio respectively ( Figure 6H ) . Next , we examined if CaV1 . 3 and MCU channel activities impact the NAD+/NADH ratio . We found that 30 min treatments with either a CaV1 . 3 or MCU channel antagonist increased the NAD+/NADH ratio ( increased Rex-YFP fluorescence ) in developing hair cells ( Figure 6H ) . Interestingly , similar 30 min treatments did not alter Rex-YFP fluorescence in mature hair cells ( Figure 6I ) . Together , our baseline MitoGCaMP3 and Rex-YFP measurements indicate that during development , CaV1 . 3 and MCU channel activities normally function to increase mito-Ca2+ and decrease the NAD+/NADH ratio . Overall , this work provides strong evidence that links NAD ( H ) redox and mito-Ca2+ with ribbon formation . Our Rex-YFP measurements suggest that in developing hair cells , CaV1 . 3 and MCU Ca2+ activities normally function to decrease the NAD+/NADH ratio; furthermore , these activities may function to restrict ribbon size . Conversely , blocking these activities increases the NAD+/NADH ratio and may increase ribbon size . If the NAD+/NADH ratio is an intermediate step between CaV1 . 3 and MCU channel activities and ribbon formation , we predicted that more NAD+ or NADH would increase or decrease ribbon size respectively . To test this prediction , we treated developing hair cells with exogenous NAD+ or NADH . After a 1 hr treatment with 100 µM NAD+ , we found that the ribbons in developing hair cells were significantly larger compared to controls ( Figure 7A–B , E ) . In contrast , after a 1 hr treatment with 5 mM NADH , ribbons were significantly smaller compared to controls ( Figure 7A , C , E ) . Neither exogenous NAD+ nor NADH were able to alter ribbon size in mature hair cells ( Figure 7F–H , J ) . These concentrations of NAD+ and NADH altered neither the number of synapses per hair cell nor postsynapse size in developing or mature hair cells ( Figure 7D , I , Figure 7—figure supplement 1 ) . These results suggest that in developing hair cells , NAD+ promotes while NADH inhibits Ribeye-Ribeye interactions or Ribeye localization to the ribbon . Overall these results support the idea that during development , the levels of NAD+ and NADH can directly regulate ribbon size in vivo . Our work outlines how during development , presynaptic activity controls the size of ribbons . When either presynaptic-Ca2+ influx or mito-Ca2+ uptake was perturbed , ribbons were significantly larger ( Figure 5A–C , E; Sheets et al . , 2012 ) . But why regulate ribbon size ? Previous work has reported variations in ribbon size and shape among hair-cell types and species ( Moser et al . , 2006 ) . In many instances ribbon size is correlated with functional properties of the synapse . For example , in the mammalian vestibular system , the ribbons of type II dimorphic hair cells in the striolar region are larger than those in the extrastriolar region ( Lysakowski and Goldberg , 1997 ) . Functionally , afferents that innervate hair cells with larger ribbons in the striolar region have lower rates of spontaneous activity compared to afferents that innervate hair cells in the extrastriolar region ( Eatock et al . , 2008; Goldberg et al . , 1984; Risner and Holt , 2006 ) . Similarly , in the mammalian auditory system , ribbon size is correlated with differences in afferent activity . Inner hair cells are populated by ribbons with a range of sizes , each of which is innervated by a unique afferent fiber . Compared to smaller ribbons , larger ribbons within inner hair cells are innervated by afferent fibers with higher thresholds of activation and lower rates of spontaneous activity ( Furman et al . , 2013; Kalluri and Monges-Hernandez , 2017; Liberman et al . , 2011; Liberman et al . , 2015; Liberman et al . , 1990; Merchan-Perez and Liberman , 1996; Song et al . , 2016; Yin et al . , 2014 ) . Interestingly , in mice differences in ribbon size can be distinguished just after the onset of hearing ( Liberman and Liberman , 2016 ) . This timing suggests that similar to our data ( Figures 4–5 ) , activity during development may help determine ribbon size . Previous work in the zebrafish-lateral line has also examined how ribbon enlargement impacts synapse function ( Sheets et al . , 2017 ) . This work overexpressed Ribeye in zebrafish hair cells to dramatically enlarge ribbons . Functionally , compared to controls , hair cells with enlarged ribbons were associated with afferent neurons with lower spontaneous activity ( Sheets et al . , 2017 ) . Furthermore , the onset encoding , or the timing of the first afferent spike upon stimulation , was significantly delayed in hair cells with enlarged ribbons . Together , both studies in zebrafish and mammals indicate that ribbon size can impact the functional properties of the synapse . Based on these studies , we predict that the alterations to ribbon size we observed in our current study would impact functional properties of the synapse in a similar manner . For example , pharmacological treatments that enlarge ribbons ( Figure 5: MCU channel block; Figure 7: exogenous NAD+ ) would also lower spontaneous spiking in afferents and delay onset encoding . In this study , we found that NAD ( H ) redox state had a dramatic effect on ribbon formation . NAD+ promotes while NADH reduces ribbon size ( Figure 7 ) . The main component of ribbons is Ribeye . Ribeye has two domains , a unique A domain and a B domain that contains an NAD ( H ) binding domain ( Schmitz et al . , 2000 ) . In vitro work on isolated A and B domains has shown that both NAD+ and NADH can affect interactions between A and B domains as well as interactions between B domains ( Magupalli et al . , 2008 ) . In the context of ribbons , the B domain has been shown to concentrate at the interface between the ribbon and the membrane opposing the postsynapse ( Sheets et al . , 2014 ) . Therefore , promoting B domain homodimerization may act to seed larger ribbons at the presynapse . In this scenario , NAD+ and NADH could increase and decrease B domain homodimerization to impact ribbon size . We also observed an increase in cytoplasmic Ribeye aggregates after MCU block ( Figure 5F–G ) . Therefore , it is alternatively possible that NAD+ and NADH could impact interactions between A and B domains more broadly . NAD ( H ) redox could alter Ribeye interactions and alter the overall accumulation or separation of Ribeye within aggregates or at the presynapse . Work in zebrafish has characterized lateral-line hair cells largely depleted of full-length Ribeye ( Lv et al . , 2016 ) . When viewed using TEM , ribbons in Ribeye-depleted hair cells are strikingly transparent , suggesting that full-length Ribeye is required for the characteristic electron-dense structure of ribbons . Although these ribbons are smaller compared to controls , they are still able to tether vesicles near the active zone . Ribeye-depleted hair cells could be used to test whether mito-Ca2+ and NAD ( H ) redox regulate ribbons size by impacting Ribeye interactions . If full-length Ribeye and its NAD ( H ) binding domain are the site of regulation , Ribeye-depleted hair cells would be unaffected by perturbations in mito-Ca2+ and NAD ( H ) . Regardless of the exact mechanism , the effect of presynaptic activity , mito-Ca2+ and related changes in NAD ( H ) redox homeostasis may extend beyond the sensory ribbon synapse . Ribeye is a splice variant of the transcriptional co-repressor CtBP2 ( Schmitz et al . , 2000 ) . While the A domain is unique to Ribeye , the B domain is nearly identical to CtBP2 minus the nuclear localization sequence ( NLS ) ( Hübler et al . , 2012 ) . In vertebrates , the CtBP family also includes CtBP1 ( Chinnadurai , 2007 ) . CtBP proteins are expressed in both hair cells and the nervous system , and there is evidence that both CtBP1 and CtBP2 may act as scaffolds at neuronal synapses . Interestingly , in cultured neurons , it has been shown that both synaptic activity and increased levels of NADH were associated with increased CtBP1 localization at the presynapse ( Ivanova et al . , 2015 ) . In our in vivo study , we also found that the NAD+/NADH ratio was lower ( more NADH ) in developing hair cells with intact presynaptic- and mito-Ca2+ activities ( Figure 6H ) . But in contrast to the in vitro work on CtBP1 in cultured neurons , we found that Ribeye localization to the presynapse and ribbon size were reduced when NADH levels were increased ( Figure 7A–C ) . It is unclear why presynaptic activity regulates Ribeye localization differently from that of CtBP1 . Ribeye and CtBP1 behavior may differ due to the divergent function of their N-terminal domains . Synaptic localization may also be influenced by external factors , such as the cell type in which the synapse operates , whether the study is performed in vitro or in vivo , as well as the maturity of the synapse . Overall , both studies demonstrate that the presynaptic localization of CtBP family members CtBP1 and Ribeye can be influenced by synaptic activity and NAD ( H ) redox state . Sensory hair cells are metabolically demanding cells – both apical mechanotransduction and basal neurotransmission are energy demanding processes ( Shin et al . , 2007; Spinelli et al . , 2012 ) . Therefore , it is likely that hair-cell mitochondria play important roles in both of these functional domains . In mammalian auditory hair cells , mito-Ca2+ uptake has been observed to buffer Ca2+ beneath mechanosensory hair bundles ( Beurg et al . , 2010; Fettiplace and Nam , 2019 ) . Blocking this uptake prolonged evoked Ca2+ rises in hair bundles . This work suggested that apical mitochondria , along with the plasma membrane Ca2+-ATPase ( PMCA ) contribute to cyto-Ca2+ clearance to maintain optimal mechanotransduction ( Beurg et al . , 2010 ) . Although the focus of our present study was on the synapse , we also found that blocking mito-Ca2+ uptake using Ru360 ( MCU antagonist ) or TRO 19622 ( VDAC antagonist ) increased mechanosensitive-Ca2+ responses in zebrafish lateral-line hair bundles ( Figure 2—figure supplement 1A–B’ ) . In the future it will be extremely interesting to explore the role apical mitochondria play in mechanotransduction . In the presynaptic region of hair cells , the link between mito-Ca2+ uptake and neurotransmission is less clear . Studies of synapses in various neuronal subtypes have demonstrated that mitochondria play multiple roles to maintain neurotransmission including: ATP production , Ca2+ buffering and signaling , and neurotransmitter synthesis ( reviewed in Kann and Kovács , 2007; Vos et al . , 2010 ) . A study on synaptic mitochondria at ribbon synapses in retinal-bipolar cells found that mito-Ca2+ uptake was sporadic and did not significantly contribute to Ca2+ clearance or the time course of evoked presynaptic-Ca2+ responses ( Zenisek and Matthews , 2000 ) . This work concluded that mitochondria may contribute indirectly to Ca2+ clearance from the synaptic terminal by providing ATP to fuel the PMCA . Our current work indicates that there is robust and reproducible mito-Ca2+ uptake at the hair-cell presynapse during stimulation . But similar to work on retinal-bipolar cell ribbons , blocking mito-Ca2+ uptake did not raise cyto-Ca2+ levels , indicating it may not be critical for Ca2+ clearance ( Figure 6A–C ) . Instead , cyto-Ca2+ levels may be maintained by the PMCA ( Beurg et al . , 2010; Bortolozzi et al . , 2010 ) . Alternatively , cyto-Ca2+ levels may be maintained by the numerous Ca2+ buffering proteins such as parvalbumin , calretinin , oncomodulin , calbindin and calmodulin that have been identified in hair cells ( Dechesne et al . , 1991; Eybalin and Ripoll , 1990; Hackney , 2005; Pack and Slepecky , 1995; Pangršič et al . , 2015; Rabié et al . , 1983; Simmons et al . , 2010 ) . In our current study on basal , synaptic mitochondria , we found that in mature zebrafish-hair cells , mito-Ca2+ uptake was critical for presynaptic-Ca2+ influx . Even partial block of evoked mito-Ca2+ uptake was sufficient to impair presynaptic-Ca2+ influx , especially during sustained stimuli ( Figure 2C–D’ , Figure 2—figure supplement 1D–E’ ) . Instead of buffering Ca2+ , our work indicates that mito-Ca2+ uptake may impact CaV1 . 3-channel density ( Figure 2E–H ) . In mature hair cells , after MCU block , impaired presynaptic-Ca2+ responses coincided with an increase in CaV1 . 3-channel density at the presynapse ( Figure 2C–H ) . Unfortunately , the majority of studies on CaV1 . 3 channels in hair cells focus on activity changes after a decrease or loss of CaV1 . 3-channel clustering . For example in Ribeye-depleted zebrafish hair cells CaV1 . 3 channels failed to cluster ( Lv et al . , 2016 ) . In addition , when ribbons were enlarged in zebrafish-hair cells , CaV1 . 3-channel density was reduced ( Sheets et al . , 2017 ) . In these studies , after a loss or reduction of CaV1 . 3-channel clustering , presynaptic-Ca2+ signals were increased . Therefore , it is possible that an increase in CaV1 . 3-channel density could incur the opposite effect and decrease presynaptic-Ca2+ responses . But how could an increase in CaV1 . 3-channel density decrease presynaptic-Ca2+ responses ? An increase in CaV1 . 3-channel density could enhance Ca2+-dependent inactivation among tightly clustered CaV1 . 3 channels . In hair cells , CaV1 . 3 channels exhibit reduced Ca2+ dependent inactivation ( Koschak et al . , 2001; Platzer et al . , 2000; Song et al . , 2003; Xu and Lipscombe , 2001 ) . This reduction is thought to be important to transmit sustained sensory stimulation ( Kollmar et al . , 1997 ) . Alternatively , an increase in CaV1 . 3-channel density could be a compensatory strategy to boost presynaptic activity after MCU block and impaired presynaptic-Ca2+ influx . If channel density is not responsible for impaired presynaptic function , mito-Ca2+ uptake could be critical to produce energy for other cellular tasks to maintain neurotransmission . Additional work is necessary to fully understand how evoked mito-Ca2+ uptake functions to sustain presynaptic-Ca2+ influx in mature zebrafish hair cells . In addition to a role in synapse function , mitochondria have been studied in the context of cellular metabolism and cell death ( Devine and Kittler , 2018; Tait and Green , 2013; Vakifahmetoglu-Norberg et al . , 2017 ) . Our work suggests that mitochondria may play distinct roles in these processes in developing and mature hair cells . We found that mitochondria spontaneously take up Ca2+ at the presynapse during hair-cell development ( Figure 4B–C ) . Blocking presynaptic- and mito-Ca2+ activities rapidly decreased the NAD+/NADH ratio and altered ribbon size in developing hair cells ( Figures 5 , 6 and 7 ) . However , in mature hair cells , blocking these activities was pathological and did not influence NAD ( H ) redox ( Figure 6I ) . Some insight into these differences can be inferred from cardiac myocytes where the relationship between mito-Ca2+ and NAD ( H ) redox has been extensively studied . Similar to our results in developing hair cells , in cardiac myocytes , mito-Ca2+ drives cellular metabolism , which reduces NAD+ to NADH ( Bertero and Maack , 2018 ) . In cardiac myocytes NADH is oxidized to NAD+ when the MCU is blocked . These results are consistent with the changes in NAD ( H ) redox we observed in developing , but not mature hair cells . Instead , after complete MCU block in mature hair cells , we observed a loss of hair cells and synapses , and an increase in ribbon size ( Figure 3 ) . This outcome may be more similar to what occurs in heart failure or after extended MCU block – in cardiac myocytes , the production of oxidized NAD+ quickly leads to energetic deficits , oxidative stress and ultimately the generation of reactive oxygen species ( ROS ) ( Bertero and Maack , 2018 ) . This is consistent with work in many cell types where mito-Ca2+ loading is associated with pathological processes such as ROS production , cell death and synapse loss ( Cai and Tammineni , 2016; Court and Coleman , 2012; DiMauro and Schon , 2008; Esterberg et al . , 2013; Esterberg et al . , 2014; Sheng and Cai , 2012 ) . Therefore , in mature hair cells , it is possible that after MCU block , changes in NAD ( H ) redox quickly become pathological . Recent work has suggested that younger hair cells may be more resilient to ototoxins , perhaps because they have not yet accumulated an excess of mitochondrial oxidation ( Pickett et al . , 2018 ) . This could explain why complete MCU block alters NAD ( H ) redox without any observable pathological consequence in developing hair cells . How could synapses be changing in mature hair cells during this pathology ? It is possible that individual ribbons in mature hair cells are not enlarging , but instead ribbons are merging together as synapses are lost . Alternatively , remaining ribbons could be enlarging in order to compensate for loss of presynaptic function or synapses . In our current work , ribbon size was measured in fixed samples; therefore , it is difficult to distinguish between these possibilities . There is considerable work that suggests that synaptic structures , including ribbons are indeed dynamic structures ( Hull et al . , 2006; Mehta et al . , 2013; Sultemeier et al . , 2017 ) . In the future , live imaging studies will help resolve whether there are different mechanisms underlying ribbon enlargement in mature and developing hair cells . Overall the pathology observed in mature hair cells has parallels in recent work on noise-induced hearing that found measurable changes in ribbon morphology and synapse number following noise insult ( Jensen et al . , 2015; Kujawa and Liberman , 2009; Liberman et al . , 2015 ) . Work studying this type of hearing loss has shown that auditory inner hair cells in the high frequency region of the mouse cochlea have enlarged ribbons immediately after noise , followed later by synapse loss ( Liberman et al . , 2015 ) . This pathology is reminiscent of our 1 hr pharmacological treatments that completely block the MCU in mature zebrafish-hair cells ( Figure 3E–F ) . After this treatment , we observed a reduction in the number of hair cells and synapses , and an increase in ribbon size . Overall , these studies and our own data in mature hair cells support the association between mito-Ca2+ and the MCU with pathological processes associated with ototoxins and noise-exposure . In further support of this idea , recent work in mice has investigated the role of the MCU in noise-related hearing loss ( Wang et al . , 2018 ) . This work demonstrated that pharmacological block or a loss of function mutation in MCU protected against synapse loss in auditory inner hair cells after noise exposure . Although this result is counter to our observed results where complete MCU block reduced synapse number ( Figure 3E ) , it highlights an association between mito-Ca2+ , noise exposure and synapse integrity . It is possible that these differences can be explained by transitory versus chronic alterations in mito-Ca2+ homeostasis . These differences may be resolved by studying hair cells in a zebrafish MCU knock out . In the future it will be interesting to examine both mito-Ca2+ uptake and ribbon morphology during other pathological conditions that enlarge ribbons such as noise exposure , ototoxicity and aging . Overall , our study has demonstrated the zebrafish-lateral line is a valuable system to study the interplay between the mitochondria , and synapse function , development and integrity . In the future it will be exciting to expand this research to explore how evoked and spontaneous mito-Ca2+ influx are impacted by pathological treatments such as age , noise and ototoxins . Adult Danio rerio ( zebrafish ) were maintained under standard conditions . Larvae 2 to 6 days post-fertilization ( dpf ) were maintained in E3 embryo medium ( in mM: 5 NaCl , 0 . 17 KCl , 0 . 33 CaCl2 and 0 . 33 MgSO4 , buffered in HEPES pH 7 . 2 ) at 28°C . All husbandry and experiments were approved by the NIH Animal Care and Use program under protocol #1362–13 . Transgenic zebrafish lines used in this study include: Tg ( myo6b:GCaMP6s-CAAX ) idc1 ( Jiang et al . , 2017 ) , Tg ( myo6b:RGECO1 ) vo10Tg ( Maeda et al . , 2014 ) , Tg ( myo6b:GCaMP3 ) w78Tg ( Esterberg et al . , 2013 ) , Tg ( myo6b:mitoGCaMP3 ) w119Tg ( Esterberg et al . , 2014 ) , and Tg ( myo6b:ribeye a-tagRFP ) idc11Tg ( Sheets , 2017 ) . Experiments were performed using Tübingen or TL wildtype strains . To create transgenic fish , plasmid construction was based on the tol2/Gateway zebrafish kit developed by the lab of Chi-Bin Chien at the University of Utah ( Kwan et al . , 2007 ) . These methods were used to create Tg ( myo6b:mitoRGECO1 ) idc12Tg and Tg ( myo6b:Rex-YFP ) idc13Tg transgenic lines . Gateway cloning was used to clone Rex-YFP ( Bilan et al . , 2014 ) and mitoRGECO1 into the middle entry vector pDONR221 . For mitochondrial matrix targeting , the sequence of cytochrome C oxidase subunit VIII ( Rizzuto et al . , 1989 ) was added to the N-terminus of RGECO1 . Vectors p3E-polyA ( Kwan et al . , 2007 ) and pDestTol2CG2 ( Kwan et al . , 2007 ) were recombined with p5E-myosinVIb ( myo6b ) ( Kindt et al . , 2012 ) and our engineered plasmids to create the following constructs: myo6b:REX-YFP and myo6b:mitoRGECO1 . To generate transgenic fish , DNA clones ( 25–50 ng/μl ) were injected along with tol2 transposase mRNA ( 25–50 ng/μl ) into zebrafish embryos at the single-cell stage . For pharmacological studies , zebrafish larvae were exposed to compounds diluted in E3 with 0 . 1% DMSO ( Isradipine , Bay K8644 , NAD+ ( Sigma-Aldrich , St . Louis , MO ) , Ru360 ( Millipore , Burlington , MA ) , TRO 19622 ( Cayman Chemical , Ann Arbor , MI ) ) or Tris-HCl ( NADH ( Cayman Chemical , Ann Arbor , MI ) ) for 30 min or 1 hr at the concentrations indicated . E3 with 0 . 1% DMSO or Tris-HCl were used as control solutions . In solution at pH 7 . 0–7 . 3 , NADH oxidizes into NAD+ by exposure to dissolved oxygen . To mitigate this , NADH was dissolved immediately before use and was exchanged with a freshly dissolved NADH solution every half hour . Dosages of isradipine , Ru360 , Bay K8644 , TRO 19622 , NAD+ and NADH did not confer excessive hair-cell death or synapse loss unless stated . After exposure to the compounds , larvae were quickly sedated on ice and transferred to fixative . To prepare larvae for imaging , larvae were immobilized as previously described ( Kindt et al . , 2012 ) . Briefly , larvae were anesthetized with tricaine ( 0 . 03% ) in E3 and pinned to a chamber lined with Sylgard 184 Silicone Elastomer ( Dow Corning , Midland , MI ) . Larvae were injected with 125 µM α-bungarotoxin ( Tocris , Bristol , UK ) into the pericardial cavity to induce paralysis . Tricaine was rinsed off the larvae and replaced with fresh E3 . For baseline measurements of Rex-YFP and CytoRGECO1 fluorescence , larvae were imaged using an upright Nikon ECLIPSE Ni-E motorized microscope ( Nikon Inc , Tokyo , Japan ) in widefield mode with a Nikon 60 × 1 . 0 NA water-immersion objective , an 480/30 nm excitation and 535/40 nm emission filter set or 520/35 nm excitation and 593/40 emission filter set , and an ORCA-D2 camera ( Hamamatsu Photonics K . K . , Hamamatsu City , Japan ) . Acquisitions were taken at 5 Hz , in 15 plane Z-stacks every 2 μm . For baseline measurements of MitoGCaMP3 , larvae were imaged using a Bruker Swept-field confocal microscope ( Bruker Inc , Billerica , MA ) , with a Nikon CFI Fluor 60 × 1 . 0 NA water-immersion objective . A Rolera EM-C2 CCD camera ( QImaging , Surrey , Canada ) was used to detect signals . Acquisitions were taken using a 70 µm slit at a frame rate of 10 Hz , in 26 plane Z-stacks every 1 μm . MitoGCaMP3 baseline intensity varied dramatically in controls between timepoints . To offset this variability , we acquired and averaged the intensity of 4 Z-stacks per time point . For all baseline measurements transgenic larvae were first imaged in E3 with 0 . 1% DMSO or 0 . 1% Tris-HCl as appropriate . Then larvae were exposed to pharmacological agents for 30 min and a second acquisition was taken . Any neuromasts with cell death after pharmacological or mock treatment were excluded from our analyses . To measure evoked Ca2+ signals in hair cells , larvae were immobilized in a similar manner as described for baseline measurements . After α-bungarotoxin paralysis , larvae were immersed in neuronal buffer solution ( in mM: 140 NaCl , 2 KCl , 2 CaCl2 , 1 MgCl2 and 10 HEPES , pH 7 . 3 ) . Evoked Ca2+ measurements were acquired using the Bruker Swept-field confocal system described above . To stimulate lateral-line hair cells , a fluid-jet was used as previously described to deliver a saturating stimulus ( Lukasz and Kindt , 2018 ) . To measure presynaptic GCaMP6sCAAX signals at ribbons , images were acquired with 1 × 1 binning using a 35 µm slit at 50 Hz in a single plane containing presynaptic ribbons ( Figure 2—figure supplement 1C–C’ ) . Ribbons were marked in live hair cells using the Tg ( myo6b:ribeye a-tagRFP ) idc11Tg transgenic line ( Figure 2—figure supplement 1C ) . Ribbons were located relative to GCaMP6s signals by acquiring a 2-color Z-stack of 5 planes every 1 μm at the base of the hair cells . To correlate presynaptic GCaMP6sCAAX signals with MitoRGECO1 signals in hair cells , 2-color imaging was performed . Images were acquired in a single plane with 2 × 2 binning at 10 Hz with a 70 µM slit . MitoGCaMP3 signals were acquired at 10 Hz in Z-stacks of 5 planes 1 μm apart with 2 × 2 binning and a 70 µM slit . High speed imaging along the Z-axis was accomplished by using a piezoelectric motor ( PICMA P-882 . 11–888 . 11 series , Physik Instrumente GmbH , Karlsruhe , Germany ) attached to the objective to allow rapid imaging at a 50 Hz frame rate yielding a 10 Hz volume rate . Due to the slow mito-Ca2+ return to baseline after stimulation ( ~5 min ) , we waited a minimum of 5 min before initiating a new evoked GCaMP6sCAAX or MitoGCaMP3 acquisition . To examine mechanotransduction , GCaMP6sCAAX signals were measured in apical hair bundles ( Figure 2—figure supplement 1A–B’; Zhang et al . , 2018 ) . Apical GCaMP6sCAAX signals were acquired in a single plane at 1 × 1 binning with a 35 µM slit at 20 Hz . For pharmacological treatment , acquisitions were made prior to drug treatment and after a 30 min incubation in the pharmacological agent . Any neuromasts with cell death after pharmacological treatment were excluded from our analyses . To measure spontaneous Ca2+ signals in hair cells , larvae were prepared in a similar manner as described for evoked Ca2+ measurements . Spontaneous Ca2+ measurements were acquired using the Bruker Swept-field confocal system described above . To measure spontaneous presynaptic GCaMP6sCAAX signals , images were acquired with 2 × 2 binning with a 70 µm slit at 0 . 33 Hz in a single plane for 900 s . For acquisition of two-color spontaneous presynaptic GCaMP6sCAAX and MitoRGECO1 signals images were acquired with 2 × 2 binning with a 70 µm slit at 0 . 2 Hz in a single plane for 900 s . Larvae were prepared for electron microscopy as described previously ( Sheets et al . , 2017 ) . Transverse serial sections ( ~60 nm thin sections ) were used to section through neuromasts . Samples were imaged on a JEOL JEM-2100 electron microscope ( JEOL Inc , Tokyo , Japan ) . The distance from the edge of a ribbon density to the edge of the nearest mitochondrion was measured ( n = 17 ribbons ) . A subset of measurements was taken from more than one ribbon within a hair cell . At 81% of ribbons , a mitochondrion could be clearly identified within 1 µm of a ribbon ( 17 out of 21 ribbons ) . All distances and perimeters were measured in FIJI ( Schindelin et al . , 2012 ) . Whole larvae were fixed with 4% paraformaldehyde in PBS at 4°C for 3 . 5–4 hr as previously described ( Zhang et al . , 2018 ) . Fixative was washed out with 0 . 01% Tween in PBS ( PBST ) in four washes , 5 min each . Larvae were then washed for 5 min with H2O . The H2O was thoroughly removed and replaced with ice-cold acetone and placed at −20°C for 3 min for 3 dpf and 5 min for 5 dpf larvae , followed by a 5 min H2O wash . The larvae were then washed for 4 × 5 min in PBST , then incubated in block overnight at 4°C in blocking solution ( 2% goat serum , 1% bovine serum albumin , 2% fish skin gelatin in PBST ) . Primary and secondary antibodies were diluted in blocking solution . Primary antibodies and their respective dilutions are: Ribbon label: Mouse anti-Ribeye b IgG2a , 1:10 , 000 ( Sheets et al . , 2011 ) ; PSD label: Mouse anti-pan-MAGUK IgG1 , 1:500 ( MABN72 , MilliporeSigma , Burlington , MA ) ; Hair-cell label: Rabbit anti-Myosin VIIa , 1:1000 ( #25–6790 , Proteus BioSciences Inc , Ramona , CA ) ; CaV1 . 3 channel label: Rabbit anti-CaV1 . 3a , 1:500 ( Sheets et al . , 2012 ) . Larvae were incubated in primary antibody solution for 2 hr at room temperature . After 4 × 5 min washes in PBST to remove the primary antibodies , diluted secondary antibodies were added and samples were incubated for 2 hr at room temperature . Secondary antibodies and their respective dilution are as follows: goat anti-mouse IgG2a , Alexa Fluor 488 , 1:1000; goat anti-rabbit IgG ( H+L ) Alexa Fluor 546 , 1:1000; goat anti-mouse IgG1 Alexa Fluor 647 , 1:1000 ( Thermo Fisher Scientific , Waltham , MA ) . Secondary antibodies were washed out with PBST for 3 × 5 min , followed by a 5 min wash with H2O . Larvae were mounted on glass slides with Prolong Gold Antifade Reagent ( Invitrogen , Carlsbad , CA ) using No . 1 . 5 coverslips . Prior to Airyscan imaging , live samples were immobilized in 2% low-melt agarose in tricaine ( 0 . 03% ) in cover-glass bottomed dishes . Live and fixed samples were imaged on an inverted Zeiss LSM 780 laser-scanning confocal microscope with an Airyscan attachment ( Carl Zeiss AG , Oberkochen , Germany ) using an 63 × 1 . 4 NA oil objective lens . The median ( ±median absolute deviation ) lateral and axial resolution of the system was measured at 198 ± 7 . 5 nm and 913 ± 50 nm ( full-width at half-maximum ) , respectively . The acquisition parameters were adjusted using the control sample such that pixels for each channel reach at least 1/10 of the dynamic range . The Airyscan Z-stacks were processed with Zeiss Zen Black software v2 . 1 using 3D filter setting of 7 . 0 . Experiments were imaged with the same acquisition settings to maintain consistency between comparisons . To quantify changes in baseline Ca2+ and NAD ( H ) homeostasis , images were processed in FIJI . For our measurements we quantified the fluorescence in the basal-most 8 μm ( four planes ) to avoid overlap between cells . The basal planes were max Z-projected , and a 24 . 0 μm ( Rex-YFP and RGECO1 ) or 26 . 8 μm ( MitoGCaMP3 ) circular region of interest ( ROI ) was drawn over the neuromast to make intensity measurements . To correct for photobleaching , a set of mock-treated control neuromasts were imaged during every trial . These mock treatments were used to normalize the post-treatment intensity values . To quantify the magnitude of evoked changes in Ca2+ , images were processed in FIJI . Images in each time series were aligned using Stackreg ( Thévenaz et al . , 1998 ) . For evoked MitoRGECO1 , MitoGCaMP3 , CytoGCaMP3 and two-color GCaMP6sCAAX and MitoRGECO1 signals , Z-stacks were max z-projected , and a 5 μm diameter circular ROI was drawn over each hair cell to make intensity measurements . For ribbon-localized measurements , GCaMP6sCAAX signals were measured within 1 . 34 μm round ROIs at individual ribbons , and intensity change at multiple ribbons per cell were averaged . For measurements of mechanotransduction , GCaMP6sCAAX signals were measured within 1 . 34 μm round ROIs at individual hair bundles , and intensity change in multiple bundles per neuromast were averaged . To plot evoked changes in Ca2+ , we subtracted the baseline ( F0 , signal during the pre-stimulus period ) was subtracted from each timepoint acquired . Then each timepoint was divided by F0 to generate the relative change in fluorescent signal from baseline or ∆F/F0 . Quantification of evoked Ca2+ signals were made on max ∆F/F0 measurements . Cells with presynaptic-Ca2+ activity are defined by max ΔF/F0 of >0 . 05 for MitoRGECO1 and MitoGCaMP3 , and max ΔF/F0 >0 . 25 for GCaMP6sCAAX for a 2 s stimulation . The method to obtain and overlay the spatial signal distribution of evoked signals as heat maps has been previously described ( Lukasz and Kindt , 2018 ) . We first computed the baseline image ( F0 or reference image ) by averaging the images over the pre-stimulus period . Then the baseline image ( F0 ) was subtracted from each image acquired , to represent the relative change in fluorescent signal from baseline or ∆F . The ∆F signal images during the stimulus period were binned , scaled and encoded by color maps with red indicating an increase in signal intensity . To quantify the average magnitude and frequency of spontaneous Ca2+ changes in GCaMP6sCAAX signals , images were processed in Matlab R2014b ( Mathworks , Natick , MA ) and ImageJ ( NIH , Bethesda , MD ) . First , images in each time series were aligned in ImageJ using Stackreg ( Thévenaz et al . , 1998 ) . To measure the average magnitude during the 900 s GCaMP6sCAAX image acquisition , a 5 μm diameter circular ROI was drawn over each hair cell and a raw intensity value was obtained from each time point . Then , in Matlab , the raw traces were bleach corrected . Next , the corrected intensity values were normalized as ∆F/F0 . For spontaneous Ca2+ signals F0 is defined as the bottom 15th percentile of fluorescence values ( Babola et al . , 2018 ) . Then , values of ∆F/F0 of less than 10% were removed . These values were considered to be noise and our threshold value for a true signal . A 10% threshold was determined by imaging spontaneous GCaMP6CAAX signals in the presence of isradipine where no signals were observed ( Figure 4—figure supplement 1 ) . The averaged magnitude of spontaneous activity per cell was obtained by dividing the integral/sum of GCaMP6sCAAX signals ( ∆F/F0 >10% ) during the whole recording period by 300 ( 300 frames in 900 s ) . The frequency of GCaMP6sCAAX signals was defined as the average number of peaks per second during the whole recording period . To quantify synapse morphology and pairing , images were first processed in ImageJ , and then synapses were paired using Python ( Python Software Foundation , Wilmington , DE ) in the Spyder Scientific Environment ( MIT , Cambridge , MA ) . In ImageJ , each Airyscan Z-stack was background subtracted using rolling-ball subtraction . Z-stacks containing the MAGUK channel were further bandpass filtered to remove details smaller than six px and larger than 20 px . A duplicate of each Z-stack was normalized for intensity . This duplicated Z-stack was used to identify individual ribbon and MAGUK using the Simple 3D Segmentation of ImageJ 3D Suite ( Ollion et al . , 2013 ) . Local intensity maxima , identified with 3D Fast Filter , and 3D watershed were used to separate close-by structures . The centroids for each identified ribbon and MAGUK puncta were obtained using 3D Manager and these coordinates were used to identify complete synapses . The max Z-projection of the segmented Z-stack was used to generate a list of 2D objects as individual ROIs corresponding to each punctum . This step also included a minimum size filter: Ribeye: 0 . 08 μm2 , MAGUK: 0 . 04 μm2 . For quantification of extrasynaptic Ribeye b puncta , the minimum size filter was not applied . The 2D puncta ROI were applied over the max Z-projection of the original Z-stack processed only with background subtraction . This step measures the intensity of the antibody label . Centroid and intensity information were exported as a CSV spreadsheet ( macro is available on https://github . com/wonghc/ImageJ-ribbon-synapse-quantification , Wong , 2019; copy archived at https://github . com/elifesciences-publications/ImageJ-ribbon-synapse-quantification ) . In Python , the 3D centroid coordinates for each ribbon punctum were measured against the coordinates of every post-synaptic MAGUK punctum to find the MAGUK punctum within a threshold distance . This threshold was calculated by taking the 2D area of the Ribeye and MAGUK punctum measured in the max Z-projection to calculate an approximate radius by dividing by π and taking the square root . The two radii were then summed to get the threshold . Puncta that were not paired were excluded from later statistical analyses of synaptic ribbon and postsynaptic MAGUK puncta . To quantify the amount of CaV1 . 3 immunolabel at ribbons , 2D ROIs generated from the Ribeye label to generate ribbon areas were applied to a max Z-projection of the CaV1 . 3 immunolabel . The integrated intensity of CaV1 . 3 immunolabel was measured within each ROI . The number of hair cells , synapses per cell , and CaV1 . 3 clusters per PSD were counted manually . Hair-cell counts were assayed with Myosin VIIa antibody label in treatments when synapse or cell numbers were reduced . Due to slight variability between clutches and immunostains we only compared experimental data taken from the same clutch , immunostain and imaging session . Statistical analyses and data plots were performed with Prism 8 ( Graphpad , San Diego , CA ) . Values of data with error bars on graphs and in text are expressed as mean ± SEM unless indicated otherwise . All experiments were performed on a minimum of 2 animals , 6 neuromasts ( posterior lateral-line neuromasts L1-L4 or anterior lateral-line neuromasts O1 and O2 [Figure 1—figure supplement 2 , Figure 3—figure supplement 1 , Figure 5—figure supplement 1] ) , on two independent days . For 3 and 5 dpf larvae each neuromast represents analysis from 8 to 12 hair cells; 24–36 synapses and 14–18 hair cells; 42–54 synapses respectively . All replicates are biological . Based on the variance and effect sizes reported previously and measured in this study , these numbers were adequate to provide statistical power to avoid both Type I and Type II error ( Sheets et al . , 2012; Zhang et al . , 2018 ) . No animals or samples were excluded from our analyses unless control experiments failed–in these cases all samples were excluded . No randomization or blinding was used for our animal studies . Where appropriate , data was confirmed for normality using a D’Agostino-Pearson normality test and for equal variances using a F test to compare variances . Statistical significance between two conditions was determined by either an unpaired t -test , an unpaired Welch’s unequal variance t-test , a Mann-Whitney U test or a Wilcoxon matched-pairs signed-rank test as appropriate . For comparison of multiple conditions , a Brown-Forsythe with Dunnett’s T3 post hoc or a Brown-Forsythe and Welch ANOVA with Holm-Sidak’s post hoc were used as appropriate . To calculate the IC50 for Ru360 block of evoked MitoGCaMP3 signals a dose response curve was plotted using 0 , 0 . 5 , 2 , 5 and 10 µM Ru360 . A non-linear fit with four parameters and a variable slope was performed to calculate an IC50 of 1 . 37 µM .	Hearing depends upon specialized cells deep within the ear called hair cells . These cells take their name from the bundles of hair-like fibers found on their surface , which move when sound waves enter the ear . This movement activates the hair cells , which send signals to nearby neurons at contact points called synapses . Hair cells must send messages to their synaptic partners rapidly and continuously in order for humans to follow complex streams of sound , such as speech . This requires large amounts of energy , which are produced by compartments inside the hair cells called mitochondria . Wong et al . show that mitochondria , which are often described as the ‘power plants’ of cells , are critical for hair cell synapses to form and work correctly . But rather than studying hair cells in the human ear , Wong et al . took advantage of the fact that another species – the zebrafish – has hair cells on its body surface . These cells detect movements in water rather than sound waves , but they work in much the same way as hair cells in the ear , and are easier to access and study . Wong et al . report that in zebrafish larvae , developing hair cells send spontaneous signals to their contact neurons even before they start receiving any sensory input . But if mitochondria in the hair cells fail to detect these signals , the synapses fail to form correctly . In older zebrafish , mature hair cells send signals to their synaptic partners whenever they detect sensory input . But if mitochondria fail to detect these signals , the synapses stop working and ultimately break down . These findings help explain why damage to mitochondria in the inner ear can lead to hearing loss . Moreover , because mitochondria are present in almost all cells , their disruption causes a wide range of diseases . Many of these involve the brain , which requires large amounts of energy and so is particularly vulnerable to mitochondrial damage . These results may provide insights into such disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2019	Synaptic mitochondria regulate hair-cell synapse size and function
Outer membrane TonB-dependent transporters facilitate the uptake of trace nutrients and carbohydrates in Gram-negative bacteria and are essential for pathogenic bacteria and the health of the microbiome . Despite this , their mechanism of transport is still unknown . Here , pulse electron paramagnetic resonance ( EPR ) measurements were made in intact cells on the Escherichia coli vitamin B12 transporter , BtuB . Substrate binding was found to alter the C-terminal region of the core and shift an extracellular substrate binding loop 2 nm toward the periplasm; moreover , this structural transition is regulated by an ionic lock that is broken upon binding of the inner membrane protein TonB . Significantly , this structural transition is not observed when BtuB is reconstituted into phospholipid bilayers . These measurements suggest an alternative to existing models of transport , and they demonstrate the importance of studying outer membrane proteins in their native environment . The passive permeation of low molecular weight solutes across the outer membrane ( OM ) of Gram-negative bacteria is typically facilitated by porins . However , many higher molecular weight solutes and trace nutrients , including carbohydrates , iron siderophores , cobalamin , copper , and nickel , are bound and transported across the OM by a family of active transporters that are TonB-dependent ( Bolam and van den Berg , 2018; Noinaj et al . , 2010 ) . These TonB-dependent transporters ( TBDTs ) derive energy from the bacterial inner membrane by interacting with TonB , a transperiplasmic protein that interacts with the inner membrane proteins ExbB and ExbD ( Celia et al . , 2016; Maki-Yonekura et al . , 2018 ) . This family of transporters has two distinct domains: a β-barrel formed from 22 anti-parallel β-strands , and a core or hatch domain that fills the interior of the barrel . The β-strands on the extracellular surface of the protein barrel are often connected by long loops , while short turns join the strands on the periplasmic interface ( see Figure 1B ) . TonB interacts with the transporter through a conserved motif on the N-terminal side of the core termed the Ton box ( Pawelek et al . , 2006; Shultis et al . , 2006 ) . The mechanism of transport in TBDTs is presently poorly understood; however , the large size of most substrates and the absence of any obvious pathway for substrate permeation ( Faraldo-Gómez et al . , 2003 ) have led to proposals that transport is mediated by a significant conformational event that involves a partial rearrangement or full removal of the core domain from the surrounding barrel ( Chimento et al . , 2005 ) . Although many high-resolution structures are available for TBDTs , there is no direct evidence for a major structural change within the core of TBDTs that might indicate a transport mechanism . In the Escherichia coli vitamin B12 ( cobalamin ) transporter , BtuB , EPR spectroscopy shows that substrate binding unfolds the Ton box at the N-terminus and extends it into the periplasm , an allosteric event that may facilitate the binding of TonB to BtuB ( Kim et al . , 2007; Xu et al . , 2006 ) ; however , no other significant structural changes have been observed in the core . High-resolution crystal structures have been obtained for a C-terminal fragment of TonB in complex with BtuB , the ferrichrome transporter FhuA , and the ferrioxamine B transporter FoxA ( Pawelek et al . , 2006; Shultis et al . , 2006; Josts et al . , 2019 ) . When TonB binds , the Ton box extends from the core and interacts with the β-sheets of TonB in an edge-to-edge manner . Except for the Ton box , the remainder of the core remains folded and is essentially unchanged . Because TonB binding does not alter the core of BtuB in the BtuB-TonB structure , it has been proposed that TonB alters the core by exerting a mechanical force on the transporter , and current models for transport favor a mechanism where TonB acts by pulling the Ton box thereby unfolding the core ( Gumbart et al . , 2007; Hickman et al . , 2017; Sverzhinsky et al . , 2015 ) . Models involving a rotation of TonB have also been proposed ( Klebba , 2016 ) ; however , in FhuA there are four to five unstructured residues between the Ton box and core when TonB is bound , making the transfer of torque from TonB to the core unlikely ( Sarver et al . , 2018 ) . Pulling models have been explored using steered molecular dynamics ( MD ) ( Gumbart et al . , 2007 ) as well as single-molecule AFM ( atomic force microscopy ) pulling experiments ( Hickman et al . , 2017 ) , and these studies indicate that an N-terminal region of the core ( up to residue 73 ) is preferentially unfolded to permit the movement of vitamin B12 into the periplasm . This work concludes that the C-terminal region of the core is static and does not unfold during transport , a result that is consistent with denaturation experiments on BtuB ( Flores Jiménez and Cafiso , 2012 ) . An important caveat to almost all the structural work on BtuB is that it has been carried out on purified or partially purified protein where the native OM environment is no longer present . Since transport in this family of transporters has never been reconstituted , it has never been established that the isolated , purified , and membrane reconstituted BtuB is capable of transport . Recently , we developed an approach to attach spin labels to either extracellular or periplasmic sites on BtuB in intact cells , thereby permitting EPR measurements to be made under conditions where the protein is known to be functional ( Joseph et al . , 2019; Nilaweera et al . , 2019 ) . Preliminary measurements made on BtuB indicate that it behaves differently in the intact cell than it does in a purified reconstituted phospholipid system . For example , a substrate-dependent change in the core domain of BtuB involving substrate binding loop 3 ( SB3 ) is observed in situ but is not seen in a detergent-treated OM preparation ( Nilaweera et al . , 2019 ) . Moreover , the extracellular loops of BtuB are also highly constrained in the intact cell , and substrate-induced structural changes and structural heterogeneity that is observed for BtuB in proteoliposomes are not observed for BtuB in situ ( Nyenhuis et al . , 2020a ) . In the present work , we perform double electron-electron resonance ( DEER ) on BtuB in intact cells to determine whether structural changes take place in the core domain that are associated with substrate binding and transport . Upon substrate binding , a movement of the core is observed involving sites 90 and 93 in SB3 . No other movements in the core are detected . When the ionic lock is broken between site R14 on the C-terminal side of the Ton box and site D316 in the barrel , long-distance components appear upon substrate binding , indicating that SB3 can assume a state where it has moved into the BtuB barrel toward the periplasmic side of the protein . Under these same conditions , other sites on the N-terminal side of the core remain static . Since this ionic interaction would normally be broken upon TonB binding , this structural transition is likely to take place during transport . This result suggests that transport may involve allosteric changes in the C-terminal side of the core upon TonB binding . Remarkably , these substrate-induced structural changes are not observed for purified , membrane reconstituted BtuB , which may be due to the absence of lipopolysaccharide ( LPS ) or other periplasmic components in the reconstituted system . The importance of the native OM environment provides an explanation for why this structural transition has not been previously observed . To investigate movements that might occur in the BtuB core region in situ , pairs of spin labels were placed into the extracellular region of BtuB , with one label located at an outer loop site that is known to be relatively fixed in the cell environment and a second label located at a site in the BtuB core . Previous work has demonstrated that several sites on the extracellular surface of BtuB may be spin labeled in vivo using site-directed cysteines and a standard methanethiosulfonate reagent to produce the side chain R1 ( Figure 1A ) . These included multiple sites on the extracellular loops of BtuB ( Nilaweera et al . , 2019; Nyenhuis et al . , 2020a; Joseph et al . , 2016 ) as well as two sites in the core region ( Nilaweera et al . , 2019 ) . Recent work has also shown that the efficient incorporation of pairs of spin labels to make distance measurements using DEER required the use of a strain deficient in the disulfide bond formation ( Dsb ) chaperone system ( Nilaweera et al . , 2019 ) . We tested several additional single cysteine mutants in BtuB using a DsbA- strain to determine whether spin labeling of additional sites in the core was possible . Shown in Figure 1c are spectra from site 90 , which was previously labeled , as wells as four additional sites in the core region . Sites 63 and 65 lie in substrate binding loop 1 ( SB1 ) , site 72 lies in substrate binding loop 2 ( SB2 ) , and sites 90 and 93 are positioned in SB3 . These spectra arise from label having more than one motional component but are dominated by a broad feature that is characteristic of a population of label with hindered motion on the ns time scale , consistent with the confined environment in the extracellular region of the core . For sites 63 , 72 , and 93 , the addition of vitamin B12 alters the spectra and increases the population of the immobile component indicating that incorporation of the label at these sites has not prevented the binding of substrate . No significant changes with substrate are seen for sites 65 and 90 . At site 90 , substrate does bind ( see below ) and the lack of a change in the EPR spectrum may reflect the fact that in the apo state the label is already highly immobile . Site 65 is also highly immobile , but we cannot exclude the possibly that incorporation of R1 at this site has blocked the binding of vitamin B12 . For distance measurements using pulse EPR , each set of spin pairs included a label at position 188 on the 3/4 extracellular loop ( the second loop connecting β-strands 3 and 4 ) . This site was chosen as a reference point because previous work in whole cells demonstrated that this loop assumed a well-defined position and exhibited minimal or no movement upon substrate addition ( Nyenhuis et al . , 2020a ) . All data were analyzed using LongDistances . Positions of both components were held constant throughout and were set to the average of an initial round of fits where distance was varied . Width and area for the two components were allowed to vary freely . Error ranges were taken from the output of the fitting routine . Shown in Figure 2 are the results for measurements on the V90R1-T188R1 spin pair in cells . Preliminary results from this pair were presented in a previous study demonstrating the use of disulfide chaperone mutants to achieve double labeling of BtuB in whole cells ( Nilaweera et al . , 2019 ) . The background corrected DEER data and resulting distance distributions are shown in Figure 2b , c . Both the apo ( blue ) and vitamin B12 bound ( red ) distributions yield two main intramolecular peaks at 2 . 4 and 3 . 2 nm , with a substrate-dependent shift observed toward the shorter component . Predicted distance distributions were generated from the apo and vitamin B12 bound in surfo crystal structures of BtuB ( PDB IDs: 1NQG and 1NQH ) using the program MMM ( Jeschke , 2018 ) . The distributions generated from these structures also show a shift toward a shorter distance in the substrate bound state , which is due to the unfolding of a helical turn in the SB3 loop ( Chimento et al . , 2003 ) . However , the magnitude of the predicted shift is smaller than that observed by DEER . Because the position of site 188 in the 3/4 extracellular loop of BtuB is not altered with substrate addition in situ ( Nyenhuis et al . , 2020a ) , this structural change must involve a movement of the SB3 apex or a change in rotamers assumed by R1 at position 90 . We also titrated this structural change by measuring the distance distribution with increasing concentrations of the vitamin B12 substrate , where the result is shown in Figure 2D . For this analysis data were processed using a model-based approach with two-Gaussian components where the position , width , and amplitude were varied . As seen in Figure 2D , there is a strong progressive response to increases in substrate until saturation is reached in the range of 30–60 μM vitamin B12 . This titration likely reflects substrate loading . Because substrate concentrations greatly exceed the affinity of vitamin B12 to BtuB ( Bradbeer et al . , 1986 ) , the saturation point likely reflects the concentration of BtuB in our sample and is roughly consistent with the spin concentrations expected from the EPR signal intensity . To determine whether the structural change observed in the apex of SB3 is limited to this site or part of a broader conformational change across the core domain , we tested additional spin pairs on the extracellular face of the protein using the core sites shown in Figure 1B . The spin pairs examined are shown in Figure 3A , B and include site 93 , which also lies in SB3 , as well as sites 63 and 65 in SB1 and site 72 in SB2 . The distance distributions that result from these spin pairs are shown in Figure 3C , along with the V90R1-T188R1 spin pair . It should be noted that the distance distributions in Figure 3 , as well as subsequent figures , have been truncated at 5 nm . As seen in the raw data ( Figure 2—figure supplement 1 and Figure 3—figure supplement 1 ) , a longer distance component is apparent in some of these dipolar evolutions . This is due to the presence of BtuB-BtuB interactions in the OM that leads to a 6 . 5 nm distance component ( Nyenhuis et al . , 2020b ) . As seen in Figure 3C , distance distributions for sites located outside SB3 show little evidence for any structural change upon the addition of substrate , with spin pairs involving sites 63 , 65 , and 72 having nearly identical distributions for both apo and vitamin B12 bound conditions . This lack of a shift in these spin pairs is consistent with work showing that site 188 does not show a substrate-dependent shift in its position in situ ( Nyenhuis et al . , 2020a ) . In the SB3 loop , however , site 93 at the edge of the loop shows a substrate-dependent change in position along with site 90 at the loop apex , although the change is much larger for site 90 ( about 8 Å ) than for site 93 ( about 4 Å ) and in an opposite direction . The structural changes measured by DEER are qualitatively consistent with the predictions from the crystal structures , which show a loss in helical structure and change in position of the loop . For the sites examined here , substrate-dependent changes in the core appear to be confined to the region of the SB3 loop . Interestingly , as we demonstrate below , this substrate-dependent change does not occur when the protein is removed from its native environment . As indicated above , BtuB is expected to bind to both substrate and TonB during transport , which will break the ionic lock between R14 in the core and D316 in the barrel ( Shultis et al . , 2006 ) . However , under the conditions of our experiment , BtuB is in large excess relative to TonB , perhaps by a factor of 10–20 or more . As a result , only a small portion of the BtuB would be bound to TonB at any time during our distance measurement . To determine whether there might be a connection between the structure of the core and this internal ionic lock , we examined the effect of disrupting the R14-D316 interaction on the core by introducing the R14A mutation into the existing pairs of labels between site 188 in the 3/4 extracellular loop and the core . Distance measurements for the apo and vitamin B12 bound states made in the presence of this mutation are shown in Figure 3D . For distance distributions involving sites 63 , 65 , and 72 in SB1 and SB2 , the core remains largely unchanged in response to substrate and unchanged by the R14A mutation . However , distance distributions involving sites 90 and 93 in SB3 are altered by breaking the D316/R14 ionic lock . For distances measured to sites 90 and 93 , the R14A mutation has two main effects . First , for the V90R1-T188R1 pair , the substrate-dependent conversion between the 2 . 4 and 3 . 2 nm distance components is absent , and the shorter distance now dominates in both apo and vitamin B12 bound conditions . Second , for both the V90R1-T188R1 and S93R1-T188R1 pairs , additional distance components are observed in the presence of substrate centered at 3 . 8 and 4 . 5 nm . These long components result in a distribution that is substantially broader than that predicted by the crystal structures , and they indicate the formation of a novel conformation of the SB3 loop and an altered substrate binding mode . The longer distance components that appear for the V90R1-T188R1 and S93R1-T188R1 spin pairs represent a substantial movement of the SB3 loop toward the periplasmic side of BtuB . A movement of SB3 toward the extracellular surface is highly unlikely , in part because movement in this direction would require a major unfolding of the core that we do not observe . For measurements to SB3 , there were relatively few positions that were both accessible and within the range of the pulse EPR measurement , but we made measurements to site 90 from site 237 , which is located near the apex of the 5/6 extracellular loop . In the apo and vitamin B12 bound states , the predicted distances from this site are shorter than 2 nm and are not within a range that can accurately be measured by DEER . The results are shown in Figure 3—figure supplement 2 . In the absence of the R14A mutation , no clear substrate-dependent shift in the position of SB3 is observed , which is likely due to the short distance involved . However , in the presence of the R14A mutation , a new distance component appears with substrate addition around 3 nm that is beyond the distance range predicted by the crystal structure . This is shorter than the 4 . 5 nm observed from position 188 , which likely reflects differences in the side chain direction and the relative positions of the 3/4 and 5/6 loops . These ionic lock mutations generate structural states that are not seen in the wild-type protein . However , disrupting the ionic lock does not appear to abolish transport . The BtuB R14A mutant is functional in transport as determined by a growth assay ( Figure 3—figure supplement 3 ) . This is consistent with earlier work where a mutant containing a dipeptide insertion into one of the barrel strands in BtuB , which should have disrupted the D316-R14A ionic lock , was shown to support transport of vitamin B12 ( Lathrop et al . , 1995 ) . It should be noted that we examined the stability of the substrate-induced conformation of SB3 for times as long as 60 min before freezing and preparing the cells for DEER . The data are shown in Figure 3—figure supplement 4 , and indicate that the conformations are stable over time , indicating that the label is stable and not being reduced , and that conversion of the transporter back to the apo state does not occur under the conditions of this experiment . Two conformations are observed by crystallography for SB3 . In the apo structure ( PDB ID: 1NQG ) , SB3 has a single helical turn and shorter conformation that we speculate may be associated with the distance observed by EPR at 3 . 2 nm for the V90R1-T188R1 spin pair ( labeled 1 in Figure 2D ) . In the vitamin B12 bound structure ( PDB ID: 1NQH ) , SB3 is more extended , and this state may be associated with the distance at 2 . 4 nm ( labeled 2 in Figure 2D ) . Computational work suggests that the extended state of SB3 requires the interaction of BtuB with LPS , whereas the shorter helical state of SB3 occurs in the presence of phospholipid ( Balusek and Gumbart , 2016 ) , suggesting that environment , specifically LPS , may be important in controlling the configuration of SB3 . To test for an environmental effect on SB3 , we reconstituted four spin pairs of BtuB into 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine ( POPC ) proteoliposomes . These included the V90R1-T188R1 and S93R1-T188R1 spin pairs both in the absence and presence of the R14A mutation . The dipolar evolution data and distance distributions from DEER measurements on these reconstituted BtuB samples are shown in Figure 4 ( the raw time domain data are presented in Figure 4—figure supplement 1 ) . In the absence of the R14A mutation , the substrate-dependent conversion to the shorter distance that was seen for the V90R1-T188R1 spin pair in whole cells ( Figure 2C and Figure 3C ) is now much more limited , with only a minor shoulder appearing around 2 . 5 nm . For the S93R1-T188R1 spin pair , the change seen in Figure 3C with substrate is largely absent . The 0 . 4 nm substrate-dependent shift to a longer distance is absent , but the small 2 . 2 nm distance component is still present . The behavior of SB3 in the presence of the R14A mutation is also altered in the phospholipid reconstituted system . Rather than increasing the population of the shorter distance component , adding the R14A mutation to the V90R1-T188R1 pair in the reconstituted system results in a single short distance , where the resulting distribution aligns almost perfectly with the predicted distribution from the 1NQH structure . For the S93R1-T188R1 pair , R14A causes a small increase in the short distance component at 2 . 2 nm . But significantly , for neither spin pair are the longer substrate-induced shifts that were seen in Figure 3D observed in the reconstituted system . Thus , when removed from the native OM environment , the structure of SB3 is altered and the large substrate-induced movement of SB3 toward the periplasmic surface in the presence of the R14A mutation is no longer seen . It should be noted that in our initial work on the V90R1-T188R1 spin pair , we failed to observe a substrate-dependent conformational change in SB3 using an isolated OM preparation where the preparation includes a sarkosyl treatment ( Nilaweera et al . , 2019 ) . This suggests that this detergent treatment of the OM to remove inner membrane components is sufficient to alter the behavior of BtuB . These observations provide an explanation for why these changes in the conformation of SB3 have not been previously observed . In earlier work , we demonstrated that breaking the ionic lock between D316 and R14 altered a conformational equilibrium in the Ton box and promoted its unfolded state ( Lukasik et al . , 2007 ) . In this work , the effect of the R14A mutation on the Ton box equilibrium was comparable to that of a D316A mutation , but slightly enhanced for the dual R14A/D316A mutant . Figure 5 shows a result of mutating one or both of R14 and D316 on the substrate-dependent changes in SB3 as measured using the V90R1-T188R1 spin pair ( the time domain data are provided in Figure 5—figure supplement 1 ) . In the apo state , the distributions for both the R14A and D316A mutants are similar , with peaks falling in the same positions , although the D316A mutant yields more residual area under the 3 . 2 nm peak than is observed for R14A . In the presence of substrate , both show significant peaks around 3 . 8 and 4 . 5 nm , and a significant peak at 2 . 4 nm in both the apo and vitamin B12 bound samples . The double R14A-D316A mutation yields a distance distribution that is more perturbed than either single mutant , with a single broad peak centered around 2 . 4 nm in the apo state and an increase in the longer distance components in the substrate bound state . Thus , disrupting the D316-R14 ionic lock by mutating one or both residues has a dramatic effect on the conformation of the apex of SB3 and populates a state where this binding loop has moved a significant distance toward the periplasmic interface . The data indicate that this ionic interaction plays a role in mediating allosteric changes within the BtuB core , affecting not only the Ton box equilibrium but the substrate binding loop SB3 on the extracellular surface . The pulse EPR measurements made here in intact E . coli indicate that SB3 , which includes residues 82–96 in the core of BtuB , undergoes a substrate-induced structural change . When the ionic interaction between R14 in the core and D316 in the barrel is broken , an alternate and more dramatic structural change in SB3 occurs upon substrate binding , where SB3 is displaced approximately 2 nm toward the periplasmic surface of BtuB . Remarkably , these structural changes do not occur when the protein is removed from the native cell environment and reconstituted into a phospholipid bilayer , indicating that features in the intact cell environment modulate the energetics of the conformational states in BtuB . Earlier experimental work provides evidence for an allosteric coupling between the substrate binding site , the R14-D316 ion pair , and the Ton box in BtuB . Measurements made by EPR in isolated OM or reconstituted phospholipid membranes demonstrated that substrate binding partially unfolded the Ton box ( Xu et al . , 2006; Merianos et al . , 2000 ) , and shifted the energy of the folded and unfolded Ton box states by about 2 kcal/mol ( Freed et al . , 2010 ) . When the ionic interaction between R14 in the core and D316 in the barrel was broken , the Ton box was also observed to unfold and the coupling between substrate binding and the Ton box was broken ( Lukasik et al . , 2007 ) . In addition , a connection between the Ton box and SB3 was seen by scintillation proximity assays where both the Ton box and SB3 were found to be necessary for a TonB-dependent retention of vitamin B12 ( Mills et al . , 2016 ) . The connection between these sites in BtuB is also suggested by computational studies . When LPS is included in MD simulations , the interaction between R14 and D316 is weakened and the energy to unfold the Ton box reduced ( Balusek and Gumbart , 2016 ) . The inclusion of LPS also alters the state of SB3 . In a symmetric phospholipid bilayer , SB3 assumes the more helical form , whereas in an asymmetric membrane containing LPS , SB3 assumes an extended form . These simulations are consistent with the results presented here , except that fully populating the extended form of SB3 in our whole cell measurement ( state 2 in Figure 2d ) requires substrate binding . Although the role of periplasmic components such as the peptidoglycan cannot be ruled out , the computational results indicate that interactions made by LPS with the extracellular loops of BtuB may alter conformational equilibria in the protein and provide an explanation for the differences in the behavior of BtuB when EPR measurements are made in cells versus reconstituted phospholipid bilayers . It should be noted that the interconversion between helical and extended forms for SB3 was also absent or diminished when the V90R1-T188R1 spin pair was examined by EPR in an isolated OM preparation ( Nilaweera et al . , 2019 ) ; as a result , the procedure to produce this OM preparation , which includes the use of sarkosyl , is apparently sufficient to modify the behavior of the protein . An unexpected observation made here is that mutation of the R14-D316 ion pair alters the structure of the SB3 loop ( Figure 3D and Figure 5 ) . In the apo state of BtuB , this mutation enhances the extended form of SB3 ( state 2 in Figure 2B ) , and upon substrate binding an alternate conformational state is generated where sites 90 and 93 are extended as much as 4 . 5 nm from site 188 on the 3/4 extracellular loop . Among the core sites examined , this more dramatic structural change involves only SB3 as labels in the first and second substrate binding loops ( SB1 and SB2 ) at sites 63 , 65 , and 72 do not exhibit any significant structural changes . Such a large structural change involving SB3 has not been previously observed , and it appears to be localized to the C-terminal side of the core . A high-resolution model obtained by crystallography for a fragment of TonB in complex with BtuB ( Shultis et al . , 2006 ) shows TonB interacting with the Ton box in an edge-to-edge manner ( Figure 6A ) . Since the core is largely unaltered by TonB binding , models for transport have focused on the idea that TonB alters the core structure by exerting a mechanical force on the Ton box . In particular , TonB has been proposed to function by pulling on the Ton box , which then results in an unfolding of the N-terminal region of the core ( Hickman et al . , 2017 ) . Single-molecule pulling experiments ( Hickman et al . , 2017 ) as well as steered MD simulations ( Gumbart et al . , 2007 ) suggest that pulling the Ton box will extract the N-terminal side of the core and eventually open a pore sufficient to allow substrate to pass . One difficulty with this model is that both experimental and computational approaches indicate that the extraction of an extended polypeptide chain longer than the width of the periplasm is required to open this pore . We do not observe any significant structural changes in the N-terminal side of the core in cells ( sites 63 , 65 , and 72 ) , suggesting that the N-terminus may not move during transport . Rather , a significant substrate-induced structural change is found to take place in SB3 on the C-terminal side of the core when the D316-R14 ionic lock is broken . At present , we have limited restraints to generate a model and we do not know the positions of many segments in the core , but a movement in SB3 that satisfies the EPR restraints is shown in Figure 6B . In this model , substrate binding moves SB3 into the barrel and toward the periplasm . We do not presently know whether the movement of SB3 is accompanied by the movement of substrate; however , it is interesting to note that the C-terminal side of the core has a lower side chain density than the N-terminal side , suggesting that there is more space for structural rearrangements within this region . Presently , the precise sequence of steps that take place during transport are not known; however , both substrate and TonB are expected to be bound to BtuB at some point during transport , which should break the ionic lock between D316 and R14 ( Figure 6A ) . This suggests that the large substrate-induced structural change observed here on the C-terminal side of the core ( Figure 6B ) will occur during transport . The energy to promote this structural change and disrupt core-barrel electrostatic interactions could be provided by the free energy of binding of TonB to BtuB , which is significant and characterized by a Kd in the nM range ( Freed et al . , 2013 ) . To complete the transport cycle , TonB must be disengaged from BtuB and the inner membrane complex of ExbB and ExbD may perform this function and restore the apo state . Whether this involves a mechanical action of TonB , such as a pulling or rotational motion , or another process such as the exchange of the strand-to-strand TonB-Ton box interaction for a strand-to-strand interaction within a TonB dimer ( Freed et al . , 2013; Gresock et al . , 2015 ) remains to be determined . In summary , EPR spectroscopy in whole cells provides evidence for a substrate-induced structural transition in BtuB involving SB3 on extracellular apex of the core . The introduction of a mutation to break the R14-D316 ionic lock acting between the core and the barrel of BtuB produces an alternate structural state upon substrate addition so that SB3 is displaced as much as 2 nm into the barrel toward the periplasmic side of the protein . Under these same conditions , no movement of the N-terminal side of the core is detected . This ionic lock will be broken upon TonB binding and mutating this ionic lock may partially mimic the TonB bound state . As a result , this substrate-induced structural transition likely represents a structural state that occurs during the transport process . Remarkably , when BtuB is reconstituted into a phospholipid bilayer , these structural changes in SB3 are no longer observed , indicating that features in the native OM environment , such as the LPS , are required to populate conformational states that are important for BtuB function . The pAG1 plasmid with WT btuB gene and the RK5016 strain ( araD139 Δ ( argF-lac ) 169 flbB5301 ptsF25 relA1 rpsL150 rbsR22 deoC1 gyrA219 non-9 metE70 argH1 btuB461 recA56 ) used for growth assays were kindly provided by late professor R Kadner , University of Virginia . The E . coli dsbA null ( dsbA- ) mutant strain , RI90 ( araD139 Δ ( araABC-leu ) 7679 galU galK Δ ( lac ) X74 rpsL thi phoR Δara714 leu+ , dsbA:: Kanr ) were obtained from the Coli Genetic Stock Center ( Yale University , New Haven , CT ) . L63C , S65C , N72C , S93C , and S237C btuB mutants were custom produced by Applied Biological Materials Inc ( Richmond , BC , Canada ) . The btuB mutants ( L63C-T188C , S65C-T188C , N72C-T188C , V90C-T188C , V90C-S237C , S93C-T188C ) with and without the R14A mutation , and V90C , V90C-T188C-D316A , and V90C-T188C-R14A-D316A were engineered using polymerase chain reaction ( PCR ) mutagenesis . The plasmids were confirmed by sequencing and were transformed into dsbA- cells . Glycerol stocks were prepared and stored at −80°C . The RK5016 strain was authenticated using phenotype assays as described previously ( Lathrop et al . , 1995 ) . This strain fails to grow in minimal media ( MM ) that is not supplemented with methionine and vitamin B12 . The RI90 strain carried kanamycin resistance and lacks DsbA function ( Rietsch et al . , 1996 ) . This cell line was authenticated by testing for kanamycin resistance and determining that cells were not able to oxidize pairs of cysteine residues that were expressed on the cell surface ( Nilaweera et al . , 2019 ) . dsbA- cells expressing V90C-T188C , L63C , S65C , N72C , V90C , and S93C BtuB were grown in MM supplemented with 200 µg/ml ampicillin , 0 . 2 % w/v glucose , 150 µM thiamine , 3 mM MgSO4 , 300 µM CaCl2 , 0 . 01 % w/v methionine , and 0 . 01% w/v arginine ( Nilaweera et al . , 2019 ) . Cells expressing BtuB with the V90C-T188C mutation was spin labeled as described ( Nilaweera et al . , 2019 ) and the aliquots of processed cell pellets were mixed with vitamin B12 ( 0 , 1 , 5 , 20 , 30 , 60 , and 100 µM final concentrations ) . The cells expressing L63C , S65C , N72C , V90C , and S93C BtuB mutants were processed as described in Nyenhuis et al . , 2020b . Glycerol stocks of dsbA- cells expressing L63C-T188C , S65C-T188C , N72C-T188C , V90C-T188C , S93C-T188C , and V90C-S237C BtuB with and without R14A , V90C-T188C-D316A , and V90C-T188C-R14A-D316A BtuB were used to directly inoculate the pre-precultures ( Luria Bertani media with 200 µg/ml ampicillin ) , grown for 8 hr at 37°C and used to inoculate the MM precultures . The main MM cultures were inoculated with precultures , grown until OD600 ~0 . 3 , and spin labeled ( Nyenhuis et al . , 2020b ) . Briefly , the cells were spin labeled with methanethiosulfonate spin label ( MTSSL ) ( ( 1-oxy-2 , 2 , 5 , 5-tetramethylpyrrolinyl-3-methyl ) methanethiosulfonate ) ( Cayman Chemical , Ann Arbor , MI ) in 100 mM HEPES buffer ( pH 7 . 0 ) containing 2 . 5% ( w/v ) glucose with the final concentration of 7 . 5 nmol/ml of cell culture at OD600 0 . 3 for 30 min at room temperature ( RT ) . Spin labeled cells were washed by resuspending in 2 . 5% ( w/v ) glucose supplemented 100 mM HEPES buffers , first , at pH 7 . 0 and then , at pD 7 . 0 . During the washing steps , cysteine double mutants without R14A were incubated for 15 min , while 2–5 min for mutants with R14A and 10 min for D316A and for R14A-D316A mutants . For the data shown in Figure 3—figure supplement 4 , aliquots of processed V90C-T188C-R14A samples were incubated with vitamin B12 for 0 , 30 , and 60 min at RT . All other samples were incubated with vitamin B12 for 20 min or less prior to freezing . It should be noted that previous work demonstrated that BtuB can be specifically labeled in these dsbA- cells and that significant labeling of other OM proteins does not occur ( Nilaweera et al . , 2019 ) . MM main cultures of V90C-T188C and S93C-T188C with and without R14A were grown for 8 hr at 37°C . The harvested cells were used to isolate intact OM ( Nyenhuis et al . , 2020b ) . After the second spin at 118 , 370 × g for 60 min at 4°C , the pellets were resuspended in 5 ml of HEPES buffer . The OMs were solubilized in 100 mM Tris pH 8 . 0 buffer with 10 mM EDTA and 0 . 5 g of octylglucoside ( OG ) ( Chem-Impex , Wood Dale , IL ) . The OM suspension was incubated at 37°C for 10 min and 2 hr at RT and then spun at 64157 × g , 60 min at 4°C . The supernatants were used to spin label BtuB with 12 mM MTSSL at RT , overnight . BtuB was purified using six column volumes ( CV ) of wash buffer ( 17 mM OG , 25 mM Tris pH 8 . 0 ) , 12 CV of 0–100% gradient of elution buffer ( 1 M NaCl , 17 mM OG , 25 mM Tris pH 8 . 0 ) and 6 CV of 100% elution buffer using a Q column and fractions containing BtuB were pooled . POPC ( Avanti Polar Lipids , Alabaster , AL ) ( 20 mg/ml ) was sonicated in reconstitution buffer ( 150 mM NaCl , 100 nM EDTA , 10 mM HEPES pH 6 . 5 ) with OG ( 100 mg/ml ) until clear , next , 1 ml from micelles mixture was added to each pooled BtuB mutant and incubated at RT , 40 min . BtuB was reconstituted into POPC by dialyzing OG over six buffer exchanges using reconstitution buffer and bio-beads ( with minimum of 6 hr dialysis per exchange ) . Reconstituted BtuB was pelleted by centrifugation at 23425 × g for 40 min at 4°C , resuspended in 200 μl of reconstituted buffer and further concentrated to 50 μl by using Beckman airfuge . The samples were frozen and stored at −80°C . For CW EPR , 6 μl of cell pellet , or 6 μl of cell pellet with 100 μM vitamin B12 were loaded into glass capillaries ( 0 . 84 OD , VitroCom , Mountain Lakes , NJ ) . Capillaries were loaded into a Bruker ER 4123D dielectric resonator ( Bruker BioSpin , Billerica , MA ) mounted to a Bruker EMX spectrometer . Data were taken at X-band and at temperature with 100 G sweep width , 1 G modulation , and 2 mW of incident microwave power . EPR spectra on live E . coli that did not produce a signal-to-noise ratio of 7–10 or greater with a single 20 s field sweep failed to produce pulse echoes at Q-band pulse of adequate amplitude and were discarded . For pulse EPR , 16 μl of cell pellet , 20% glycerol , and 100 μM CNCbl ( vitamin B12 ) when applicable were combined and loaded into quartz capillaries ( 1 . 6 mm OD . , VitroCom , Mountain Lakes , NJ ) . Samples were flash-frozen in liquid nitrogen and loaded into an EN5107D2 resonator ( Bruker BioSpin , Billerica , MA ) . Data were collected on a Bruker E580 at Q-band and 50 K using a 300 W TWT Amplifier ( Applied Systems Engineering , Benbrook , TX ) . The dead-time free 4-pulse DEER sequence was used for all experiments , with rectangular pulses of typical lengths π/2 = 10 ns and π = 20 ns , and a 75 MHz frequency separation . It should be noted that modulation depths obtained from whole cell samples were generally highly variable . We suspect that this is a result of the cells actively metabolizing during the labeling and washing steps in their preparation . Labeled whole cell samples that did not label well or produce sufficiently large modulation depths ( approximately 4% ) were re-grown and re-labeled . For the instrument settings and specific resonator used in this work , the modulation depth for a well-labeled protein is approximately 20% . Based upon this , labeling efficiencies for most samples were estimated to be approximately 40–60% . CW EPR spectra were normalized by dividing by the spectral second integral using in-house python scripts . All pulse EPR data , except for data in Figure 3—figure supplement 2 , were processed using LongDistances v 932 ( Christian Altenbach , UCLA ) . Data were fit to a variable dimension background , after which the model-free mode was used for distance fitting . The value of the smoothing parameter was selected based on the L-curve , ensuring that the selected value passed through the major oscillations present in the data . Error analysis used 100 variations at the default values for background noise , start time , dimensionality , and regularization error . For data in Figure 2d , the data were instead fit using a model-based mode with two-Gaussian components with free position , width , and amplitude to investigate the dosage dependence of the substrate-dependent shift toward a shorter distance component for the 90–188 pair . Supplementary file 1 contains the error analysis for these data . Data in Figure 3—figure supplement 2 were processed using the DeerNet routine ( Worswick et al . , 2018 ) in DeerAnalysis ( Jeschke et al . , 2006 ) . A folder of all Source Data ( raw , unprocessed DEER data ) has been provided . All EPR figures were generated using python scripts and the matplotlib plotting library . Protein structure images were generated using Pymol ( DeLano , 2002 ) . Simulated distance distributions were generated using the software package MMM v . 2018 . 2 and the default rotamer library ( Jeschke , 2018; Polyhach et al . , 2011 ) . The 90–188 and 90–237 distributions with the 14A mutation and in the presence of vitamin B12 were used in the generation of a model ( Figure 6b ) for the motion of the apical hatch loop using the software package Xplor-NIH ( v . 3 . 2 ) . The starting structure for modeling was the in surfo structure crystal structure in the presence of cobalamin ( PDB ID: 1NQH ) , which we previously determined to be closest to the native state of the extracellular loops in the native environment ( Nyenhuis et al . , 2020a ) . In that work , we found minimal evidence for motion of the extracellular loops , and we assumed that motion was localized to the SB3 loop . The template structure was labeled in silico with the R1 side chain at the 90 , 188 , and 237 positions using the software package MMM ( v . 2017 . 2 ) and the default rotamer library . The top three rotamers were selected from the in silico labeling and used to generate three input structures for ensemble calculations , with the relative weights of the three rotamers conserved from the in silico labeling calculation . During the calculation , the reference R1 sites in the barrel ( 188 and 237 ) were held entirely fixed . The R1 side chain at site 90 , the underlying SB3 loop element comprising residues 81–104 , and the adjoining , unstructured hatch region comprising residues 112–124 were fully mobile during runs , while all remaining residues had fixed backbone atoms and mobile side chains . The standard Xplor potentials BOND , ANGL , and IMPR were used in conjunction with the torsionDB and repel potentials for all elements , and DEER restraints were encoded as square well potentials using the noePot potential term . All potential terms were ensemble averaged across the three input structures . The peak positions used in the modeling were the peak centered at 4 . 5 nm for 90–188 , and the peak at 2 . 8 nm for 90–237 . Full peak widths were used , with the square well stopping at 5% of maximum intensity . Randomization was introduced to the calculations using the randomizeTorsions function on the starting side chain positions of the mobile hatch elements . Following this , 10 rounds of 400 step Powell minimization using all potentials and with the mobile hatch elements were used to satisfy the experimental restraints . Structure calculation was repeated until both restraints were within error .	Bacteria must obtain nutrients from their surrounding environment in order to survive . In Gram-negative bacteria , proteins in the outer membrane surrounding the cell actively transport carbohydrates and trace nutrients like iron into the cell’s interior . Although the structures of many of these transport proteins have been determined , the mechanism they use to move molecules across the membrane is poorly understood . To better understand this process , Nilaweera , Nyenhuis and Cafiso examined the structure of BtuB , a transport protein found in the outer membrane of Escherichia coli that is responsible for absorbing vitamin B12 . Previous experiments analyzing the structure of BtuB , and other similar transporters , have been carried out on purified proteins that were extracted from the outer membrane . However , these isolated proteins fail to replicate the transport activity observed in bacterial cells . Nilaweera , Nyenhuis and Cafiso therefore wanted to see how the structure of BtuB changes when it is still enclosed in the membrane of E . coli . This revealed that BtuB undergoes large structural changes when it binds to vitamin B12 , suggesting that this is an important part of the transport process . However , when purified BtuB was placed into an artificial membrane , these structural changes did not occur . This indicates that the cellular environment in the bacteria is needed for BtuB to carry out its transport role , and explains why previous experiments using purified proteins struggled to see this structural shift . This work highlights the importance of studying bacterial membrane proteins in their native cell environment . BtuB and similar transporters represent a large family of proteins unique to Gram-negative bacteria that have an impact on human health . Since these proteins are structurally alike , the results of this study may help resolve the transport mechanisms of other proteins , ultimately leading to new ways to control bacterial growth .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "microbiology", "and", "infectious", "disease" ]	2021	Structural intermediates observed only in intact Escherichia coli indicate a mechanism for TonB-dependent transport
A prerequisite for the systems biology analysis of tissues is an accurate digital three-dimensional reconstruction of tissue structure based on images of markers covering multiple scales . Here , we designed a flexible pipeline for the multi-scale reconstruction and quantitative morphological analysis of tissue architecture from microscopy images . Our pipeline includes newly developed algorithms that address specific challenges of thick dense tissue reconstruction . Our implementation allows for a flexible workflow , scalable to high-throughput analysis and applicable to various mammalian tissues . We applied it to the analysis of liver tissue and extracted quantitative parameters of sinusoids , bile canaliculi and cell shapes , recognizing different liver cell types with high accuracy . Using our platform , we uncovered an unexpected zonation pattern of hepatocytes with different size , nuclei and DNA content , thus revealing new features of liver tissue organization . The pipeline also proved effective to analyse lung and kidney tissue , demonstrating its generality and robustness . A major challenge for the understanding of mammalian tissue structure and function is the ability to monitor cellular processes across different levels of complexity , from the subcellular to the tissue scale ( Megason and Fraser , 2007 ) . This information can then be used to develop quantitative functional models that describe and predict the system behaviour under perturbed conditions ( Hunter et al . , 2008; Smith et al . , 2011; Fonseca et al . , 2011; Sbalzarini , 2013 ) . The development of such multi-scale models requires first a geometrical model of the tissue , that is , an accurate three-dimensional ( 3D ) digital representation of the cells in the tissue as well as their critical subcellular components ( Peng et al . , 2010; Boehm et al . , 2010; Mayer et al . , 2012 ) . This can be constructed from high-resolution microscopy images with multiple fluorescent markers , either fusion proteins or components detected by antibody staining . Since organelles can be as small as ~0 . 1 µm in size , the geometrical model has also to cover a wide range of scales spanning over three orders of magnitude . However , substantial limitations persist with respect to availability of markers , volume of tissue to reconstruct , scale of measurements , computational methods to perform the analysis and sample throughput . Although a few existing platforms provide standard tools for 3D segmentation and methods to process 2D surface layers of cells [ImageJ/Fiji ( Girish and Vijayalakshmi , 2004; Collins , 2007 ) , ICY ( de Chaumont et al . , 2012 ) and MorphoGraphX ( Barbier de Reuille et al . , 2015 ) ] , the challenges posed by dense and thick tissue specimens require the development of new algorithms . Therefore , there is a demand for a platform that can provide the required set of methods for the reconstruction of multi-scale digital 3D geometrical models of mammalian tissues from confocal microscopy images . The number of fluorescent markers that can be used simultaneously is limited to 4–5 , making the reconstruction of tissue models a challenging problem . For a meaningful model , it is necessary to properly identify the different cell types within the tissue but also to detect subcellular and extracellular structures , for example , nuclei , plasma membrane or cell cortex , extracellular matrix ( ECM ) and cell polarity . Automated morphological cell recognition is a possible way to reconstruct dense tissue with limited number of markers . Geometrical digital models of tissues also require 3D information over large volumes . Validated fluorescent protein chimeras are not always available , especially in the appropriate combination of fluorescence emission spectral profiles . On the other hand , in dense tissues immunostaining is inhomogeneous due to restricted antibody penetration . The development of protocols that render tissues optically transparent and permeable to macromolecules without significantly compromising their general structure enables the imaging of relatively thick specimens ( Chung and Deisseroth , 2013; Ke et al . , 2013 ) . However , in the case of a densely packed tissue , for example , liver , homogeneous staining is still limited to a thickness of ~100 µm . Therefore , obtaining high-resolution data from large volumes of tissue ( typically from 0 . 1 mm to a few centimetres ) requires sectioning the sample into serial 100-µm-thick slices that are stained and imaged separately . Furthermore , the cutting process introduces artefacts , such as bending , uneven section surfaces and partial damage of tissue , that require corrections during tissue model reconstruction . Unfortunately , the publicly available generic image processing software is unable to deal with such problems . In this study , we addressed these challenges by developing a set of new algorithms as well as implementing established ones in an adjustable pipeline implemented in stand-alone freely available software ( http://motiontracking . mpi-cbg . de ) . As proof of principle , we tested the pipeline on the reconstruction of a geometrical model of liver tissue . We chose this particular tissue due to its utmost importance for basic research , medicine and pharmacology . In order to test the accuracy of the pipeline , we created a benchmark for the evaluation of dense tissue reconstruction algorithms comprising a set of realistic 3D images generated from the digital model of liver tissue . Furthermore , we applied the platform to the analysis of lung and kidney tissue , demonstrating its generality and robustness . Mouse livers were fixed by trans-cardial perfusion instead of the conventional immersion fixation ( Burton et al . , 1987 ) to minimize the time lag between the termination of blood flow and fixation ( Gage et al . , 2012 ) . This proved to be absolutely essential to preserve the tissue architecture and the epitopes for immunostaining . Serial sections of fixed tissues were prepared at a thickness of 100 µm to maximize antibody penetration and limit laser light scattering . Liver sections were stained to visualize key subcellular and tissue structures , namely nuclei ( DAPI ) , the apical surfaces of hepatocytes ( CD13 ) , the sinusoidal endothelial cells ( Flk1 ) or ECM ( Laminin and Fibronectin ) and the cell cortex ( F-actin stained by phalloidin ) . We tested various reagents and protocols to clear the liver tissue , such as glycerol and 2 , 2′thiodiethanol ( TDE ) , and found that SeeDB ( Ke et al . , 2013 ) yielded the best results . Stained sections were imaged sequentially ( generating Z-stacks ) by one- and two-photon laser scanning confocal microscopy to maximize the number of fluorescent channels available . The same section was imaged twice , at low and high magnification , using 25×/0 . 8 and 63×/1 . 3 objectives , respectively . The first covers a large volume to reconstruct the whole lobule and the latter focuses on a small area to reconstruct the tissue at high resolution . The registration of 3D high-resolution images within low-resolution ones provides tissue-scale context information that is essential for the interpretation of the data at the cellular and subcellular level . A major problem for the image analysis of thick tissue sections is the low signal-to-noise ratio deep into the tissue , especially for stainings that yield high and diffuse background ( e . g . actin staining with phalloidin throughout the cytoplasm ) . To address this problem , we developed a new Bayesian de-noising algorithm that first makes a probabilistic estimation of the background and separates it from the foreground ( see ‘Methods’ ) . Subsequently , the estimated background and foreground signals are independently smoothed and summed to generate a new de-noised image ( Figure 1—figure supplement 2 ) . We applied BFBD de-noising to both low- and high-resolution images . BFBD de-noising provides better results than the standard ones in the field , such as median filtering , Gauss low-pass filtering and anisotropic diffusion ( Figure 1—figure supplement 4 ) , but also outperforms ( by quality and computational performance ) other algorithms , known to be more elaborate , such as the ‘Pure Denoise’ ( Luisier et al . , 2010 ) and ‘edge preserving de-noising and smoothing’ ( Beck and Teboulle , 2009 ) ( see ‘Methods’ ) ( Figure 1—figure supplement 5 ) . The tissue was imaged at low- and high-resolution for the multi-scale reconstruction . The reconstruction was performed in three steps: ( 1 ) images of physical sections were assembled as mosaics of low-resolution images , ( 2 ) all mosaics were corrected for physical distortions and combined in a single 3D image ( image stitching ) and ( 3 ) the high-resolution images were registered into the low-resolution one . In more detail , the partially overlapping ( ~10% overlap ) low-resolution images of each physical section were combined in 3D mosaics ( Figure 2A and Figure 2—figure supplement 1A ) using the normalized cross-correlation ( NCC ) approach ( see ‘Methods’ ) . NCC was chosen because it allows finding accurate shifts given a coarse initial match between 3D images ( Emmenlauer et al . , 2009; Peng et al . , 2010; Bria and Iannello , 2012 ) . Then , the 3D image mosaics were combined into a single 3D image . The mechanical distortion and tissue damage produced by sectioning are such ( as illustrated in Figure 2B and Figure 2—figure supplement 1C ) that even advanced and well-established methods for image stitching ( Preibisch et al . , 2009; Saalfeld et al . , 2012; Hayworth et al . , 2015 ) fail due to the lack of texture correlations between adjacent sections . To address this problem , we developed a Bayesian algorithm for stitching images of bended and partially damaged soft tissue sections . The algorithm first corrects section bending and then uses the empty space at the interior of large structures ( e . g . vessels ) within adjacent sections to register and stitch them . 10 . 7554/eLife . 11214 . 009Figure 2 . Reconstruction of a multi-scale lobule image . ( A ) Schematic representing a single serial section obtained from a grid of M × N partially overlapping 3D images ( tiles ) . The cross-correlation between two neighbouring tiles in the grid provides a local metric , which describes the value of their relative shifts . The reconstruction of each section was performed by maximizing the sum correlations of each tile to all adjacent tiles ( see ‘Methods’ for details ) . ( B , C ) Correction of tissue deformations ( introduced during the sample preparation process ) using a surface detection algorithm and β-spline transformation . ( B ) Output of the surface detection algorithm . The proposed Bayesian approach uses prior information about expected bending of the section , its thickness and measurement error ( see ‘Methods’ for details ) to determine the volume of the image belonging to the tissue and to the out-of-field region . ( C ) The tissue section after correcting its bending by using quadratic β-splines . ( D ) Tissue section before ( left ) and after ( right ) the correction of the mechanical distortions and the tissue damage . ( E ) Full lobule-level reconstruction established by the alignment of six low-resolution sections ( 1 μm × 1 μm × 1 μm per voxel ) and the interpolation of blood vessels . Two high-resolution images ( 0 . 3 μm × 0 . 3 μm × 0 . 3 μm per voxel ) were registered in the low-resolution reconstruction and are shown in grey ( see Video 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 00910 . 7554/eLife . 11214 . 010Figure 2—figure supplement 1 . Reconstruction of multi-scale tissue images . Tissue section reconstruction: ( A ) Schematic representation of an M × N grid of partially overlapping 3D images . The regions in light blue and light red represent the overlapping areas between neighbouring images . The colour-coded maps show the cross-correlation matrixes between neighbouring images . ( B ) Reconstructed tissue section from 4×4 a grid of low-resolution images . The pattern of DAPI staining ( nuclei ) at the intersection of two neighbouring images is shown . Correction mechanical distortion and tissue damage on serial sections: ( C ) x–z section of the image of a tissue section showing the main obstacles for the tissue surface detection: unstained volume of blood vessels ( C' ) and blurring ( C'' ) . Probabilities ( D ) p ( ym1 , ym2 , \|y1 , y2 ) , ( E ) p ( y2\|y1 ) and ( F ) p ( y1 ) calculated from the maximum entropy segmentation ( red ) , model equations ( blue ) and manual solution ( green ) . All distributions in the figure were averaged over all tissue sections in the benchmark . ( G ) Comparison of manual and automated surfaces calculated for two tissue sections from P16 ( upper ) and adult ( lower ) mice datasets . ( H ) Accuracy of surface detection . Plot presenting the mean absolute deviation calculated between manually and automatically detected surfaces for 33 different tissue sections in 4 data sets . Since tissue section segmentation is ambiguous , the control experiment was conducted by segmenting the same tissue sections manually three times . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 01010 . 7554/eLife . 11214 . 011Figure 2—figure supplement 2 . Reconstruction of multi-scale tissue images . Tissue-level network segmentation: ( A ) Reconstructed image of a tissue section . Large vessels appear as empty space in the image . ( B ) Spatial distribution of the local maximum entropy threshold value . ( C ) Segmentation of large vessel in a single tissue section . Registration of high-resolution images into low-resolution ones: Representative region of a 2D plane of ( D ) a low-resolution ( yellow ) and ( E ) a high-resolution ( red ) image stained with Flk1 for sinusoids . ( F ) Superimposed images after the registration . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 011 A prerequisite for the correction of section bending is the detection of its upper and lower surfaces ( Figure 2B ) . The high degree of image axial blurring in thick samples ( Nasse and Woehl , 2010 ) and the presence of large vessels pose problems for the detection of surfaces ( see Figure 2—figure supplement 1C ) . The algorithm reconstructed the probability distribution of the surface excursion ( deviation from the mean position over the neighbourhood ) and then used it to predict the localization of each point at the surface ( see ‘Methods’ ) . The surface predicted by the algorithm closely matched the surface detected manually ( Figure 2—figure supplement 1G ) . Then , the bending correction was performed by standard β-spline transformation ( Figure 2C , D ) . Next , the individual sections were combined . Since approximately one cell layer is removed upon sectioning , direct matching of two adjacent sections is impossible . Therefore , we first segmented the large vessels and then aligned the sections by matching them ( Figure 2D ) . The vessels were segmented by using the local maximum entropy ( LME ) approach ( Brink , 1996 ) ( see ‘Methods’ ) . Subsequently , the segmented vessels were classified ( marked as PV or CV ) revealing the precise arrangement of lobule-level structures . Finally , we interpolated these vessels within the gaps caused by tissue removal by tri-linear intensity approximation . Following the assembly of the low-resolution model , we registered the high-resolution images within it using rigid body transformation . To accelerate the search for registration parameters , we built a hierarchy of binned images and performed registration sequentially from the coarsest to the finest one ( see ‘Methods’ ) . This method was used for the reconstruction of a liver tissue model from six serial sections , each imaged as a 3 × 3 mosaic grid with 10% overlap and resolution of 1 μm × 1 μm × 1 μm per voxel . Then , two sections , each imaged as a 2 × 2 mosaic grid at high-resolution ( 0 . 3 μm × 0 . 3 μm × 0 . 3 μm per voxel ) were registered within the low-resolution model . The reconstruction covers about 1300 μm × 1300 μm × 600 μm of the tissue and is presented in Figure 2E and Video 1 . 10 . 7554/eLife . 11214 . 012Video 1 . 3D image visualization of a multi-resolution geometrical model of liver tissue . A set of six low-resolution ( 1 . 0 μm × 1 . 0 μm × 2 . 0 μm per voxel ) and two high-resolution tissue sections ( 0 . 3 μm × 0 . 3 μm × 0 . 3 μm per voxel ) were used . Central veins are shown in light blue , portal veins in orange and high-resolution cubes in grey . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 012 The next step was to reconstruct the tubular structures present in the tissue , that is , sinusoidal and BC networks . One of the most popular tools for image segmentation is global thresholding ( Pal and Pal , 1993 ) . In particular , the maximum entropy approach has been widely applied to image reconstruction problems , including the segmentation of fluorescent microscopy images ( Dima et al . , 2011; Pecot et al . , 2012 ) . However , since 3D confocal images are usually heterogeneous in intensity due to staining unevenness and light scattering in the tissue ( Lee and Bajcsy , 2006 ) , global thresholding approaches may produce segmentation artefacts . In contrast , local thresholding allows adjusting the segmentation threshold to the spatial variability . We applied the LME method to find segmentation thresholds in the de-noised images . For this , we split the 3D image into a set of cubes and calculated the maximum entropy segmentation threshold ( Brink , 1996 ) within each cube . The threshold values were tri-linearly interpolated to the entire 3D image . However , this segmentation approach produced two major artefacts . The objects were moderately swollen and contained holes resulting from local uneven staining . We used standard approaches to close the holes by morphological operations ( opening/closing ) , which unfortunately led to even higher overestimation of the diameter of thin structures , such as sinusoids and BC . To correct this , we extended the segmentation algorithm by including the following steps . We generated a triangulation mesh of the segmented surfaces by the cube marching algorithm ( Lorensen and Cline , 1987 ) ( Figure 3A ) . Then , we tuned the active mesh so that the triangle mesh vertexes aligned to the maximum gradient of fluorescence intensity in the original image ( Figure 3A ) . Finally , we generated a representation of the skeletonized image via a 3D graph describing the geometrical and topological features of the BC and sinusoidal networks . The reconstruction of sinusoidal and BC networks are shown in Figure 3B , C , respectively . 10 . 7554/eLife . 11214 . 013Figure 3 . Reconstruction of tubular structures , nuclei and cells . ( A ) A single 2D image section is shown with the contours of the sinusoidal network reconstruction overlaid on the de-noised image . The contours of the initial mesh are drawn in yellow , and the ones of the tuned mesh are drawn in cyan . ( B–E ) 3D representation of the different structures segmented in a sample of liver tissue: sinusoids ( B ) , BC ( C ) , nuclei ( D ) and cells ( E ) . All the reconstructed structures are shown together in ( F ) . The reconstructed triangle meshes are drawn inside the inner box and the raw images are outside . In the case of tubular networks ( i . e . sinusoids and BC ) , the central lines of the structures are shown together with the raw images . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 01310 . 7554/eLife . 11214 . 014Figure 3—figure supplement 1 . Nuclei splitting . ( A ) 3D visualization of a confocal image of closely packed nuclei ( DAPI ) . ( B ) Objects resulting from the initial segmentation and reconstruction: triangle meshes of the artificially merged structures . The approximation of different structures ( C ) by ( D ) one or ( E ) two overlapping ellipsoids is shown . Prediction of multi-nuclear structures: ( F ) distribution of the In ( MSE ) values obtained from the nuclei approximation by one and double ellipsoids . The distribution was fitted by a sum of two Gaussian distributions . The fitting curve is shown in blue ( solid line ) and the components in magenta and red ( dash lines ) . ( G ) Calculated threshold that discriminates between bi/mono-nuclear and multi-nuclear structures . The graphs were obtained from the analysis of a sample of liver tissue , which covers the entire central vein ( CV ) -portal vein ( PV ) axis . Multi-nuclei splitting: ( H ) original confocal image where the nuclei seeds were detected ( I ) and expanded to the real nuclei shape ( J ) . ( K ) The performance of the splitting algorithm was evaluated in both synthetic and real 3D images . The synthetic image consisted of 150 nuclei , which included single nuclei , double- and triple-nucleated structures . The individual nuclei had a radius between 5 and 7 µm . The multi-nucleated structures were generated with different degrees of overlap . A global background of 10% of the intensity of the nuclei was added to the whole image , then it was blurred using a Gaussian filter and finally salt and paper noise was added . The real image corresponds to an adult mouse tissue sample of 2 . 3 ×10–3 mm3 volume . The initial segmentation yielded 281 structures , which were analysed ( the nuclei touching the borders of the sample were excluded from the analysis ) . The performance was evaluated in terms of true positive ( TP ) , false positive ( FP ) , true negative ( TN ) and false negative ( FN ) values . TP = correctly split , FP = over-splitting , TN = correctly not split , FN = under-splitting . Precision ( PR ) = TP/ ( TP FP ) , sensitivity ( SN ) = TP/ ( TP FN ) , specificity ( SP ) = TN/ ( TN FP ) , F-score = 2 × ( PR × SN ) / ( PR SN ) and accuracy ( AC ) = ( TP TN ) / ( TP TN FP FN ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 01410 . 7554/eLife . 11214 . 015Figure 3—figure supplement 2 . Cell classification . ( A ) Example of an image used to generate the training set for the classifier . The different types of nuclei forming in liver tissue where manually classified using the specific markers , that is , Flk1 ( magenta ) for sinusoidal endothelial cells ( SECs ) , the macrophage antibody F/4/80 ( yellow ) for Kupffer cells and the intermediate filament Desmin ( green ) for stellate cells . The training set was extracted from three samples covering the entire central-portal vein axis . ( B ) Selection of the set of parameters for the linear discriminant analysis ( LDA ) . The 74 calculated parameters were sorted by the Fisher score and the top five ranked parameters with the largest Fisher scores are shown . The classifier accuracy in dependency of the number of parameters used for the classification is plotted . The set of parameters that yielded the highest accuracy of the classifier was chosen . ( C ) Features dependency obtained in the Bayesian network classifier . The Bayesian network structure learning from the experimental data revealed that 15 parameters were relevant for the nuclei classification . The five parameters with the highest mutual information to the nuclei type are shown in inset . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 01510 . 7554/eLife . 11214 . 016Figure 3—figure supplement 3 . Cell classification accuracy . Confusion matrixes obtained with the ( A ) linear discriminant analysis and ( B ) the Bayesian network classifier . The instances ( e . g . nuclei ) in each predicted class are represented in the columns of the matrix , while the instances in an actual class ( manually identified ) are represented in the rows . 3D representation of the different nuclei types identified in a representative sample of liver tissue: ( C ) hepatocytes , ( D ) sinusoidal endothelial cells ( SECs ) , ( E ) stellate and ( F ) Kupffer cells . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 01610 . 7554/eLife . 11214 . 017Figure 3—figure supplement 4 . Reconstruction of tubular structures , nuclei and cells . Single 2D image planes are shown with contours of ( A ) sinusoidal and ( B ) and bile canalicular ( BC ) networks , ( C ) nuclei and ( D ) cells reconstructions overlaid on raw data . Insets show zoomed areas of the image . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 01710 . 7554/eLife . 11214 . 018Figure 3—figure supplement 5 . Generation of realistic 3D images of liver tissue . ( A ) Generation of images with uneven staining . The image of the idealized structure ( homogeneous tubes ) created for the bile canalicular ( BC ) network is shown in the top left image . The initial coarse grained sampling ( 6 × 6 ×6 binning ) of intensities is shown in the top-right image . The fine sampling ( unbinned image ) of intensities is shown in the bottom-right image and final result in the bottom-left one . ( B ) 3D representation and 2D projections of a model image of BC with uneven staining . ( C ) Characteristic point spread function ( PSF ) of a confocal microscope . ( D ) 3D representation and 2D projections of a model image of BC convolved with the PSF . ( E ) Mean variance of each intensity level for different depth ( z-direction ) levels of a confocal image . ( F ) Linear increase of the intensity scaling factor ( alpha ) with the sample depth for different channels . The error bars represent the standard deviation between three samples . ( G ) 3D representation and 2D projections of a final model image of BC after adding spatially variable Poisson noise . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 01810 . 7554/eLife . 11214 . 019Figure 3—figure supplement 6 . Benchmark of images to evaluate 3D reconstructions of dense tissue . Example of a realistic 3D image of liver tissue . 3D representation and 2D projections ( xy and xz ) of a high-resolution image created for bile canalicular ( BC ) ( A ) and sinusoidal ( B ) networks as well as nuclei ( C ) and cell borders ( D ) . The images size is 256 ×256 ×256 voxels with a resolution of 0 . 3 μm × 0 . 3 μm × 0 . 3 μm per voxel . The image shown corresponds to a 2:1 signal-to-noise ratio . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 01910 . 7554/eLife . 11214 . 020Figure 3—figure supplement 7 . Model validation: Evaluation of the accuracy of our pipeline for the 3D reconstruction of dense tissue . The reconstructions of the different structures forming the tissue were evaluated in terms of true positive ( TP ) , false positive ( FP ) , true negative ( TN ) and false negative ( FN ) values extracted from the comparison of the reconstructed image and the ground truth ( image without distortions ) . The precision ( PR ) and sensitivity ( SN ) are defined as TP/ ( TP FP ) and TP/ ( TP FN ) , respectively . F-score is given 2 × ( PR × SN ) / ( PR SN ) . The tests were performed in three sets of images ( three images per set ) with different signal-to-noise ratio ( 10:1 , 4:1 , 2:1 ) . ( A–C ) and ( D–F ) The results for the bile canalicular ( BC ) and sinusoidal networks , respectively . ( G–I ) and ( J–L ) The ones for nuclei and cells , respectively . Whereas in the case of BC , sinusoids and nuclei , the error bar corresponds to standard deviations of the values between three images , for the cells the error bar corresponds to the standard deviation of the values over all the cells in the samples ( 32 cells ) . Only the cells that were not in contact with the boundary of the image were analysed . ( M–N ) The mean values for the radius of BC and sinusoidal networks . ( O ) The mean error in the estimation of the cell volume . The error was calculated as 100×Vs-VgtVgt , where Vs and Vgt are the volumes of the reconstructed and ground truth cells , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 020 Nuclei were reconstructed similar to the tubular structures . However , as shown in Figure 3—figure supplement 1A , B , closely packed nuclei are optically not well-resolved in 3D confocal images , resulting in artificially merged structures . Since 30–60% ( depending on the animal strain and age ) of hepatocytes in adult liver are bi-nucleated , artificial nuclei merging compromises the tissue analysis . To address this problem , we used a probabilistic algorithm for double- and multi-nuclei splitting ( Figure 3—figure supplement 1 ) . Briefly , the algorithm first discriminated between mono- , double and multi-nuclear structures by learning the misfit distribution of triangulation mesh and nuclei approximation by single and double ellipsoids ( Figure 3—figure supplement 1A–G ) . Then , the seed points for the multi-nuclear structures were detected using the Laplacian-of-Gaussian ( LoG ) scale-space maximum intensity projection ( Stegmaier et al . , 2014 ) and , finally , the real nuclear shapes were found using an active mesh expansion starting from the nuclei seeds ( see ‘Methods’ for details ) . Tested in both synthetic and real 3D images , the algorithm proved capable of splitting clustered nuclei with different degrees of overlap ( Figure 3—figure supplement 1K ) with an accuracy of over 90% . Although this approach is based on active triangulation mesh , it achieved similar accuracy values to other recently published voxel-based methods for nuclei segmentation ( Amat et al . , 2014; Chittajallu et al . , 2015 ) . Generating geometrical models of tissues requires the proper recognition of different cell types . A previous automated classification system discriminated hepatocytes from non-parenchymal cells in 2D human liver images with a 97 . 8% accuracy ( O'Gorman et al . , 1985 ) . However , the automatic classification of non-parenchymal cells in 3D liver tissue is more challenging . Given their importance in physiology and disease ( Bouwens et al . , 1992; Kmiec , 2001; Malik et al . , 2002 ) and the limitation on the number of fluorescent markers that can be simultaneously imaged , we designed an algorithm to automatically classify different cell types in the tissue based on nuclear morphological features . We chose two deterministic supervised classifiers , linear discriminant analysis ( LDA ) and Bayesian network classifier ( BNC ) . LDA , also known as Fisher LDA ( Fisher , 1936 ) , is a fundamental and widely used technique to classify data into several mutually exclusive groups ( Duda et al . , 2001 ) . It has been successfully applied for nuclei discrimination in microscopy images ( Huisman et al . , 2007; Lin et al . , 2007 ) . On the other hand , BNCs are more recently developed classifiers which not only show good performance but also allow for probabilistic classification . In addition , BNCs reveal the hierarchy of parameters used for the classification ( Friedman et al . , 1997 ) , which may provide insights into underlying biological processes . As input for the classifiers , we manually built a training set of 2301 nuclei using specific cellular markers ( Figure 3—figure supplement 2A ) and computed for each nucleus a profile of 74 parameters ( Table 1 ) describing nuclei morphology , texture and localization relative to sinusoids and cell borders ( density of actin in vicinity of nuclei ) ( see ‘Methods’ ) . For the LDA , the parameters were ranked using Fisher score ( Duda et al . , 2001 ) , and the most relevant ones were selected based on the classification accuracy ( Figure 3—figure supplement 2B and ‘Methods’ ) . Independently , the most relevant parameters were selected on the basis of Bayesian network structure reconstruction ( Friedman et al . , 1999 ) ( Figure 3—figure supplement 2C ) . 10 . 7554/eLife . 11214 . 021Table 1 . List of the 74 parameters calculated for the nuclei classification . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 021ParameterF-scoreParameterF-scoreFLK1 surface intensity 1 vx4 . 802Mean radius0 . 920FLK1 surface intensity 0 vx4 . 737FLK1 KURT0 . 915FLK1 mean4 . 674MB Frac Dim0 . 904FLK1 surface intensity 2 vx4 . 570Log Lac20 . 885FLK1 surface intensity 3 vx4 . 100HF20 . 833Phallo surface intensity 2 vx3 . 477HF130 . 825FLK1 surface intensity 4 vx3 . 453HF30 . 817Phallo surface intensity 1 vx3 . 430Phallo surface intensity 9 vx0 . 787FLK1 SKEW3 . 351Surface area0 . 768Phallo surface intensity 3 vx3 . 253Log lac 30 . 718Phallo surface intensity 0 vx3 . 236Radius variance0 . 669Norm lac 32 . 930Volume0 . 668Norm lac 22 . 913BC Frac Dim0 . 649FLK1 surface intensity 5 vx2 . 857Log lac 40 . 612Norm lac 42 . 847Phallo surface intensity 10 vx0 . 554Phallo surface intensity 4 vx2 . 838Log lac 50 . 536Norm lac 52 . 753Sphericity0 . 423Phallo surface intensity 5 vx2 . 347HF70 . 408FLK1 surface intensity 6 vx2 . 310Shape index0 . 402HF92 . 141Lacunarity 10 . 381FLK1 surface intensity 7 vx1 . 893b/c0 . 342Phallo surface intensity 6 vx1 . 868Lacunarity 20 . 333HF51 . 575Lacunarity 30 . 309HF81 . 554Lacunarity 40 . 295FLK1 surface intensity 8 vx1 . 552HF40 . 287HF111 . 471Lacunarity 50 . 285Phallo surface intensity 7 vx1 . 444HF120 . 153a/c1 . 406DAPI Sd0 . 123Log lac 11 . 287DAPI gradient surface0 . 094FLK1 surface intensity 9 vx1 . 265Log norm lac 20 . 087HF61 . 158CVM0 . 076Phallo surface intensity 8 vx1 . 084Log norm lac 30 . 062FLK1 surface intensity 10 vx1 . 018Log norm lac 40 . 045HF10 . 978DAPI SKEW0 . 035FLK1 Sd0 . 942Log norm lac 50 . 033HF100 . 939DAPI mean0 . 029a/b0 . 937DAPI KURT0 . 022Note: The parameters are sorted based on their Fisher score , which is a measure of the discriminative power of the parameter . The performance of the classifiers was measured using the leave-one-out cross-validation method on the training set . Both classifiers recognized hepatocytes with ~100% accuracy , thus further improving the previous performance ( O'Gorman et al . , 1985 ) . The overall cell-type classification yielded 95 . 4% and 92 . 6% accuracy for the LDA and BNC , respectively . Although discriminating non-parenchymal cells is difficult even for a person skilled in the art , our algorithms achieved accuracy higher than 90% . The predictive performance of the classifiers is shown in Figure 3—figure supplement 3A , B . As expected , the first largest population of cells corresponds to hepatocytes ( 44 . 6% ± 2 . 7% , mean ± SEM ) followed by sinusoidal endothelial cells ( 29 . 8% ± 2 . 5% ) . Surprisingly , we found important quantitative differences for Kupffer and stellate cells . The percentage of Kupffer cells ( 8 . 7% ± 0 . 7% ) was lower than that of stellate cells ( 11 . 2% ± 1 . 0% ) , against previous estimates on 2D images ( Baratta et al . , 2009 ) . The percentage of other cells was 5 . 7% ± 0 . 8% . A 3D visualization of the localization of the nuclei of the different cell types is shown in Figure 3—figure supplement 3C–F . Finally , cells were segmented by expansion of the active mesh from the nuclei to the cell surface . The expansion was either limited to the cell cortex ( i . e . the maximum density of actin ) or to contacts with neighbouring cells or tubular structures ( Figure 3E ) . The active mesh expansion was parameterized by inner pressure and mesh rigidity . However , this algorithm over-segmented bi-nucleated cells into two cells with a single nucleus . Therefore , we used phalloidin intensity and nucleus-to-nucleus distance to recognize over segmented multinuclear cells and merge them . A manual check of segmentation of 2559 cells revealed only ~2% error for hepatocyte segmentation that is a further improvement of the state-of-the-art achievements by voxel-based segmentation methods ( Mosaliganti et al . , 2012 ) . The results of the segmentation of all imaged cellular and subcellular structures in the liver tissue ( i . e . cells , nuclei , sinusoidal and BC networks ) are presented in Figure 3E , Figure 3—figure supplement 4 , and Videos 2 and 3 . 10 . 7554/eLife . 11214 . 022Video 2 . Reconstruction of all imaged structures in a high-resolution image . A 2x2 stitched ( ~ 400 μm × 400 μm × 100 μm ) high-resolution image ( 0 . 3 μm × 0 . 3 μm × 0 . 3 μm per voxel ) was used . First , the reconstruction of the large vessels , that is , central vein ( CV ) ( cyan ) , portal vein ( PV ) ( orange ) and bile duct ( green ) are shown . Then , raw images and the corresponding reconstructed objects of the different structures are shown sequentially: sinusoids ( magenta ) , BC ( green ) , nuclei ( random colours ) and cells ( random colours ) . Additionally , central lines are shown for the tubular structures . Finally , all segmented structures are shown . This video provides a complete over view of the reconstructed objects in a typical high-resolution image . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 02210 . 7554/eLife . 11214 . 023Video 3 . Detailed reconstruction of all imaged structures in a high-resolution image . In order to highlight the details of the reconstruction of small structures [e . g . nuclei , bile canalicular ( BC ) network , etc . ] , a video of a small , cropped ( ~125 μm × 125 μm × 75 μm ) high-resolution image ( 3 μm × 0 . 3 μm × 0 . 3 μm per voxel ) was generated . Similarly to Video 2 , the raw image and the corresponding reconstructed structures of sinusoids ( magenta ) , BC ( green ) , nuclei ( random colours ) and cells ( random colours ) are shown sequentially . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 023 To evaluate the performance of the pipeline for the reconstruction of dense tissues , we generated a benchmark comprising a set of realistic 3D images of liver tissue . Each synthetic image consisted of four channels for the main structures forming the tissue , that is , cell nuclei , cell borders , sinusoidal and BC networks . We first generated 3D models of liver tissue based on experimental data ( see ‘Methods’ ) . The benchmark models had levels of complexity similar to that of the real tissue ( Figure 3—figure supplement 5 , 6 ) . Second , we imposed uneven staining to the models in order to resemble the experimental data . Third , the artificial microscopy images were simulated by convolving the models according to the 3D confocal microscope point spread function ( PSF ) ( Nasse et al . , 2007; Nasse and Woehl , 2010 ) and adding z-dependent Poisson noise . The resulting benchmark image statistics were similar to those from the images acquired in our experimental setup ( see ‘Methods’ ) ( Figure 3—figure supplement 5 ) . Given their general usefulness for testing image analysis software , the benchmark images and models are provided as supplementary material ( Supplementary file 1 , Morales-Navarrete et al . , 2016 ) . Finally , we applied our 3D tissue reconstruction pipeline to the benchmark images and quantified the accuracy of the reconstructed models using the precision-sensitivity framework ( Powers , 2011 ) . The overall quality was expressed as F-score , the harmonic mean between precision and sensitivity . The benchmark tests were performed in three sets of images with different signal-to-noise ratio ( 10:1 , 4:1 , 2:1 ) . For tubular structures , we achieved average ( over the different noise level sets ) F-scores of 0 . 90 ± 0 . 04 and 0 . 73 ± 0 . 06 for sinusoidal and BC networks , respectively . In the case of the nuclei and cell segmentation , we found average F-scores 0 . 91 ± 0 . 03 and 0 . 92 ± 0 . 03 , respectively . The detailed quantifications are shown in Figure 3—figure supplement 7A–L . Additionally , we measured morphometric parameters of the reconstructed structures such as the average radius of the tubular structures ( BC and sinusoidal networks ) and cell volumes . We obtained values of 2 . 72 ± 0 . 13 µm ( ground truth value = 3 . 0 µm ) and 0 . 58 ± 0 . 05 µm ( ground truth value = 0 . 5 µm ) for sinusoidal and BC networks , respectively ( Figure 3—figure supplement 7M , N ) . The average error for cell volume estimation was found to be 5 . 17% ± 1 . 97% ( Figure 3—figure supplement 7O ) . The benchmark experiments showed high accuracy for the reconstruction of the ‘ground truth’ models of all the morphologically different structures forming the liver tissue ( Figure 3—figure supplement 7 ) . Next , we applied our software to quantitatively analyse the geometric features of liver tissue from three adult mice . Geometric features have important implications , for example , for the development of models of fluid exchange between blood and hepatocytes ( Wisse et al . , 1985 ) . A critical parameter for blood flux models is the radius of sinusoids . We measured a radius of 4 . 0 ± 1 . 1 μm , a value close to quantifications by electron microscopy ( EM ) analysis ( Wisse et al . , 1985; Oda et al . , 2003; McCuskey , 2008 ) . In the sinusoidal networks , we determined the angles between two branching arms to be 111 . 6° ± 12 . 37° ( Figure 4—figure supplement 1B ) , against previous estimates ( Hammad et al . , 2014 ) . Moreover , the values for the BC network are similar to the sinusoidal network ( 110 . 36° ± 9 . 85° , Figure 4—figure supplement 1B ) . Additionally , we provided new geometric information such as the cardinality of the branching nodes ( Figure 4—figure supplement 1C ) . Recent studies on the morphometric parameters of the liver tissue ( Hammad et al . , 2014; Friebel et al . , 2015 ) provided either average values or limited data measurements of the hepatocytes volume , omitting information on their heterogeneity . We could not only perform accurate measurements of hepatocytes volumes and poly-nucleation , but also correlate them with polyploidy and spatial localization within the tissue . Interestingly , we found a multi-modal distribution of hepatocyte volumes ( Figure 4A ) in line with measurements on isolated hepatocytes ( Martin et al . , 2002 ) . A trivial explanation is that it reflects the presence of mono- and bi-nucleated hepatocytes . However , we found that this was not the case . The distribution of volumes of both mono- and bi-nucleated hepatocytes can be independently described by a mixture of two populations with mean volumes 3126 ± 1302 µm3 ( ~14% of cells ) and 5313 ± 1175 µm3 ( ~10% of cells ) , and 5678 ± 1176 µm3 ( ~45% of cells ) and 10606 ± 1532 µm3 ( ~30% of cells ) , respectively ( Figure 4B , C ) . Hence , surprisingly , although the bi-nucleated hepatocytes are assumed to be larger than the mono-nucleated , we found that a population of mono-nucleated hepatocytes can have a volume that does not differ from that of bi-nucleated hepatocytes ( Figure 4A–C ) . 10 . 7554/eLife . 11214 . 024Figure 4 . Distribution of hepatocyte volumes and DAPI integral intensity per cell for all hepatocytes ( A , B ) and separated by number of nuclei ( B , C and E , F ) . Whereas experimental data are shown by dots , the log-normal components fitted to data are shown by solid lines . ( A ) Cell volume distribution of all hepatocytes . ( B , C ) Cell volume distribution obtained for mono and bi-nucleated hepatocytes , respectively . ( D ) Distribution of DAPI integral intensity ( proportional to the content of DNA ) of all hepatocytes . ( E , F ) Distributions of DAPI integral intensity obtained for mono and bi-nucleated hepatocytes , respectively . The analysis was performed on 2559 hepatocytes ( excluding boundary cells ) from three adult mice . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 02410 . 7554/eLife . 11214 . 025Figure 4—figure supplement 1 . Morphometric features of the sinusoidal and bile canalicular ( BC ) networks . ( A ) Radius distribution of the sinusoidal capillary network . ( B ) Distributions of the angles between branches of BC and sinusoidal networks . ( C ) Cardinality of branching nodes of BC and sinusoidal networks . The data shown here correspond to a representative sample of adult mouse liver . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 02510 . 7554/eLife . 11214 . 026Figure 4—figure supplement 2 . ( A , B ) DAPI integral intensity normalization . Distribution of DAPI integral intensity per nucleus calculated for each sample ( A ) before and ( B ) after normalization . We found scaling ( stretching ) factors 1 . 19 and 0 . 93 for the second and third samples , respectively . ( C , D ) DNA content in bi-nuclear hepatocytes . DAPI integral intensity per nucleus was calculated for each nucleus of the cells . ( C ) The distribution of the ratio between DAPI integral intensity of the two nuclei in each cell . It follows a normal distribution with mean value 1 . 0 ± 0 . 21 ( mean ± SD ) . ( D ) The dependency between DAPI integral intensity of the two nuclei in bi-nuclear cells . They show a linear dependency ( R2 = 0 . 945433 ) with a slope of 0 . 995 , showing that both nuclei have the same DNA content in bi-nuclear hepatocytes . ( E , F ) Scatter plot of the volume versus DAPI integral intensity of ( E ) mono-nuclear and ( F ) bi-nuclear hepatocytes . The results of the hierarchical clustering of ( E ) mono-nuclear and ( F ) bi-nuclear hepatocytes are shown . Four ( 2n , 4n , 8n , 16n ) and three ( 2×2n , 2×4n , 2×8n ) populations were found for mono-nuclear and bi-nuclear hepatocytes , respectively . The classification was performed using volume and DAPI integral intensity per cell . We used an agglomerative hierarchical cluster algorithm and tested several distances for the dissimilarity calculation and different methods for the clustering . We found that the standardized Euclidean distance with the Ward method yielded the best results . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 026 Having found such a peculiar size distribution of bi-nucleated hepatocytes , we measured the total content of DNA per nucleus in every cell sub-population as the integral intensity of DAPI ( Coleman et al . , 1981; Xing and Lawrence , 1991; Dmitrieva et al . , 2011; Zhao and Darzynkiewicz , 2013 ) ( see ‘Methods’ ) . The resulting distribution ( Figure 4D ) shows three well-separated peaks . These presumably correspond to the 2n ( diploid nuclei ) , 4n and 8n ( polyploid nuclei ) DNA content previously reported ( Guidotti et al . , 2003; Martin et al . , 2002 ) ( note that this analysis does not resolve the aneuploidy of specific chromosomes ( Faggioli et al . , 2011 ) ) . Next , we asked how the nuclei are distributed between the mono- and bi-nucleated cell populations . Interestingly , in the small bi-nucleated hepatocytes ( volume < 8000 µm3 ) both nuclei had 2n DNA content , whereas in the large hepatocytes ( volume > 8000 µm3 ) both had 4n DNA content . Almost no bi-nuclear hepatocytes ( <1 . 0% ) with different amount of DNA per nucleus ( e . g . one nucleus with 2n and one with 4n ) were observed ( Figure 4—figure supplement 2C , D ) . These results suggest that the hepatocyte volume does not depend on the number of nuclei but rather on their polyploidy , in agreement with previous reports ( Miyaoka and Miyajima , 2013 ) . Therefore , we classified hepatocytes with respect to number of nuclei , volume and DNA content using a hierarchical cluster algorithm . We identified seven populations , namely 2n , 4n , 8n , 16n for mono-nuclear and 2×2n , 2×4n , 2×8n for bi-nuclear hepatocytes ( Figure 4—figure supplement 2E , F ) . Four populations ( mono-nucleated 2n and 4n , and bi-nucleated 2×2n and 2×4n ) were major , representing around 97% of all hepatocytes . The reports on the spatial distribution of polyploid hepatocytes are controversial ( Gentric and Desdouets , 2014 ) . Whereas some suggest that periportal hepatocytes show a lower polyploidy than the perivenous ones ( Gandillet et al . , 2003; Asahina et al . , 2006 ) , others suggest that both regions have similar polyploid compositions ( Margall-Ducos et al . , 2007; Pandit et al . , 2012 ) . These discrepancies prompted us to analyse the spatial distribution of mono- and bi-nucleated hepatocytes within the lobule . We particularly analysed the largest populations of hepatocytes , 2n , 4n , 2×2n and 2×4n . Strikingly , we found a pronounced zonation in their localization . Whereas the 2n mono-nucleated were enriched in the PC and PV regions , mono-nucleated 4n showed a homogeneous distribution between PV and PC regions ( Figure 5 ) . The 2×2n bi-nucleated hepatocytes have a similar pattern as the 2n mono-nucleated ( highly enriched in the CV and PV regions ) , but the density of 2×4n bi-nucleated was lower in those regions and increased in the middle region ( Figure 5 ) . As far as we know , this is the first time that polyploidy and poly-nuclearity are found to be zonated and follow a specific pattern . These findings have important implications for both the structural organization of liver tissue and its proliferating and metabolic activities . 10 . 7554/eLife . 11214 . 027Figure 5 . Relative density of different sub-populations of hepatocytes as function of central vein ( CV ) -portal vein ( PV ) axis coordinate . ( A , C , E , G ) Relative density of 2n mono-nucleated , 2x×2n bi-nucleated , 4n mono-nucleated , 2x×4n bi-nucleated hepatocytes , respectively . ( B , D , F , H ) 3D visualization of the corresponding sub-populations of hepatocytes . The analysis was performed on 2559 hepatocytes ( excluding boundary cells ) from three adult mice . The CV-PV axis is determined by the coordinate χ , which describes the position of a point relative to the closest CV and PV . χ=50× ( \|D-dpv\|-\|D-dcv\|D +1 ) , where dcv and dpv are the distances to the closest CV and PV respectively , and D is the CV-PV distance . χ takes values between 0 and 100 , where 0 and 100 represents a localization at the CV and PV surfaces , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 027 To test the general applicability of the pipeline as well as the robustness of our algorithms , we applied it to two morphologically distinct tissues , lung and kidney . Lung and kidney sections were stained for nuclei ( DAPI ) and the cell cortex ( F-actin by phalloidin ) . Kidney samples were additionally stained for the apical ( CD13 ) and basal ( fibronectin and laminin ) cell surface . The pipeline allowed us to generate geometrical reconstructions of the tissues ( Figure 6 and Videos 4 and 5 , respectively ) without fine-tuning of the parameters . As proof of principle , we extracted some statistics of the most relevant structures from each tissue . Structural information from both relatively large structures like alveoli in lung or glomerulus in kidney , and smaller ones like cells and nuclei were extracted from the geometrical models . Figure 6—figure supplement 1 , 2 show the statistical distributions of some interesting tissue features , such as cell volume and elongation , number of neighbouring cells , etc . Information about the spatial organization of the alveolar cells ( i . e . their localization relative to the alveoli ) in the lung was extracted as well . 10 . 7554/eLife . 11214 . 028Figure 6 . Reconstruction of geometrical models of lung and kidney tissues . 3D representation of the different structures segmented in each tissue: ( A , C ) nuclei and ( B , D ) cells in the lung and kidney tissues , respectively . The triangle meshes are drawn inside the inner box and the raw images outside . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 02810 . 7554/eLife . 11214 . 029Figure 6—figure supplement 1 . Morphometric features of lung tissue . Distributions of ( A ) volume , ( B ) elongation and ( C ) number of neighbouring cells for the lung cells . ( D ) Distribution of the cell position ( centre of the cell ) relative to the closest alveoli . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 02910 . 7554/eLife . 11214 . 030Figure 6—figure supplement 2 . Morphometric features of kidney tissue . ( A ) and ( B ) The size and volume distribution of the two cell types identified in the kidney tissue , proximal and distal tubular structures . It was observed that the two cell populations have different characteristic sizes , proximal cells were found to be larger than distal ones . ( C ) and ( D ) The distribution for the cells elongation and the number of neighbouring cells , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 03010 . 7554/eLife . 11214 . 031Video 4 . 3D reconstruction of lung tissue . Nuclei and cells reconstructed from a high-resolution image ( ∼220 μm × 220 μm × 80 μm ) . First , the raw images of the cell cortex ( F-actin by phalloidin ) and nuclei ( DAPI ) staining are displayed . Then , the reconstruction of the nuclei ( random colours ) and the cells ( random colours ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 03110 . 7554/eLife . 11214 . 032Video 5 . 3D reconstruction of kidney tissue . Nuclei and cells reconstructed from a high-resolution image ( ∼220 μm × 220 μm × 80 μm ) . First , the raw images of the cell cortex ( F-actin by phalloidin ) and nuclei ( DAPI ) staining are displayed . Then , the reconstruction of the nuclei ( random colours ) and the cells ( random colours ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11214 . 032 For example , in the lung , we found that the alveolar cells constitute around 19% of the volume , consistent with previous measurements ( Irvin and Bates , 2003 ) . In the kidney , we found that proximal tubule cells have larger volumes than distal tubule cells ( Figure 6—figure supplement 2 ) , also in agreement with previous studies ( Nyengaard et al . , 1993; Rasch and Dørup , 1997 ) . Altogether , the new data show that our pipeline is versatile and able to reconstruct geometrical models of tissues with fairly different architectures . We developed a versatile pipeline that combines new algorithms with established ones aimed to reconstruct geometrical models of dense tissues from confocal microscopy images acquired at different levels of resolution . Our pipeline is implemented in a freely available platform designed to address unmet computational needs . Despite many efforts , the reconstruction of digital geometrical models of tissues suffers from critical bottlenecks such as lack of automation , limited accuracy and low throughput analysis ( Peng et al . , 2010 ) . The platform developed here overcomes such bottlenecks in that it ( 1 ) achieves high accuracy of geometric reconstruction , ( 2 ) can process large volumes of imaged tissue , for example , a full liver lobule , ( 3 ) increases the image analysis performance to such an extent that the model reconstruction time is shorter than the biological experimental time and compatible with middle-throughput ( this is achieved by combining the computational efficiency of C++ with the CPU/GPU multi-threading capabilities ) , ( 4 ) can be run on a regular PC and ( 5 ) provides a flexible tool for constructing image processing pipelines that are tuneable for specific tissue and imaging conditions . For the automatic recognition of different cell types , we included morphological classifiers into the software . The user-friendly pipeline assembly mechanism allows adjusting the platform for specific tissue analysis demands . The newly developed algorithms both increase the quality of the results ( e . g . 3D image de-noising , LME method , active mesh tuning , cell classification ) and deal with problems for which there appears currently to be no real good solutions available ( e . g . correction of tissue deformation and combination of individual sections in the case of partial tissue removal ) ( Figure 1—figure supplement 1 ) . Our platform is implemented as stand-alone free to download software ( http://motiontracking . mpi-cbg . de ) . Furthermore , we created a benchmark of realistic images ( with the underlying ground truth model ) for the evaluation of 3D segmentation algorithms in biological images ( Supplementary file 1 , Morales-Navarrete et al . , 2016 ) . To test its efficacy , we applied it towards the generation of a multi-resolution geometrical model of liver tissue . The resulting model was used to extract quantitative measurements of various features of liver tissue organization , such as radius , branching angles and cardinality of the sinusoidal and BC networks , and to recognize different cell types based on their morphological parameters . Our analysis revealed an unexpected zonation pattern of hepatocytes with different size , nuclei and DNA content within the liver lobule . Furthermore , we extended the analysis to two additional tissues , lung and kidney , demonstrating the general applicability and robustness of our platform . In building our pipeline , we spent considerable effort to improve the accuracy of the measurements of cell and tissue parameters and preserve their contextual information . The new algorithms allow correcting major defects originating from tissue sectioning , improve the segmentation of cellular , subcellular and tissue-level structures , and extract morphological features and distributions in space . A major limiting factor in the development of a comprehensive geometrical model is the trade-off between imaging large volumes of samples to gain a view of the overall tissue architecture and imaging at high-resolution to achieve an accurate description of the structures at the limit of resolution of the light microscope , for example , the apical surface of hepatocytes forming the BC network . We solved this problem by imaging the tissue at low-resolution and registering within it the parts of tissue ( the PV-PC area in the case of the liver lobule ) imaged at high resolution . In this way , the measured morphological features ( e . g . BC ) and parameters ( e . g . cell size ) are embedded in their proper context of tissue architecture . For example , the hepatocyte volume is a parameter that has little value as average without considering the distribution of parameter values in the lobule ( Figure 5 ) . In general , the diversity of geometric features of the cells within the liver lobule could provide new insights into the regulation of metabolic zonation ( see below ) . Our nuclei reconstruction approach achieved accuracy higher than 90% . As shown in Figure 3—figure supplement 1K , the major source of errors is over-segmented nuclei . Additional steps to improve nuclei reconstruction , such as the region-merging algorithm ( Chittajallu et al . , 2015 ) to correct for over-segmentation , could reduce such errors . Even though our cell segmentation method proved able to identify and reconstruct cells with high accuracy , in a few cases ( ~2% ) , binuclear cells were mistaken for two separate cells due to weak staining of the cell cortex . Therefore , implementation of additional methods for enhancing the staining of the cell surface , such as the anisotropic plate diffusion filters ( Mosaliganti et al . , 2010; 2012 ) , could help reduce further the over-segmentation of multi-nuclear cells . The active mesh tuning allowed improving the accuracy of segmentation of the BC and sinusoidal networks . This is important since the accuracy of a geometrical model is indispensable for the development of predictive models of tissue function . For example , a model of blood flow through the sinusoidal network and exchange with hepatocytes via the space of Disse ( Ohtani and Ohtani , 2008; Wisse et al . , 1985 ) critically depends on the estimation of the sinusoid diameter . An overestimation of the sinusoidal tube radius would have major consequences for the predictions of blood cells flow through the sinusoidal network . Our geometrical model yielded a diameter of the sinusoidal-walled tube equal to the typical size of erythrocytes and lymphocytes . Therefore , it supports the model of active exchange of blood serum and lymph in the space of Disse , whereby blood flux propels cells through the sinusoids causing waves of capillary walls deformation ( McCuskey , 2008; Wisse et al . , 1985 ) . The active mesh tuning algorithm yielded a distribution of the radius of sinusoid capillaries with a mean value that was 20% lower ( Figure 4—figure supplement 1A ) than previously estimated by similar approaches ( Hammad et al . , 2014; Hoehme et al . , 2010 ) , but in line with the values reported by EM ( Wisse et al . , 1985 ) . The reconstruction also revealed a large difference with the previously reported angles between two arms of branching sinusoids ( 112° vs . 32° , Figure 4—figure supplement 1B ) . Moreover , the geometrical model provides correct values for other sinusoidal network parameters such as number of intersection nodes per mm3 ( 8 . 3 × 104 ± 1 . 9 × 104 ) and network length per mm3 ( 3 . 1 × 106 ± 0 . 3 × 106 µm ) , which appear to have been overestimated in a recent report ( Hammad et al . , 2014 ) ( see ‘Methods’ ) . The discrepancy between our geometrical model and others ( Hoehme et al . , 2010; Hammad et al . , 2014 ) could be due to differences in image processing and/or experimental procedures ( tissue fixation , image acquisition , etc . ) . One possible explanation for this discrepancy is that our platform applies the active mesh approach to the segmentation of structures on different scales ( from the apical surface of hepatocytes forming the BC to cells ) and this may yield a more precise geometrical reconstruction in comparison with voxel-based methods ( Figure 3A ) . For the marker-less cell-type recognition , we compared two approaches , the classical LDA and the more recent BNC , applied to nuclei morphology . The accuracy of both approaches was comparable , reaching higher than 99% efficiency for hepatocyte recognition and about 92–95% for all cell types . The latter value is highly significant since the distinction between stellate and sinusoid endothelial cells in the absence of specific markers is challenging even for a skilled person . The analysis of parameters that were mostly informative for cell type discrimination yielded some unexpected results . Although nuclear size and roundness were traditionally considered a priori as the most relevant parameters to discriminate hepatocytes from non-parenchymal cells ( Baratta et al . , 2009; O'Gorman et al . , 1985 ) , we found that they are less informative than the parameters related to nuclear texture ( e . g . moments of lacunarity ) . The analysis of parameters relevant for cell classification can shed light on the differences in cell morphology that are difficult to grasp by the naked eye . The accurate active mesh-based cell shape estimation led to well-separated peaks of cell volume distribution ( Figure 4A–C ) , which failed to be discriminated by approximation through Voronoi tessellation ( Bock et al . , 2010 ) ( data not shown ) . The analysis of liver tissue using our software platform revealed some unexpected biological findings . It is well established that hepatocytes are heterogeneous in size , number of nuclei ( mono and bi-nucleated cells ) and DNA content ( polyploidy ) . However , we observed that these features are not randomly distributed but follow a specific zonation pattern within the liver lobule . Surprisingly , the mono-nucleated 2n and bi-nucleated 2×2n hepatocytes were enriched in the CV and PV regions , whereas bi-nucleated 2×4n were more frequent in the middle region . This particular distribution suggests that polyploidy is spatially regulated and follows a gradient between CV and PV . Zonation of metabolic activities in the liver is well known , but zonation of mono- and bi-nucleated cells and total DNA content ( polyploidy ) remains controversial . The spatial distribution of hepatocytes according to their ploidy in the CV-PV axes correlates with the metabolic zonation . This correlation suggests a possible role of polyploidy in regulating hepatocyte functions in the liver lobule . Interestingly , two unique populations of cells with stem cell-like properties and the capacity to repopulate the liver have been recently identified ( Ray , 2015; Wang et al . , 2015; Font-Burgada et al . , 2015 ) . One population located close to the CV , which has been implicated in homeostatic hepatocyte renewal ( Wang et al . , 2015 ) , coincides with the mono-nucleated 2n cells we identified . The other population of hepatocytes located near the PV , which was found to repopulate the liver after injury ( Font-Burgada et al . , 2015 ) , could correspond to the low ploidy cells ( 2n and 2×2n ) we observed . These results inspire future studies aimed at exploring the mechanisms underlying regulation of mono- versus bi-nuclearity and polyploidy in the context of liver tissue structure , function and regeneration ( Zaret , 2015; Ray , 2015 ) . In this context , the accurate digital geometrical model of tissue is a valuable resource . Geometrical models provide the means of extracting structural information as a precondition for the development of functional models of tissues . They can be a tool for acquiring accurate quantitative measurements of morphological features and , as such , have the potential of uncovering the fundamental rules underlying tissue organization . In addition , the measurement of specific parameters , such as BC and sinusoid diameters , network cardinality , cell volume and shape , etc . , can serve as diagnostic markers of early stages of tissue dysfunction/repairing , thus providing new tools for clinical research and drug development . Six- to nine-week-old C57BL/6JOlaHsd mice were purchased from Charles River Laboratory . All animal studies were conducted in accordance with German animal welfare legislation and in strict pathogen-free conditions in the animal facility of the Max Planck Institute of Molecular Cell Biology and Genetics , Dresden , Germany . Protocols were approved by the Institutional Animal Welfare Officer ( Tierschutzbeauftragter ) and all necessary licenses were obtained from the regional Ethical Commission for Animal Experimentation of Dresden , Germany ( Tierversuchskommission , Landesdirektion Dresden ) ( license number: AZ 24-9168 . 24-9/2012-1 , AZ 24-9168 . 11-9/2012-3 ) . We took advantage of the fact that point-spread-function of confocal microscopes is strongly elongated in z-axis and developed a new de-noising algorithm based on the linear approximation of the image background intensity in the z-direction . Since confocal microscopy images are photon-limited and therefore obey Poisson statistics , we first found the parameters α and β that convert the photon counts ( N ) into the intensity ( I ) units , such that:I=αN+β where the operator . represents the average , α is the conversion coefficient from number of photons to intensity values and β is the offset of the microscope digitization system ( dark current ) . For this , we calculated the variance of the intensities between sequential optical z-sections for each X–Y pixel and binned them according to the pixel intensities . Then , the mean variance was calculated within each bin and , as a result , the dependency of mean variance upon the intensities was found ( Figure 1—figure supplement 2G ) . This dependency was found to be linear , as expected for a Poisson noise model:V ( I ) =α2N=α ( I-β ) where V ( I ) is the variance for each intensity level I . Moreover , when thick 3D tissue samples are imaged , it is required to use different laser intensity and microscope gain . This results in an increase of the intensity scaling factor α with the image depth . Therefore , we calculated the Poisson noise model for different image depths ( z-direction ) and then , we used α and β to estimate the variance for every pixel . After that , we estimated the background intensity of every pixel . Briefly , for each pixel a set of sequential intensities in z-direction was extracted ( Figure 1—figure supplement 2H , left ) . Then , the intensities were fitted by a straight line using the outlier-tolerant algorithm described in ( Sivia , 1996 ) ( Figure 1—figure supplement 2H , right ) . The prediction of the straight line was considered as the background intensity , and the difference between the measured intensity and background was considered as candidate foreground intensity . The candidate foreground intensities below a defined threshold ( expressed in variance units ) were excluded . Finally , the background was added to the foreground to form the de-noised image . To evaluate the performance of our algorithm , we applied it to a set of three artificial images of BC from our benchmark ( 2:1 signal-to-noise ratio ) . Additionally , we applied other methods such as median filtering , Gauss low-pass filtering and anisotropic diffusion , ‘pure denoise’ ( PD ) ( Luisier et al . , 2010 ) and ‘edge preserving de-noising and smoothing’ ( EPDS ) ( Beck and Teboulle , 2009 ) for comparison . The performance of each method was quantitatively evaluated using the metrics mean square error ( MSE ) and coefficient of correlation ( CoC ) , defined as follows:MSE=∑i∈Ω ( Ii-Ii* ) 2\|Ω\|CoC=∑i∈Ω ( Ii-I ) · ( Ii-I ) ( ∑i∈Ω ( Ii-I ) 2·∑i∈Ω ( Ii-I ) 2 ) 1/2 where Ω is the region of interest in the image , Ii and Ii* are the intensities at voxel i of the de-noised and noise-free ( ground truth ) images respectively , I and I* are the mean intensities of the de-noised and noise-free images , respectively . We calculated the MSE and CoC over the whole images ( global ) as well as in the vicinity of the objects ( Figure 1—figure supplement 3A ) . For PD and EPDS , we selected the best parameters for their performance before the comparison ( Figure 1—figure supplement 3B , C ) . The results of our quantifications are shown in Figure 1—figure supplement 4–5 .	Understanding how individual cells interact to form tissues in animals and plants is a key problem in cell and developmental biology . To be able to answer this question researchers need to use microscopy to observe the cells in a tissue , extract structural information from the images , and then generate three-dimensional digital models of the tissue . However , the software solutions that are currently available are limited , and reconstructing three-dimensional tissue from microscopy images remains problematic . To meet this challenge , Morales-Navarrete et al . extended the free software platform called MotionTracking , which had been used previously for two-dimensional work . The software now combines a series of new and established algorithms for analysing fluorescence microscopy images that make it possible to identify the different structures that make up a tissue and then create and analyse a three-dimensional model . Morales-Navarrete et al . used the software to analyse liver tissue from mice . The resulting model revealed that liver cells called hepatocytes are arranged in particular zones within the tissue according to their size and DNA content . The software was also applied successfully to analyse lung and kidney tissue , which demonstrates that the approach can be used to create three-dimensional models of a variety of tissues . Morales-Navarrete et al . ’s approach can rapidly generate accurate models of larger tissues than were previously possible . Therefore , it provides researchers with a powerful tool to analyse the different features of tissues . This tool will be useful for many areas of research: from understanding of how cells form tissues , to diagnosing diseases based on the changes to features in particular tissues .	[ "Abstract", "Introduction", "Results", "Discussion", "Methods" ]	[ "computational", "and", "systems", "biology", "developmental", "biology", "tools", "and", "resources" ]	2015	A versatile pipeline for the multi-scale digital reconstruction and quantitative analysis of 3D tissue architecture
Evolutionary studies are often limited by missing data that are critical to understanding the history of selection . Selection experiments , which reproduce rapid evolution under controlled conditions , are excellent tools to study how genomes evolve under selection . Here we present a genomic dissection of the Longshanks selection experiment , in which mice were selectively bred over 20 generations for longer tibiae relative to body mass , resulting in 13% longer tibiae in two replicates . We synthesized evolutionary theory , genome sequences and molecular genetics to understand the selection response and found that it involved both polygenic adaptation and discrete loci of major effect , with the strongest loci tending to be selected in parallel between replicates . We show that selection may favor de-repression of bone growth through inactivating two limb enhancers of an inhibitor , Nkx3-2 . Our integrative genomic analyses thus show that it is possible to connect individual base-pair changes to the overall selection response . Understanding how populations adapt to a changing environment is an urgent challenge of global significance . The problem is especially acute for mammal populations , which are often small and fragmented due to widespread habitat loss . Such populations often show increased inbreeding , leading to the loss of genetic diversity ( Hoffmann and Sgrò , 2011 ) . Because beneficial alleles in mammals typically come from standing genetic variation rather than new mutations like in microbes , this loss of diversity would ultimately impose a limit on the ability of small populations to adapt . Nonetheless , mammals respond readily to selection in many traits , both in nature and in the laboratory ( Darwin , 1859; Gingerich , 2001; Garland and Rose , 2009; Keightley et al . , 2001 ) . In quantitative genetics , such traits are interpreted as the overall effect from a large set of loci , each with an infinitesimally small ( and undetectable ) effect ( ‘infinitesimal model’ ) . Broadly speaking , the infinitesimal model has performed remarkably well across a wide range of selection experiments , and the model is the basis for commercial breeding ( Walsh and Lynch , 2018 ) . However , it remains unclear what type of genomic change is associated with rapid response to selection , especially in small populations where allele frequency changes can be dominated by random genetic drift . While a large body of theory exists to describe the birth , rise and eventual fixation of adaptive variants under diverse selection scenarios ( Maynard Smith and Haigh , 1974; Barton , 1995; Otto and Barton , 2001; Weissman and Barton , 2012; Crow and Kimura , 1965; Hill and Robertson , 1966 ) , few empirical datasets capture sufficient detail on the founding conditions and selection regime to allow full reconstruction of the selection response . This is particularly problematic in nature , where historical samples , environmental measurements and replicates are often missing . Selection experiments , which reproduce rapid evolution under controlled conditions , are therefore excellent tools to understand response to selection—and by extension—adaptive evolution in nature ( Garland and Rose , 2009 ) . Here we describe an integrative , multi-faceted investigation into an artificial selection experiment , called Longshanks , in which mice were selected for increased tibia length relative to body mass ( Marchini et al . , 2014 ) . The mammalian limb is an ideal model to study the dynamics of complex traits under selection: it is both morphologically complex and functionally diverse , reflecting its adaptive value; and limb development has been studied extensively in mammals , birds and fishes as a genetic and evolutionary paradigm ( Petit et al . , 2017 ) . The Longshanks selection experiment thus offers the opportunity to study selection response not only from a quantitative and population genetics perspective , but also from a developmental ( Marchini and Rolian , 2018 ) and genomic perspective . By design , the Longshanks experiment preserves a nearly complete archive of the phenotype ( trait measurements ) and genotype ( via tissue samples ) in the pedigree . Previously , Marchini et al . investigated how selection was able to overcome correlation between tibia length and body mass and produced independent changes in tibia length during the first 14 generations of the Longshanks experiment ( Marchini et al . , 2014 ) . Importantly , that study focused on the phenotypes and inferred genetic correlations indirectly using the pedigree . The current genomic analysis was initiated when the on-going experiment reached generation 17 and extends the previous study by integrating both phenotypic and genetic aspects of the Longshanks experiment . By sequencing the initial and final genomes , the current analysis benefits from direct and highly resolved genetic information . Here , with essentially complete information , we wish to answer a number of important questions regarding the factors that determine and constrain rapid adaptation: Are the observed changes in gene frequency due to selection or random drift ? Does rapid selection response of a complex trait proceed through innumerable loci of infinitesimally small effect , or through a few loci of large effect ? What type of signature of selection may be associated with this process ? Finally , when the same trait changes occur independently , do these depend on changes in the same gene ( s ) or the same pathways ( parallelism ) ? At the start of the Longshanks experiment , we established three base populations with 14 pairs each by sampling from a genetically diverse , commercial mouse stock ( Hsd:ICR , also known as CD-1; derived from mixed breeding of classical laboratory mice [Yalcin et al . , 2010] ) . In two replicate ‘Longshanks’ lines ( LS1 and LS2 ) , we bred mice by pairing 16 males and females ( and excluding sibling pairs ) with the longest tibia relative to the cube root of body mass for each sex . This corresponds to 15–20% of all offspring ( only details essential to understanding our analysis are summarized here . See Marchini et al . , 2014 for a detailed description of the breeding scheme ) . We kept a third Control line ( Ctrl ) using an identical breeding scheme , except that breeders were selected at random . In LS1 and LS2 , we observed a strong and significant response to selection in tibia length ( 0 . 29 and 0 . 26 Haldane or standard deviations ( s . d . ) per generation , from a selection differential of 0 . 73 s . d . in LS1 and 0 . 62 s . d . in LS2 ) . Over 20 generations , selection for longer relative tibia length produced increases of 5 . 27 and 4 . 81 s . d . in LS1 and LS2 , respectively ( or 12 . 7% and 13 . 1% in tibia length ) , with a modest decrease in body mass ( −1 . 5% in LS1 and −3 . 7% in LS2; Student’s t-test , p<2 × 10−4 and p<1 × 10−8 , respectively; Figure 1B and C; Figure 1—figure supplement 1; n . b . this relationship was in part biased by the F1 generation , which were fed a different diet and phenotyped three weeks later than later generations , see Marchini et al . , 2014 for details ) . By contrast , Ctrl showed no directional change in tibia length or body mass ( Figure 1C; Student’s t-test , p>0 . 05 ) . This approximately 5 s . d . change in 20 generations is rapid compared to typical rates observed in nature ( Hendry and Kinnison , 1999 , but see Grant and Grant , 2002 ) but is in line with responses seen in selection experiments ( Gingerich , 2001; Keightley et al . , 2001; Falconer and Mackay , 1996; Pitchers et al . , 2014 ) . The rapid but generally smooth increase in tibia length in Longshanks is typically interpreted as evidence for a highly dispersed genetic architecture with no individually important loci contributing to the selection response . This is classically described under quantitative genetics as the infinitesimal model . Crucially , the appropriate null hypothesis for the genomic response here should capture “polygenic adaptation” rather than a neutral model . We therefore developed a simulation that faithfully recapitulates the artificial selection experiment by integrating the trait measurements , selection regime , pedigree and genetic diversity of the Longshanks selection experiment , in order to generate an accurate expectation for the genomic response . Using the actual pedigree and trait measurements , we mapped fitness onto tibia length T and cube-root body mass B as a single composite trait lnTBϕ . We estimated ϕ from actual data as −0 . 57 , such that the ranking of breeders closely matched the actual composite ranking used to select breeders in the selection experiment , based on T and B separately ( Marchini et al . , 2014 ) ( Figure 1—figure supplement 2A ) . We assumed a maximally polygenic genetic architecture using an “infinitesimal model with linkage” ( abbreviated here as HINF ) , under which the trait is controlled by very many loci , each of infinitesimally small effect ( see Appendix for details ) . Results from simulations seeded with actual genotypes or haplotypes showed that overall , the predicted increase in inbreeding closely matched the observed data ( Figure 1—figure supplement 2B ) . We tested models with varying selection intensity and initial linkage disequilibrium ( LD ) , and for each , ran 100 simulated replicates to determine the significance of changes in allele frequency ( Figure 1—figure supplement 2C-E ) . This flexible quantitative genetics framework allowed us to explore possible changes in genetic diversity over 17 generations of breeding under strong selection . In simulations , we followed blocks of genomes as they were passed down the pedigree . In order to compare with observations , we seeded the initial genomes with single nucleotide polymorphisms ( SNPs ) in the same number and initial frequencies as the data . We observed much more variation between chromosomes in overall inbreeding ( Figure 1—figure supplement 2B ) and in the distribution of allele frequencies ( Figure 2—figure supplement 1B ) than expected from simulations in which the ancestral SNPs were initially in linkage equilibrium . This can be explained by linkage disequilibrium ( LD ) between the ancestral SNPs , which greatly increases random variation . Therefore , we based our significance threshold tests on simulations that were seeded with SNPs drawn with LD consistent with the initial haplotypes ( Figure 1—figure supplement 2C and E; see Appendix ) . Because our simulations assume infinitesimal effects of loci , allele frequency shifts exceeding this stringent threshold would suggest that discrete loci contribute significantly to the selection response . An excess of such loci in either a single LS replicate or in parallel would thus imply a mixed genetic architecture of a few large-effect loci amid an infinitesimal background . To detect the genomic changes in the actual Longshanks experiment , we sequenced all individuals of the founder ( F0 ) and 17th generation ( F17 ) to an average of 2 . 91-fold coverage ( range: 0 . 73–20 . 6×; n = 169 with <10% missing F0 individuals; Supplementary file 1 ) . Across the three lines , we found similar levels of diversity , with an average of 6 . 7 million ( M ) segregating SNPs ( approximately 0 . 025% , or 1 SNP per four kbp; Supplementary file 2; Figure 2—figure supplement 1A and Figure 2—figure supplement 2 ) . We checked the founder populations to confirm negligible divergence between the three founder populations ( across-line FST on the order of 1 × 10−4 ) , which increased to 0 . 18 at F17 ( Supplementary file 2 ) . This is consistent with random sampling from an outbred breeding stock . By F17 , the number of segregating SNPs dropped to around 5 . 8 M ( Supplementary file 2 ) . This 13% drop in diversity ( 0 . 9M SNPs genome-wide ) is predicted by drift . Our simulations confirmed this and moreover , showed that selection contributed negligibly to the drop in diversity ( Appendix , Figure 1—figure supplement 2B , D ) . We conclude that despite the strong selection on the LS lines , there was little perturbation to genome-wide diversity . Indeed , the changes in diversity in 17 generations were remarkably similar in all three lines , despite Ctrl not having experienced selection on relative tibia length ( Figure 2—figure supplement 1A ) . Hence , and consistent with our simulation results ( Figure 1—figure supplement 2B , D ) , changes in global genome diversity had little power to distinguish selection from neutral drift despite the strong phenotypic selection response . We next asked whether specific loci reveal more definitive differences between the LS replicates and Ctrl ( and from infinitesimal predictions ) . We calculated ∆z2 , the square of an arcsine transformed allele frequency difference between F0 and F17; this has an expected variance of 1/2Ne per generation , independent of starting frequency , and ranges from 0 to π2 . We averaged ∆z2 within 10 kbp windows ( see Methods for details ) , and found 169 windows belonging to eight clusters ( i . e . , loci ) that had significant shifts in allele frequency in LS1 and/or LS2 ( corresponding to 9 and 164 clustered windows respectively at p≤0 . 05 under HINF , max LD; ∆z2 ≥0 . 33 π2; genome-wide ∆z2 = 0 . 02 ± 0 . 03 π2; Figure 2; Figure 1—figure supplement 2D , Figure 2—figure supplement 2 , Figure 2—figure supplement 3; see Methods for details ) and 8 windows in three clusters in Ctrl ( genome-wide ∆z2 = 0 . 01 ± 0 . 02 π2 ) . The eight loci in Longshanks each overlapped between 2 to 179 genes and together contained 11 candidate genes with known roles in bone , cartilage and/or limb development ( e . g . , Nkx3-2 and Sox9; Table 1; Figure 2—figure supplement 3 , Figure 2—figure supplement 4 ) . Four out of the eight loci contain genes with a ‘short tibia’ or ‘short limb’ knockout phenotype ( Table 1; p≤0 . 032 from 1000 permutations , see Methods for details ) . Of the broader set of genes at these loci with any limb knockout phenotypes , only fibrillin 2 ( Fbn2 ) is polymorphic for SNPs coding for different amino acids , suggesting that for the majority of loci with large shifts in allele frequency , gene regulation was likely important in the selection response ( Figure 2—figure supplement 4; Supplementary file 3; see Appendix for further analyses on enrichment in gene functions , protein-coding vs . cis-acting changes and clustering with loci affecting human height ) . Taken together , two major observations stand out from our genomic survey . One , a polygenic , infinitesimal selection model with strong LD among marker SNPs performed better than moderate LD or no LD ( Figure 1—figure supplement 2E ) ; and two , we nevertheless find more discrete loci in LS1 and LS2 than in Ctrl , beyond the significance threshold set by the infinitesimal model ( Figure 2; Figure 2—figure supplement 2 ) . Thus , we conclude that although the genetic basis of the selection response in the Longshanks experiment may be largely polygenic , evidence strongly suggests discrete loci with major effect , even when each line is considered separately . We next tested the repeatability of the selection response at the gene/locus level using the two LS replicates . If the founding populations shared the same selectively favored variants , we may observe parallelism or co-incident selective sweeps , as long as selection could overcome random drift . Indeed , the ∆z2 profiles of LS1 and LS2 were more similar to each other than to Ctrl ( Figure 2 and 3A; Figure 3—figure supplement 1; Pearson’s correlation in ∆z2 from 10 kbp windows: LS1–LS2: 0 . 21 vs . LS1–Ctrl: 0 . 06 and LS2–Ctrl: 0 . 05 ) . Whereas previous genomic studies with multiple natural or artificial selection replicates focused mainly on detecting parallel loci ( Burke et al . , 2010; Jones et al . , 2012; Chan et al . , 2012; Kelly and Hughes , 2018 ) , here we have the possibility to quantify parallelism and determine the selection value of a given locus . Six out of eight significant loci at the HINF , max LD threshold were line-specific , even though all eight selected alleles were present in the F0 generation in both lines . This prevalence of line-specific loci was consistent under different significance thresholds . However , the two remaining loci that ranked first and second by selection coefficient were parallel , both with s > 0 . 3 ( Figure 3B; note that as outliers , the selection coefficient may be substantially overestimated , but their rank order should remain the same ) , supporting the idea that the probability of parallelism can be high among those loci with the greatest selection advantage ( Orr , 2005 ) . Finding just two parallel loci out of 8 discrete loci may appear to be low , given the genetic similarity in the founding generation and the identical selection applied to both Longshanks replicates . However , one should bear in mind the very many genetic paths to increasing tibia length under an infinitesimal model , and that the effect of drift is expected to be very strong in these small populations . In larger populations , the shift in the balance from drift to selection should result in selection being able to favor increasingly subtle variants and thus produce a greater proportion of parallel loci . However , we expect the trend of parallelism being enriched among the top loci to hold . In contrast to the subtle differences within each line in changes in global diversity over 17 generations ( Figure 2 and Figure 2—figure supplement 2 ) , we found the signature of parallelism to be significantly enriched in the comparison between the selected replicates ( χ2 test , LS1–LS2: p≤1 × 10−10 ) , as opposed to comparisons between each selected line and Ctrl ( LS1–Ctrl: p>0 . 01 and LS2–Ctrl: p>0 . 2 , both non-significant after correcting for multiple testing ) , or between simulated replicates ( Figure 3—figure supplement 1; see Appendix for details ) . Because the parallel selected loci between LS1 and LS2 have the highest selection coefficients and parallelism is not generally expected in our populations , these loci provide the strongest evidence for the role of discrete major loci . As such , the top-ranked parallel locus is the prime candidate for molecular dissection ( see Figure 4—figure supplement 1 and Appendix ‘Molecular dissection of Gli3’ for an additional a priori candidate locus with known limb function ) . Between the two major parallel loci , we chose the locus on chromosome 5 ( Chr5 ) at 41–42 Mbp for functional validation because it showed the strongest estimated selection coefficient , its signature of selection was clear , and crucially for functional characterization , it contains only three genes , including Nkx3-2 ( also known as Bapx1 ) , a known regulator of bone maturation ( Figure 2 and 4A ) ( Provot et al . , 2006 ) . At this locus , the pattern of variation resembles a selective sweep spanning 1 Mbp ( Figure 4A ) . Comparison between F0 and F17 individuals revealed no recombinant in this entire region ( Figure 5—figure supplement 1A , top panel ) , precluding fine-mapping using recombinants . We then analyzed the genes in this region to identify the likely target ( s ) of selection . First , we determined that no coding changes existed for either Rab28 or Nkx3-2 , the two genes located within the topologically associating domain ( TAD , which mark chromosome segments with shared gene regulatory logic ) ( Dixon et al . , 2012 ) . We then performed in situ hybridization and detected robust expression of Nkx3-2 and Rab28 in the developing fore- and hind limb buds of Ctrl , LS1 and LS2 E12 . 5 , in a domain broadly overlapping the presumptive zeugopod , the region including the tibia ( Figure 4—figure supplement 2B ) . A third gene , Bod1l , straddled the TAD boundary with its promoter located in the neighboring TAD , making its regulation by sequences in the selected locus unlikely . Consistent with this , Bod1l showed only weak or undetectable expression in the developing limb bud ( Figure 4—figure supplement 2A ) . We next combined ENCODE chromatin profiles and our own ATAC-Seq data to identify limb enhancers in the focal TAD . Here we found three novel enhancer candidates ( N1 , N2 and N3 ) carrying three , one and three SNPs respectively , all of which showed significant allele frequency shifts in LS1 and LS2 ( Figure 4B and C; Figure 5—figure supplement 1A ) . Chromosome conformation capture assays showed that the N1 and N3 sequences formed long-range looping contacts with the Nkx3-2 promoter—a hallmark of enhancers—despite as much as 600 kbp of intervening sequence ( Figure 4B ) . We next used transgenic reporter assays to determine whether these sequences could drive expression in the limbs . Here , we were not only interested in whether the sequence encoded enhancer activity , but specifically whether the SNPs would affect the activity ( Figure 4C and D ) . An examination of the predicted transcription factor binding sites showed that both the N1 and N3 enhancers contain multiple SNPs with consistent directional impact on the putative enhancer activity ( Figure 4C ) . In contrast , the N2 enhancer contains only a single SNP and is predicted to have inconsistent effect on its activity . We therefore excluded the N2 enhancer from further testing . We found that the F0 alleles of the N1 and N3 enhancers ( three SNPs each in about one kbp ) drove robust and consistent lacZ expression in the developing limb buds ( N1 and N3 ) as well as in expanded trunk domains ( N3 ) at E12 . 5 ( Figure 4E ) . In contrast , transgenic reporters carrying the selected F17 alleles of the N1 and N3 enhancers drove consistently weak , nearly undetectable lacZ expression ( Figure 4E ) . Thus , switching from the F0 to the F17 enhancer alleles led to a nearly complete loss in activity ( ‘loss-of-function’ ) at developmental stage E12 . 5 . This is consistent with the role of Nkx3-2 as a repressor in long bone maturation ( Provot et al . , 2006 ) . It should be noted that even though our selective regime favored an increase in the target phenotype ( tibia length ) , at the molecular level we expect advantageous loss- and gain-of-function variants to be equally likely favored by selection . In fact , in an additional functional validation example at the Gli3 locus , we found a gain-of-function enhancer variant that may have been favored at that locus ( see Figure 4—figure supplement 1 and Appendix ‘Molecular dissection of Gli3’ ) . At the Nkx3-2 locus , we hypothesize that the F17 allele causes de-repression of bone and/or cartilage formation by reducing enhancer activity and Nkx3-2 expression . Crucially , the F0 N1 enhancer showed activity that presages future long bone cartilage condensation in the limb ( Figure 4E ) . That is , the observed expression pattern recalls previous results that suggest that undetected early expression of Nkx3-2 may mark the boundaries and size of limb bone precursors , including the tibia ( Sivakamasundari et al . , 2012 ) . Conversely , over-expression of Nkx3-2 has been shown to cause shortened tibia ( even loss ) in mice ( Bren-Mattison et al . , 2011; Tribioli and Lufkin , 2006 ) . In humans , homozygous frameshift mutations in NKX3-2 cause the rare disorder spondylo-megaepiphyseal-metaphyseal dysplasia ( SMMD; OMIM: 613330 ) , which is characterized by short-trunk , long-limbed dwarfism and bow-leggedness ( Hellemans et al . , 2009 ) . The affected bones in SMMD patients broadly correspond to the expression domains of the two novel N1 ( limbs ) and N3 ( limbs and trunk ) enhancers . Instead of wholesale loss of Nkx3-2 expression , which would have been lethal in mice ( Akazawa et al . , 2000 ) or likely cause major defects similar to SMMD patients ( Hellemans et al . , 2009 ) , our in situ hybridization data did not reveal qualitative differences in Nkx3-2 expression domains between Ctrl or LS embryos ( Figure 4—figure supplement 2B ) . Taken together , our results recapitulate the key features of a cis-acting mode of adaptation: Nkx3-2 is a broadly expressed pleiotropic transcription factor that is lethal when knocked out ( Akazawa et al . , 2000 ) . We found no amino acid changes between the F0 and F17 alleles that could impact protein function . Rather , selection favored changes in tissue-specific expression by modular enhancers . By combining population genetics , functional genomics and developmental genetic techniques , we were able to dissect a megabase-long locus and present data supporting the identification of up to six candidate quantitative trait nucleotides ( QTNs ) . In mice , this represents a rare example of genetic dissection of a trait to the base-pair level . We next aimed to determine the evolutionary relevance of the Nkx3-2 enhancer variants at the molecular and the population levels . At the strongly expressed N3/F0 ‘trunk and limb’ enhancer , we note that the SNPs in the F17 selected allele lead to disrupted Nkx3-1 and Nkx3-2 binding sites ( Figure 4C and 5A; UNIPROBE database [Berger et al . , 2008] ) . This suggests that the selected SNPs may disrupt an auto-feedback loop to decrease Nkx3-2 activity in the limb bud and trunk domains ( Figure 5A ) . Using a GFP transgenic reporter assay in stickleback fish embryos , we found that the mouse N1/F0 enhancer allele was capable of driving expression in the distal cells but not in the fin rays of the developing fins ( Figure 5A ) . This pattern recapitulates fin expression of nkx3 . 2 in fish , which gives rise to endochondral radials ( homologous to ulna/tibia in mice ) ( Crotwell and Mabee , 2007 ) . Our results suggest that strong selection may have favored the weaker N1/F17 and N3/F17 enhancer alleles in the context of the Longshanks selection regime despite the deep functional conservation of the F0 variants . Using theory and simulations , we went beyond qualitative molecular dissection to quantitatively estimate the selection coefficient at the Nkx3-2 locus and its contribution to the total selection response in the Longshanks mice . We retraced the selective sweep of the Nkx3-2 N1 and N3 alleles through targeted genotyping in 1569 mice across all 20 generations . The selected allele steadily increased from around 0 . 17 to 0 . 85 in LS1 and 0 . 98 in LS2 but fluctuated around 0 . 25 in Ctrl ( Figure 5B ) . We estimated that such a change of around 0 . 7 in allele frequency would correspond to a selection coefficient s of ~0 . 24 ± 0 . 12 at this locus ( Figure 5—figure supplement 1B; see Appendix section on ‘Estimating selection coefficient of the top-ranking locus , Nkx3-2 , from changes in allele frequency’ ) . By extending our simulation framework to allow for a major locus against an infinitesimal background , we find that the Nkx3-2 locus would contribute 9 . 4% of the total selection response ( limits 3 . 6–15 . 5%; see Appendix section ‘Estimating selection coefficient’ for details ) in order to produce a shift of 0 . 7 in allele frequency over 17 generations . To avoid inflation stemming from estimating from outliers , we also independently estimated the contribution of the Nkx3-2 locus using a linear mixed animal model based on the full genotyped series mentioned above ( see Appendix section ‘Estimating selection coefficient , animal model’ for details ) . Using this alternative approach , we estimated that each selected allele increases tibia length by 0 . 36% ( N = 1569 , 95% conf . int . : 0 . 07–0 . 64% , p=0 . 0171 ) . Multiplying the effect with the increase in the allele frequency suggests that the Nkx3-2 locus alone would account for approximately 4% of the overall 12 . 9% increase in tibia length . This lower estimate of around 4% is nonetheless within the bounds of the estimate from simulations . Together , both approaches indicate that the Nkx3-2 locus contributes substantially to the selection response . Using the Longshanks selection experiment and synthesizing theory , empirical data and molecular genetics , we show that it is possible to identify some of the individual SNPs that have contributed to the response to selection on morphology . In particular , discrete , large-effect loci are revealed by their parallel response . Further work should focus on dissecting the mechanisms behind the dynamics of selective sweeps and/or polygenic adaptation by re-sequencing the entire selection pedigree , testing how the selection response depends on the genetic architecture , and the extent to which linkage places a fundamental limit on our inference of selection . Improved understanding in these areas may have broad implications for conservation , rapid adaptation to climate change and quantitative genetics in medicine , agriculture and in nature . All experimental procedures described in this study have been approved by the applicable University institutional ethics committee for animal welfare at the University of Calgary ( HSACC Protocols M08146 and AC13-0077 ) ; or local competent authority: Landesdirektion Sachsen , Germany , permit number 24–9168 . 11-9/2012-5 . All co-ordinates in the mouse genome refer to Mus musculus reference mm10 , which is derived from GRCm38 . Sequence data have been deposited in the SRA database under accession number SRP165718 and GEO under GSE121564 , GSE121565 and GSE121566 . Non-sequence data have been deposited at Dryad , doi:10 . 5061/dryad . 0q2h6tk . Analytical code and additional notes have been deposited in the following repository: https://github . com/evolgenomics/Longshanks ( Evolgenomics , 2019; copy archived at https://github . com/elifesciences-publications/Longshanks ) . Additional raw data and code are hosted via our institute's FTP servers at http://ftp . tuebingen . mpg . de/fml/ag-chan/Longshanks/ . Tibia length and body weight phenotypes were measured as previously described ( Marchini et al . , 2014 ) . A total of 1332 Control , 3054 LS1 , and 3101 LS2 individuals were recorded . Five outlier individuals with a skeletal dysplasia of unknown etiology were removed from LS2 and excluded from further analysis . Missing data in LS2 were filled in with random individuals that best matched the pedigree . Trait data were analyzed to determine response to selection based on the measured traits and their rank orders based on the selection index . Simulations were based on the actual pedigree and selection scheme , following one chromosome at a time . Each chromosome was represented by a set of junctions , which recorded the boundaries between genomes originating from different founder genomes; at the end , the SNP genotype was reconstructed by seeding each block of genome with the appropriate ancestral haplotype . This procedure is much more efficient than following each of the very large number of SNP markers . Crossovers were uniformly distributed , at a rate equal to the map length ( Cox et al . , 2009 ) . Trait value was determined by a component due to an infinitesimal background ( Vg ) ; a component determined by the sum of effects of 104 evenly spaced discrete loci ( Vs ) ; and a Gaussian non-genetic component ( Ve ) . The two genetic components had variance proportional to the corresponding map length , and the heritability was estimated from the observed trait values ( see Appendix section ‘Major considerations’ ) . In each generation , the actual number of male and female offspring were generated from each breeding pair , and the male and female with the largest trait value were chosen to breed . SNP genotypes were assigned to the founder genomes with their observed frequencies . However , to reproduce the correct variability requires that we assign founder haplotypes . This is not straightforward , because low-coverage individual genotypes cannot be phased reliably , and heterozygotes are frequently mis-called as homozygotes . We compared three procedures , which were applied within intervals that share the same ancestry: assigning haplotypes in linkage equilibrium ( LE , or ‘no LD’ ) ; assigning the two alleles at heterozygous sites in each individual to its two haplotypes at random , which minimizes linkage disequilibrium but is consistent with observed diploid genotypes ( ‘min LD’ ) ; and assigning alleles at heterozygous sites in each individual to the ‘reference’ and ‘alternate’ haplotype consistently within an interval , which maximizes linkage disequilibrium ( ‘max LD’ ) ( Figure 1—figure supplement 2C ) . For details , see legend in Figure 1—figure supplement 2 . To obtain significance thresholds , we summarized the genome-wide maximum ∆z2 shift for each replicate of the simulated LS1 and LS2 lines , averaged within 10 kb windows , and grouped by the selection intensity and extent of linkage disequilibrium ( LD ) . From this distribution of genome-wide maximum ∆z2 we obtained the critical value for the corresponding significance threshold ( typically the 95th quantile or p=0 . 05 ) under each selection and LD model ( Figure 3A; Figure 1—figure supplement 2E ) . This procedure controls for the effect of linkage and hitchhiking , line-specific pedigree structure , and selection strength . Sequencing libraries for high-throughput sequencing were generated using TruSeq or Nextera DNA Library Prep Kits ( Illumina , Inc , San Diego , USA ) according to manufacturer’s recommendations or using equivalent Tn5 transposase expressed in-house as previously described ( Picelli et al . , 2014 ) . Briefly , genomic DNA was extracted from ear clips by standard Protease K digestion ( New England Biolabs GmbH , Frankfurt am Main , Germany ) followed by AmpureXP bead ( Beckman Coulter GmbH , Krefeld , Germany ) purification . Extracted high-molecular weight DNA was sheared with a Covaris S2 ( Woburn , MA , USA ) or ‘tagmented’ by commercial or purified Tn5-transposase according to manufacturer’s recommendations . Each sample was individually barcoded ( single-indexed as N501 with N7XX variable barcodes; all oligonucleotides used in this study were synthesized by Integrated DNA Technologies , Coralville , Iowa , USA ) and pooled for high-throughput sequencing by a HiSeq 3000 ( Illumina ) at the Genome Core Facility at the MPI Tübingen Campus . Sequenced data were pre-processed using a pipeline consisting of data clean-up , mapping , base-calling and analysis from software fastQC v0 . 10 . 1 ( Andrews , 2016 ) ; trimmomatic v0 . 33 ( Bolger et al . , 2014 ) ; bwa v0 . 7 . 10-r789 ( Li and Durbin , 2010 ) ; GATK v3 . 4–0-gf196186 modules BQSR , MarkDuplicates , IndelRealignment ( McKenna et al . , 2010; DePristo et al . , 2011 ) . Genotype calls were performed using the GATK HaplotypeCaller under the GENOTYPE_GIVEN_ALLELES mode using a set of high-quality SNP calls made available by the Wellcome Trust Sanger Centre ( Mouse Genomes Project version three dbSNP v137 release [Keane et al . , 2011] ) , after filtering for sites segregating among inbred lines that may have contributed to the original seven female and two male CD-1 founders , namely 129S1/SvImJ , AKR/J , BALB/cJ , BTBR T+Itpr3 tf/J , C3H/HeJ , C57BL/6NJ , CAST/EiJ , DBA/2J , FVB/NJ , KK/HiJ , MOLF/EiJ , NOD/ShiLtJ , NZO/HlLtJ , NZW/LacJ , PWK/PhJ and WSB/EiJ based on ( Yalcin et al . , 2010 ) . We consider a combined ~100x coverage sufficient to recover any of the 18 CD-1 founding haplotypes still segregating at a given locus . The raw genotypes were phased with Beagle v4 . 1 ( Browning and Browning , 2016 ) based on genotype posterior likelihoods using a genetic map interpolated from the mouse reference map ( Cox et al . , 2009 ) and imputed from the same putative CD-1 source lines as the reference panel . The site frequency spectra ( SFS ) were evaluated to ensure genotype quality ( Figure 2—figure supplement 1A ) . Summary statistics of the F0 and F17 samples were calculated genome-wide ( Weir–Cockerham FST , π , heterozygosity , allele frequencies p and q ) in adjacent 10 kbp windows or on a per site basis using VCFtools v0 . 1 . 14 ( Danecek et al . , 2011 ) . The summary statistic ∆z2 was the squared within-line difference in arcsine square-root transformed MAF q; it ranges from 0 to π2 . The resulting data were further processed by custom bash , Perl and R v3 . 2 . 0 ( R Development Core Team , 2015 ) scripts . Peak loci were defined by a descending rank ordering of all 10 kbp windows , and from each peak signal the windows were extended by 100 SNPs to each side , until no single SNP rising above a ∆z2 shift of 0 . 2 π2 was detected . A total of 810 peaks were found with a ∆z2 shift ≥0 . 2 for LS1 and LS2 . Following the same procedure , we found 766 peaks in Ctrl . To determine whether genes with related developmental roles were associated with the selected variants , the topologically associating domains ( TADs ) derived from mouse embryonic stem cells as defined elsewhere ( Dixon et al . , 2012 ) were re-mapped onto mm10 co-ordinates . Genes within the TAD overlapping within 500 kbp of the peak window ( ‘core span’ ) were then cross-referenced against annotated knockout phenotypes ( Mouse Genome Informatics , http://www . informatics . jax . org ) . This broader overlap was chosen to include genes whose regulatory sequences ( e . g . , enhancers ) , but not necessarily their gene bodies , fall close to the peak window . We highlight candidate genes showing limb- and bone-related phenotypes , e . g . , with altered limb bone lengths or epiphyseal growth plate morphology , as observed in Longshanks mice ( Marchini and Rolian , 2018 ) , of the following categories ( along with their Mammalian Phenotype Ontology term and the number of genes ) : ‘abnormal tibia morphology/MP:0000558’ ( 212 genes ) , ‘short limbs/MP:0000547’ and ‘short tibia/MP:0002764’ ( 223 genes ) , ‘abnormal cartilage morphology/MP:0000163’ ( 321 genes ) , ‘abnormal osteoblast morphology/MP:0004986’ ( 122 genes ) . Note that we excluded compound mutants or those conditional mutant phenotypes involving transgenes . To determine if the overlap with these genes wassignificant , we performed 1000 permutations of the core span using bedtools v2 . 22 . 1 shuffle with the -noOverlapping option ( Quinlan and Hall , 2010 ) and excluding ChrY , ChrM and unassembled scaffolds . We then followed the exact procedure as above to determine the number of genes in the overlapping TAD belonging to each category . We reported the quantile rank as the P-value , ignoring ties . To determine other genes in the region , we list all genes falling within the entire hitchhiking window ( Supplementary file 3 ) . We downloaded publicly available chromatin profiles , derived from E14 . 5 limbs , for the histone H3 lysine-4 ( K4 ) or lysine-27 ( K27 ) mono-/tri-methylation or acetylation marks ( H3K4me1 , H3K4me3 and H3K27ac ) generated by the ENCODE Consortium ( Shen et al . , 2012 ) . We intersected the peak calls for the enhancer-associated marks H3K4me1 and H3K27ac and filtered out peaks overlapping promoters ( H3K4me3 and promoter annotation according to the FANTOM5 Consortium [Forrest et al . , 2014] ) . To calculate enrichment through the whole range of ∆z2 , a similar procedure was taken as in Candidate genes above . For knockout gene functions , genes contained in TADs within 500 kbp of peak windows were included in the analysis . We used the complete database of annotated knockout phenotypes for genes or spontaneous mutations , after removing phenotypes reported under conditional or polygenic mutants . For gene expression data , we retained all genes which have been reported as being expressed in any of the limb structures , by tracing each anatomy ontological term through its parent terms , up to the top-level groupings , e . g . , ‘limb’ , in the Mouse Genomic Informatics Gene Expression Database ( Finger et al . , 2017 ) . For E14 . 5 enhancers , we used a raw 500 kbp overlap with the peak windows because enhancers , unlike genes , may not have intermediaries and may instead represent direct selection targets . For coding mutations , we first annotated all SNPs for their putative effects using snpEff v4 . 0e ( Cingolani et al . , 2012 ) . To accurately capture the per-site impact of coding mutations , we used per-site ∆z2 instead of the averaged 10 kbp window . For each population , we divided all segregating SNPs into up to 0 . 02 bands based on per-site ∆z2 . We then tracked the impact of coding mutations in genes known to be expressed in limbs , as above . We reported the sum of all missense ( ‘moderate’ impact ) , frame-shift , stop codon gain or loss sites ( ‘high impact’ ) . A linear regression was used to evaluate the relationship between ∆z2 and the average impact of coding SNPs ( SNPs with high or moderate impact to all coding SNPs ) . For regulatory mutations , we used the same bins spanning the range of ∆z2 , but focused on the subset of SNPs falling within the ENCODE E14 . 5 limb enhancers . We then obtained a weighted average conservation score based on an averaged phastCons ( Pollard et al . , 2010 ) or phyloP ( Siepel et al . , 2005 ) score in ±250 bp flanking the SNP , calculated from a 60-way alignment between placental mammal genomes ( downloaded from the UCSC Genome Browser [Kent et al . , 2002] ) . We reported the average conservation score of all SNPs within the bin and fitted a linear regression on log-scale . In particular , phastCons scores range from 0 ( un-conserved ) to 1 ( fully conserved ) , whereas phyloP is the log10 of the P-value of the phylogenetic tree , expressed as a positive score for conservation and a negative score for lineage-specific accelerated change . We favored using phastCons for its simpler interpretation . Using the same SNP effect annotations described in the section above , we checked whether any specific SNP with significant site-wise ∆z2 in either LS1 or LS2 cause amino acid changes or protein disruptions and are known to cause limb defects when knocked out . For each position we examined outgroup sequences using the 60-way placental mammal alignment to determine the ancestral amino acid state and whether the selected variant was consistent with purifying vs . diversifying selection . The resulting 12 genes that matched these criteria are listed in Supplementary file 4 . We downloaded the set of 697 SNPs associated with loci for human height ( Wood et al . , 2014 ) to test if these loci cluster with the selected loci in the Longshanks lines . In order to facilitate mapping to mouse co-ordinates , each human SNP was expanded to 100 kbp centering on the SNP and converted to mm10 positions using the liftOver tool with the multiple mapping option disabled ( Kent et al . , 2002 ) . We were able to assign positions in 655 out of the 697 total SNPs . Then for each of the 810 loci above the HINF , no LD threshold in the selected Longshanks lines , the minimal distance to any of the mapped human loci was determined using bedtools closest with the -d option ( Quinlan and Hall , 2010 ) . When a region actually overlapped , a distance of 0 bp was assigned . To generate a permuted set , the 810 loci were randomly shuffled across the mouse autosomes using the bedtools shuffle program with the -noOverlapping option . Then the exact same procedure as the actual data was followed to determine the closest interval . The resulting permuted intervals followed an approximately normal distribution , with observed results falling completely below the range of permuted results , that is , closer to height-associated human SNPs . Detection of specific gene transcripts were performed as previously described in Brown et al . , 2005 . Probes against Nkx3-2 , Rab28 , Bod1l and Gli3 were amplified from cDNA from wildtype C57BL/6NJ mouse embryos ( Supplementary file 5 ) . Amplified fragments were cloned into pJET1 . 2/blunt plasmid backbones in both sense and anti-sense orientations using the CloneJET PCR Kit ( Thermo Fisher Scientific , Schwerte , Germany ) and confirmed by Sanger sequencing using the included forward and reverse primers . Probe plasmids have also been deposited with Addgene . In vitro transcription from the T7 promoter was performed using the MAXIscript T7 in vitro Transcription Kit ( Thermo Fisher Scientific ) supplemented with Digoxigenin-11-UTP ( Sigma-Aldrich ) ( MPI Tübingen ) , or with T7 RNA polymerase ( Promega ) in the presence of DIG RNA labeling mix ( Roche ) ( University of Calgary ) . Following TURBO DNase ( Thermo Fisher Scientific ) digestion , probes were cleaned using SigmaSpin Sequencing Reaction Clean-Up columns ( Sigma-Aldrich ) ( MPI Tübingen ) , or using Illustra MicroSpin G-50 columns ( GE Healthcare ) ( University of Calgary ) . During testing of probe designs , sense controls were used in parallel reactions to establish background non-specific binding . ATAC-seq was performed on dissected C57BL/6NJ E14 . 5 forelimb and hindlimb . Nuclei preparation and tagmentation were performed as previously described in Buenrostro et al . ( 2013 ) , with the following modifications . To minimize endogenous protease activity , cells were strictly limited to 5 + 5 min of collagenase A treatment at 37°C , with frequent pipetting to aid dissociation into single-cell suspensions . Following wash steps and cell lysis , 50 , 000 nuclei were tagmented with expressed Tn5 transposase . Each tagmented sample was then purified by MinElute columns ( Qiagen ) and amplified with Q5 High-Fidelity DNA Polymerase ( New England Biolabs ) using a uniquely barcoded i7-index primer ( N701-N7XX ) and the N501 i5-index primer . PCR thermocycler programs were 72°C for 4 min , 98°C for 30 s , 6 cycles of 98°C for 10 s , 65°C for 30 s , 72°C for 1 min , and final extension at 72°C for 4 min . PCR-enriched samples were taken through a double size selection with PEG-based SPRI beads ( Beckman Coulter ) first with 0 . 5X ratio of PEG/beads to remove DNA fragments longer than 600 bp , followed by 1 . 8X PEG/beads ratio in order to select for Fraction A as described in Milani et al . ( 2016 ) . Pooled libraries were run on the HiSeq 3000 ( Illumina ) at the Genome Core Facility at the MPI Tübingen Campus to obtain 150 bp paired end reads , which were aligned to mouse mm10 genome using bowtie2 v . 2 . 1 . 0 ( Langmead and Salzberg , 2012 ) . Peaks were called using MACS14 v . 2 . 1 ( Zhang et al . , 2008 ) . Chromosome conformation capture ( 3C ) template was prepared from pooled E14 . 5 liver , forelimb and hindlimb buds ( n = 5–6 C57BL/6NJ embryos per replicate ) , with improvements to the primer extension and library amplification steps following ( Sexton et al . , 2012 ) . The template was amplified with Q5 High-Fidelity Polymerase ( New England Biolabs GmbH , Frankfurt am Main , Germany ) using a 4C adapter-specific primer and a pool of 6 Nkx3-2 enhancer viewpoint primers ( and , in a separate experiment , a pool of 8 Gli3 enhancer-specific viewpoint primers; Supplementary file 6 ) . Amplified fragments were prepared for Illumina sequencing by ligation of TruSeq adapters , followed by PCR enrichment . Pooled libraries were sequenced by a HiSeq 3000 ( Illumina ) at the Genome Core Facility at the MPI Tübingen Campus with single-end , 150 bp reads . Sequence data were processed using a pipeline consisting of data clean-up , mapping , and analysis based upon cutadapt v1 . 10 ( Martin , 2011 ) ; bwa v0 . 7 . 10-r789 ( Li and Durbin , 2010 ) ; samtools v1 . 2 ( Li et al . , 2009 ) ; bedtools ( Quinlan and Hall , 2010 ) and R v3 . 2 . 0 ( R Development Core Team , 2015 ) . Alignments were filtered for ENCODE blacklisted regions ( ENCODE Project Consortium , 2012 ) and those with MAPQ scores below 30 were excluded from analysis . Filtered alignments were binned into genome-wide BglII fragments , normalized to Reads Per Kilobase of transcript per Million mapped reads ( RPKM ) , and plotted and visualized in R . Putative limb enhancers corresponding to the F0 and F17 alleles of the Gli3 G2 and Nkx3-2 N1 and N3 enhancers were amplified from genomic DNA of Longshanks mice from the LS1 F0 ( nine mice ) and F17 ( 10 mice ) generations and sub-cloned into pJET1 . 2/blunt plasmid backbone using the CloneJET PCR Kit ( Thermo Fisher Scientific ) and alleles were confirmed by Sanger sequencing using the included forward and reverse primers ( Supplementary file 7 ) . Each allele of each enhancer was then cloned as tandem duplicates with junction SalI and XhoI sites upstream of a β-globin minimal promoter in our reporter vector ( see below ) . Constructs were screened for the enhancer variant using Sanger sequencing . All SNPs were further confirmed against the rest of the population through direct amplicon sequencing . The base reporter construct pBeta-lacZ-attBx2 consists of a β-globin minimal promoter followed by a lacZ reporter gene derived from pRS16 , with the entire reporter cassette flanked by double attB sites . The pBeta-lacZ-attBx2 plasmid and its full sequence have been deposited and is available at Addgene . The reporter constructs containing the appropriate allele of each of the three enhancers were linearized with ScaI ( or BsaI in the case of the N3 F0 allele due to the gain of a ScaI site ) and purified . Microinjection into mouse zygotes was performed essentially as described ( DiLeone et al . , 2000 ) . At 12 d after the embryo transfer , the gestation was terminated and embryos were individually dissected , fixed in 4% paraformaldehyde for 45 min and stored in PBS . All manipulations were performed by R . N . or under R . N . ’s supervision at the Transgenic Core Facility at the Max Planck Institute of Molecular Cell Biology and Genetics , Dresden , Germany . Yolk sacs from embryos were separately collected for genotyping and all embryos were stained for lacZ expression as previously described ( Mortlock et al . , 2003 ) . Embryos were scored for lacZ staining , with positive expression assigned if the pattern was consistently observed in at least two embryos . Allele-specific primers terminating on SNPs that discriminate between the F0 from the F17 N3 enhancer alleles were designed ( rs33219710 and rs33600994; Supplementary file 8 ) . The amplicons were optimized as a qPCR reaction to give allele-specific , present/absent amplifications ( typically no amplification for the absent allele , otherwise average ∆Ct >10 ) . Genotyping on the entire breeding pedigree of LS1 ( n = 602 ) , LS2 ( n = 579 ) and Ctrl ( n = 389 ) was performed in duplicates for each allele on a Bio-Rad CFX384 Touch instrument ( Bio-Rad Laboratories GmbH , Munich , Germany ) with SYBR Select Master Mix for CFX ( Thermo Fisher Scientific ) and the following qPCR program: 50°C for 2 min , 95°C for 2 min , 40 cycles of 95°C for 15 s , 58°C for 10 s , 72°C for 10 s . In each qPCR run we included individuals of each genotype ( LS F17 selected homozygotes , heterozygotes and F0 major allele homozygotes ) . For the few samples with discordant results between replicates , DNA was re-extracted and re-genotyped or otherwise excluded . In sticklebacks , transgenic reporter assays were carried out using the reporter construct pBHR ( Chan et al . , 2010 ) . The reporter consists of a zebrafish heat shock protein 70 ( Hsp70 ) promoter followed by an eGFP reporter gene , with the entire reporter cassette flanked by tol2 transposon sequences for transposase-directed genomic integration . The Nkx3-2 N1/F0 enhancer allele was cloned as tandem duplicates using the NheI and EcoRV restriction sites upstream of the Hsp70 promoter . Enhancer orientation and sequence was confirmed by Sanger sequencing . Transient transgenic stickleback embryos were generated by co-microinjecting the plasmid ( final concentration: 10 ng/µl ) and tol2 transposase mRNA ( 40 ng/µl ) into freshly fertilized eggs at the one-cell stage as described in Chan et al . ( 2010 ) .	Humans have been making use of artificial selection for thousands of years . Much of what we eat , for example , from beef to poultry to cereals , comes from a collection of organisms with genomes that have been completely reshaped by the actions of generations of farmers and breeders . Yet , despite decades of research in evolutionary biology , it remains difficult to predict what will happen to an organism’s genes when selective pressure is applied . Traits that at first seem simple often arise from layers upon layers of complexity . It can take hundreds if not thousands of tiny changes to many genes , plus just the right alterations to a few key ones , to have a desired effect on a single trait . Also , if you consider that often the genomes of the starting population are unknown and that many traits are under simultaneous selection in wild populations , it becomes clear why many questions remain unanswered . Castro , Yancoskie , et al . have analyzed an on-going laboratory experiment dubbed “the Longshanks experiment” to explore how an animal’s genome changes under strong selection . Over five years , two independent populations of mice were selectively bred to have longer legs . In each generation , the mice were measured and those with the longest tibia – a bone in the shin – relative to their body mass were allowed to breed . Genetic data were also recorded . Now , Castro , Yancoskie , et al . have analyzed the genetic data up to the first 17 generations in the Longshanks experiment to find out what kind of genes may be relevant to the 13% increase in leg length seen in the mice so far . This analysis uncovered many genes , possibly thousands , all acting in concert to increase tibia length . But the gene with the largest effect by far was a key developmental gene called Nkx3-2 . Mutations in this gene cause a disease called spondylo-megaepiphyseal-metaphyseal dysplasia in people , which can lead to long limbs and a short trunk . Although inactivating this gene completely in mice is lethal , among the founding group of mice in the Longshanks experiment was a rare copy of Nkx3-2 . This copy of the gene worked perfectly in all tissues with the exception of the legs , where a genetic switch that controls it was left in the “off” state . Mice inheriting this short stretch of DNA ended up having longer tibia . In effect , these mice held the winning ticket in the genetic lottery that was the Longshanks experiment . Even in highly controlled experiments that record a great deal of information about the organisms involved , predicting how the genome will change and which genes will be involved is not a straightforward question . Finding out how the genome may change in response to selection is important because scientists can build better models to help with breeding farm animals or crops , or with predicting the consequences of climate change . As a result , experiments such as these may have broad applications in conservation , genomic medicine and agriculture .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2019	An integrative genomic analysis of the Longshanks selection experiment for longer limbs in mice
Neural representations of behaviorally relevant stimulus features displaying invariance with respect to different contexts are essential for perception . However , the mechanisms mediating their emergence and subsequent refinement remain poorly understood in general . Here , we demonstrate that correlated neural activity allows for the emergence of an invariant representation of natural communication stimuli that is further refined across successive stages of processing in the weakly electric fish Apteronotus leptorhynchus . Importantly , different patterns of input resulting from the same natural communication stimulus occurring in different contexts all gave rise to similar behavioral responses . Our results thus reveal how a generic neural circuit performs an elegant computation that mediates the emergence and refinement of an invariant neural representation of natural stimuli that most likely constitutes a neural correlate of perception . Understanding how the brain processes sensory input in order to generate behavioral responses remains a central problem in neuroscience . Several studies have shown that neurons located in more central brain areas tend to respond more selectively to stimuli than neurons located in more peripheral brain areas ( Rolls and Tovee , 1995; Perez-Orive et al . , 2002; Olshausen and Field , 2004; Hromadka et al . , 2008 ) . At the same time , however , neural representations in more central brain areas become more robust to the differential patterns of sensory input associated with a given stimulus ( e . g . those experienced when looking at the same object under different levels of illumination ) . Such invariant representations are thought to mediate context-independent object recognition and have been observed across sensory modalities ( auditory: [Bendor and Wang , 2005; Barbour , 2011; Bizley and Cohen , 2013; Rabinowitz et al . , 2013]; visual: [Quiroga et al . , 2005; Zoccolan et al . , 2007; Rust and Dicarlo , 2010; DiCarlo et al . , 2012; Sharpee et al . , 2013]; somatosensory: [DiCarlo and Johnson , 1999; Pei et al . , 2010]; olfactory: [Stopfer et al . , 2003; Cleland et al . , 2007] ) . Modeling studies have proposed that such representations are built and refined in a feedforward manner across successive brain areas by using OR-like operations ( Hubel and Wiesel , 1962; Riesenhuber and Poggio , 1999; Kouh and Poggio , 2008 ) . Indeed , these studies suggest that combining neural activities that are locally invariant to complementary subsets of stimulus waveforms associated with a given object should then give rise to a more globally invariant representation . However , the actual computations used by the brain that give rise to invariant neural representations remain poorly understood to this day . Wave-type gymnotiform weakly electric fish offer an attractive model system for studying the emergence and refinement of feature invariant neural representations because of well-characterized anatomy and relatively simple natural sensory stimuli that can easily be mimicked in the laboratory ( Chacron et al . , 2011; Marsat et al . , 2012; Marquez et al . , 2013; Stamper et al . , 2013; Krahe and Maler , 2014; Clarke et al . , 2015b ) . These fish generate a quasi-sinusoidal electric organ discharge ( EOD ) at a typical individual frequency , resulting in an electric field that surrounds the body . EOD perturbations are sensed by an array of electroreceptors that synapse onto pyramidal neurons within the hindbrain electrosensory lateral line lobe ( ELL ) that in turn synapse onto neurons within the midbrain Torus semicircularis ( TS ) . When two conspecifics come into contact ( <1 m ) , each fish experiences a sinusoidal amplitude modulation ( i . e . beat ) with a frequency that is equal to the difference between the two EOD frequencies . Natural communication signals or chirps consist of transient increases in EOD frequency and always occur simultaneously with the beat under natural conditions . Stimulus waveforms associated with chirps can be uniquely characterized by the increase in EOD frequency , duration , and the beat phase at onset . It has long been recognized that chirp stimulus waveforms are highly heterogeneous ( Zupanc and Maler , 1993; Hupé and lewis . , 2008 ) . Interestingly , a recent study has shown that , while duration and frequency increase of chirps were both narrowly distributed , the beat phase at chirp onset was instead uniformly distributed over the entire beat cycle ( Aumentado-Armstrong et al . , 2015 ) . This suggests that the observed heterogeneities in stimulus waveforms are , at least in part , caused by the fact that a chirp with given frequency increase and duration can occur with equal probability at any phase of the beat cycle . Previous studies have shown that weakly electric fish perceive chirps as evidenced from behavioral responses ( Hupé and Lewis , 2008 ) . Electrophysiological studies have also characterized how chirps are encoded across successive stages of processing and notably , both afferents and ELL pyramidal neurons tend to respond differentially to different chirp stimulus waveforms ( Benda et al . , 2005; 2006; Marsat et al . , 2009; Marsat and Maler , 2010 ) . TS neurons display large heterogeneities in their responses: while some respond in a manner that is similar to that of ELL neurons , others instead respond selectively to chirps through reliable and precisely timed action potentials ( Vonderschen and Chacron , 2011 ) . A recent study has , furthermore , shown that some TS neurons can respond similarly to different chirp stimulus waveforms ( Aumentado-Armstrong et al . , 2015 ) . However , the critical questions of whether weakly electric fish can perceive that different waveforms might constitute signatures of the same chirp stimulus and , if so , the computations leading to the emergence and refinement of an invariant neural representation have not been investigated to date . We first recorded from single peripheral receptor afferents ( Figure 2A ) in response to different patterns of sensory input resulting from the same chirp stimulus occurring at different phases during the beat cycle ( Figure 1E ) . Single afferent responses consisted of patterns of excitation and inhibition that strongly depended on the stimulus waveform that was presented ( Figure 2B , blue curves ) . Indeed , population-averaged responses strongly varied depending on where the chirp occurred within the beat cycle ( Figure 2C ) . Afferents were excited when the chirp occurred at beat phases lower than 180° and inhibited when it occurred at beat phases greater or equal to 180° ( Figure 2D , blue curve ) . We shall henceforth refer to the stimulus waveforms that gave rise to excitation and inhibition as '+ chirps' and '- chirps' , respectively . The similarity between single afferent firing rate responses to different waveforms during a 30 ms time window commensurate with chirp duration was then compared to that obtained between the stimulus waveforms during the same time window in order to measure phase invariance ( see Methods ) . An invariance score near zero indicates that the response waveforms display heterogeneities that are similar to those displayed by the stimulus waveforms while a value near one implies that the response waveforms are all the same . We found that the invariance score for single afferents was near zero ( mean: 0 . 11 ± 0 . 01; max: 0 . 15; min: 0 . 07; population: 0 . 08 ) ( Figure 2E , inset ) : this is because the distance between responses was always more or less equal to the distance between the corresponding stimulus waveforms ( Figure 2—figure supplement 1 ) . We note that this low value was not due to variability in the response of single afferents , as the population responses obtained by linearly summating the individual neural responses also displayed invariance scores near zero ( Figure 2E , inset ) . We further note that computing invariance scores using only responses to subsets of chirps ( i . e . those computed using only same type '++/- -' or opposite type '+-' chirp combinations ) resulted in low values as well ( Figure 2F , blue ) . Thus , we conclude that single peripheral afferent responses were minimally phase invariant as these neurons faithfully encoded the different patterns of sensory input resulting from the same chirp occurring at different phases within the beat cycle . 10 . 7554/eLife . 12993 . 006Figure 2 . Correlated but not single peripheral afferent activity provides a representation that is invariant to the different patterns of stimulation resulting from the same chirp occurring at different phases within the beat cycle . ( A ) Schematic showing the different stages of processing in the electrosensory system . Recordings were made from single ( N = 18 ) as well as pairs ( N = 8 ) of afferents . ( B ) Example spike trains ( blue , top ) , firing rates averaged across trials ( middle , blue ) , and correlation coefficient ( purple ) from an example afferent pair to different stimulus waveforms associated with the same electrosensory object . Note that each stimulus was presented at least 20 times . The horizontal bars ( shaded gray ) represent the chirp window ( 30 ms ) used for evaluation . ( C ) Population-averaged firing rate responses to the different stimulus waveforms during the 30 ms chirp window shown in B . ( D ) Population-averaged tuning curves obtained from single neuron firing rates ( blue ) and from the correlation coefficient ( purple ) . '+ chirps' and '- chirps' were defined as the stimulus waveforms that gave rise to positive ( i . e . increases in firing rate ) and negative ( i . e . decreases in firing rate ) responses in afferents , respectively . ( E ) Correlation coefficient as a function of time in response to all stimulus waveforms . Inset: Invariance score computed from single afferents ( left ) , from the population-averaged activity ( center ) , and from the correlation coefficient ( right ) . ( F ) Invariance score computed from single afferent activity ( blue ) and from correlated activity ( purple ) computed for all stimulus waveforms , '++/- - chirps' , and '+- chirps' . '' indicates statistical significance at the p = 0 . 05 level using a one-way ANOVA with Bonferroni correction . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 00610 . 7554/eLife . 12993 . 007Figure 2—source data 1 . Source data for Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 00710 . 7554/eLife . 12993 . 008Figure 2—figure supplement 1 . Phase invariant coding by single and correlated activity . ( A ) Plot of the distance between responses as a function of the distance between the corresponding stimulus waveforms for an example single afferent ( blue ) and for the correlated activity of an example afferent pair ( purple ) . ( B ) Time varying correlation coefficients in response to eight different stimulus waveforms for different time windows ( see legend ) . ( C ) Invariance as a function of time window length . Robust invariance was seen for time windows up to ˜60 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 00810 . 7554/eLife . 12993 . 009Figure 2—figure supplement 1—source data 1 . Source data for Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 009 We next investigated phase invariance for multiple afferents . To do so , we performed simultaneous recordings from afferent pairs , as well as recombined recorded activities from single afferents , in response to the patterns of sensory input resulting from the same chirp occurring at different phases within the beat cycle used above for single afferents . Similar results were seen when considering simultaneously recorded or recombined pairs: afferent activities were always more similar after chirp onset as afferent spiking activities were more synchronized either through excitation or inhibition ( Figure 2B , blue curves compare Firing Rate 1 to Firing Rate 2 ) . We thus quantified similarity by computing a time varying correlation coefficient between spike trains at short timescales ( see Methods ) . Our results show that correlation transiently increased following chirp onset ( Figure 2B , purple curves ) . Importantly , the time course of the correlation coefficient was similar for all stimulus waveforms ( Figure 2E ) , as quantified by a large invariance score ( mean: 0 . 53 ± 0 . 02; max: 0 . 67; min: 0 . 35; population: 0 . 61 ) ( Figure 2E , inset & Figure 2F ) that was not significantly different for either simultaneously recorded ( 0 . 51 ± 0 . 02 ) or recombined ( 0 . 53 ± 0 . 03 ) afferent activities ( t-test , p=0 . 4 ) . These high invariance scores were obtained because the distance between the time-varying correlation coefficient was always less than that between the corresponding stimulus waveforms ( Figure 2—figure supplement 1 ) . Similar results were observed when systematically varying the time scale at which correlations were computed up to ∼60 ms ( Figure 2—figure supplement 1 ) . We conclude that correlated activity at short timescales in receptor afferents displays phase invariance . Information transmitted by neural activity is only useful to the organism if it is actually decoded downstream . To investigate whether the electrosensory system actually uses the fact that correlated afferent activity provides a phase invariant representation of natural communication signals , we recorded from hindbrain ELL pyramidal neurons ( Figure 3A ) . These receive synaptic input from afferents and furthermore constitute the sole output neurons of the ELL ( Maler , 1979; Maler et al . , 1981 ) . Pyramidal neurons can be classified as either ON or OFF-type based on whether they respond to increases in stimulus amplitude through increases or decreases in firing rate , respectively ( Saunders and Bastian , 1984 ) ( see [Clarke et al . , 2015b] for review ) . Previous results have shown that pyramidal neurons respond to correlated afferent activity ( Berman and Maler , 1999; Middleton et al . , 2009 ) . Thus , our working hypothesis is that , if single ON and OFF-type ELL pyramidal neurons actually decode increases in correlated afferent activity in response to natural communication stimuli , then they should: 1 ) display significantly more phase invariance than single afferents and 2 ) respond in a complementary fashion to different waveform subsets . Specifically , ON and OFF-type pyramidal neurons should respond more similarly with excitation and inhibition to the stimulus waveforms giving rise to synchronized excitation ( i . e . '+ chirps' ) and inhibition ( i . e . '- chirps' ) in afferents , respectively . 10 . 7554/eLife . 12993 . 010Figure 3 . ON and OFF-type hindbrain ELL pyramidal neurons display more phase invariance by decoding correlated activity from peripheral afferents . ( A ) Schematic showing the different stages of processing in the electrosensory system . Recordings were made from ELL ON- ( N = 22 ) and OFF-type ( N = 9 ) pyramidal neurons . ( B ) Example responses of an ON-type ( green ) and an OFF-type neuron ( magenta ) to stimulus waveforms associated with the same electrosensory object . Shown are the stimulus waveforms , raster plots showing responses to 20 presentations of each stimulus , and trial-averaged time varying firing rates . Stimulus waveforms corresponding to '+ chirps' are shown above while those corresponding to '- chirps' are shown below . Note that excitatory responses of the ON-type neuron were mainly observed during + chirps while excitatory responses of the OFF-type neuron were found for - chirps . ( C , D ) Averaged time dependent firing rates for ON ( C ) and OFF ( D ) –type neurons in response to '+ chirps' ( dark green ) and '- chirps' ( light green ) . ( E ) Tuning curves averaged from individual responses for ON- ( green ) and OFF ( magenta ) -type neurons . ( F ) Phase invariance scores computed from individual ON and OFF-type pyramidal neurons for all stimulus waveforms , same-type chirp combinations only ( '++/- -' ) , and opposite-type chirp combinations only '+- chirps' . '' indicates statistical significance at the p = 0 . 05 level using a one-way ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 01010 . 7554/eLife . 12993 . 011Figure 3—source data 1 . Source data for Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 01110 . 7554/eLife . 12993 . 012Figure 3—figure supplement 1 . Responses of ELL pyramidal ON and OFF type populations to chirp stimuli . ( A ) Averaged responses of ON ( green; N = 22 ) and OFF- ( magenta; N = 9 ) type pyramidal neurons to '+ chirps' ( top row ) and '- chirps' ( lower row ) . Note that '+ chirps' elicited increased response of ON-type units , while '- chirps' instead elicited increased response in OFF-type units . ( B ) Correlation coefficients for linear correlations of the ON- and OFF-type population responses for the chirps occurring at different phases in the beat cycle . Note that for all chirp waveforms the population responses were significantly negatively correlated ( r<-0 . 45; p<<10–3 in all cases ) . ( C ) Plots of the distance between responses as a function of the distance between the corresponding stimulus waveforms for example ON-type ( green ) and OFF-type ( magenta ) ELL pyramidal cell . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 01210 . 7554/eLife . 12993 . 013Figure 3—figure supplement 1—source data 1 . Source data for Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 013 Our results were largely consistent with these predictions as '+ chirps' mostly gave rise to similar excitatory activity patterns in ON-type cells ( Figure 3B , top rows , green; Figure 3C , dark green ) and inhibitory activity patterns in OFF-type cells ( Figure 3B , top rows , magenta; Figure 3D , light magenta ) . In contrast , '- chirps' mostly gave rise to patterns of inhibition in ON-type cells ( Figure 3B , lower rows , green; Figure 3C , light green ) and excitation in OFF-type cells ( Figure 3B , lower rows , magenta; Figure 3D , dark magenta ) . The complementarity of responses was furthermore reflected by the fact that the population responses of ON and OFF cells were negatively correlated during responses to all waveforms ( Figure 3—figure supplement 1 ) . Consequently , the average tuning curves of ON and OFF-type cells were mirror images of one another ( Figure 3E ) . Consistent with our predictions , ELL pyramidal neurons were significantly more phase invariant than single afferents ( ELL ON; mean: 0 . 19 ± 0 . 01; max: 0 . 32; min: 0 . 03; population: 0 . 19; ELL OFF; mean: 0 . 16 ± 0 . 01; max: 0 . 21; min: 0 . 07; population: 0 . 16 ) ( ANOVA with Bonferroni correction , p<0 . 01 ) ( Figure 3F ) . Further analysis revealed that this was because the distance between responses was , on average , lower than that between the corresponding stimulus waveforms ( Figure 3—figure supplement 1 ) . Why are ELL pyramidal neurons more phase invariant than single afferents ? We hypothesized that the increased phase invariance is primarily due to the fact that ON cells respond similarly with excitation to '+ chirps' and with inhibition to '- chirps' while OFF cells instead respond similarly with inhibition to '+ chirps' and with excitation to '- chirps' . These opposite responses to different subsets of chirp stimulus waveforms should limit phase invariance as the distances between the evoked responses will then be larger . To test this hypothesis , we computed the invariance score only for same-type chirps combinations ( i . e . '++/- -' ) as well as for opposite-type chirps combinations ( i . e . '+-' ) . We found that invariance scores computed for same-type chirp combinations were significantly larger than those computed for opposite-type chirp combinations for both ON ( ANOVA; p<0 . 01 ) and OFF-type ( ANOVA; p<0 . 018 ) pyramidal cells ( Figure 3F ) . Further , average invariance score values computed for opposite-type chirp combinations in ON ( 0 . 15 ± 0 . 02 ) and OFF-type ( 0 . 11 ± 0 . 01 ) ELL pyramidal cells were similar to those obtained for single afferents ( 0 . 16 ± 0 . 002 ) ( compare Figures 3F and 2F ) . These results thus confirm our hypothesis that the increased phase invariance of ELL pyramidal cells is due to the fact that they respond more similarly to '+ chirps' and '- chirps' than single afferents . However , phase invariance in ELL is limited by the fact that ON and OFF-type pyramidal cells both respond in opposite fashion to different subsets of chirp stimulus waveforms ( i . e . '+ chirps' and '- chirps' ) . So far , we have shown that correlated but not single afferent activity provided a phase invariant representation of natural communication stimuli . Recordings from ELL pyramidal neurons revealed that these responded primarily through excitation to one half of the input patterns resulting from the same chirp occurring at different phases within the beat cycle and through inhibition to the other half , which limits phase invariance . Perhaps the simplest way to increase phase invariance is to sum the activities of ON and OFF cells . These responses should then be more similar and thus increase phase invariance . To test this hypothesis , we recorded from midbrain TS neurons ( Figure 4A ) . Anatomical studies have shown that neurons within the midbrain TS receive only direct excitatory synaptic input from multiple ON and OFF-type ELL pyramidal neurons ( Carr and Maler , 1985 ) , as reflected by their response properties ( McGillivray et al . , 2012 ) . We thus hypothesized that single TS neurons should display significantly higher phase invariance than ELL neurons on average by responding primarily with excitation to most if not all patterns of stimulation input resulting from the same chirp occurring at different phases within the beat cycle . 10 . 7554/eLife . 12993 . 014Figure 4 . Midbrain TS neurons display invariant responses to the different patterns of stimulation resulting from the same chirp occurring at different phases within the beat cycle . ( A ) Schematic showing the different stages of processing in the electrosensory system . Recordings were made from TS neurons that receive balanced excitatory input from both ON and OFF-type ELL hindbrain neurons ( N = 25 ) . ( B ) Example responses of a TS neuron to stimulus waveforms associated with the same electrosensory object . Shown are the stimulus waveforms , raster plots showing responses to 20 presentations of each stimulus , and trial-averaged time varying firing rates . Stimulus waveforms corresponding to '+ chirps' are shown above while those corresponding to '- chirps' are shown below . ( C ) Averaged responses of TS neurons were similar when presented with stimulus waveforms associated with a given electrosensory object . ( D ) Tuning curve averaged from individual responses ( mean ± SEM ) for TS neurons . E . Invariance scores from single TS neurons for all stimulus waveforms , same-type chirp combinations only ( '++/- -' ) , and opposite-type chirp combinations only '+-' . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 01410 . 7554/eLife . 12993 . 015Figure 4—source data 1 . Source data for Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 01510 . 7554/eLife . 12993 . 016Figure 4—figure supplement 1 . TS neurons respond more similarly and with excitation to all chirp waveforms . ( A ) Population-averaged response tuning curve for TS neurons with invariance score greater than 0 . 25 ( N = 13 ) . ( B ) Plot of the distance between responses as a function of the distance between the corresponding stimulus waveforms for an example TS neuron . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 01610 . 7554/eLife . 12993 . 017Figure 4—figure supplement 1—source data 1 . Source data for Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 017 Our results show that TS neurons indeed responded with excitation to most stimulus waveforms ( Figure 4B ) and that , consequently , responses were more similar than those of either single afferents or ELL pyramidal neurons ( compare Figure 4C with Figures 3C , D , and 2C ) . The response similarity was reflected in the population-averaged tuning curves ( Figure 4D ) as well as in the phase invariance score ( mean: 0 . 26 ± 0 . 02; max: 0 . 49; min: 0 . 16; population: 0 . 4 ) that was significantly higher than that of ELL neurons ( ANOVA with Bonferroni correction , p<0 . 01 ) when considering all waveforms ( Figure 4E ) . This is because the distance between responses is considerably lower than the distance between the corresponding stimulus waveforms ( Figure 4—figure supplement 1 ) . Phase invariance score values obtained for the TS neuron population ( 0 . 4 ) furthermore approached those observed for correlated peripheral afferent activity ( 0 . 53 ) . We note that the population-averaged response of TS neurons decreased for '- chirps' ( Figure 4D ) : this is because some TS neurons responded similarly to ELL ON-type pyramidal cells and were thus inhibited by '- chirps' . However , TS neurons with larger phase invariance scores ( N=13 ) were excited by all chirp waveforms as quantified by positive responses for all chirp waveforms ( Figure 4—figure supplement 1 ) . Further , computing phase invariance scores over either same type ( '++/- -' ) or opposite type ( '+-' ) chirp combinations led to values that were not significantly different from one another ( ++/- -: 0 . 24 ± 0 . 02; +-: 0 . 28 ± 0 . 02 ) ( ANOVA; F = 1 . 3 , p=0 . 28 ) and that were furthermore similar to those obtained using all waveforms ( 0 . 26 ± 0 . 02 ) . This confirms our hypothesis that TS neurons are , on average , more phase invariant than ELL pyramidal cells because a significant fraction ( ∼50% ) responds with excitation to both '+ chirps' and '- chirps' . The simplest explanation for this observation is that these neurons receive excitatory input from ELL ON and OFF-type pyramidal cells . This point is discussed further below . Does the organism actually make use of the increased phase invariance observed when moving from peripheral to more central brain areas ? To answer this important question , we recorded the animal’s behavioral responses to patterns of stimulation resulting from the same chirp occurring at different phases within the beat cycle ( Figure 5A ) using a previously established behavioral paradigm ( Zupanc et al . , 2006; Hupé et al . , 2008 ) ( see thods ) ( Figure 5B ) . Specifically , we used the echo response in which an animal will reliably respond to a communication signal from a conspecific with a communication signal of its own ( Zupanc et al . , 2006 ) . These occur during aggressive interactions between two conspecifics under natural conditions ( Hupé et al . , 2008 ) . If our hypothesis is true , then we expect that all the presented waveforms will elicit similar behavioral responses . Confirming our hypothesis , the elicited behavioral responses were very similar ( Figure 5C , D ) . Consequently , the response tuning was effectively independent of beat phase ( Figure 5E , top ) . Importantly , other measures such as response latency and rate of occurrence also did not depend on beat phase ( Figure 5E , bottom ) . Invariance values computed from behavioral responses were near unity ( Figure 5F ) . This is because the distance between responses is considerably lower than the distance between the corresponding stimulus waveforms ( Figure 5—figure supplement 1 ) . We conclude that the organism perceives the different patterns of stimulation input resulting from the same chirp occurring at different phases within the beat cycle in a similar fashion and thus displays perceptual phase invarianceFigure 6 . 10 . 7554/eLife . 12993 . 018Figure 5 . Behavioral responses indicate that perception is invariant to the different patterns of stimulation resulting from the same chirp occurring at different phases within the beat cycle . ( A ) Schematic showing the different stages of processing in the electrosensory system . We recorded behavioral responses from N = 29 fish . ( B ) Experimental setup . The fish was placed in an enclosure within a tank ( chirp chamber ) . Stimuli were applied via two electrodes ( S1 & S2 ) perpendicular to the fish’s rostro-caudal axis . The fish’s EOD frequency was recorded by a pair of electrodes positioned at the head and tail of the animal ( E1 & E2 ) . Behavioral responses consisted of communication stimuli characterized by transient increases in EOD frequency in response to the presented stimulus . ( C ) Raster plots showing behavioral responses ( brown ) to repeated presentations of different stimulus waveforms ( black ) associated with the same electrosensory object . ( D ) Population-averaged time dependent behavioral response rates in response to '+ chirps' ( dark brown ) and '- chirps' ( light brown ) . ( E ) Population-averaged tuning curve ( top ) , as well time-averaged behavioral response rates ( bottom , filled circles ) and latency to first chirp ( bottom , open circles ) as a function of beat phase . ( F ) Population-averaged invariance score computed from behavioral responses . EOD , electric organ discharge . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 01810 . 7554/eLife . 12993 . 019Figure 5—source data 1 . Source data for Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 01910 . 7554/eLife . 12993 . 020Figure 5—figure supplement 1 . Behavioral responses are phase invariant . Plot of the distance between responses as a function of the distance between the corresponding stimulus waveforms for behavioral echo chirp responses . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 02010 . 7554/eLife . 12993 . 021Figure 5—figure supplement 1—source data 1 . Source data for Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 021 Taken together , our results provide the first evidence as well as an explanation as to how a phase invariant neural representation of natural communication stimuli first emerges in the hindbrain and is then refined in the midbrain in order to give rise to perception and behavior ( Figure 6A ) . The tuning curves became progressively more independent of beat phase when comparing peripheral afferents , ON and OFF-type ELL pyramidal neurons , TS neurons , and behavior ( Figure 6B ) , thereby leading to increased levels of phase invariance ( Figure 6C ) . The implications of these results for electrosensory processing and for other systems are discussed below . 10 . 7554/eLife . 12993 . 022Figure 6 . An invariant neural representation of natural communication stimuli emerges and is refined along ascending electrosensory brain areas , thereby giving rise to perception . ( A ) Schematic showing how invariance increases across successive stages of electrosensory processing , thereby leading to behavior . ( B ) Normalized response tuning curves obtained for each stage of sensory processing as well as behavior . Note that tuning curves become progressively more independent of beat phase . ( C ) Invariance progressively increases across successive stages of electrosensory processing as well as behavior downstream brain areas and behavior . '*' indicates statistical significance at the p = 0 . 01 level using a one-way ANOVA with Bonferroni correction . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 02210 . 7554/eLife . 12993 . 023Figure 6—source data 1 . Source data for Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 12993 . 023 We investigated the mechanisms by which neural representations of natural communication stimuli become progressively more phase invariant in the electrosensory system of weakly electric fish . We found that the responses of single peripheral afferents displayed minimal phase invariance , as these neurons faithfully encoded all the different waveforms resulting from the same chirp stimulus occurring at different phases within the beat cycle . In contrast , simultaneous recordings from afferent pairs revealed that these waveforms all gave rise to similar increases in correlated activity . ON and OFF-type ELL neurons displayed significantly higher phase invariance than single peripheral afferents . Indeed , both the ON and OFF-type subpopulations showed the greatest degree of phase invariance to waveforms giving rise to either excitation or inhibition in afferents . Responses from both subpopulations are most likely combined in the midbrain TS , thereby giving rise to a more phase invariant representation that , at the population level , approached that seen for correlated afferent activity . Finally , we showed that weakly electric fish displayed similar behavioral responses to stimulus waveforms generated by applying an identity-preserving transformation , which strongly suggests that the globally invariant representation observed in midbrain constitutes a neural correlate of perception . Our results thus reveal generic mechanisms that mediate the emergence and refinement of a neural representation that is invariant to the different patterns of stimulation input resulting from the same chirp occurring at different phases within the beat cycle . Previous studies have reported that the waveforms associated with natural communication stimuli were highly heterogeneous with respect to their time course ( Zupanc and Maler , 1993 ) and that single peripheral afferents ( Benda et al . , 2005 , 2006 ) as well as ELL pyramidal neurons ( Marsat et al . , 2009; Marsat and Maler , 2010; Vonderschen and Chacron , 2011 ) display differential responses to these . While several studies have characterized the distributions of various natural communication stimulus attributes such as frequency increase , duration , or beat phase , for different beat frequencies ( Bastian et al . , 2001; Zakon et al . , 2002; Kolodziejski et al . , 2005; Aumentado-Armstrong et al . , 2015 ) , the tuning curves of electrosensory neurons to these have not been systematically investigated to date . Our results show that the different waveforms resulting from a chirp with given frequency increase and duration occurring at different phases within the beat cycle tend to give rise to more similar neural responses centrally . These most likely constitute a neural correlate of perception . Our electrophysiological and behavioral results thus support our initial hypothesis that the identity of a given chirp is determined by frequency increase and duration but not the phase of the beat at which it occurs at . Our results further show that TS neurons displayed significantly less phase invariance than behavioral responses , indicating that further refinement in downstream brain areas is likely necessary . Such refinement is likely to occur in the nucleus electrosensorius , a diencephalic brain structure that receives direct input from TS ( Carr et al . , 1981 ) , and likely involves integrating the activities of multiple TS neurons . The fact that the invariance score of the TS neuron population was on average higher than that of single neurons supports the latter statement . Importantly , we note here that the TS in weakly electric fish consists of 11 layers with roughly 50 cell types ( Carr and Maler , 1985 ) . Previous electrophysiological studies have revealed that TS neurons are highly heterogeneous and can be roughly classified in two groups: 'dense' neurons that respond to stimuli in a manner that is strongly reminiscent of that displayed by ELL pyramidal neurons; and 'sparse' neurons that respond selectively to a given class of natural electrosensory stimuli ( Vonderschen and Chacron , 2011; Sproule et al . , 2015 ) . Within the sparse group , neurons can either respond selectively to movement ( Chacron et al . , 2009; Vonderschen and Chacron , 2011 ) , increases in beat amplitude ( i . e . envelopes ) ( McGillivray et al . , 2012 ) , or natural communication stimuli ( Vonderschen and Chacron , 2011 ) , indicating that parallel processing of behaviorally relevant stimulus features occurs in this area . Interestingly , some TS neurons displayed phase invariance scores that were higher than those observed in single receptors or in ELL neurons . It is likely that these are members of the sparse group as they responded selectively to communication stimuli rather than the underlying beat . A previous study has shown that neurons with similar response profiles receive balanced excitatory input from ON and OFF-type pyramidal cells ( Aumentado-Armstrong et al . , 2015 ) and , as such , most likely correspond to previously described 'ON-OFF' neurons that respond to both increases and decreases in the stimulus ( Partridge et al . , 1981; Rose and Call , 1993 ) . We , however , note that these neurons correspond to a few if not a single cell type within TS ( Carr et al . , 1981; Carr and Maler , 1985 ) , as partially reflected by the broad distribution of phase invariance scores observed in the current study in TS neurons . In fact , confirming previous results ( Vonderschen and Chacron , 2011 ) , a significant fraction of TS neurons displayed low-phase invariance scores comparable to those seen in ELL , indicating that they responded differentially ( i . e . non-invariantly ) to the different waveforms resulting from a given chirp occurring at different phases within the cycle of a beat . Interestingly , it was shown recently that both dense and sparse TS neurons project to higher brain areas ( Sproule et al . , 2015 ) , indicating that these actually receive parallel sources of information including those transmitted by both phase invariant and non-phase invariant TS neurons . Phase invariant TS neurons would then give information that a chirp has occurred while non-phase invariant TS neurons would instead transmit contextual information about this chirp . While our behavioral results do not demonstrate whether the animal can actually perceive information transmitted by non-phase invariant TS neurons , it is likely that this information is behaviorally relevant . However , further studies are needed to understand how TS neurons are tuned to other attributes of natural communication stimuli ( i . e . frequency increase , duration , beat phase , beat frequency ) . This is important , as these have all been shown to carry behaviorally relevant information ( Bastian et al . , 2001; Zakon et al . , 2002; Kolodziejski et al . , 2005; Hupé et al . , 2008 ) . Thus , we predict that the organism will perceive communication signals whose attributes other than beat phase differ . We further hypothesize that , in more central brain areas , neural representations become more selective to chirp identity while at the same time becoming more tolerant to changes in waveform resulting from a given chirp occurring at different phases during the beat cycle . Specifically , we predict that peripheral afferents will faithfully encode the different waveforms associated with changes in a given feature irrespective of behavioral relevance . However , when moving more centrally from receptors to midbrain , neural responses should become more selective when varying chirp identity . If true , it will then be very interesting to understand the mechanisms by which neurons become selective to given stimulus attributes yet tolerant to others . While a previous study has shown that some TS neurons could respond differentially to communication signals with different frequency increases ( Vonderschen and Chacron , 2011 ) , the tuning properties of TS neurons to the behaviorally relevant features of natural communication stimuli have not been systematically investigated to date and should be the focus of future studies . Finally , we note that we focused on small chirps or type II communication stimuli that tend to occur on top of low frequency beats ( Bastian et al . , 2001 ) because the associated waveforms displayed the largest heterogeneities ( Aumentado-Armstrong et al . , 2015 ) . Recent studies have shown that type II communication stimuli can also occur on top of high frequency beats ( Walz , 2013 ) and further studies are needed to understand whether the encoding strategies for these would differ from those described above . Also , big chirps or type I communication stimuli , as they mostly occur in high-frequency beat contexts , tend to display much less heterogeneity ( Bastian et al . , 2001; Aumentado-Armstrong et al . , 2015 ) . Previous studies have shown that the encoding strategies for type I and II communication stimuli strongly differ at the level of peripheral afferents ( Benda et al . , 2005 , 2006 ) , ELL pyramidal cells ( Marsat et al . , 2009; Marsat and Maler , 2010; Vonderschen and Chacron , 2011 ) , and it is likely that such differences are preserved in the midbrain ( Vonderschen and Chacron , 2011 ) . Further studies are needed to uncover how heterogeneities in type I communication stimuli are encoded in the electrosensory system . Also , we note that the large heterogeneities seen in stimulus waveforms associated with type II communication stimuli occurring at different beat phases might not be found in other weakly electric fish species . Indeed , the duration of type II communication stimuli in Apteronotus albifrons tend to be much larger than those of Apteronotus leptorhynchus considered in this study ( Kolodziejski et al . , 2005 ) and , as such , might not display the large degree of heterogeneity seen here . We have shown that the electrosensory system of weakly electric fish actually exploits the feature invariant representation that is given by correlated neural activity through ON and OFF-type ELL pyramidal neurons . We note that ON and OFF-type cells have been observed across species and sensory modalities ( Wassle et al . , 1981a; Wassle et al . , 1981b; Wassle , 2004; Chalasani et al . , 2007; Scholl et al . , 2010; Gallio et al . , 2011; Frank et al . , 2015 ) and , further , that downstream neurons receive excitatory inputs from both cell types ( Fairhall et al . , 2006; Gollisch and Meister , 2008 ) . Thus , the discovered computations for generating and refining phase invariance uncovered in the electrosensory system that involve decoding correlated activity through ON and OFF-type neurons and their subsequent combination are also likely to be applicable . This is particularly the case in the primary visual cortex . Indeed , while simple cells respond selectively ( Movshon , 1975 ) , complex cells instead display responses that are invariant to the spatial phase of a drifting grating ( Movshon et al . , 1978 ) . It is thought that this invariance arises , at least in part , through integration of input from multiple ON and OFF-type simple cells ( Hubel and Wiesel , 1962; Martinez and Alonso , 2003 ) . It is thus conceivable that correlated activity from simple cells could convey a spatial phase invariant representation that is then exploited by complex cells . Further , correlations between the activities of neighboring neurons are observed ubiquitously in the brain ( Averbeck et al . , 2006 ) and are dynamically regulated by several factors such as attention ( Cohen and Maunsell , 2009 ) , behavioral state ( Vaadia et al . , 1995 ) , and stimulus statistics ( Chacron and Bastian , 2008; Litwin-Kumar et al . , 2012; Ponce-Alvarez et al . , 2013; Simmonds and Chacron , 2015 ) . Previous studies have found that correlated but not single neuron activity can carry information about stimulus attributes ( deCharms and Merzenich , 1996; Ishikane et al . , 2005; Metzen et al . , 2015a; Metzen et al . , 2015b ) . Here , we have instead shown that correlated neural activity can mediate the emergence of a neural representation of behaviorally relevant stimuli that is phase invariant . We also note that , since previous studies have shown that noise correlations were negligible in electroreceptor afferents ( Chacron et al . , 2005b; Metzen et al . , 2015b ) , our results focus on signal correlations that give rise to feature invariant representations of natural communication stimuli . Nevertheless , noise correlations are also dynamically regulated by stimulus attributes ( Chacron and Bastian , 2008 ) and thus might provide information about sensory input . Future studies are needed to understand whether noise correlations can also transmit information about behaviorally relevant stimulus features . Finally , we note that the natural communication stimuli considered here can be seen as high frequency transients that must be distinguished from a low frequency background ( i . e . the beat ) . Thus , it is very likely that our results will be applicable to the problem of separating a fast moving object from a slow moving background as in the visual system ( Olveczky et al . , 2003 ) or to separating a high frequency transient sound occurring simultaneously with a low-frequency background in the auditory system ( Berti , 2013 ) . We investigated the neural coding strategies used by the electrosensory system to process natural communication stimuli of heterogeneous waveforms . Our results show that this system exploits the fact that correlated peripheral neuron activity provides an invariant representation through decoding synchronized excitation and inhibition by ON and OFF-type ELL pyramidal neurons , respectively . Midbrain neurons receiving input from both ON and OFF-type ELL pyramidal neurons displayed the most invariant responses and likely provide a neural correlate of the invariant perception seen at the organismal level . Apteronotus leptorhynchus specimens were acquired from tropical fish suppliers and acclimated to laboratory conditions according to published guidelines ( Hitschfeld et al . , 2009 ) . Surgical procedures have been described in detail previously ( Toporikova and Chacron , 2009; Vonderschen and Chacron , 2011; McGillivray et al . , 2012; Deemyad et al . , 2013 ) . Briefly , animals ( N = 16 ) were injected with tubocurarine chloride hydrate ( 0 . 1 – 0 . 5 mg ) before being transferred to an experimental tank and respirated with a constant flow of water over their gills ( ~10 ml/min ) . A small craniotomy ( ~5 mm2 ) was made above the hindbrain ( for afferent and ELL pyramidal neuron recordings ) or above the midbrain ( for TS recordings ) , respectively . We used 3M KCl-filled glass micropipettes ( 30 MΩ resistance ) to record from electroreceptor afferent axons as they enter the ELL ( Savard et al . , 2011; Metzen and Chacron , 2015; Metzen et al . , 2015b ) either simultaneously from pairs ( N = 8 ) or from single units ( N = 18 ) . As similar results were obtained when considering pairs of simultaneously and non-simultaneously recorded afferents , data were pooled when computing correlated activity , these sample sizes were sufficient to access differences between responses to different stimulus waveforms based on previous studies ( Benda et al . , 2005; Chacron et al . , 2005b; Benda et al . , 2006 ) . Importantly , it was previously shown that afferents do not display noise correlations ( Chacron et al . , 2005b; Metzen et al . , 2015b ) . Extracellular recordings from ELL pyramidal ( N = 31 ) and TS neurons ( N = 25 ) were performed with metal-filled micropipettes ( Frank and Becker , 1964; Chacron et al . , 2009; Chacron and Fortune , 2010; Simmonds and Chacron , 2015 ) . These sample sizes are similar to those used in previous studies . Baseline ( i . e . in the absence of stimulation ) firing rates for afferents , pyramidal cells , and TS neurons were 380 ± 72 Hz , 12 . 10 ± 1 . 74 Hz , and 3 . 35 ± 3 . 27 Hz . These are similar to previously reported values ( Chacron et al . , 2005a; McGillivray et al . , 2012; Clarke et al . , 2015a; Metzen et al . , 2015b ) . We only recorded from neurons that responded to at least one chirp stimulus waveform . Recordings were digitized at 10 kHz ( CED Power 1401 & Spike 2 software , Cambridge Electronic Design ) and stored on a computer for subsequent analysis . The neurogenic electric organ of A . leptorhynchus is not affected by injection of curare-like drugs . Stimuli thus consisted of amplitude modulations of the animal’s own EOD and were produced by first generating a sinusoidal waveform train with frequency slightly ( 20–30 Hz ) greater than the EOD frequency that was triggered by the EOD zero crossing . This train is synchronized to the animal’s EOD and will either increase or decrease EOD amplitude based on polarity and intensity . This train is then multiplied ( MT3 multiplier , Tucker Davis Technologies ) with an amplitude modulated waveform ( i . e . the stimulus ) . The resultant signal is then isolated from ground ( A395 linear stimulus isolator , World Precision Instruments ) and delivered to the experimental tank via two chloritized silver wire electrodes located ~15 cm on each side of the animal ( Bastian et al . , 2002 ) . Chirps consisting of Gaussian increases in frequency of 60 Hz and standard deviation of 14 ms , were generated on top of a beat with frequency fbeat = 4 Hz and the resulting AM stimulus waveforms were used as done previously ( Vonderschen and Chacron , 2011 ) . Previous studies have shown that chirps occur with uniform probability over all phases of the beat ( Aumentado-Armstrong et al . , 2015 ) . For experimental purposes , we systematically varied the beat phase at which the chirp occurred between 0 and 315 deg in increments of 45 deg , the corresponding AM waveforms ( S1 to S8 ) are shown in Figure 1E . The same chirp stimuli were also used in the behavioral experiments . All analyses were performed using custom-built routines in Matlab ( The Mathworks , Natick , MA ) , these routines are freely available online ( Metzen et al . , 2016 ) . Action potential times were defined as the times at which the signal crossed a suitably chosen threshold value . From the spike time sequence we created a binary sequence R ( t ) with binwidth ∆t = 0 . 5 ms and set the content of each bin to equal the number of spikes which fell within that bin . PSTHs were obtained by averaging the neural responses across repeated presentations of a given stimulus with binwidth 0 . 1 ms and were smoothed with a 6 ms long boxcar filter . We computed the cross-correlation coefficient between the spiking responses Ri ( t ) and Rj ( t ) of neurons i and j as was done previously ( Shea-Brown et al . , 2008; Metzen et al . , 2015b ) . The time varying correlation coefficient was computed during a time window with duration 31 . 25 ms unless otherwise indicated that was translated in steps of 0 . 25 ms using: ( 1 ) ρ=\|PRiRj ( 0 ) \|PRiRi ( 0 ) PRjRj ( 0 ) Here , PRiRi is the cross-spectrum between Ri ( t ) and Rj ( t ) , PRiRi ( f ) and PRiRi ( f ) are the power spectra of Ri ( t ) and Rj ( t ) , respectively , and \|…\| denotes the absolute value . We note that our definition of the cross-correlation coefficient is equivalent to that used in previous studies ( Shadlen and Newsome , 1998; Chacron and Bastian , 2008 ) and furthermore gives results similar to those obtained when computing the cross-correlation coefficient between spike counts during a time window of 40 ms ( de la Rocha et al . , 2007; Metzen et al . , 2015b ) . For each cell , the response to a given chirp stimulus waveform as: ( 2 ) Response=log ( FRchirpFRbeat ) where FRchirp is the average firing rate over a time window of duration Tchirp centered 15 ms after chirp onset and FRbeat is the average firing rate during the beat . Note that response values can vary between - ∞ and + ∞ . Negative response values imply inhibition and positive values imply excitation by the stimulus waveform , respectively . We used Tchirp = 15 ms for electrosensory afferents and ELL pyramidal cells and Tchirp = 20 ms for TS neurons . To obtain the tuning curve , the responses obtained for each chirp waveform were plotted as a function of beat phase for each cell and then averaged across the respective population . The phase invariance score was defined as: ( 3 ) Invariance = 1 -∑i≠jD ( FRi ( t ) , FRj ( t ) ) ) D ( Si ( t ) , Sj ( t ) ) ( Nchirps ) ( Nchirps-1 ) where Nchirps is the number of chirp stimulus waveforms , and D ( x , y ) is a distance metric between x and y that was computed as ( Aumentado-Armstrong et al . , 2015 ) : ( 4 ) D ( x , y ) = ( x-x-y+y ) 2max[max ( x ) −min ( x ) 2 , max ( y ) −min ( y ) 2] where <…> denotes an average over a time window of 30 ms after chirp onset , FRi ( t ) is the PSTH response of a given cell to chirp stimulus waveform Si ( t ) , and max ( … ) , min ( … ) denote the maximum and minimum values , respectively . All responses were normalized prior to computing the distance metric . We note that , according to equation [3] , the distance between responses to two different stimulus waveforms is actually normalized by the distance between the stimulus waveforms themselves . Thus , a neuron whose response faithfully encodes the detailed timecourse of the different chirp waveforms will not be considered invariant according to our definition . Invariance scores were computed for each individual cell and subsequently averaged across the respective populations . Some analyses were performed on TS neurons ( N = 13 ) whose phase invariance score was greater than 0 . 25 , which was chosen as the mean plus one standard deviation from our ELL data . We computed phase invariance for correlated activity as described above except that we used the timecourse of the varying correlation coefficient as an input . Alternatively , to test that the low values obtained for receptor afferents were not due to variability , we also computed phase invariance scores for these by first averaging responses across the population . To measure the chirp stimulus waveforms shown in Figure 1—figure supplement 1 , fish were restrained by placing them in a 'chirp chamber' ( Deemyad et al . , 2013 , Metzen and Chacron , 2014 ) . Chirps were elicited by sinusoidal waveforms mimicking another fish’s EOD whose frequency was set 10 Hz above the animal’ s own EOD frequency . The fish’s EOD was recorded and the instantaneous EOD frequency was computed from the inverse of the timing difference between successive zero crossings . Chirps were then identified as increases in the animal’s own EOD frequency that exceeded 30 Hz ( Bastian et al . , 2001 ) . To measure the chirp echo response , each fish ( N = 29 ) was restrained in a 'chirp chamber' as described previously ( Deemyad et al . , 2013 , Metzen and Chacron , 2014 ) . The EOD was measured between electrodes placed near the head and tail , amplified ( model 1700 amplifier , A-M systems ) , digitized at 10 kHz sampling rate using CED 1401plus hardware and Spike2 software ( Cambridge Electronic Design ) ( Figure 5B , E1 and E2 ) , and stored on a computer hard disc for offline analysis . To analyze the chirp echo response , we first extracted the time varying EOD frequency of each fish tested . The fish’s EOD was recorded and the instantaneous EOD frequency was computed from the inverse of the timing difference between successive zero crossings . To measure echo responses to small chirp stimuli delivered at different beat phases , we initially habituated the animal to a 4 Hz beat stimulus lasting 60 s in order to minimize the probability of chirp responses being elicited by the beat alone: the chirp rate during a 200 ms window preceding stimulus chirp onset was null ( 0 ± 0 Hz ) . We then randomly interspersed chirp stimuli at variable intervals ( 15 ms ± 3 ms ) and the recording was started 200 ms before chirp onset . In total , for each fish , we played 40 chirp stimuli such that each chirp phase was presented five times . Echo response chirps occurring during a 1 s window after stimulus chirp onset were identified as increases in the animal’s own EOD frequency that exceeded 30 Hz and were segregated into small ( type II ) and big ( type I ) chirps as done previously ( Bastian et al . , 2001 ) . Based on this criterion , almost all echo response chirps were classified as type II ( 98 . 7% ) while the rest were classified as type I ( 1 . 3% ) . All echo response chirps were included in the analysis . The time of occurrence of echo response chirps was defined as the time at which the EOD frequency excursion was maximal . The echo response chirp rate CRchirp was computed as the number of echo response chirps during a time window of 1 s following the stimulus chirp onset since previous studies have shown that the majority of responses occur during this time window ( Zupanc et al . , 2006 ) . The echo response latency was determined as the timing of the first echo response chirp after chirp stimulus onset . Since the probability of eliciting an echo response chirp was low in general , we averaged the responses to a particular chirp phase obtained from all fish tested to build peri-stimulus time histograms ( PSTHs ) . PSTHs for all single chirp phases were smoothed using a 1 s long boxcar filter . Behavioral responses were quantified as log ( CRchirp/CRbeat ) , where CRbeat is the average echo response chirp rate during the beat between 1 and 2 s following stimulus chirp onset , where relatively fewer echo responses tend to occur ( Zupanc et al . , 2006 ) . Invariance scores for behavior were computed as described above for neural responses except that we used the behavioral PSTHs as responses . Error bars were estimated using a sequential bootstrapping method ( Efron , 1979 ) with block size 27 . Statistical significance was assessed through one-way analysis of variance ( ANOVA ) with the Bonferroni method of correcting for multiple comparisons at the p=0 . 05 level . Values are reported as mean ± SEM . All data and codes are freely available online ( Metzen et al . , 2016 ) .	We can effortlessly recognize an object – a car , for example – in many different contexts such as when seen from behind , under different lighting levels or even from different viewpoints . This phenomenon is known as perceptual invariance: objects are correctly recognized , despite variations in exactly what is seen ( or otherwise sensed ) . However , it is still not clear how the brain processes perceptual information to recognize the same object under a wide variety of contexts . Some fish , such as the brown ghost knifefish , produce a weak electric signal that they can alter to communicate with other members of their species . A communication call may be produced in a variety of contexts that alter which aspects of the signal nearby fish detect . Despite this , fish tend to respond to a given communication call in the same way regardless of its context; this suggests that these fish also have perceptual invariance . The communication calls of weakly electric fish can be easily mimicked in a laboratory and produce reliable behavioral responses , which makes these fish a good model for understanding how perceptual invariance might be coded in the brain . Therefore , Metzen et al . recorded the activity of the receptor neurons that first respond to communication calls in weakly electric fish . The results revealed that a given communication signal made the firing patterns of all receptor neurons in the fish’s brain more similar to each other , regardless of the signal’s context . This occurs despite the changes in context causing single receptor neurons to respond in different ways . At each stage of the process by which information is transmitted from the receptor neurons to neurons deeper in the brain , the similarity in the neurons’ firing patterns is refined , thereby giving rise to perceptual invariance . While perceptual invariance to a given object in different contexts is desirable , it is also important to be able to distinguish between different objects . This implies that neurons should respond similarly to stimuli associated with the same object and differently to stimuli associated with different objects . Further studies are now needed to confirm whether this is the case .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2016	Neural correlations enable invariant coding and perception of natural stimuli in weakly electric fish
Quantifying pathogen transmission in multi-host systems is difficult , as exemplified in bovine tuberculosis ( bTB ) systems , but is crucial for control . The agent of bTB , Mycobacterium bovis , persists in cattle populations worldwide , often where potential wildlife reservoirs exist . However , the relative contribution of different host species to bTB persistence is generally unknown . In Britain , the role of badgers in infection persistence in cattle is highly contentious , despite decades of research and control efforts . We applied Bayesian phylogenetic and machine-learning approaches to bacterial genome data to quantify the roles of badgers and cattle in M . bovis infection dynamics in the presence of data biases . Our results suggest that transmission occurs more frequently from badgers to cattle than vice versa ( 10 . 4x in the most likely model ) and that within-species transmission occurs at higher rates than between-species transmission for both . If representative , our results suggest that control operations should target both cattle and badgers . Control of a pathogen in a system where it can infect multiple species requires an understanding of the role of each host species in the infection dynamics ( Haydon et al . , 2002 ) . For example , when each host species is capable of maintaining infection independently , control operations in one species can be rendered ineffective as a result of spillover from another . Mycobacterium bovis infection in cattle populations ( resulting in bovine tuberculosis - bTB ) is a problem around the world ( Ayele et al . , 2004; Cousins and Roberts , 2001; de Kantor and Ritacco , 2006; Godfray et al . , 2013; Reviriego Gordejo and Vermeersch , 2006; Schmitt et al . , 2002 ) , with many wildlife species implicated in its spread and persistence in different bTB systems ( Delahay et al . , 2002; Gortazar et al . , 2003; Miller and Sweeney , 2013; Nugent , 2005; Nugent et al . , 2015 ) . On the islands of Britain and Ireland , the current evidence suggests that effective control of infection in cattle is hindered by transmission from an infected wildlife population – the European badger ( Meles meles ) ( Godfray et al . , 2013 ) . Although a considerable amount of research demonstrates an association between M . bovis found in sympatric cattle and badger populations ( Balseiro et al . , 2013; Goodchild et al . , 2012; Olea-Popelka et al . , 2005; Vial et al . , 2011; Woodroffe et al . , 2005 ) , quantification of the direction and extent of transmission remains elusive . Recent studies using whole genome sequences ( WGS ) have demonstrated a close genetic relationship among M . bovis isolates taken from sympatric cattle and wildlife populations ( Biek et al . , 2012; Glaser et al . , 2016; Patané et al . , 2017 ) . However , the low genomic variability of M . bovis and imbalanced sampling across host species has limited the ability to identify the direction of transmission . Evidence to date suggests that , even with access to pathogen sequence data , obtaining directional estimates of transmission might only be possible at the population level and will require dense targeted sampling and fine-grained epidemiological metadata ( Kao et al . , 2016; Kao et al . , 2014 ) , as has previously been demonstrated in investigations of M . tuberculosis outbreaks in humans ( Bryant et al . , 2013; Gardy et al . , 2011; Guthrie et al . , 2018; Walker et al . , 2012; Walker et al . , 2018; Yang et al . , 2017 ) and in tracing between cattle herds for outbreaks of M . bovis ( Biek et al . , 2012; Salvador et al . , 2019 ) . However , these approaches have yet to be applied to situations where dense multi-host pathogen data are available . Since the 1970s , a high-density naturally infected badger population at Woodchester Park in southwest England has been the subject of detailed study ( Delahay et al . , 2013 ) . Both the resident badgers and sympatric cattle herds are frequently infected with M . bovis , providing the potential for inter-species transmission of infection to occur in either direction ( DEFRA , 2017; Delahay et al . , 2013 ) . The data and samples associated with bTB occurrence in and around Woodchester Park are uniquely detailed , with individual-level host life history data and archived M . bovis isolates available for both the cattle ( Orton et al . , 2018 ) and badger ( Delahay et al . , 2013 ) populations . By combining WGS of selected cattle and badger isolates , with detailed local population data from this exceptionally in-depth study system , our work aimed to quantify the relative roles of the local badger and cattle populations in the spread and persistence of M . bovis in an endemic area . Based on previous evidence of transmission between cattle and badgers , and the success of combining detailed tracing methods with WGS for M . tuberculosis , our hypothesis is that M . bovis circulation in our endemic setting is not limited to a single maintenance host and that it involves bi-directional transmission between the two host populations . Our research aimed to test this hypothesis and to quantify transmission patterns by analysing the Woodchester Park data using a series of statistical and observational analyses linking pathogen genome data with diagnostic testing and population movement and demographic data for both cattle and badgers . Archived M . bovis isolates were available from 116 badgers and 189 cattle living in and around Woodchester Park . Multiple isolates were available from the sampled badgers , resulting in a total of 230 isolates sourced from badgers . These isolates were whole genome sequenced , and , after quality assessments , 193 badger-derived ( from 98 individual badgers taken from 2000 to 2011 ) and 159 cattle-derived sequences ( from 1988 to 2013 ) were retained for further analyses . To investigate the presence of spatial , temporal , and network signatures associated with infection dynamics in the M . bovis genomic data , inter-sequence genetic distances were calculated between all the cattle- and badger-derived sequences and compared to population metrics . The metrics described the spatial- , temporal- , and network-based relationships that were expected to be associated with pathogen transmission . The genetic and epidemiological data were compared using Random Forest ( Liaw and Wiener , 2002 ) and Boosted Regression ( Elith et al . , 2008 ) models in R ( v3 . 4 . 3; R Development Core Team , 2016 ) to separately analyse badger–badger ( n = 12483 ) , cattle–cattle ( n = 1927 ) , and badger–cattle ( n = 4838 ) comparisons . The Random Forest ( and Boosted Regression ) models were able to explain approximately 67% ( 62% ) , 60% ( 54% ) and 75% ( 70% ) of the variation observed in the inter-sequence genetic distance distributions associated with the badger–badger , cattle–cattle , and badger–cattle comparisons , respectively . For each of these models , metrics based on spatial and temporal distances were the most informative in explaining the variation in the genetic distances . Generally , as the temporal and spatial distances associated with the sampled animals decreased , the number of differences between the M . bovis genomes decreased ( Appendix 1—figures 5 , 6 and 7 ) . There was substantial agreement in the variable rankings between the Random Forest and Boosted Regression models ( Appendix 1—figures 2 , 3 and 4 ) . For the within-species comparisons metrics , the network data were also highly informative . Generally , the number of differences between the genomes associated with a pair of animals of the same species decreased as the connectedness of their social groups ( badgers ) or herds ( cattle ) increased . The variation explained by the Random Forest models and the high ranking of spatial- , temporal- , and network-based metrics was robust to the presence of highly correlated or non-informative metrics and those with missing data ( data not shown ) . The relatedness of M . bovis genomes sampled from the cattle and badgers was evaluated by constructing a phylogenetic tree ( Figure 1 ) using RAxML ( v8 . 2 . 11; Stamatakis , 2014 ) . Genetic diversity was observed between the cattle- and badger-derived M . bovis sequences , with the number of Single Nucleotide Variants ( SNVs ) between sequences ranging from 0 to 150 ( median = 20 ) . Five clades including cattle- and badger-derived sequences were identified ( Figure 1 and Figure 1—figure supplement 1 ) , using a 10 SNV threshold ( informed by thresholds used for M . tuberculosis [Bryant et al . , 2013; Jajou et al . , 2018; Roetzer et al . , 2013; Yang et al . , 2017] ) . Four of the five clades ( 1–4 ) identified contained highly similar ( within three SNVs ) badger- and cattle-derived M . bovis sequences . The badger-derived M . bovis sequence in clade 5 was six SNVs away from its closest cattle-derived sequence . The similarities between the cattle-derived and badger-derived M . bovis sequences in clades 1–4 indicate recent shared transmission histories ( Meehan et al . , 2018 ) . Clade 4 ( highlighted in purple in Figure 1 ) contained the majority ( 156/193 ) of the badger-derived M . bovis sequences and represents the main lineage circulating within the Woodchester Park badger population . In addition , the presence of 16 cattle-derived sequences in clade 4 , 15 of which were distant ( up to 12 SNVs ) from the clade root is consistent with multiple badger-to-cattle transmission events . In contrast , the presence of cattle-derived sequences close to the roots of clades 1–5 suggests that these lineages might have originated in cattle , although these patterns could also be explained by the cattle population being sampled up to 12 years prior to the badger population ( cattle were sampled from 1988 to 2013 and badgers from 2000 to 2011 ) . Although clades 1 and 5 contained highly similar sequences originating from cattle and badgers , each clade was associated with only eight animals , making meaningful inference of inter-species transmission patterns difficult . In addition to inter-species clades , several cattle-only clades were identified ( Figure 1 ) . Consistent with our hypothesis , the close proximity of M . bovis genomes sourced from cattle and badgers suggests that inter-species transmission occurred in the sampled system . In addition , the presence of clades dominated by a single species suggests that sustained within-species transmission has been occurring in both the cattle and badger populations . The life histories of the sampled cattle and badgers and in-contact animals associated with the inter-species clades ( clades 1–5 ) identified in Figure 1 were interrogated . In this manuscript , a badger or cow is considered ‘sampled’ , if one of the M . bovis genomes analysed here was sourced from it . In-contact animals were defined as those that lived in the same herd ( for cattle ) or social group ( for badgers ) at the same time as one or more of the sampled animals , according to the available data . From the interrogations of the life history data , further evidence indicative of inter-species transmission and disease maintenance in the Woodchester Park badger population was identified for the animals associated with clade 4 ( Figure 2; equivalent figures for the remaining clades can be found in Figure 2—figure supplements 1 , 2 , 3 , and 4 ) . Infection was detected in the majority of the sampled badgers before it was detected in the majority of the sampled cattle . Sampled badgers were present in Woodchester Park at least from 1993 until 2011 , based on the available capture and sampling data ( Figure 2c ) . The sampled badgers were in contact with 575 captured badgers , 291 ( 51% ) of which had tested positive for M . bovis infection at some point in their lives ( Figure 2a ) . In contrast , the sampled cattle were in contact with 1760 cattle , of which only 312 ( 18% ) tested positive for M . bovis ( Figure 2b ) . In the animals associated with clade 4 , infection was detected earlier in badgers , except in the case of one cow , despite the cattle population being sampled over a broader temporal and spatial window ( see Materials and methods section: ‘Selecting the isolates’ for more details ) . In addition , the badgers were the most represented species in clade 4 . These two observations suggest that the clade 4 lineage was being maintained in the badger population . The single cattle-derived sequence that was found closest to the root node of clade 4 ( Figure 2c ) was sourced from an animal sampled six years prior to any sequences derived from badgers being available . Across all inter-species clades investigated , the sampled cattle ( n = 71 ) were in contact with approximately 11 , 732 animals , 1356 of which tested positive for M . bovis infection , whereas the sampled badgers ( n = 97 ) were in contact with approximately 650 badgers , over half of which ( 329 ) tested positive . Although the patterns observed in the phylogenetic and animal life history data were consistent with inter-species transmission in both directions , further analyses were required to quantify the inter-species transmission rates . These further analyses should account for the temporal and spatial sampling biases resulting from the broader sampling window applied to the cattle population in time ( 1988 to 2013 versus 2000 to 2011 ) and space ( cattle were sampled from up to 100 km away from the Woodchester Park area , whereas the badgers were only sampled from within Woodchester Park ) . A series of analyses were conducted using the Bayesian Structured coalescent Approximation , or BASTA , package ( De Maio et al . , 2018 ) available as part of Bayesian evolutionary analyses platform BEAST2 ( Bayesian Evolutionary Analysis by Sampling Trees; Bouckaert et al . , 2014 ) . These analyses aimed to estimate the M . bovis inter-species transmission rates between the sampled badger and cattle populations . BASTA is capable of estimating evolutionary dynamics in a structured population and accounting for sampling biases . Here the sampled M . bovis population was structured as it was circulating largely separately in the sampled cattle and badger populations , as seen in Figure 1 and the strong population-specific epidemiological signatures found by the Random Forest and Boosted Regression analyses . In addition , further structure exists within the cattle and badger populations as these were subdivided into herds and social groups , respectively . A series of increasingly spatially structured population models were defined to determine whether the inter-species transmission rates estimated using BASTA were affected by the spatial patterns evident from the Random Forest and Boosted Regression analyses . Structured population models were also chosen to address the spatial sampling biases , by introducing an increasingly structured unsampled badger population . Previous analyses have used BASTA in a similar fashion to estimate evolutionary dynamics in the presence of unsampled populations ( De Maio et al . , 2015 ) . To further reduce the influence of the spatial and temporal biases and the computational load , the BASTA analyses used a subset of the cattle- ( n = 83 ) and badger-derived ( n = 97 ) M . bovis sequences obtained between 1999 and 2014 within 10 km of Woodchester Park . The AICM ( Akaike’s Information Criterion Markov Chain Monte Carlo ) score ( Baele et al . , 2013 ) was used to compare the BASTA analyses based on different structured populations ( Figure 3a ) . The structured population with two demes ( M . bovis populations in badgers and cattle ) had the best ( lowest ) AICM score , although there was considerable overlap with the bootstrapped AICM score interval for one of the four deme models ( splitting the M . bovis populations in badgers and cattle into inner and outer populations based on being within or beyond 3 . 5 km from Woodchester Park [Figure 3a] ) . The estimated inter-species transition rates provided from each BASTA analysis demonstrated considerable variation , with some estimated cattle-to-badger transition rates bounding zero ( Figure 3b ) . The estimated transition rates can be considered equivalent to the transmission rates , because the states ( between which the transition rates were estimated ) considered here represented different species . The estimates of the inter-species transition rates from the two-deme model with the best AICM score support the existence of both badger-to-cattle transmission ( 0 . 045 times per lineage per year , lower 2 . 5%: 0 . 028 , upper 97 . 5%: 0 . 069 ) and cattle-to-badger transmission ( 0 . 0044 times per lineage per year , lower 2 . 5%: 0 . 00021 , upper 97 . 5%: 0 . 017 ) . Figure 3b shows the order of magnitude differences between the estimated inter-species transmission rates , with the highest supported two-deme model estimating that badger-to-cattle transmission events occurred on average 10 . 4 times more frequently than cattle-to-badger transmission events in the sample population . Figure 3c represents the lower bound on the number of times ( according to the analyses based on the favoured two-deme model ) that the sampled M . bovis population was transmitted from one animal to another ( regardless of sub-population and , where possible , assuming the ancestral node and one of its daughter nodes represent infection in the same animal [Figure 3—figure supplement 1] ) . The estimated counts of these transmission events are consistent with the estimated inter-species transition rates and demonstrate that within-species transmission occurs at a higher rate . Specifically , badger-to-badger transmission was estimated to occur at least 2 . 7 times more frequently than badger-to-cattle transmission ( lower 2 . 5%: 2 . 2 , upper 97 . 5%: 3 . 8 ) . In cattle , analyses estimated that at least 46 cattle-to-cattle transmission events occurred ( lower 2 . 5%: 40 , upper 97 . 5%: 56 ) , whereas the estimated number of cattle-to-badger events bounded zero ( lower 2 . 5%: 0 , upper 97 . 5%: 4 , with a median value of zero ) . The counts of events between individual animals outputted by BASTA represent the lower bound of the number of transmission events that occurred over the evolutionary history of the sampled M . bovis population because they are estimated on the transmission chains between the sampled and ancestral host animals and do not account for missing individuals in these chains . Taken together , the results from the BASTA analyses are consistent with the hypothesis that circulation of M . bovis in our study populations involved transmission within and between the badgers and cattle . In addition , the directional inter-species transmission rates indicate that transmission from badgers to cattle occurred more frequently than transmission from cattle to badgers and inter-species transmission rates were estimated to be considerably lower than intra-species transmission rates . We hypothesised that the sampled M . bovis population was circulating within and between the sampled cattle and badger populations . Testing our hypothesis across multiple analyses , we found that , while none of these analyses are definitive in their own right , our results are consistent with our hypothesis and suggest that there has been a long history of within- and between-species transmission in the Woodchester Park area , and an important role for badgers in disease persistence . Our choice of analytical methods was based in part on our awareness of underlying data biases . Ideally , sampling should be proportionate to prevalence in the host populations and matched over the same spatial and temporal ranges . Here , the combination of poor sensitivities of the standard tests for cattle ( ~50–80%; de la Rua-Domenech et al . , 2006 ) and badgers ( ~50–70%; Chambers et al . , 2009 ) and a reliance on historical archived isolates , meant data biases were unavoidable . Counterbalancing this weakness are the dense sampling of both host populations and the exceptionally detailed metadata . Random Forest and Boosted Regression models identified strong epidemiological signatures of M . bovis transmission within and between host populations . Within species , metrics capturing the spatial , temporal , and network dynamics were all highly informative , indicative of M . bovis circulation being dependent on these factors . Between species , the variation observed between M . bovis sourced from cattle and badgers was found to be well explained by where the animals resided and when they were infected . Changes in these relationships could be exploited to rapidly identify changes in the epidemiology , as might be caused by badger social perturbation under culling operations ( Tuyttens et al . , 2000; Woodroffe et al . , 2006 ) . The present study identified further evidence of within- and between-species transmission in the phylogenetic relationships between the M . bovis genomes ( Figure 1 ) . Five clades containing highly similar M . bovis genomes derived from infected cattle and badgers were identified , suggesting that substantial inter-species transmission had occurred . The presence of clades dominated by a single host species was also consistent with sustained within-species transmission . However , these phylogenetic relationships are particularly sensitive to sampling biases and should be interpreted with caution . For example , one interpretation of the basal location of the cattle-derived M . bovis genomes in the clades shown in Figure 1 is that they originated in cattle . Alternatively , this pattern could be the result of sampling the cattle population over a broader temporal range ( from 1988 to 2013 ) than the badgers ( 2000 to 2011 ) . Further interrogation of the cattle and badger life histories associated with clade 4 ( Figure 1 ) revealed evidence of prolonged persistence of this lineage in the badger population ( Figure 2 ) . Despite the cattle population being sampled over a longer time period , the badgers associated with clade 4 were predominantly infected earlier than the cattle and that strain persisted in the badgers for over 10 years . The remaining clades examined suggested that cattle could have been infected before badgers; however , it was not possible to determine whether badgers outside of Woodchester Park could be driving these interactions . Our results do suggest that inter-badger transmission is likely to be dominated by short-range interactions , given that short spatial distances ( all less than 3 . 7 km ) were highly informative in describing the genetic relationships examined in the machine learning analyses . Therefore , badgers further away from Woodchester Park are unlikely to be directly driving the patterns observed in our sampled badger population , and the ‘invading’ clades observed here are more parsimoniously explained by introductions of M . bovis from cattle . An additional limitation of these analyses is that no other wildlife species were sampled . Previous research by Delahay et al . ( 2007 ) found other mammal species infected with M . bovis in the area , albeit at lower prevalence ( 7 . 2% in Fallow deer and 6 . 8% in Muntjac deer ) than the sampled badger population ( ~30%; Delahay et al . , 2013 ) . Given considerable evidence in the present study for inter-species transmission of M . bovis , we next used BASTA , an analysis platform that can account for sampling biases ( De Maio et al . , 2018 ) , to quantify these processes ( Figure 3b ) . The BASTA analyses estimated transition rates between demes within a structured population . As the demes within the structured model were species-specific , the estimated between-species transition rates can be considered equivalent to transmission rates between populations of badgers and cattle . The most favoured two-deme model estimated badgers-to-cattle transmission rates were , on average , 10 . 4 times higher than cattle-to-badgers transmission rates ( Figure 3a and b ) . However , the second most favoured four-deme model ( which included a more complex population structure ) estimated that inter-species transmission rates were close to equal . Although even structured coalescent models do not accurately reflect spatial contact patterns , that the simplest ‘two-deme’ model is favoured is encouraging ( i . e . more spatially structured models do not perform better ) . However , the two-deme model may also have been favoured because of the limited genetic diversity available to estimate the evolutionary parameters and therefore further exploration with explicitly spatial approaches is an important next step . In the process of quantifying inter-species transmission rates , the BASTA analyses also provide counts of the number of transmission events within and between the sampled badgers and cattle ( Figure 3c ) . These counts provide a conservative estimate of the minimum number of transitions between the sampled animals and their ancestors . Although it is not appropriate to directly compare the counts within- and between-species , they do demonstrate that , at a minimum , within-species transmission occurs at least twice as frequently as between-species transmission . The high degree of within-species transmission estimated here is consistent both with the results of other studies that highlight the importance of cattle-to-cattle transmission ( Costello et al . , 1998; Gilbert et al . , 2005; Goodchild and Clifton-Hadley , 2001; Green et al . , 2008; Menzies and Neill , 2000 ) , and the persistent long-term infection observed in the Woodchester Park badger population ( Delahay et al . , 2013 ) . The high-density badger population in Woodchester Park is likely to be similar to populations found in other parts of southwest England ( Judge et al . , 2017 ) . However , broader representativeness should be confirmed by comparison to sympatric cattle and badger populations elsewhere in Britain and Ireland , particularly in areas with high bTB incidence . In addition , we selected only isolates of spoligotype SB0263 , as this was the dominant type in the badger population . The selection of SB0263 could artificially inflate the badgers-to-cattle transition rates estimated here , as the high prevalence of this spoligotype in the badgers could be a reflection of host preference . However , though there are known phenotypic differences between spoligotypes , there is no evidence that these fundamentally change the epidemiology ( Garbaccio et al . , 2014; Wright et al . , 2013 ) . In addition , many different M . bovis spoligotypes have been observed in sympatric badger and cattle populations ( Smith et al . , 2003 ) and SB0263 is not only one of the most common spoligotypes in the UK ( Smith et al . , 2003 ) , it is also highly prevalent in the cattle around Woodchester Park . If the transmission interactions estimated in our research are replicated elsewhere , this could help to explain the failure of efforts to address recurrent and persistent infection in cattle herds that co-exist with badger populations ( Gallagher et al . , 2013; Karolemeas et al . , 2011 ) . In addition , the bi-directional transmission of M . bovis between species has the potential to combine local persistence in badgers with the long-distance mobility of the cattle . In line with a recent evidence review ( Godfray et al . , 2018 ) , our research also suggests that coordinated bTB control in both cattle and badgers may be necessary to control infection in cattle . More generally , our analyses illustrate the complex interplay that underpins multi-host pathogen problems and demonstrate that , despite this complexity , appropriately defined suites of methods can be used to overcome issues of data biases and identify important epidemiological properties of these systems . Figure 4 describes the complete set of analyses conducted on the M . bovis whole genome sequences sourced from infected cattle and badgers living in and around Woodchester Park . These analyses are described in the sections that follow . Since 1976 , the Woodchester Park badger population has been the subject of a capture-mark-recapture study whereby each badger social group is trapped four times a year ( Delahay et al . , 2013 ) . Social group territories are delineated annually using bait-marking ( Delahay et al . , 2000 ) . During trapping operations , each captured badger is given a unique tattoo and at each capture event a number of samples are obtained to determine M . bovis infection status ( full details described in Delahay et al . , 2013 ) . From 1990 onwards , any M . bovis isolated from samples taken during trapping were spoligotyped ( spacer-oligo typing ) using conventional methods ( Aranaz et al . , 1996 ) and archived . Spoligotyping reports the presence or absence of 43 known spacer sequences within a single direct repeat region of the M . bovis genome . In total , 230 isolates were available from the archive , which originated from samples taken from 116 different badgers from 2000 to 2011 . The cattle herds surrounding Woodchester Park undergo statutory annual testing for M . bovis infection as a part of routine surveillance , and results are stored in APHA’s cattle testing ( SAM ) database ( Lawes et al . , 2016 ) . Test-positive cattle are slaughtered , selected tissues taken for culture and any M . bovis isolates are spoligotyped and archived . In addition , the movements of every cow in the UK are recorded in the Cattle Tracing System ( CTS ) . For the present study 124 cattle-derived M . bovis isolates , each collected from an individual cow between 1988 and 2013 , were selected from the archives . Cattle isolates were selected if they were of the same spoligotype as the badger isolates and were from herds within 10 km of Woodchester Park . More than 90% of the badger-derived isolates were spoligotype SB0263 . More than 75% ( 1096/1442 ) of the isolates available from cattle within 10 km of Woodchester Park shared the same spoligotype and it is the second most common type found across England ( Smith et al . , 2003; Smith et al . , 2006 ) . To increase the chances of sequencing strains that were shared with the badgers in Woodchester Park , rather than circulating in the cattle population independently , only cattle-derived isolates of spoligotype SB0263 were selected . Additional spoligotype SB0263 isolates from cattle that lived in herds within 100 km of Woodchester Park ( n = 65 ) were included to provide a broader spatio-temporal context , resulting in a total of 189 isolates . Badger-derived M . bovis isolates were prepared for sequencing by the Agri-Food and Biosciences Institute in Northern Ireland ( AFBI-NI ) and for the cattle-derived isolates by APHA . M . bovis isolates were selected from the frozen archives and re-cultured on Löwenstein-Jensen medium . Prior to DNA extraction the isolates were heat killed in a water bath at 80°C for a minimum of 30 min . DNA was extracted from these cultures using standard high salt and cationic detergent cetyl hexadeycl trimethyl ammonium bromide ( CTAB ) and solvent extraction protocols ( Parish and Stoker , 2001; van Soolingen et al . , 2001 ) . Extracted DNA was sequenced at the Glasgow Polyomics facility using an Illumina Miseq producing 2 × 300 bp paired end reads ( badger derived isolates ) and at the APHA central sequencing unit in Weybridge using an Illumina Miseq producing 2 × 150 bp paired end reads ( cattle derived isolates ) . The 65 additional cattle-derived isolates were sequenced at the APHA central sequencing unit in Weybridge using an Illumina NextSeq producing 2 × 150 bp paired end reads ( cattle-derived isolates ) . Following quality assessments in FASTQC ( v0 . 11 . 2; Andrews , 2010; RRID:SCR_014583 ) , the raw WGS data were trimmed using PRINSEQ ( v0 . 20 . 4; Schmieder and Edwards , 2011; RRID:SCR_005454 ) and adapters were removed using TRIMGALORE ( v0 . 4 . 1; Krueger , 2015; RRID:SCR_016946 ) . The trimmed data were aligned to the M . bovis reference genome ( AF2122/97; Malone et al . , 2017 ) using the Burrows-Wheeler aligner ( BWA , v0 . 7 . 17; Li and Durbin , 2009; RRID:SCR_010910 ) . Regions encoding proline-glutamate and proline-proline-glutamate surface proteins , or annotated repeat regions were excluded ( Sampson , 2011 ) . Mapping quality information on all the SNVs identified was retained for each isolate . The allele frequencies at each position in the aligned ( against reference ) sequence from each isolate were examined . For a haploid organism these frequencies are expected to be either 0 or 1 , with some random variation expected from sequencing errors ( Sobkowiak et al . , 2018 ) . A heterozygous site was defined as one where the allele frequencies were >0 . 05 and <0 . 95 . Four cattle-derived sequences that had more than 150 heterozygous sites , and allele frequencies that were clustered and non-random ( data not shown ) , were removed . In addition , 26 badger-derived and 16 cattle-derived M . bovis sequences were removed because of suspected errors in the metadata ( Appendix 1: Investigating isolate metadata discrepancies ) . For the sequences from the remaining isolates ( 204 badger- and 169 cattle-derived isolates ) , alleles were called at each variant position if they had mapping quality ≥30 , high-quality base depth ≥4 ( applied to reverse and forward reads separately ) , read depth ≥30 , and allele support ≥0 . 95 . For any site that failed these criteria , if the allele called had been observed in a different isolate that had passed , a second round of filtering was conducted using a high-quality base depth of 5 ( total across forward and reverse reads ) and the same allele support . As recombination is thought to be extremely rare for mycobacteria ( Namouchi et al . , 2012 ) , variants in close proximity could indicate a region that is difficult to sequence or under high selection . To avoid calling variants in these regions , variant positions within 10 bp of one another were removed . Following filtering , sequences from 11 badger and 10 cattle isolates that had insufficient coverage ( <95% ) of the variant positions were removed . Once the alignment was generated , sites with a consistency index less than 1 , generally considered homoplasies ( Farris , 1989 ) , were removed ( n = 4 , of 14 , 991 sites ) using HomoplasyFinder ( v0 . 0 . 0 . 9; Crispell et al . , 2019; RRID: SCR_017300 ) . All the scripts necessary for the processing of the WGS data are freely available online . Our research hypothesized that within- and between-species transmission was occurring in the study system . If bi-directional transmission was occurring , then there should be epidemiological signatures in the genomic data linked to these events . These signatures are likely to relate to the spatial , temporal , and network dynamics of the sampled badger and cattle populations , as these will determine their contact patterns . To investigate whether there were any epidemiological signatures of within- and between-species transmission of the sampled M . bovis isolates , the genetic distances between sequences were compared to epidemiological metrics describing the spatial , temporal , and network relationships between the animals associated with each sequence . Inter-sequence genetic distances were calculated , for every pair of sequences , by dividing the number of differences present between the pair of sequences by the total number of sites considered ( n = 14 , 987 ) . In addition , epidemiological metrics were calculated to identify any similarities among animals associated with a particular pair of isolates . Epidemiological metrics were calculated using the data , where available , on each animal obtained from its capture or movement and testing history ( further details in Appendix 1: Defining the epidemiological metrics ) . Two additional dummy metrics , samples from a uniform distribution and a Boolean distribution , were included to determine a threshold of importance that distinguishes noise from signal . Inter-isolate genetic distances and associated epidemiological metrics were compared using Random Forest ( RRID:SCR_015718; Liaw and Wiener , 2002 ) regression and Boosted Regression ( RRID:SCR_017301; Elith et al . , 2008 ) models in R ( v3 . 4 . 3; R Development Core Team , 2016 ) . These machine learning approaches were used to separately analyse badger–badger , badger–cattle , and cattle–cattle comparisons . For each set of comparisons , a training dataset was constructed using 50% of the data available and , following training using these data , the model was tested on the remaining 50% of the data . Genetic distances ≤ 15 SNVs were used for these analyses to avoid larger inter-sequence distances that were not likely to relate to the fine resolution epidemiological relationships of interest . Random Forest and Boosted Regression approaches were selected as these methods can deal with large datasets with many highly correlated variables whose relationship to the response variable ( genetic distances ) cannot readily be defined ( Auret and Aldrich , 2012 ) . A broad range of epidemiological metrics were defined as the Random Forest and Boosted Regression models are robust to non-informative and/or highly correlated variables ( Auret and Aldrich , 2012; Elith et al . , 2008; Liaw and Wiener , 2002 ) . The two independent approaches were used to ensure that any patterns observed were robust . The influence of including highly correlated and non-informative predictor variables and variables with a large amount of missing data in the machine learning approaches was investigated using the Random Forest models . For highly correlated variables , clusters of correlated variables were defined and the least informative variable from each cluster was incrementally removed and the impact on the fitted Random Forest regression models was examined . A similar approach was used twice more to evaluate the influence of retaining non-informative predictor variables and of including predictor variables with large amounts of missing data in the models . Following investigation of population level epidemiological signatures in the sequence data , a phylogenetic tree was constructed to describe the evolutionary relationships among our set of M . bovis genome sequences . If inter- and intra-species transmission events were occurring in the sampled system , there should be evolutionary signatures in the phylogenetic tree . For example , if M . bovis sequences sourced from cattle and badgers have a very close phylogenetic relationship , this suggests that inter-species transmission has occurred . The phylogeny was constructed with the maximum likelihood algorithm in RAxML ( v8 . 2 . 11; Stamatakis , 2014; RRID:SCR_006086 ) using a GTR ( generalized time reversible ) substitution model with 100 bootstraps . The maximum likelihood algorithm was selected as a fast alternative to Bayesian approaches . Although Bayesian approaches will better explore the phylogenetic tree space , this space is expected to be small for phylogenies based on M . bovis data given its highly conserved genome . The GTR model was the most appropriate based on analyses using the modelTest ( ) function in the R package PHANGORN ( v2 . 3 . 1; Schliep , 2011; RRID:SCR_017302 ) . Based on the range of SNV thresholds ( 3–12 ) used to define recent M . tuberculosis transmission ( Bryant et al . , 2013; Jajou et al . , 2018; Roetzer et al . , 2013; Yang et al . , 2017 ) , clades containing highly related ( <10 SNVs apart ) cattle-derived and badger-derived sequences ( inter-species clades ) were identified ( Figure 1 ) . The testing histories and recorded movements ( for cattle ) , and capture information ( for badgers ) of the sampled and in-contact animals associated with each cluster were available . These data were investigated to determine whether they provided any additional evidence to support the phylogenetic relationships indicative of inter-species transmission . ‘In-contact’ animals were defined as those badgers that resided in the same badger social group , or those cattle that lived in the same herd , at the same time as one or more of the sampled badgers or cattle ( respectively ) associated with a particular inter-species clade . To further investigate patterns of inter- and intra-species transmission , additional evolutionary analyses were completed to estimate directional inter-species transmission rates and quantify their frequency relative to intra-species transmission events . A subset of the sequences available ( from 97 badger- and 83 cattle-derived isolates ) was selected to estimate the transmission rate of M . bovis between the sampled cattle and badger populations . The selected sequences were within the parent clade containing all the inter-species clades ( shown in Figure 1 ) and were sampled from within 10 km of Woodchester Park between 1999 and 2014 . The subset of sequences was split into ‘inner’ and ‘outer’ groups , based on a 3 . 5 km radius from Woodchester Park ( Figure 5 ) . The 3 . 5 km radius size was selected to contain the sampling locations associated with all the badger-derived sequences and the closest cattle-derived sequences , based on the reported home-ranges of badgers in southern England being <1 km2 ( Garnett et al . , 2005; Macdonald et al . , 2008; Roper et al . , 2003 ) . The presence of a temporal signal among the selected M . bovis sequences was examined ( Appendix 2: Testing the presence of a temporal signal ) . A temporal signal was supported by a positive trend , calculated within TEMPEST ( v1 . 5; Rambaut et al . , 2016; RRID:SCR_017304 ) , between each sequence’s root-to-tip distance and its sampling time and the results of a tip-date randomisation procedure ( Firth et al . , 2010 ) . The Bayesian Structured coalescent Approximation ( BASTA v2 . 3 . 1; De Maio et al . , 2015; RRID:SCR_017303 ) tool , available in BEAST2 ( Bayesian Evolutionary Analysis by Sampling Trees – v2 . 4 . 4 ( Bouckaert et al . , 2014 ) , RRID:SCR_017307 ) , uses an approximation of the structured coalescent approach ( Vaughan et al . , 2014 ) to estimate migration rates within a structured population . The structured population in the current context is the M . bovis population , whose structure was likely to relate to host species and their spatial relationships . BASTA , in contrast to previously popular methods such as discrete trait analyses ( Lemey et al . , 2009; Pagel et al . , 2004 ) , can estimate the ancestral structure of the population in the presence of biased sampling ( De Maio et al . , 2015 ) . There were two biases associated with the set of sequences available . First , the prevalence of M . bovis in the sampled cattle and badger populations was likely to be different as a result of the on-going control operations in the cattle , therefore the sampling proportions of these different populations relative to the prevalence of M . bovis were likely to be unequal . Second , although the badger population within Woodchester Park has been intensively monitored and sampled , the surrounding badger population is less well understood and unsampled , whereas cattle both within and outside the Woodchester Park area have been sampled . Based on the ‘inner’ and ‘outer’ populations of the sampled cattle and badgers ( shown in Figure 5 ) , a series of BASTA analyses , splitting the sampled M . bovis population into different demes , were designed to estimate the inter-species transition rates while accounting for the two sampling biases discussed ( Figure 6 ) . For each of the nine separate population structures , two separate analyses were conducted , one where the deme sizes were constrained to be equal and another where they were allowed to vary . Each of these 18 analyses was repeated three times and estimates were combined across replicates . The inter-species transition rates from each model were compared using the Akaike’s Information Criterion through Markov Chain Monte Carlo ( AICM; Baele et al . , 2013 ) , for further details see Appendix 2: Structured coalescent analyses using BASTA . All the code generated for this manuscript is freely available on GitHub . General scripts are available within the ‘WoodchesterPark’ of the GeneralTools repository ( https://github . com/JosephCrispell/GeneralTools; Crispell , 2019a; copy archived at https://github . com/elifesciences-publications/GeneralTools ) . The Java source code files can be found in a separate respository ( https://github . com/JosephCrispell/Java; Crispell , 2019b; copy archived at https://github . com/elifesciences-publications/Java ) . These scripts are licenced under the General Public Licence v3 . 0 . All WGS data used for these analyses have been uploaded to the National Centre for Biotechnology Information Short Read Archive ( NCBI-SRA: PRJNA523164 ) . Because of the sensitivity of the associated metadata , only the sampling date and species will be provided with these sequences .	Disease-causing microbes that infect more than one type of animal can be difficult to control . This is especially true when they infect wildlife . For example , Mycobacterium bovis is a bacterium that causes tuberculosis in tens of thousands of cattle in Britain every year and also infects badgers and other wildlife . Controlling the infections in cattle is essential , as it helps prevent the bacteria from infecting humans , improves cattle welfare and reduces the substantial costs to the livestock industry . Analysing the relatedness of M . bovis genomes from infected cattle and badgers may help scientists work out how often badgers infect cattle and vice versa . Scientists have collected data and M . bovis samples from infected badgers in Woodchester Park , in England , for over three decades . Using these data and additional information about M . bovis infecting nearby cattle may help scientists learn how the bacteria spreads and how to stop it . Now , Crispell et al . show that complex patterns of contact between cattle and badgers likely drive the persistence of tuberculosis in cattle , also known as bovine tuberculosis . In three separate analyses , Crispell et al . compared the genomes of M . bovis found in cattle and badgers , the animals' locations , when they were infected , and whether they could have been in contact . The analyses found that M . bovis was likely to have been transmitted more frequently from badgers to cattle rather than from cattle to badgers . They also showed that transmission within each species happened more often than transmission between species . If these results are confirmed by other studies , they may help scientists develop better strategies for controlling tuberculosis in British cattle . In particular , controversial control strategies – such as badger culls – could be more targeted to better combat tuberculosis in cattle but have less of an impact on badgers . These insights might also aid control efforts in other countries where bovine tuberculosis is a problem and an important source of human tuberculosis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "microbiology", "and", "infectious", "disease" ]	2019	Combining genomics and epidemiology to analyse bi-directional transmission of Mycobacterium bovis in a multi-host system
The Aedes aegypti mosquito shows extreme sexual dimorphism in feeding . Only females are attracted to and obtain a blood-meal from humans , which they use to stimulate egg production . The fruitless gene is sex-specifically spliced and encodes a BTB zinc-finger transcription factor proposed to be a master regulator of male courtship and mating behavior across insects . We generated fruitless mutant mosquitoes and showed that males failed to mate , confirming the ancestral function of this gene in male sexual behavior . Remarkably , fruitless males also gain strong attraction to a live human host , a behavior that wild-type males never display , suggesting that male mosquitoes possess the central or peripheral neural circuits required to host-seek and that removing fruitless reveals this latent behavior in males . Our results highlight an unexpected repurposing of a master regulator of male-specific sexual behavior to control one module of female-specific blood-feeding behavior in a deadly vector of infectious diseases . Across animals , males and females of the same species show striking differences in behavior . Male Paradisaeidae birds-of-paradise perform an elaborate courtship dance to seduce prospective female partners , contorting their bodies in forms resembling flowers , ballerinas , and smiling faces ( Scholes , 2008 ) . Female Serromyia femorata midges pierce and suck conspecific males dry during mating , breaking off his genitalia inside her , thereby supplying the female with both nutrition and sperm ( Edwards , 1920 ) . Although an astonishing diversity of sexually dimorphic behaviors exists across species , most insights into the genetic and neural basis of sex-specific behaviors have come from a limited set of model organisms ( Matthews and Vosshall , 2020 ) . Which genes control sexual dimorphism in specialist species that have evolved novel behaviors ? Do conserved genes control sexual dimorphism in species-specific behaviors , or do novel genes evolve to control new behaviors ? Many advances in understanding the genetics of sexually dimorphic behaviors have come from the study of Drosophila melanogaster fly courtship , where a male fly orients toward , taps , and follows a female fly , extending a wing to produce a courtship song before tasting , mounting , and copulating with her ( Hall , 1994 ) . Courtship comprises behavioral modules , which are simple discrete behaviors that must be combined to perform a complex behavior and are elicited by different sensory modalities and subsets of fruitless-expressing neurons ( Clowney et al . , 2015; Clyne and Miesenböck , 2008; Kohl et al . , 2013; Ruta et al . , 2010; von Philipsborn et al . , 2011 ) . Courtship modules include orienting , which is driven by visual information ( Ribeiro et al . , 2018 ) and persistent following and singing , which are triggered by chemical cues on a female fly ( Clowney et al . , 2015 ) and guided by vision ( Ribeiro et al . , 2018; Sten et al . , 2020 ) . The fruitless gene is sex-specifically spliced in the brain of multiple insect species including mosquitoes ( Bertossa et al . , 2009; Gailey et al . , 2006; Salvemini et al . , 2013 ) and has been proposed to be a master regulator of male courtship and mating behavior across insects ( Clowney et al . , 2015; Demir and Dickson , 2005; Hall , 1994; Ryner et al . , 1996; Seeholzer et al . , 2018; Tanaka et al . , 2017 ) . Sex-specific splicing of the fruitless gene controls several aspects of courtship behavior . Male flies mutant for fruitless promiscuously court other males and cannot successfully mate with females ( Ito et al . , 1996; Ryner et al . , 1996 ) . Forcing male fruitless splicing in females triggers orientation and singing behaviors normally only performed by males ( Demir and Dickson , 2005 ) . In addition , fruitless is required for sex-specific aggressive behaviors ( Vrontou et al . , 2006 ) . Fruitless encodes a BTB zinc-finger transcription factor that is thought to control cell identity and connectivity during development ( Ito et al . , 2016; Neville et al . , 2014 ) , as well as the functional properties of neurons in adulthood ( Sethi et al . , 2019 ) . Fruitless modulates the expression of a number of potential downstream target genes ( Neville et al . , 2014; Sato and Yamamoto , 2020; Vernes , 2015 ) in a cell-type-specific manner ( Brovkina et al . , 2020 ) . Moreover , fruitless has a conserved role controlling courtship in multiple Drosophila species ( Seeholzer et al . , 2018; Tanaka et al . , 2017 ) , and sex-specific fruitless splicing is conserved across wasps ( Bertossa et al . , 2009 ) and mosquitoes ( Gailey et al . , 2006; Salvemini et al . , 2013 ) , suggesting that fruitless may act as a master regulator of sexually dimorphic mating behaviors across insects . Mosquitoes display striking sexually dimorphic mating and feeding behaviors . Only male mosquitoes initiate mating , and only females drink blood , which they require to develop their eggs ( Bowen , 1991; Galun et al . , 1963; Jové et al . , 2020; Klowden , 1995 ) . Sexual dimorphism in blood-feeding is one of the only instances of a completely sexually dimorphic feeding behavior because male mosquitoes never pierce skin or engorge on blood . While part of this dimorphism is enforced by sex-specific genitalia ( Spielman , 1964 ) or feeding appendages ( Jones and Pilitt , 1973 ) , there is also a dramatic difference in the drive to hunt hosts between males and female mosquitoes ( Bowen , 1991; Roth , 1948 ) . To blood-feed , females combine multiple behavioral modules ( Bowen , 1991 ) . Female Aedes aegypti mosquitoes take flight when exposed to carbon dioxide ( Bowen , 1991; McMeniman et al . , 2014 ) and are attracted to human olfactory ( DeGennaro et al . , 2013; Dekker et al . , 2005; Zwiebel and Takken , 2004 ) , thermal , and visual cues ( Liu and Vosshall , 2019; McMeniman et al . , 2014; van Breugel et al . , 2015 ) , and integrate at least two of these cues to orient toward and land on human skin . Engorging on blood is triggered by specific sensory cues tasted by the female ( Galun et al . , 1963; Jové et al . , 2020 ) . It is not known which genes have evolved to control this unique sexually dimorphic and mosquito-specific feeding behavior . Here , we generate fruitless mutant Aedes aegypti mosquitoes and show that consistent with observations in Drosophila , fruitless is required for male mating behavior . Unexpectedly , fruitless mutant male mosquitoes gain the ability to host-seek , specifically driven by an attraction to human odor . Our results demonstrate that sexual dimorphism in a single module of a mosquito-specific behavior is controlled by a conserved gene that we speculate has gained a new function in the course of evolution . We used an arm-next-to-cage assay ( Figure 1A ) to monitor attraction of male and female Aedes aegypti mosquitoes to a live human arm . Consistent with their sexually dimorphic blood-feeding behavior , only females were strongly attracted to the arm ( Figure 1B–C ) . What is the genetic basis for this extreme sexual dimorphism ? We reasoned that fruitless , which is alternatively spliced in a sex-specific manner and promotes male courtship and copulation in D . melanogaster flies ( Ito et al . , 1996; Ryner et al . , 1996 ) , may play similar roles in controlling sexually dimorphic behaviors in Aedes aegypti . Fruitless is a complex gene with multiple promoters and multiple alternatively spliced exons . Downstream promoters drive broadly expressed non-sex-specific fruitless transcripts and proteins ( Lee et al . , 2000 ) . A previous study showed ( Salvemini et al . , 2013 ) and we confirmed that transcripts from the upstream neuron-specific ( P1 ) promoter in the Aedes aegypti fruitless gene are sex-specifically spliced ( Figure 1D–H ) . Both male and female transcripts include a short male ‘m’ exon , and female transcripts additionally include a longer female ‘f’ exon with an early stop codon , predicted to yield a truncated Fruitless protein in the female . However , it is unlikely that any sex-specific Fruitless protein or peptides are stably expressed in adult females . In Drosophila , sex-specific female fruitless peptides are not detected ( Lee et al . , 2000 ) , and transformer is thought to inhibit translation by binding to female fruitless P1 transcripts ( Usui-Aoki et al . , 2000 ) . P1 transcripts of both sexes splice to the first common ‘c1’ exon , but only male transcripts are predicted to encode full-length Fruitless protein with BTB and zinc-finger domains ( Figure 1E ) . By analyzing previously published tissue-specific RNA-seq data ( Matthews et al . , 2016 ) , we verified that of all the fruitless exons , only the f exon was sex-specific in Aedes aegypti brains ( Figure 1G ) . Moreover , while the c1 exon was broadly expressed through downstream promoters , P1 transcripts were specifically expressed in the brain and the antenna , the major olfactory organ of the mosquito ( Figure 1H ) , consistent with fruitless expression in Drosophila ( Stockinger et al . , 2005 ) . To ask if fruitless splicing was conserved across mosquitoes , we sequenced RNA from male and female brains of five different species and assembled de novo transcriptomes for each sex . Three of these species are important arboviral disease vectors because their females blood feed on humans , whereas the two other species only feed on plants ( Bradshaw et al . , 2018; Zhou et al . , 2014; Figure 1F ) . We identified orthologues of fruitless in each species and found that all had conserved ‘m’ and ‘c1’ exons and distinct ‘f’ exons . Fruitless was sex-specifically spliced in each of these species with a female-specific ‘f’ exon and early stop codon , predicted to produce a full-length fruitless protein only in males . We used CRISPR-Cas9 genome editing ( Kistler et al . , 2015 ) to disrupt P1 neural-specific fruitless transcripts in Aedes aegypti to investigate a possible role of fruitless in sexually dimorphic mosquito behaviors . We generated two alleles , fruitless∆M , which introduces a frameshift that is predicted to produce a truncated protein in males , and fruitless∆M-tdTomato , in which the fruitless gene is disrupted by a knocked-in CsChrimson:tdTomato fusion protein ( Figure 1I ) . In both alleles , the protein is truncated before the downstream BTB and zinc-finger domains . The fruitless∆M-tdTomato line allowed us to visualize cells that express the fluorescent tdTomato reporter under the control of the endogenous fruitless regulatory elements . To control for independent background mutations , we used the heteroallelic fruitless∆M/fruitless∆M-tdTomato mutant strain in all subsequent behavior assays ( Figure 1I ) . In this heteroallelic mutant , fruitless P1 transcripts are disrupted in both males and females . Since full-length fruitless protein is male-specific , we expected that only fruitless∆M/fruitless∆M-tdTomato male mosquitoes would display altered behavioral phenotypes . In D . melanogaster , P1 fruitless transcripts are expressed in several thousand cells comprising about ~2% of the neurons in the adult brain ( Stockinger et al . , 2005 ) . To examine the distribution of cells expressing fruitless in male and female Aedes aegypti mosquitoes , we carried out whole mount brain staining to reveal the tdTomato marker expressed from the fruitless locus . Fruitless >tdTomato is expressed in a large number of cells in both male and female brains ( Figure 2A–F ) , as well as in the ventral nerve cord ( Figure 2—figure supplement 1A–B ) . Fruitless >tdTomato expressing cells innervate multiple regions of the mosquito brain , including the suboesophageal zone , the lateral protocerebral complex , and the lateral horn . These areas have been implicated in feeding ( Jové et al . , 2020 ) , mating ( Seeholzer et al . , 2018 ) , and innate olfactory behaviors ( Datta et al . , 2008 ) respectively , and also receive projections from fruitless-expressing neurons in Drosophila ( Seeholzer et al . , 2018; Stockinger et al . , 2005 ) . The projections of fruitless >tdTomato neurons are dramatically sexually dimorphic , with denser innervation in the female suboesophageal zone and the male lateral protocerebral complex ( Figure 2A–F ) . We did not detect any gross anatomical differences between heterozygous and heteroallelic fruitless mutant male brains or the pattern of fruitless >tdTomato expression ( Figure 2B , C , E , F ) . We cannot exclude the possibility that there are subtle differences that can only be observed with sparse reporter expression in subsets of cells . We also examined fruitless expression in the periphery . Odors are sensed by olfactory sensory neurons in the mosquito antenna , and each type of neuron projects to a single glomerulus in the antennal lobe of the mosquito brain ( Figure 3A ) . We found that , as is the case in Drosophila ( Stockinger et al . , 2005 ) , fruitless >tdTomato is expressed in olfactory sensory neurons in the antenna of both male and female mosquitoes , and that some of these neurons co-express the olfactory receptor co-receptor Orco ( Figure 3B–E ) . Fruitless >tdTomato labels a subset of glomeruli in the antennal lobe , with females having about twice as many positive glomeruli compared to males of either genotype ( Figure 3F–L ) . There was no difference in the number of fruitless >tdTomato labeled glomeruli between wild-type and fruitless mutant males ( Figure 3F ) , suggesting that fruitless does not control sexual dimorphism in the number of glomeruli labeled by fruitless >tdTomato . Given the broad neural expression and sexual dimorphism in fruitless circuits , we asked if fruitless mutant males showed any defects in sexually dimorphic feeding and mating behaviors . Since only female mosquitoes have the anatomical capacity to pierce skin and artificial membranes ( Jové et al . , 2020; Klowden , 1995 ) , we developed a feeding assay in which both females and males are able to feed from warmed liquids through a net without having to pierce a membrane to access the meal ( Figure 4A ) . Both wild-type males and females reliably fed on sucrose and did not feed on water . Only wild-type females fed on blood . Even when warm blood was offered and available to males for ready feeding , they still did not find it appetizing ( Figure 4B ) . Fruitless mutant males fed similarly to their wild-type male counterparts on all meals , suggesting that this behavioral preference is not under the control of fruitless in males ( Figure 4B ) . Because fruitless plays a key role in male courtship and mating in Drosophila , we asked if it is similarly required in Aedes aegypti . Since mosquitoes show extremely rapid in-flight mating behavior that is completed in less than 30 s , it is difficult to directly observe or quantify ( Hartberg , 1971 ) . We used previously developed insemination assays ( Degner and Harrington , 2016; Duvall et al . , 2017 ) to quantify the ability of males to successfully mate ( Figure 4C ) . We found that fruitless mutant males appeared to contact females but were unable to successfully inseminate wild-type females ( Figure 4D ) . This mating failure is consistent with the established role of fruitless in Drosophila male sexual behavior ( Demir and Dickson , 2005; Ryner et al . , 1996 ) . We then turned to innate olfactory behaviors that govern the search for nectar , which is used as a source for metabolic energy by both males and females , and blood , which is required only by females for egg production . Consistent with the use of these meals , nectar-seeking behavior is not sexually dimorphic , but human host-seeking behavior is sexually dimorphic . To measure these behaviors , we adapted the Uniport olfactometer ( Liesch et al . , 2013 ) , which is only able to test one stimulus at a time , and developed the Quattroport , an olfactometer that tests attraction to four separate stimuli in parallel ( Figure 4E ) . The Quattroport measures both the activation , the participation of the animals in the assay , and attraction , short range attraction to the stimulus ( Figure 4F ) . In control experiments , we examined activation responses of wild-type male and females offered a blank , CO2 , a human arm , or the floral odor of honey . While males and females showed equivalent activation with a blank and honey , females were more strongly activated to the host-related cues of CO2 and the human arm ( Figure 4G ) . To model nectar-seeking behavior , we used honey as a floral odor and glycerol as a control odor as previously described ( Figure 4H; DeGennaro et al . , 2013 ) . There was no difference in nectar-seeking as defined by attraction in the Quattroport between wild-type females , males , and fruitless mutant males ( Figure 4I ) . We next used the Quattroport with a live human host as a stimulus ( Figure 4J ) . As expected , wild-type females robustly and reliably entered traps in response to a live human forearm . In contrast , zero wild-type males entered the trap , consistent with our observations in the arm-next-to-cage assay ( Figure 1A–C ) . If fruitless function in Aedes aegypti were limited to mating and aggression as it is in Drosophila , we would expect fruitless mutant males to show no interest in a live human host . Unexpectedly , fruitless mutant males were as attracted to a live human host as wild-type females ( Figure 4K ) . This indicates that fruitless males have gained the ability to host-seek , displaying the signature sexually dimorphic behavior of the female mosquito . A live human arm gives off multiple sensory cues that are known to attract female mosquitoes , the most salient of which are body odor and heat . Fruitless mutant males might be attracted by heat alone or only the human odor , or to the simultaneous presentation of both cues . To disentangle the contribution of these complex sensory cues to the phenotype we observed , we tested the response of fruitless mutant males to each cue in isolation . We first used a heat-seeking assay ( Corfas and Vosshall , 2015; McMeniman et al . , 2014 ) to present heat to mosquitoes in the absence of human odor ( Figure 5A ) . Neither fruitless mutant nor wild-type males were attracted to the heat cue at any temperature ( Figure 5B ) . In contrast , wild-type females showed typical heat-seeking behavior that peaked near human skin temperature ( Figure 5B ) . To ask if fruitless mutant males are attracted to human host odor alone , we collected human scent on nylon stockings and presented this stimulus in the Quattroport ( Figure 5C ) to both male and female mosquitoes . Although wild-type males and heterozygous fruitless mutant males showed no response to human odor , wild-type females and fruitless∆M/fruitless∆M-tdTomato females were strongly attracted to human odor ( Figure 5D ) . Normal host-seeking in fruitless mutant females is expected since full-length fruitless protein is translated only in males . These females also showed normal blood-feeding , egg-laying , and mating behaviors ( Figure 5—figure supplement 1A–C ) , confirming our prediction that fruitless acts specifically in the male . We attempted to test the effect of forcing male fruitless splicing on female host-seeking ( Figure 5—figure supplement 2A–C ) , but found that these animals were inviable due to blood-feeding and egg-laying defects ( Figure 5—figure supplement 2D–I , Supplementary file 1 ) . We then returned to the fruitless mutant males . Remarkably , heteroallelic fruitless mutant males were strongly attracted to human scent , at levels comparable to wild-type females ( Figure 5D ) . These results demonstrate that fruitless mutant males have gained a specific attraction to human odor , which drives them to host-seek . Only female Aedes aegypti mosquitoes host-seek , and we have shown that mutating fruitless reveals an attraction to human odor in the male mosquito ( Figure 5E ) . Previously , fruitless was shown to be required for male mating behavior in both Drosophila ( Demir and Dickson , 2005 ) and Bombyx silkmoths ( Xu et al . , 2020 ) . Our work demonstrates that in Aedes aegypti mosquitoes , fruitless has acquired a novel role in inhibiting female host-seeking behavior in the male ( Figure 5F ) . Interestingly , fruitless also acts to suppress female-specific aggressive behaviors in male Drosophila in addition to its role in promoting male-specific courtship and aggression ( Vrontou et al . , 2006 ) , suggesting a common theme where this gene can repress specific aspects of female-specific behavior . We cannot exclude the possibility that fruitless had a broader ancestral role in repressing male-specific host-seeking or feeding but consider this extremely unlikely given the rarity of sexually dimorphic feeding behaviors relative to sexually dimorphic mating behaviors . Our results suggest that the neural circuits that promote female attraction to human scent are latent in males and suppressed by expression of fruitless either during development , or during adulthood . This is in contrast to a model where the ability to host-seek develops exclusively in females . Since males are able to host-seek in the absence of fruitless , other components of the sex-determination pathway do not intrinsically regulate the development and function of brain circuits controlling host-seeking behavior , even though this behavior is normally sex-specific . The concept that a latent sex-specific behavior can be revealed by knocking out a single gene was elegantly demonstrated in the mouse ( Mus musculus ) . Only male mice court and initiate sexual contact with females and yet knocking out the Trpc2 gene causes female mice to display these male-specific behaviors ( Kimchi et al . , 2007 ) . There are field reports of Aedes aegypti males being collected near human hosts ( Hartberg , 1971 ) , which the experimenters interpreted as male Aedes aegypti attraction to humans . We note , however , that these field experiments did not control for the presence of females , suggesting that the collected males may have been attracted to the female mosquitoes that attempt to bite humans . In our well-controlled laboratory assays , we were unable to find any evidence of strong attraction to humans in mosquito males at close-range ( Figure 1C , Figure 4K ) or at long distances ( Figure 4G ) . In these same assays , wild-type male mosquitoes showed strong attraction to floral cues ( Figure 4I ) . We cannot exclude that male mosquitoes in the field show some attraction to a human host , but suggest that any attraction in males would be weaker than in wild-type females , or the attraction we demonstrate in fruitless mutant males here . Host-seeking is the first step in a complex sequence of behaviors that lead to blood-feeding . After detecting and flying toward a human host , the female mosquito must land on the human , pierce the skin , and ultimately engorge on blood ( Bowen , 1991 ) . We have shown that fruitless has evolved to control sexual dimorphism in one module of this specialized behavior , the ability to host-seek . Females integrate multiple sensory cues to identify and approach human hosts , and we show that fruitless controls the response to just one of those cues , human odor . Sexual dimorphism in thermosensation , or in subsequent feeding behaviors does not appear to be controlled by fruitless in the male mosquito , since neither wild-type nor mutant fruitless males will drink warm blood . The modular genetic organization of mosquito behavior sparks intriguing parallels to other complex sexually dimorphic behaviors like mouse parenting ( Kohl et al . , 2018 ) . To be effective parents , female mice must build nests , and then retrieve , groom , and nurse their pups . In the deer mouse Peromyscus , the conserved peptide vasopressin has evolved to control nest building ( Bendesky et al . , 2017 ) . In both Peromyscus and Aedes , a conserved gene has gained control over a single aspect of a complex behavior . It has been hypothesized that certain classes of genes like neuromodulators and transcription factors are more likely to underlie phenotypic differences between species ( Bendesky and Bargmann , 2011; Martin and Orgogozo , 2013; Tosches , 2017 ) , and our study demonstrates that this is true even for an entirely novel behavior . Where in the nervous system is fruitless required to suppress host seeking in male mosquitoes ? fruitless might function in the antenna to modulate the detection of human odor in male mosquitoes , perhaps by tuning the functional or anatomical properties of olfactory sensory neurons . Such a role has been recently demonstrated in the aging-dependent sensitization of the male Drosophila antennal response to pheromones ( Sethi et al . , 2019; Zhang and Su , 2020; Zhao et al . , 2020 ) . Alternatively , both wild-type males and females might detect human odor , and fruitless could function in the central brain to reroute these signals to drive different motor outputs , as has been demonstrated with the sexually dimorphic response to Drosophila pheromones ( Datta et al . , 2008; Kohl et al . , 2013; Ruta et al . , 2010 ) . To distinguish between these two models , we would require significant advances in technology and mosquito genetics , including a fruitless driver line to image neural responses to human odor . Despite significant effort , we were unable to generate a viable fruitless driver line both because of tight genetic linkage of fruitless to the sex-determining M locus ( Hall et al . , 2015 ) and because gene-targeted females failed to blood-feed and were therefore sterile ( Supplementary file 1 ) . To explore central brain fruitless+ circuits , we would need to be able to subset expression to label and drive reporters or rescue fruitless expression in sparse populations of neurons , a technology that is still out of reach . Advances in mosquito genetic tools , such as the successful implementation of orthogonal transcriptional activator reagents , combined with sparse labeling approaches will be required to gain mechanistic insight into fruitless function within mosquito host-seeking circuits . We note that these advances were not trivial in D . melanogaster , requiring efforts from multiple laboratories over the past decade ( Datta et al . , 2008; Kohl et al . , 2013; Ruta et al . , 2010 ) , and expect that the generation of these tools and the subsequent characterization of the circuit will be significantly more challenging in the mosquito , a non-model organism . Our work suggests that fruitless has evolved the novel function of enforcing female-specific host-seeking while maintaining its presumably ancestral male mating function . How might fruitless have evolved to control host-seeking ? One possibility is that non-sex-specific host-seeking neural circuits first emerged in the ancestral mosquito , and then secondarily began to express fruitless to suppress the development or adult function of host-seeking circuits specifically in males . Another possibility is that mosquitoes duplicated and co-opted the ancestral fruitless-expressing mating neural circuits and retuned the inputs and outputs into this circuit to drive host-seeking . Circuit duplication is one of the mechanisms by which neural circuits are proposed to evolve ( Tosches , 2017 ) and has been demonstrated in the case of vocal learning ( Chakraborty and Jarvis , 2015 ) and in the evolution of cerebellar nuclei ( Kebschull et al . , 2020 ) . In these duplicated mosquito circuits , fruitless function would have switched from promoting mating to inhibiting host-seeking in males . We speculate that both possibilities allow for fruitless to control both sex-specific host-seeking and mating behaviors , and identification and molecular profiling of the fruitless cells controlling host-seeking and mating will help distinguish between these models . Our work highlights fruitless as a potential means to investigate the circuit basis of Aedes aegypti host seeking , a behavior that is responsible for infecting millions of people with life-threatening pathogens . Aedes aegypti wild-type laboratory strains ( Liverpool-IB12 ) were maintained and reared at 25–28°C , 70–80% relative humidity with a photoperiod of 14 hr light: 10 hr dark ( lights on at 7 a . m . ) as previously described ( DeGennaro et al . , 2013 ) . All behavioral assays were performed at these conditions of temperature and humidity . Adult females were blood-fed on mice for stock maintenance and on human subjects for initial stages of mutant generation . Anopheles gambiae ( G3 strain ) , Wyeomyia smithii ( PB strain ) , Toxorhynchites amboinensis , and Culex quinquefasciatus ( JHB strain ) were reared in similar conditions , following previously described protocols for each species ( Bradshaw et al . , 2018; Werling et al . , 2019; Zhou et al . , 2014 ) . Adult mosquitoes of each species were provided constant access to 10% sucrose . Seven- to 14-day-old mosquitoes of each species were cold-anesthetized and kept on ice for up to 1 hr or until dissections were complete . Brains were dissected in ice-cold RNase-free phosphate-buffered saline ( PBS ) ( Invitrogen AM9625 ) on ice , moved into a microfuge tube with forceps , and immediately snap frozen in a cold block ( Simport S700-14 ) chilled to −80°C on dry ice . Each sample group was dissected in parallel to avoid artefacts and batch effects , and five brains were used per sample . Dissected tissue was stored at −80°C until RNA extraction . RNA extraction was performed using the PicoPure Kit ( ThermoFisher #KIT0204 ) following the manufacturer’s instructions , including DNase treatment . Samples were run on a Bioanalyzer RNA Pico Chip ( Agilent #5067–1513 ) to determine RNA quantity and quality . Libraries were prepared using the Illumina TruSeq Stranded mRNA kit #20020594 , following manufacturer’s instructions . Library quantity and quality were evaluated using High Sensitivity DNA ScreenTape Analysis ( Agilent #5067–5585 ) prior to pooling . Barcoded samples from all non-Aedes tissues were pooled in an equal ratio before distributing the pool across two sequencing lanes . Sequencing was performed at The Rockefeller University Genomics Resource Center on a NextSeq 500 sequencer ( Illumina ) . All reads were 2 × 150 bp . Data were de-multiplexed and delivered as fastq files for each library . Sequencing reads have been deposited at the NCBI Sequence Read Archive ( SRA ) under BioProject PRJNA612100 . Reads from individual Aedes libraries were mapped to the AaegL5 genome ( Matthews et al . , 2018 ) using STAR version 2 . 5 . 2a with default settings ( Dobin et al . , 2013 ) . Raw counts were used for differential splicing analysis in Aedes aegypti using DEXSeq version 1 . 32 . 0 ( Anders et al . , 2012 ) as per author instructions . For the other mosquito species without genomes or incomplete genome annotations , we assembled sex-specific de novo transcriptomes using Trinity version 2013-03-25 with default settings ( Grabherr et al . , 2011 ) . We then searched for fruitless orthologues in each species using BLAST 2 . 6 . 0 ( Altschul et al . , 1990 ) , aligned hits to Aedes aegypti fruitless P1 transcripts using MacVector version 15 . 0 . 3 , and picked the best match for each exon , species , and sex . Fruitless exon sequences are found in Figure 1—source data 1 . The fruitless gene was targeted using CRISPR-Cas9 methods as previously described ( Kistler et al . , 2015 ) . Gene-targeting reagents were injected into wild-type Liverpool-IB12 embryos at the Insect Transformation Facility at the University of Maryland Institute for Bioscience and Biotechnology Research . For each line , either 2000 or 1000 embryos were injected with 600 ng/µL plasmid , 300 ng/µL Cas9 protein ( PNABio CP01-200 ) , and 40 ng/µL sgRNA . Proper integration was confirmed in each strain using polymerase chain reaction ( PCR ) and sequencing . Animals were then back-crossed to wild-type Liverpool-IB12 for at least four generations . All homology arms for homology-directed integration were isolated by PCR using Liverpool-IB12 genomic DNA . sgRNA DNA template was prepared by annealing oligonucleotides as previously described ( Kistler et al . , 2015 ) . In vitro transcription of sgRNA template was performed using HiScribe Quick T7 kit ( New England Biolabs #E2050S ) following the manufacturer’s directions and incubating for 4 hr at 37°C . Following transcription and DNAse treatment for 15 min at 37°C , sgRNA was purified using Ampure RNAse-free SPRI beads ( Beckman-Coulter #A63987 ) and eluted in Ultrapure water ( Invitrogen #10977–015 ) . For all plasmids , fragments were generated by PCR from the indicated template with the indicated primers ( Supplementary file 2 ) and assembled using NEBuilder HiFi DNA Assembly ( NEB E5520S ) . Plasmids were transformed into NEB competent cells ( NEB C2987I ) , purified with the NucleoBond Xtra Midi Endotoxin Free kit ( Clontech 740420 . 50 ) , and sequence verified . The fruitless∆M mutant was generated in the course of attempting to generate a fruitless QF2 knock-in mutant ( Supplementary file 1 ) ( see below ) . One of the families had viable 3xP3-dsRed-positive offspring and an out-of-frame QF2 insertion , which was predicted to produce a truncated fruitless protein in males . This was the fruitless∆M mutant allele we used in the study . The fruitless∆M-tdTomato knock-in/knock-out strain was generated by inserting a cassette containing T2A followed by CsChrimson fused to the fluorescent protein tdTomato and the 3xP3-EYFP strain marker . We obtained two independent viable lines and selected one for use in this study . We used the CsChrimson:tdTomato protein expressed from the fruitless locus in fruitless∆M-tdTomato animals as a marker for fruitless expression in these studies . CsChrimson is a red-light-activated cation channel , and we originally generated this animal with the intention of optogenetically manipulating behavior . However , CsChrimson:tdTomato intrinsic fluorescence was not visible under a confocal microscope , even at high laser intensities , and required immunofluorescent amplification in all our images . When animals were fed with retinal , the necessary co-factor which was absent in all other experiments , and we attempted to substitute human odor and CO2 with optogenetic activation of fruitless+ neurons in a blood-feeding assay , we did not observe increased feeding or any other behaviors ( preliminary data not shown ) . Although we were unable to see evidence of CsChrimson activity in these optogenetic experiments , potential background levels of CsChrimson-driven activation of fruitless-expressing neurons is an important concern to rule out when considering the behavioral data in this paper . We note that animals were not fed trans-retinal , the necessary cofactor for CsChrimson activity . Fruitless∆M-CsChrimson-tdTomato/+ males and females were able to mate normally , and fruitless∆M-CsChrimson-tdTomato/+ females show normal blood-feeding and egg-laying behavior ( Figure 5—figure supplement 1A–C ) . We have not examined fruitless∆M /∆M animals due to the difficulty of obtaining these animals without molecular genotyping of each individual . However , given the weak expression of CsChrimson and the robust behavior of heterozygote animals , we consider it unlikely that this allele is significantly affecting mosquito behavior . We attempted to generate fruitless P1-specific driver lines by knocking in a cassette containing the ribosomal-skipping peptide T2A followed by the transcriptional activator QF2 ( Riabinina et al . , 2015 ) , with 3xP3-dsRed as an insertion marker as previously described ( Matthews et al . , 2019 ) . In this knock-in/knock-out strain , we aimed to disrupt the fruitless gene as well as generate a driver line that would allow us to label and manipulate fruitless-expressing neurons . We recovered seven independent 3xP3-dsRed positive G1 families . However , all females with one copy of the correct integration did not blood-feed after many attempts using multiple different human hosts . Males with one copy of this insertion did not mate with wild-type females . Since blood-meals are required for Aedes aegypti egg-development , this line could not be maintained . We next tried to knock-in the weaker QF2w transcriptional activator , and recovered six independent families , all of which showed the same blood-feeding and mating defects ( Supplementary file 1 ) . We speculate that toxicity of QF2 or QF2w may affect the function or viability of fruitless-expressing neurons , leading to the behavioral defects we observed . The cause of Q-system toxicity , even attenuated from Q to QF2 to QF2w is unknown ( Riabinina et al . , 2015 ) . We speculate that this toxicity is unrelated to the fruitless locus , because fruitless∆M/+ animals had no phenotype as heterozygotes , unlike fruitless∆QF2/+ and fruitless∆QF2w/+ animals . To ask if fruitless protein was sufficient to inhibit host-seeking behavior in females , we attempted to force females to express male fruitless protein by deleting the female exon of fruitless P1 transcripts and forcing male fruitless splicing in female brains ( Supplementary file 1 ) . For the fruitless∆F line , embryos were injected with 300 ng/µL Cas9 protein , 125 ng/µL oligonucleotide with template repairing the splice site , 40 ng/µL each of two sgRNAs targeting the beginning and end of the female-specific exon . We recovered multiple G1 animals with the correct integration , as verified by PCR and sequencing . Male fruitless splicing in fruitless∆ females was verified with reverse-transcription PCR ( data not shown ) . G2 fruitless∆F females did not fully blood-feed or lay eggs even though they were successfully inseminated by wild-type males ( Figure 5—figure supplement 2D–I ) . It was therefore impossible to maintain these lines . We do not know if the blood-feeding defect was due to a failure to respond to the host or some other behavioral or anatomical defect . Since fruitless is tightly linked to the male-determining locus , it was not an option to maintain this targeted allele in males . Integrations on the male chromosome would yield ~1/500 females with the recombinant allele , and integrations on the female chromosome yield inviable females and rare recombinant males . In either scenario , the fruitless∆F insertion is unmarked and would need to be followed by PCR genotyping . We also attempted to generate a line where we both deleted the female fruitless exon and knocked-in an intronic 3xP3 fluorescent marker , which would allow us to maintain this line in males and use the marker to select rare recombinants for behavioral analysis . However , females with this integration did not have any behavioral phenotypes , suggesting that the intronic 3xP3 marker interfered with regular fruitless splicing in both males and females . These difficulties precluded any further investigation of the phenotype of expressing full-length fruitless protein in females . Dissection of adult brains and immunostaining was carried out as previously described ( Jové et al . , 2020; Matthews et al . , 2019 ) . Six- to 14-day-old mosquitoes were anesthetized on ice and decapitated . Heads were fixed in 4% paraformaldehyde ( Electron Microscopy Sciences 15713 s ) , 1X Ca+2 , Mg+2 free PBS ( Thermo 14190144 ) , 0 . 25% Triton X-100 ( Sigma 93443 ) , and nutated for 3 hr at 4°C . Brains were then dissected and placed in cell-strainer caps ( Falcon #352235 ) in a 24-well plate . All subsequent steps were performed on a low-speed orbital shaker . Brains were washed for 15 min at room temperature in 1x PBS with 0 . 25% Triton X-100 ( 0 . 25% PBT ) at least six times . Brains were permeabilized with 4% Triton X-100 with 2% normal goat serum ( Jackson ImmunoResearch #005-000-121 ) in PBS at 4°C for 2 days . Brains were washed for 15 min at least sixtimes with 0 . 25%PBT at room temperature . Brains were incubated in 0 . 25% PBT plus 2% normal goat serum with primary antibodies at the following dilutions: rabbit anti-RFP ( Rockland 600-401-379 ) 1:200 and mouse anti-Drosophila melanogaster Brp ( nc82 ) 1:5000 . The nc82 hybridoma developed by Erich Buchner of Universitätsklinikum Würzburg 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 52242 . Primary antibodies were incubated for 2 nights at 4°C and then washed at least six times for 15 min with 0 . 25% PBT at room temperature . Brains were incubated with secondary antibody for 2 nights at 4°C with secondary antibodies at 1:500% and 2% normal goat serum . Secondary antibodies used were goat anti-rabbit Alexa Fluor 555 ( Thermo A32732 ) and goat anti-mouse Alexa Fluor 647 ( Thermo A-21235 ) . Brains were then washed for 15 min at least six times with 0 . 25% PBT at room temperature and mounted in Slowfade Diamond ( Thermo S36972 ) using #1 . 5 coverslips as spacers before confocal imaging . This protocol was adapted from a Drosophila embryo staining protocol ( Manning and Doe , 2017 ) . Six- to 10-day-old mosquitoes were anesthetized , decapitated , and placed in 1 . 5 mL 5 U/mL chitinase ( Sigma C6137 ) and 100 U/mL chymotrypsin ( Sigma CHY5S ) in 119 mM NaCl , 48 mM KCl , 2 mM CaCl2 , 2 mM MgCl2 , 25 mM HEPES buffer on ice . Male heads were incubated for 5 min on a ThermoMixer ( Eppendorf 5382000023 ) , and 25 min in a rotating hybridization oven , and female heads were incubated for 10 min on the ThermoMixer and 50 min in rotating oven , all at 37°C . Heads were then rinsed once and fixed in 4% paraformaldehyde , 1X Ca+2 , Mg+2 free PBS , and 0 . 25% Triton X-100 for 24 hr at room temperature on a rotator . All subsequent 4°C steps used a nutator , and room temperature steps used a rotator . Heads were washed for 30 min at room temperature at least three times in 1X PBS with 0 . 03% Triton X-100 ( 0 . 03% PBT ) . Antennae were then dissected into 0 . 5-mL microfuge tubes and dehydrated in 80% methanol/20% DMSO for 1 hr at room temperature . Antennae were washed in 0 . 03% PBT for 30 min at room temperature , and blocked/permeabilized in 1X PBS , 1% DMSO ( Sigma 472301 ) , 5% normal goat serum , 4% Triton X-100 for 24 hr at 4°C . Antennae were washed for 30 min at least five times with 0 . 03% PBT , 1% DMSO at room temperature , and then moved to primary antibody in 1X PBS , 1% DMSO , 5% normal goat serum , 0 . 03% Triton X-100 for 72 hr at 4°C . Primary antibodies used were mouse anti-Apocrypta bakeri Orco monoclonal antibody #15B2 ( 1:50 dilution , gift of Joel Butterwick and Vanessa Ruta ) , and rabbit anti-RFP ( 1:100 , Rockland 600-401-379 ) . Orco monoclonal antibody specificity was verified in Aedes aegypti by staining orco mutant antennae , which showed no staining ( data not shown ) . Antennae were washed for 30 min at least five times with 0 . 03% PBT , 1% DMSO at room temperature , and then washed overnight in the same solution . Antennae were then moved to secondary antibody ( 1:200 ) and DAPI ( 1:10000 , Sigma D9542 ) in 1X PBS , 1% DMSO , 5% normal goat serum , 0 . 03% Triton X-100 for 72 hr at 4°C . Secondary antibodies used were goat anti-mouse Alexa Fluor 488 ( Thermo A-11001 ) and goat anti-rabbit Alexa Fluor 555 Plus ( Thermo A32732 ) . Antennae were washed for 30 min at least five times with 0 . 03% PBT , 1% DMSO at room temperature , and then washed overnight in the same solution . Antennae were rinsed in 1X PBS , rinsed three times in Slowfade Diamond ( Thermo S36972 ) , and mounted in Slowfade Diamond . Images were acquired with a Zeiss Axio Observer Z1 Inverted LSM 880 NLO laser scanning confocal microscope ( Zeiss ) with either 25x/0 . 8 NA ( whole brains ) or 40x/1 . 4 NA ( antennal lobes , antennae ) immersion-corrected objective at a resolution of 1024 × 1024 or 2048 × 2048 ( brains ) or 3024 × 1024 ( antennae ) pixels . Confocal images were processed in ImageJ ( NIH ) . This assay was performed as described previously ( DeGennaro et al . , 2013 ) . Briefly , for each trial , 20 adult mosquitoes were sorted under cold anesthesia ( 4°C ) and placed in a cage and allowed to acclimate for 30 min . A human arm was placed 2 . 5 cm from one side of a standard 28 × 28×28 cm cage . Mosquitoes could not directly contact the human arm . A Logitech C920s HD Pro Webcam was positioned to take images of mosquitoes responding to the human arm . Trials ran for 10 min and images were acquired at a rate of 1 frame/s . To quantify mosquito responses , we manually counted the number of mosquitoes resting on the lower portion of the screen closest to the human arm . Mosquitoes were cold-anesthetized , and 20 mosquitoes were sorted into 250 mL bottles covered with a taut net secured with rubber bands . They were allowed to acclimate for 24 hr with access to water through cotton balls . The following meals were presented: water , 10% sucrose , or sheep’s blood ( Hemostat DSB100 ) supplemented with 1 mM ATP ( Sigma A6419 ) . Meals were warmed to 45°C before being used in the assay . 10 mL of a given meal was pipetted into the bottle caps , animals were activated with a 30 s pulse of 4% CO2 , and bottles were inverted on top of the caps . Mosquitoes were allowed to feed on each meal through the net for 10 min and were then anesthetized at 4°C and scored as fed if any level of feeding was observed , as assessed by visual inspection of the abdomen of the animal . Mosquitoes were separated by sex at the pupal stage and sex was confirmed within 24 hr of eclosion . For each trial , 10 female Liverpool-IB12 virgin mosquitoes were crossed to 11 virgin male mosquitoes of either Liverpool-IB12 or fruitless∆M/fruitless∆M-tdTomato genotype in a bucket cage for 24 hr , with access to 10% sucrose . Mosquitoes were then anesthetized at 4°C , females separated from males , and female spermathecae were dissected to score for insemination as a sign of successful mating ( Degner and Harrington , 2016 ) . Control virgin females were dissected in parallel to verify absence of insemination . Details of Quattroport fabrication and operation are available at https://github . com/VosshallLab/Basrur_Vosshall2020 . Briefly , the Quattroport consists of four tubes , each with its own stimulus box , trap , and mosquito start chamber . There are adjustable gates between each chamber . The stimulus was placed upstream of a trap , and mosquitoes are prevented from contacting the stimulus by a mesh barrier . In each trial , four stimuli were run in parallel , with the positions of stimuli randomized and rotated between each trial . Air was filtered and pumped into each box , either at a final concentration of 1% CO2 ( for host-seeking assays ) or at ambient CO2 ( honey-seeking assays ) . For all assays , ~20 mosquitoes were sorted and placed into canisters the day of behavior . Mosquitoes were allowed to acclimate in the assay for 10 min , then exposed to the stimulus for 30 s , after which gates were opened and animals allowed to fly for 5 min . After this time , gates were closed and mosquitoes were counted to quantify the percent of mosquitoes in the trap . For honey assays , 3- to 7-day-old mosquitoes were fasted for 24 hr before the experiment by replacing 10% sucrose with a water source . CO2 was not added for honey assays . Either 1 g of leatherwood honey ( Tasmanian Honey Company , Tasmania , Australia ) or glycerol ( Sigma G5516 ) was applied to a 55 mm diameter Whatman filter paper circle ( GE Healthcare , Buckinghamshire , UK ) and placed in a Petri dish . For host-seeking assays , mosquitoes were allowed access to sucrose before the experiment . A final concentration of 1% CO2 was supplied in the airstream for the duration of the 5 min 30 s assay in all host-seeking assays ( for both human forearm/odor stimuli and blank/unworn nylon controls ) . For live human host-seeking assays , a human subject placed their forearm on an acrylic box , exposing a 2 . 5 × 5 cm rectangle of skin to the airstream . For human odor host-seeking assays , the same human subject wore a tan nylon sleeve ( L'eggs Women's Comfortable Everyday Knee Highs Reinforced Toe Panty Hose , modified with scissors to remove the toe area ) on their forearm . A second black nylon sleeve was worn on top of the tan nylon odor sampling sleeve to protect it from external odors . After 6 hr of continuous wear , the black nylon sleeve was discarded , and the tan nylon sleeve was frozen at −20°C . Nylons were used within 1 week of being worn . On the day of the assay , nylons were thawed for at least 1 hr at room temperature . A 10 × 14 cm piece of the sleeve was presented with the skin-contacting surface facing upward in the stimulus box along with CO2 . Unworn nylons were similarly frozen , thawed , and cut to serve as negative controls . Experiments were performed as previously described ( Corfas and Vosshall , 2015; McMeniman et al . , 2014 ) . Briefly , 45–50 mosquitoes were fasted for 3 hr before the experiment by replacing 10% sucrose with a water source and were then transferred into a custom-made Plexiglass box ( 30 × 30×30 cm ) , with carbon-filtered air pumped continuously into the box via a diffusion pad installed on the ceiling of the enclosure . All stimulus periods lasted 3 min and were presented on a single Peltier element ( 6 × 9 cm , Tellurex ) covered with a piece of standard white letter-size printer paper ( NMP1120 , Navigator ) cut to 15 × 17 cm and held taut by a magnetic frame . CO2 pulses ( 20 s , to >1000 ppm above background levels ) were added to the air stream and accompanied all stimulus period onsets . Mosquito landings on the Peltier were monitored by fixed cameras ( FFMV-03M2M-CS , Point Grey Research ) with images acquired at 1 Hz . Images were analyzed using custom MATLAB scripts to count mosquito landings within a fixed target region . Mosquito occupancy on the Peltier was quantified during seconds 90–180 of each stimulus period . All statistical analyses were performed using GraphPad Prism Version 8 . Data collected as percent of total are shown as mean ± s . e . m . Details of statistical methods are reported in the figure legends . Preliminary experiments were used to assess variance and determine sample sizes before carrying out experiments . Typically , sample sizes were n = 6–14 groups of 10-20 mosquitoes in behavioral assays . We used similar sample sizes for all experiments where the same variable was being compared . No data were excluded from this study . Since mosquito behavior is variable , all olfactometer experiments with a human arm were carried out repeatedly to assess the effect of external environmental conditions on behavior . No experiments were performed on days when < 40% of wild-type females responded to a live human arm . No data met these exclusion criteria . All attempts at replication over multiple days were successful . We carried out all experiments with different groups of animals hatched up to 4 weeks apart , and over multiple days . Several experiments were carried out repeatedly over the course of this study , namely the wild-type female response to a live human arm in Quattroport olfactometer assays . These results were robust and reliable over the course of the many years it took to complete this study . For all experiments , mosquitoes from a cage were randomly selected and sorted by sex into groups for behavioral assays . All stimuli and genotypes were interspersed , and positions were randomized when possible . Every experiment involved replicates collected over multiple days , to ensure that there was no effect of daily environmental or experimental conditions . We also collected a similar sample size for each variable every time the experiment was run , to ensure no effect of external conditions . Blinding to genotype was performed in the heat-seeking assays . The experimenter was not blinded to genotype in the host-seeking assays because the mutant phenotype we describe is so robust it was impossible to be blinded in these assays . All raw data are provided in the accompanying Source Data files . Plasmids are available at Addgene ( #141099 , #141100 ) . RNA-seq data are available in the Short Read Archive at Genbank ( Bioproject: PRJNA612100 ) . Details of Quattroport fabrication and operation are available at Github: https://github . com/VosshallLab/Basrur_Vosshall2020 .	Sexual dimorphism is a phenomenon among animals , insects and plants where the two sexes of a species show differences in body size , physical features or colors . The bushy mane of a male lion , for example , is nowhere to be seen on a female lioness , and only male peacocks have extravagant tails . Most examples of sexual dimorphism , such as elaborate visual displays or courtship behaviors , are linked to mating . However , there are a few species where behavioral differences between the sexes are not connected to mating . Mosquitoes are an example: while female mosquitoes feed on humans , and are attracted to a person’s body heat and odor , male mosquitoes have little interest in biting humans for their blood . Therefore , female mosquitoes are the ones responsible for transmitting the viruses that cause certain blood-borne diseases such as dengue fever or Zika . Determining which genes are linked to feeding behaviors in mosquitoes could allow researchers to genetically engineer females so they no longer bite people , thus stopping the spread of these diseases . Unfortunately , the genes that control mosquito feeding behaviors have not been well studied . In other insects , some of the genes that control mating behaviors that depend on sex have been identified . For example , a gene called fruitless controls courtship behaviors in male flies and silkworms , and is thought to be the ‘master regulator’ of male sexual behavior across insects . Yet it remains to be seen whether the fruitless gene has any effect in mosquitoes , where sex differences relate to feeding habits . To investigate this , Basrur et al . removed the fruitless gene from Aedes aegypti mosquitoes . The genetically altered male mosquitoes became unable to mate successfully , but – similar to unmodified males – still preferred sugar water over blood when feeding . Unlike unmodified males , however , the male mosquitoes lacking fruitless were attracted to the body odor of a person’s arm ( like females ) . These results reveal that fruitless , a gene that controls sex-specific mating behaviors in other insects , controls a sex-specific feeding behavior in mosquitoes . The fruitless gene , Basrur et al . speculate , likely gained this role controlling mosquito feeding behavior in the course of evolution . More research is required to fully understand the effects of the fruitless gene in male and female mosquitoes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience", "genetics", "and", "genomics" ]	2020	Fruitless mutant male mosquitoes gain attraction to human odor
Phototrophic microorganisms adjust photosystem synthesis in response to changes in light intensity and wavelength . A variety of different photoreceptors regulate this process . Purple photosynthetic bacteria synthesize a novel photoreceptor AerR that uses cobalamin ( B12 ) as a blue-light absorbing chromophore to control photosystem synthesis . AerR directly interacts with the redox responding transcription factor CrtJ , affecting CrtJ’s interaction with photosystem promoters . In this study , we show that AerR is translated as two isoforms that differ by 41 amino acids at the amino terminus . The ratio of these isoforms was affected by light and cell growth phase with the long variant predominating during photosynthetic exponential growth and the short variant predominating in dark conditions and/or stationary phase . Pigmentation and transcriptomic analyses show that the short AerR variant represses , while long variant activates , photosynthesis genes . The long form of AerR also activates many genes involved in cellular metabolism and motility . Purple photosynthetic bacteria are an interesting group of diverse bacteria that preferentially synthesizes a photosystem under anaerobic conditions ( Imhoff and Hiraishi , 2015 ) . This is partially due to the fact that photopigments can generate singlet oxygen as a byproduct of light excitation ( Berghoff et al . , 2011; Ziegelhoffer and Donohue , 2009; Pospísil , 2009 ) . Thus , many purple photosynthetic bacteria repress synthesis of their photosystem in response to the presence of oxygen , and in this case , generate energy using respiration ( Crofts et al . , 1983; Swem et al . , 2006; Wu and Bauer , 2010 ) . In species where it has been examined , there is an aerobic repressor called CrtJ ( also called PpsR in some species ) ( Elsen et al . , 2005; Gomelsky and Kaplan , 1995; Penfold and Pemberton , 1991; Ponnampalam et al . , 1995 ) that senses oxygen via oxidation of a Cys present in CrtJ’s DNA binding domain ( Cheng et al . , 2012; Masuda and Bauer , 2002; Masuda et al . , 2002 ) . Initially , it was thought that CrtJ/PpsR only binds and represses photosystem promoters under aerobic conditions ( Elsen et al . , 2005; Ponnampalam et al . , 1995; Cheng et al . , 2012; Masuda and Bauer , 2002; Masuda et al . , 2002 ) . However , a recent study showed that CrtJ is bound to a photosystem promoter ( bchC ) under both aerobic and anaerobic conditions ( Fang and Bauer , 2017 ) . Interestingly , the extent of DNA protection by CrtJ to this promoter is significantly altered in vivo under aerobic verses anaerobic conditions ( Fang and Bauer , 2017 ) . In addition to environmental changes in oxygen tension , phototrophic microorganisms are subjected to daily variations in light intensity and wavelength . To optimize light energy absorption for photosynthesis , and to better adapt their physiology to dark periods , these cells also control photosystem synthesis using various photoreceptors ( Gomelsky and Hoff , 2011; Masuda , 2013 ) . A variety of chromophores such as FMN ( used in Light , Oxygen or Voltage , LOV domain containing proteins ) , FAD ( used in sensors of Blue-Light Using FAD , BLUF domain containing proteins ) , 4-hydroxycinnamic acid ( used in Photoactive Yellow Protein , PYP ) , phytochromobilin ( used in bacteriophytochromes ) and cobalamin ( B12 ) are used as light absorbing chromophores by photoreceptors present in purple bacteria ( Masuda and Bauer , 2002; Gomelsky and Hoff , 2011; Masuda , 2013; Gomelsky and Klug , 2002; Hendriks , 2009; Giraud et al . , 2002; Jiang et al . , 1999; Cheng et al . , 2014; Cheng et al . , 2016 ) . Photoreceptor proteins need to either contain a linked output domain or to interact with other proteins that transmit photoreceptor sensed light signals into physiological changes . In some purple bacteria , photoreceptors can interact with transcription factors to alter the transcription of genes ( Hendriks , 2009 ) . For example , in the purple non-sulfur alpha-proteobacterium Rhodobacter sphaeroides , the interaction of the FAD-binding blue light sensor AppA with the transcription factor PpsR is well studied ( Masuda and Bauer , 2002; Giraud et al . , 2002; Gomelsky and Kaplan , 1997; Metz et al . , 2012 ) . In Rhodopseudomonas palutris , PpsR repression is also regulated via an interaction with a red-light sensing bacteriophytochrome ( Braatsch et al . , 2006; Braatsch et al . , 2007 ) . These light dependent interactions have also been used as optogenetic tools ( Kaberniuk et al . , 2016; Redchuk et al . , 2017 ) . In all sequenced purple photosynthetic bacterial genomes , there is a gene coding for a cobalamin binding photoreceptor called aerR that is located upstream of ppsR/crtJ ( Cheng et al . , 2014; Vermeulen and Bauer , 2015 ) . The discovery that AerR binds cobalamin in a light-dependent manner explained a long-standing observation that this group of bacteria needs cobalamin to undergo synthesis of their photosystem ( Cheng et al . , 2014 ) . However , it remains unknown how AerR and CrtJ coordinate global changes in R . capsulatus physiology in response to changes in redox and light . We have shown that AerR can directly interact with CrtJ both in vivo and in vitro ( Fang and Bauer , 2017 ) and that AerR can affect the DNA binding characteristics of CrtJ at target promoters ( Fang and Bauer , 2017 ) . In this study , we demonstrate that R . capsulatus uses two offset promoters , and an alternative internal Leu start codon , to synthesize long and short isoforms of AerR . We further show that the long AerR variant converts CrtJ into an activator while the short AerR variant converts CrtJ into a repressor of photosystem gene expression . We also show that the long AerR variant coordinates synthesis of photosystem with expression of a large variety of additional genes that affect cellular metabolism and motility . We analyzed the in vivo presence of AerR by Western blot analysis using an R . capsulatus strain in which a 3xFLAG-tag was chromosomally inserted at the 3’ terminus of the aerR open reading frame . A previous study established that a 3xFLAG-tag at the carboxyl end of AerR did not measurably affect AerR activity ( Cheng et al . , 2014 ) . As indicated in Figure 1B , Western blot analysis shows the unexpected presence of two AerR-FLAG isoforms with one at ~30 kDa ( LAerR ) and the other at ~25 kDa ( SAerR ) based on electrophoretic mobility . The larger 30 kDa isoform present in R . capsulatus extracts co-migrates with full-length ( based on mass-spectrophotometry analysis ) AerR-FLAG protein that was expressed and purified from Escherichia coli ( Cheng et al . , 2014 ) . This indicates that the 25 kDa isoform ( SAerR ) either represents a proteolytic processing event or an alternative translational start site . Interestingly , steady-state amounts of these two AerR isoforms consistently changed depending on what point in the R . capsulatus cultivation growth curve that AerR-FLAG isoforms were analyzed ( Figure 1A , B ) . When dark semi-aerobically grown cells were harvested at different points in mid-exponential phase , these two AerR isoforms are present at approximately equal amounts . However , when harvested in stationary phase , the 25 kDa SAerR isoform was constantly the dominant form . When grown under anaerobic photosynthetic ( constant illumination ) conditions , the large isoform predominates in early and mid-exponential phase and then decreases in concentration as the culture transitions to late exponential and stationary phases , leading to a predominance of the small isoform ( Figure 1—figure supplement 1 ) . In addition to growth culture changes in the ratio of LAerR and SAerR , a change of the LAerR/SAerR ratio is also observed upon a light shift . For example , switching dark exponentially grown cells at an optical density of 0 . 3 to anaerobic illuminated conditions , results in a significant increase in LAerR relative to a parallel fraction of cells that remained in the dark ( Figure 1C ) . Conversely , switching anaerobic photosynthetically cells from light to dark conditions , reduced the level of LAerR ( Figure 1C ) . The difference in electrophoretic mobility between LAerR and SAerR indicates that there are ~40–50 fewer amino acid residues in the N-terminal region of the SAerR isoform . We first addressed whether SAerR is a product of proteolytic truncation of LAerR by constructing an LAerR frameshift mutation via the insertion of one nucleotide immediately downstream of the LAerR Met initiation codon ( Met1a ) ( Figure 2A ) . The resultant aerR-Met1a construct , containing a carboxyl FLAG tag sequence , was then ectopically expressed on a plasmid using its own native promoter . Western blot analysis demonstrated that the aerR-Met1a plasmid only expressed the 25 kDa SAerR isoform ( Figure 2A ) . This indicates that the small isoform is not derived by proteolytic processing of LAerR , and must be generated from a second internal translation initiation site present in the LAerR coding sequence . There are several additional in-frame Met codons in the 5’ aerR coding sequence ( Met35 , Met49 and Met61 ) that could potentially function as a second initiation codon for SAerR . We accessed the possibility that one of these internal Met codons was functioning as an internal SAerR initiation codon by constructing three deletions within the LAerR coding region ( Figure 2A ) . The first deletion extended from the LAerR initiation codon to Met35 , the second deletion extended from the LAerR initiation codon to Met49 and the third deletion extended from the LAerR initiation codon to Met61 . Each of these AerR deletion constructs also contained a C-terminal FLAG epitope and all were ectopically expressed from the aerR native promoter in R . capsulatus on a plasmid . As shown in Figure 2A , the Met1 through Met35 truncated strain still expressed two forms with the 30 kDa isoform shifted to ~27 kDa as a result of the deletion/truncation while the 25 kDa isoform showed same mobility as the WT control . The second deletion from Met1 through Met49 and the third deletion that extended to Met61 lost expression of both the longer and shorter isoforms indicating that the SAerR isoform starts between the M35 and M49 codons . One nucleotide frame shift insertions were subsequently created at several points between the Met35 and Met49 codons to further determine the SAerR initiation site ( Figure 2B ) . While M35+1 ( one nucleotide inserted after M35 codon ) and V38+1 ( one nucleotide inserted after V38 codon ) strains still expressed the 25 kDa isoform , strains with nucleotide insertions after the L41 , T43 , and V44 codons did not express SAerR ( Figure 2B ) . This result indicates the second initiation starts at the A39 , E40 , or L41 codon . To find the exact location of the SAerR initiation codon , we constructed plasmids that contained the aerR promoter region and a partial aerR sequence fused to the bchE open reading frame that also had a FLAG epitope tag for use as a reporter ( Figure 2C ) . To focus on identification of the second initiation codon , these constructs also contained a one base insertion downstream of the LAerR initiation codon , resulting in disruption of LAerR peptide synthesis . Three constructs were made; one having the FLAG-bchE gene fused downstream of the A39 aerR codon , a second construct with FLAG-bchE fused to aerR codon E40 , and a third with FLAG-bchE fused to L41 . When these aerR-bchE-FLAG reporter plasmids were introduced into R . capsulatus , only the plasmid that had bchE-FLAG fused to L41 resulted in expression of a BchE-FLAG protein ( Figure 2C ) . This indicates that the L41 codon ( CTG ) is functioning as a second start site for the SAerR isoform . Six through sixteen bases upstream of L41 is a putative ribosome binding site ( 5’-GAAcGGAGtgG-3’ ) that exhibits considerable complementarity with the cognate R . capsulatus 16S rRNA sequence ( 3’-CUUuCCUCcaC-5’ ) . Finally , we also analyzed whether there was an additional internal transcription start site for SAerR by amplification of 5’ mRNA end ( s ) using cDNA 5’ RACE ( Rapid Amplification of cDNA Ends ) ( GeneRacer kit , Invitrogen ) . Note that a terminator exonuclease ( Epicentre ) treatment step was introduced in the 5' RACE experiments to degrade processed RNA , allowing only the selection for primary transcripts . A 24 base cDNA primer was also designed to be complementary to a region of the aerR transcript that is 299 bases downstream of the LAerR initiation codon . Consequently , this primer should amplify both a large transcript responsible for transcribing LAerR as well as a shorter transcript responsible for transcribing SAerR , should such a shorter transcript exist . As shown in Figure 2—figure supplement 1A , two transcription start sites were indeed detected by sequencing: one with a start site 25 bp upstream of the LAerR initiation codon and a second initiating at the 8th codon of LAerR ( Figure 2—figure supplement 1A ) . The longer transcript could potentially be responsible for both LAerR and SAerR expression while the shorter transcript would be dedicated to the expression of SAerR . Inspection of the sequence upstream of these transcript initiation sites shows the presence of putative promoter recognition sequences with the SAerR −35 and −10 regions exhibiting good sequence similarity to previously characterized R . capsulatus promoters ( Swem et al . , 2001 ) . However , this is not the case with the upstream LAerR promoter region which shows poor promoter sequence conservation indicating that the upstream LAerR promoter may utilize an alternative sigma subunit ( Figure 2—figure supplement 1A ) . Furthermore , there are FnrL binding sites located near both promoter regions , which is not unexpected given that FnrL is known to regulate AerR expression ( Kumka and Bauer , 2015 ) and that ChIP-seq analysis also has revealed FnrL binds at both of these promoter regions in vivo ( Kumka and Bauer , 2015 ) . To effectively evaluate the function of each AerR isoform required that strains be constructed that only express the long or short isoforms of AerR . To disrupt SAerR synthesis , we constructed a single silent mutation of L41 ( Leu CTG to Leu CTC ) as well as a second silent mutation that also disrupted the upstream ribosome binding site ( RBS ) ( Figure 2—figure supplement 1B ) . The RBS silent mutation changed the Glu37 codon from GGA to an alternate Glu codon GGT resulting in conversion of the GAAcGGAGtgG ribosome recognition sequence to GAAcGGtGtgG . The combination of both of these silent mutations resulted in a strain ( termed ∆SAerR ) that produces normal amounts of LAerR without any detectable amounts of SAerR ( Figure 2—figure supplement 1C ) . To obtain a strain that only synthesizes SAerR , we constructed a strain that had a one nucleotide chromosomal insertion immediately downstream of the LAerR Met initiation codon ( Figure 2—figure supplement 1B ) . This frameshift insertion results in the generation of an in frame stop codon 15 bp downstream of the LAerR initiating Met codon resulting in the synthesis of just a five aa peptide . Western blot analysis shows that this strain ( termed ∆LAerR ) only synthesizes SAerR ( Figure 2—figure supplement 1C ) . Finally , a negative control strain was constructed ( AerR null ) that lacks both the short and long forms of AerR . This strain has a frameshift after codon L45 that results in undetectable amounts of both the LAerR and SAerR peptides ( Figure 2—figure supplement 1C ) . All of these constructions were recombined into the R . capsulatus chromosome at their native location under control of the described AerR promoters . We measured the in vivo production of photopigments on strains containing a deletion of either the long or short AerR isoforms , relative to wild-type and the aerR null mutant strains . For this analysis , cells were grown in rich PY medium under dark semi-aerobic conditions , harvested at late-exponential phase ( OD = 0 . 6 to 0 . 7 ) and analyzed for the level of pigments after organic extraction . The bar graph in Figure 3A shows that relative to wild-type cells , the strain that lacks the short isoform ( ∆SAerR ) has ~1 . 6 and 1 . 75-fold enhancement of carotenoid and bacteriochlorophyll ( Bchl ) levels , respectively . This is contrasted by the strain which lacks the long isoform of AerR ( ∆LAerR ) that exhibits significantly reduced amounts of carotenoids and Bchl ( 20% and 2% of WT , respectively ) . The amount of Bchl biosynthesis exhibited by the strain lacking LAerR is also lower than that observed with the AerR null strain ( 20% of WT ) , which is thought to undergo constitutive CrtJ mediated repression of bch gene expression ( Fang and Bauer , 2017 ) . In this growth condition , it appears that the long and short isoforms of AerR have opposite functions with SAerR involved in suppressing photopigment synthesis and LAerR involved in enhancing photopigment biosynthesis . When these same four strains were grown under anaerobic photosynthetic illuminated conditions , the ∆SAerR strain exhibited a slight 30% reduction in Bchl and carotenoid production while the ∆LAerR strain showed a more severe reduction in these pigments ( 78% and 65% reduction , respectively ) relative to WT cells ( Figure 3B ) . Photosynthetic pigment reduction observed upon an absence of LAerR is slightly lower than observed with the AerR null strain that had 72% and 63% reduction in Bchl and carotenoids relative to WT cells . When shifted from aerobic to anaerobic photosynthetic growth conditions , the WT and ∆SAerR strains both exhibited an 8 hr lag followed by comparable growth rates ( Figure 3C ) . In contrast to good photosynthetic growth by the ∆SAerR strain , the AerR null and ∆LAerR strains both exhibited a more severe lag upon the shift to photosynthetic growth conditions ( ~30 and 60 hr , respectively ) as compared to the WT strain ( Figure 3C ) . This lag in growth is presumably a result of decreased pigment biosynthesis by the ∆LAerR and AerR null strains . Studies have shown that AerR alone does not appear to contain the ability to directly interact with DNA ( Fang and Bauer , 2017; Cheng et al . , 2014 ) . However , these studies have also demonstrated that AerR does tightly interact with the photosystem regulator CrtJ in vivo and in vitro . This interaction also subsequently affects the DNA binding characteristics of CrtJ ( Fang and Bauer , 2017; Cheng et al . , 2014 ) . To further understand the role of short and long forms of AerR , we checked whether both isoforms are indeed both capable of interacting with CrtJ . We also checked the dependency of CrtJ on the observed phenotypes of strains expressing only SAerR or LAerR . In vitro binding affinities measuring the interaction of each AerR isoform with CrtJ were obtained using microscale thermophoresis ( MST ) as previously reported ( Cheng et al . , 2014 ) . The observed binding affinity of CrtJ to LAerR was slightly lower than observed with CrtJ to SAerR with an EC50 = 7 . 8 ± 1 . 2 µM for LAerR containing bound hydroxyl-Cbl ( OH-Cbl ) , 2 . 1 ± 1 µM for SAerR with bound OH-Cbl and 2 . 1 ± 0 . 8 µM SAerR with bound Adenosyl-Cbl ( Ado-Cbl ) . Thus , both AerR isoforms are indeed capable of forming a complex with CrtJ . The type of Cbl bound to SAerR also does not seem to appear affect its interaction with CrtJ . We next addressed whether LAerR and SAerR both interact with CrtJ in vivo by addressing CrtJ dependency on the observed phenotypes exhibited by strains that lacked either LAerR or SAerR . A 1 . 5-fold increase in dark semi-aerobic Bchl production has previously been reported for the crtJ deletion strain ( Cheng et al . , 2012 ) so the question we addressed is whether increased Bchl production exhibited by a CrtJ deletion mutation is dominant over the observed reduction in Bchl production by the ∆LAerR and AerR null strains when grown under dark semi-aerobic conditions . For this analysis , we constructed an in-frame deletion of crtJ in the relevant aerR mutant strains giving rise to ∆SaerR-∆crtJ , ∆LaerR-∆crtJ and null aerR-∆crtJ strains . In dark semi-aerobic growth conditions , all three strains that contained the crtJ deletion ( ∆crtJ , null aerR-∆crtJ , ∆LaerR-∆crtJ and ∆SaerR-∆crtJ ) exhibited 1 . 5 to 2-fold higher amounts of Bchl relative to that observed by the WT control ( Figure 3—figure supplement 1A ) . Interestingly , this increase in pigment production is very similar to the increase in pigment production observed when the ∆SAerR strain just expresses the LAerR isoform ( Figure 3A ) . The 1 . 5-fold increase in Bchl production over that of WT strain exhibited by the ∆LaerR-∆crtJ strain ( Figure 3—figure supplement 1 ) is also a stark contrast to the severe ~80% reduction in pigment production exhibited by the ∆LAerR strain ( Figure 3A ) . This result demonstrates that the increased pigment production exhibited by a deletion of CrtJ is dominant over reduced pigment production exhibited by the loss of LAerR . Furthermore , the observed increased pigment production exhibited by the ∆SAerR strain appears to be indistinguishable from the increased pigmentation phenotype exhibited by a deletion of CrtJ . One conclusion in comparing the results in Figure 3 and in Figure 3—figure supplement 1 is that ( i ) both LAerR and SAerR works in CrtJ dependent manner and ( ii ) an absence of SAerR results in CrtJ no longer being able to repress bacteriochlorophyll gene expression whereas a loss of the LAerR isoform leads to the inability of CrtJ to enhanced pigment production . When pigment production was analyzed under anaerobic photosynthetic conditions , we observed that the introduction of the ∆crtJ deletion into the AerR null , ∆SAerR and ∆LAerR strains largely suppresses these AerR mutant phenotypes ( Figure 3—figure supplement 1B ) indicating that the ∆crtJ phenotype is dominant over that of the ∆SAerR and ∆LAerR phenotypes under photosynthetic conditions . Finally , the severe lag in photosynthetic growth exhibited by the ∆LAerR strain and the null AerR strains ( Figure 3C ) was also suppressed by introduction of a ∆crtJ deletion ( Figure 3—figure supplement 1C ) . A previous study showed that LAerR binds cobalamin ( Cbl ) in a light-dependent manner ( Cheng et al . , 2014 ) . Specifically , it was shown that LAerR tightly binds OH-Cbl which is generated as a byproduct of light excitation of Ado-Cbl ( Cheng et al . , 2014 ) . A Cbl deficiency was also shown to result in a reduction in pigmentation and photosystem gene expression in a manner that mimics the phenotype of an AerR null mutation . Furthermore , in vitro studies have shown that LAerR forms upper and lower axial ligands with the Co metal in OH-Cbl using two histidine residues ( His10 and His145 ) . Alanine substitutions on one of these His ligands alters the Co spectrum with Ala mutations in both of these His residues abolishing the ability of LAerR to bind OH-Cbl ( Cheng et al . , 2014 ) . Given that SAerR lacks the His10 upper Co ligand , we also addressed whether SAerR is indeed capable of light-dependent binding OH-Cbl in vitro as described for LAerR . For this analysis , purified SAerR protein was incubated with several Cbl derivatives under dark or illuminated conditions followed by removal of unbound cobalamins with a desalting column ( Figure 4A ) . Spectral analysis of SAerR surprisingly showed that SAerR is able to bind all tested cobalamin derivatives ( OH-Cbl , Ado-Cbl , cyano-Cbl , and methyl-Cbl ) under both dark and illuminated conditions ( Figure 4B and Figure 4—figure supplement 1 ) . This result indicates that His10 in the amino-terminal region of LAerR appears to be responsible for the selectivity of light generated OH-Cbl . An SAerR construct incapable of binding Cbl was also constructed to evaluate Cbl dependency on SAerR activity . For this analysis , we constructed an Ala mutation at H145 in the cobalamins binding motif ( E143xH145xxG148 ) that in LAerR is known to form a lower axial ligand with the Cbl bound Co ( Cheng et al . , 2014 ) . As shown in Figure 4C , the H145A substitution largely disrupted SAerR ability to bind Cbl . To completely disrupt Cbl binding we also constructed a second mutation in Gly148 in the cobalamins binding motif to Glu . The SAerR_H145A , G148E double mutant protein no longer bound any detectable amounts of Cbl ( Figure 4C ) . To evaluate the Cbl dependency on SAerR repressor activity in vivo , this SAerR double mutant protein was also expressed in R . capsulatus cells . As shown in Figure 5 , a WT strain harboring an SAerR expression plasmid exhibited much less Bchl and carotenoid synthesis than did a WT strain without the SAerR expression plasmid ( red versus blue spectrum in Figure 5A , respectively ) . This pigment reduction mirrors the results in Figure 3 which show that a strain that just harbors SAerR has a severe reduction in pigment synthesis . In contrast , a strain harboring an SAerR_H145A , G148E expression plasmid contained only slightly less than WT amounts of Bchl and carotenoid ( orange spectrum in Figure 5A ) . This indicates that disruption of SAerR’s ability to bind B12 impairs CrtJ mediated repression activity . The growth rate observed when shifted from semi-aerobic to anaerobic photosynthetic conditions also supports this conclusion ( Figure 5B ) . Specifically , the SAerR_H145A , G148E expressing strain grew at nearly the same growth rate as the strain that did not express SAerR while the strain that expressed SAerR exhibited a significant delay in photosynthetic growth . Collectively , these results indicate that unlike LAerR , SAerR promiscuously binds Cbl with differing upper ligands in a light-independent manner and that Cbl binding has an important role for SAerR activity . That said , we do note that the Cbl binding mutant does not reach the same optical density as does the WT strain , and also has pigment levels that do not reach the same level as that of the WT strain . This suggests that apo-SAerR may also have a yet undefined role in these cells . We next explored changes in gene expression patterns in the strains lacking LAerR or SAerR by differentially comparing their transcriptomes with the transcriptome of the WT strain using RNA-seq . The heat map in Figure 6A , and the quantitated fold-changes in Supplementary file 1 , show photosynthesis gene expression changes under dark semi-aerobic conditions . Overall the ∆LAerR strain that lacks LAerR , and the AerR null mutation strain that does not contain either isoforms of AerR , both exhibited reduced photosystem gene expression profiles relative to that observed by the WT strain . However , there are several important differences . For example , the ∆LAerR strain has significantly reduced expression of the puc operon coding for light harvesting II ( LHII ) structural peptides which is not observed by the AerR null strain where puc expression is unchanged from the WT strain ( Figure 6A ) . This result indicates that SAerR has a role in repressing puc ( LHII ) expression . A second difference is that the ∆LAerR strain also has significantly reduced expression of the bchEJGP operon and reduced expression of the divergent crtA-bchIDO and crtIB operons relative to the AerR null and WT strains ( Figure 6A , Supplementary file 1 ) . This also indicates a role of SAerR in repressing expression of these operons . Note that the crtI-crtB gene products code for phytoene dehydrogenase and phytoene synthase , respectively , which are involved in the first two committed steps of carotenoid biosynthesis ( Armstrong , 1995 ) . The same is true for bchID which code for the ATPase subunits in Mg-chelatase , the first committed step in Bchl biosynthesis ( Senge and Smith , 1995 ) . Thus , the rather significant ‘super’ repression of these two divergent operons that occurs upon loss of LAerR likely causes the observed severe reduction of Bchl and carotenoids synthesis in the dark semi-aerobically grown ∆LAerR strain ( Figure 3A ) . Evidence for the involvement of CrtJ in LAerR mediated activation of these photosystem genes is also observed by the rather significant increase in expression exhibited by the ∆crtJ and ∆aerR∆crtJ double mutant strains ( Figure 6A ) . This is also congruent with the observed photopigment phenotype suppression in the SAerR and LAerR strains by the addition of a CrtJ mutation as seen in Figure 3—figure supplement 1 . Finally , in regards to the ∆SAerR strain that lacks SAerR , there is increased photosystem transcripts relative to WT cells indicating that SAerR likely has a repressing role under dark semi-aerobic growth conditions ( Figure 6A ) . When analyzing the transcriptome under illuminated anaerobic photosynthetic conditions ( Figure 6B , Supplementary file 2 ) , the transcriptome profiles again largely mimic the pigment levels observed by these strains ( Figure 3B ) . Specifically , the ∆LAerR strain that lacks LAerR shows a reduced photosystem expression profile relative to WT cells indicating that LAerR also has an important role in activating photosystem gene expression anaerobically . Again , this is particularly evident for the bchEJGP operon and the divergent crtA-bchIDO and crtIB operons . In contrast , the ∆SAerR strain lacking SAerR has only a minor reduction in photosystem gene expression relative to WT cells indicating that SAerR has a minor , or even no role , in controlling photosystem gene expression under anaerobic photosynthetic growth conditions . Additional analysis of RNA-seq results from the AerR null strain under dark semi-aerobic growth conditions revealed that this strain had only eight genes , in addition to bch , crt and photosystem structural genes ( puf , puc , puh ) , that exhibited significant differential expression relative to WT cells ( Supplementary file 3 , tab 1 ) . This indicates that the primary role of AerR under dark semi-aerobic growth conditions , a condition where SAerR predominates , is to control the expression of photosynthesis genes . However , the limited dark semi-aerobic regulatory role is contrasted by analysis under illuminated anaerobic photosynthetic conditions where deletion of AerR affects the expression of >1500 genes ( Supplementary file 3 , tab 2 ) . In this growth mode LAerR predominates and seems to have a significant role in controlling cellular physiology well beyond that of photosynthesis . We specifically addressed the involvement of the LAerR isoform in controlling global cellular physiology under photosynthetic conditions by analyzing the RNA-seq transcriptome profile in the ∆LAerR strain . As shown in Supplementary file 4 and Figure 7 , the AerR null strain that lacks both isoforms and the ∆LAerR strain that lacks only LAerR , both exhibited differential expression changes ( relative to the WT strain ) that were very similar to each other . Interestingly , in the few cases where loss of SAerR has an effect on gene expression under photosynthetic conditions , there is an inverse effect relative to that observed upon loss of LAerR . Thus , in cases where LAerR functions as an activator , SAerR appears to function as a repressor and vice versa ( Figure 7 , Supplementary file 4 ) . When assessing the role of individual genes that are regulated by LAerR photosynthetically ( Supplementary file 4 , tab subcategories and summarized in Figure 8 ) , we observed the following . A loss of the LAerR isoform results in lower expression of genes involved in such diverse cellular processes as photosynthesis , carbon fixation , chemotaxis and motility , cobalamin biosynthesis , glycolysis and TCA cycle , heme biosynthesis , ribosomal proteins and several transporters . In each these cases , LAerR appears to be functioning as an activator as its loss leads to a reduction of expression of these genes . One stark exception is an operon coding for a bacterial microcompartment that metabolizes 1 , 2 propanediol for ATP production where LAerR appears to function as a repressor as a loss of LAerR leads to a rather dramatic increase in the expression of these bacterial microcompartment genes ( Supplementary file 4 ) . In the central metabolism category , many genes involved in glycolysis are reduced in both the AerR Null and LAerR depleted strains but increased in the SAerR depleted strain ( Supplementary file 4 ) . Specifically , these results indicate that LAerR activates expression of many glycolysis enzymes such as pyruvate dehydrogenase , fructose-bisphosphate aldolase ( fba ) , glyceraldehyde-3-phosphate dehydrogenase ( gap1 ) , phosphoglycerate kinase ( pgk ) and pyruvate kinase ( pykA2 ) ( Figures 7 and 8 ) . LAerR mediated increase in glycolysis likely lead to increased synthesis of pyruvate that feeds into the TCA cycle potentially increasing synthesis of a number of important molecules such as isoprenoids , tetrapyrroles and branched-chain amino acids ( BCAA ) ( Figure 8 ) . Conversely , increased expression of pyruvate dehydrogenase in the ΔSAerR strain indicates that the SAerR isoform has a role in decreasing the flow of metabolites into these cellular processes . Interestingly , the same pattern of decreased expression by SAerR and increased expression by LAerR is observed for several genes involved in branched-chain amino acid transportation , carbon fixation mediated by form I ( cbbLS ) and form II RubisCO ( cbbM ) ( Figure 7 ) . Finally , LAerR stimulates expression of almost all the 50S and 30S ribosomal proteins , numerous genes involved in motility and chemotaxis , and numerous genes involved in cobalamin biosynthesis ( Figure 8 and Supplementary file 4 ) . This latter effect is notable as LAerR itself uses cobalamin as a cofactor in a light-dependent manner . An earlier in vitro study using small DNA templates for DNA binding analysis indicated that AerR ( corresponding to LAerR in this study ) likely functioned as an anti-repressor that dissociated CrtJ from photosystem promoters ( Cheng et al . , 2014 ) . However , more recent in vitro studies , using much larger DNA segments , demonstrated that LAerR does not disassociate CrtJ from target promoters , but instead , alters CrtJ’s interaction with the DNA template by significantly increasing the extent of the DNA that it interacts with ( Fang and Bauer , 2017 ) . Additional in vivo analysis using ChIP-seq also revealed that CrtJ does not significantly disassociate from target promoters under aerobic versus anaerobic conditions ( Fang and Bauer , 2017 ) . These results indicate that the control of photosystem gene expression by CrtJ is much more complex and nuanced than previously thought . Indeed in this study , we demonstrate that R . capsulatus synthesizes two isoforms of AerR , LAerR and SAerR , which have opposite effects on gene expression in a CrtJ dependent manner . As summarized in Figure 9 , SAerR is the predominant variant in stationary phase under dark semi-aerobic conditions . Its interaction with CrtJ promotes CrtJ mediated super repression of the bchEJGP , bchODI-crtA , crtIB and puc operons leading to reduced synthesis of the photosystem ( Figure 9 ) . While CrtJ is capable of repressing photosystem gene expression on its own , its repression without SAerR seems partial or weaker than when CrtJ is complexed with SAerR . Previous studies have shown that CrtJ cooperatively binds to target promoters at tandem CrtJ binding motifs ( TGT-N12-ACA ) ( Ponnampalam and Bauer , 1997; Elsen et al . , 1998; Ponnampalam et al . , 1998 ) . The CrtJ binding motifs are either located close together such as what occurs in the bchC promoter where they are 8 bp apart ( Ponnampalam et al . , 1998 ) , or present at more distant locations 45 to 500 bp apart such as what occurs in the puc , bchEJGP , bchODI-crtA , and crtIB promoters ( Elsen et al . , 1998 ) . In this regard , it’s interesting that our RNA-seq results show that SAerR mediated enhancement of CrtJ repression is greater at promoters where CrtJ binding motifs are more distantly located than at the bchC promoter where CrtJ binding sites are only 8 bp apart . In stark contrast to the ability of SAerR to enhanced CrtJ’s ability to promote aerobic repression , our study also indicates that LAerR , which is the predominate isoform under photosynthetic conditions , switches CrtJ to an anaerobic photosynthetic activator ( Figure 9 ) . One clue to how this may occur is provided by previous in vitro and in vivo studies with LAerR which demonstrated that LAerR can dramatically affect CrtJ binding to the bchC promoter region ( Fang and Bauer , 2017 ) . Specifically , it was observed that CrtJ alone only bound to the two tandem bchC CrtJ binding motifs . However when in the presence of LAerR , then CrtJ interacted with an extended region that spanned several hundred base pairs beyond the tandem CrtJ binding motifs ( Fang and Bauer , 2017 ) . We propose that LAerR mediated extension of CrtJ interaction to target promoters is likely responsible for switching CrtJ from an aerobic repressor to an anaerobic photosynthetic activator ( Figure 9 ) . LAerR and SAerR also have clear differences in cobalamin binding characteristics with LAerR only binding OH-Cbl in a light-dependent manner while SAerR can bind all tested biologically relevant forms of cobalamin irrespective of light exposure . This result indicates that His10 , which is only present in LAerR , is responsible for providing cobalamine specificity . LAerR also likely has a predominant role under illuminated conditions which is the growth condition that will form OH-Cbl as a product of photolysis of the upper ligand in Cbl . Conversely , the ability of SAerR to bind all tested forms of Cbl lends to its dominant role under dark conditions . Differing roles of the LAerR and SAerR isoforms is also evident from RNA-seq results which show that LAerR controls expression of many more cellular process ( photosystem synthesis and the expression of numerous glycolysis genes involved in central metabolism ) than does SAerR . In WT cells the high LAerR/SAerR ratio during exponential phase presumably enhances energy production and protein biosynthesis needed for fast cellular growth . Conversely , the low LAerR/SAerR ratio observed in cells entering stationary phase would lead to reduced synthesis of the photosystem and reduced glycolysis as these cells are not actively replicating and thus require less energy production . Finally , while LAerR likely has a direct effect on gene expression by interacting with CrtJ it also likely has an indirect role in controlling gene expression . For example , as diagramed in Figure 8 , deletion of LAerR leads to reduced synthesis of genes involved in energy production as well as reduced expression of many ribosomal genes . Indeed , reduced expression of numerous ribosomal genes in the ∆LAerR strain is puzzling as ribosomal genes show no in vivo binding of CrtJ ( Fang and Bauer , 2017 ) . One possible explanation is that the absence of LAerR in the ∆LAerR strain actually causes an energy limiting stringent like growth condition which is known to lead to reduced expression of ribosomal genes . Along this vein , the branch chain amino acid ( BCAA ) transporter genes ( rcc03426-03433 ) , that code for an importer of BCAA’s , were found to be highly directly downregulated in the AerR null and in the ∆LAerR strains ( Figure 7—figure supplement 1 ) . Furthermore , several BCAA were recently shown to stimulate the degradation of the cellular alarmone molecules ( p ) ppGpp which regulate the stringent response that is known to influence expression of ribosomal genes ( Fang and Bauer , 2018 ) . Another consequence of reduced synthesis of the photosystem by the ∆LAerR strain would be fewer electrons flowing from the photosystem to the quinone pool . Such an alteration in the redox state of ubiquinones would be sensed by the ubiquinone responding sensor kinase RegB leading to downstream alterations in global gene expression by RegA ( Wu and Bauer , 2010; Schindel and Bauer , 2016 ) . This might be the reason why part of RegA regulon ( chemotaxis and motility genes , bacterial microcompartment etc . ) is also observed to be indirectly affected upon deletion of AerR . It is known that the AerR gene is present and linked to the CrtJ gene in all sequenced purple non-sulfur bacteria ( Cheng et al . , 2014; Vermeulen and Bauer , 2015 ) . The long isoform of AerR has also been isolated from numerous species and shown to bind cobalamin ( Vermeulen and Bauer , 2015 ) . In a previous study , AerR and CrtJ from Rhodospirillum centenum were also disrupted with resulting phenotypes indicating that AerR and/or CrtJ likely have dual functions for both activation and repression of photosystem gene expression ( Masuda et al . , 2008 ) . Our results are in good agreement with the R . centenum study . Furthermore , we have also observed that R . centenum also synthesize both long and short isoforms of AerR ( Figure 1—figure supplement 2 ) . While the presence of two isoforms in other species suggests that diverse species of purple bacteria may use similar LAerR/SAerR isoforms to control gene expression , it does not indicate that they all do so . For example , the AerR homolog from Rhodobacter sphaeroides ( also called PpaA ) appears to only have one large isoform ( Figure 1—figure supplement 2 ) . Interestingly , this species synthesizes a second photoreceptor protein called AppA that contains extensive homology to AerR with the caveat that AppA uses flavin as a chromophore instead of cobalamin ( Moskvin et al . , 2010 ) . A study has also shown that AerR and AppA have opposing functions in regulating the activity of CrtJ ( Vermeulen and Bauer , 2015 ) suggesting that this species may have replaced the short AerR isoform with a gene duplication event that gave rise to AppA . The Rhodobacter capsulatus strain SB1003 was used as the WT parental strain and was also the host strain from which LaerR and SaerR expression strains were constructed . R . capsulatus strains were first grown semi-aerobically overnight as a 3 ml PY medium in tubes at 34˚C with shaking at 200 rpm . The overnight cultures were then transferred to flasks shaking at 200 rpm for aerobic conditions or into screw-caped vials for anaerobic conditions . 75 W tungsten filament light bulbs were used as a light source under anaerobic photosynthetic conditions . E . coli strains , HST08 and S17-1λpir were used for cloning and for the conjugation of plasmids to R . capsulatus , respectively . AerR overexpression was carried out using E . coli strain BL21 ( DE3 ) grown in LB medium . UNICO 1100RS Spectrometer was used to check growth curves under photosynthetic conditions . To express AerR-FLAG protein in R . capsulatus cells , we constructed an in-frame chromosomal FLAG-tagged aerR strain ( Fang and Bauer , 2017 ) as well as expressed AerR from a low copy broad-host range vector , pBBR-MSC2 . For the plasmid construction , DNA fragments containing the aerR coding region with an appended FLAG-tag sequence ( Rc_aerR-f and pSRK-pBBR-r ) were amplified along with 500 bp upstream and downstream of the aerR gene ( Rc_aerRup-f and Rc_aerRup-r ) using the primers in Supplementary file 5 from pSRKGm-aerR ( Fang and Bauer , 2017 ) and genomic DNA from R . capsulatus as a template . These two DNA fragments were connected and cloned into pBBR-MSC2 EcoRV and HindIII site using In Fusion cloning kit ( Clonetech ) . To construct an N-terminal truncated AerR expression plasmid , Rc_aerR-insA-f ( one nucleotide insertion after M1 codon ) , Rc_aerR_MNG-f ( M35-AerR ) , Rc_aerR_MVE-f ( M49-AerR ) , or Rc_aerR_MDL-f ( M61-AerR ) primers were used instead of Rc_aerR-f . Each point mutation was introduced into the pBBR-aerR-FLAG plasmid by PCR amplification using specific primer pairs ( Supplementary file 5 ) . For a reporter assay , FLAG-bchE fragment was amplified from pSRKGm-bchE plasmid with Flag-bchE-f and M13+ pBBR-MSC2-r and pBBR plasmid that included a partial aerR sequence . This was amplified with pBBR-MSC2-f and Rc_aerRA39/E40/L41-FbchE-r , respectively from the pBBR-aerR-FLAG plasmid . Then , these two fragments were connected using an In-Fusion kit ( Clonetech ) , resulting pBBR-aerR_A39-FLAG-bchE , pBBR-aerR_E40-FLAG-bchE , and pBBR-aerR_L41-FLAG-bchE , respectively . Chromosomal aerR mutations were generated using the suicide plasmid pZJD29a containing 1 kb fragment covering aerR gene with the point mutation as previously reported ( Cheng et al . , 2014 ) . The aerR fragment with select mutations were amplified from the corresponding pBBR-aerR-FLAG plasmid using the primer pairs Rc_aerRup-f2 and Rc_aerR1130-r ) and cloned into pZJD29a using the In Fusion kit . For in vitro analysis , pSUMO-AerR ( Cheng et al . , 2014 ) was used to express LAerR isoform in E . coli . A 120 bp DNA sequence including M1 to E40 codons of aerR was removed from pSUMO-AerR plasmid to express the SAerR isoform . The deletion was made by PCR amplification using Rc_saerR + SUMO f and Rc_saerR + SUMO r primers using pSUMO-AerR as a template with the resulting plasmid , pSUMO-SAerR transformed BL21 ( DE3 ) to express the SAerR isoform . In vivo expression of AerR proteins in R . capsulatus cells were measured by Western blot analysis after the addition FLAG epitope to the carboxyl terminus of AerR . For this analysis , collected R . capsulatus cells were resuspended in TBS buffer and then disrupted by sonication . Disrupted cell extracts were clarified by centrifugation 20 , 000 x g for 10 min at 4˚C . Clarified proteins in the supernatant were separated by SDS-PAGE followed by Western blot analysis that detected the FLAG epitope using commercial FLAG epitope-specific monoclonal antibodies containing an HRP conjugate ( Sigma ) . R . capsulatus strains were grown to early exponential phase ( OD6600 . 3–0 . 35 ) from which 1 . 5 ml of cell cultures were quickly chilled to 4°C , harvested by centrifugation and stored as a cell pellet at −80°C until needed . Triplicate biological replicates ( independent cell cultures grown under similar conditions at different times ) were used for each RNA-Seq analysis for each described condition . Total RNA was extracted using ISOLATE II RNA Mini Kit ( Bioline ) followed by TURBO DNase ( Ambion ) treatment . The reaction mixture was cleaned and concentrated by RNeasy MinElute Cleanup Kit ( QIAGEN ) and assayed for DNA contamination by PCR amplification on samples with or without reverse transcriptase treatment . Final RNA concentrations were measured using NanoDrop spectroscopy ( Thermo Scientific ) . Further quality control was performed with a 2200 TapeStation using RNA ScreenTape ( Agilent Technologies ) . Library construction and RNA-sequencing were performed by the Center for Genomics and Bioinformatics at Indiana University-Bloomington . Ribosomal RNA was depleted and libraries were created using a ScriptSeq Complete Kit ( Illumina ) for bacteria according to manufacturer’s protocol . Single-end sequencing reactions ( >75 × coverage ) were performed on Illumina NextSeq sequencer with raw sequence read files deposited in Sequence Read Archive ( SRA ) with the accession number SRP136743 . The raw reads were trimmed and aligned to the R . capsulatus SB1003 annotated genome ( GenBank accession no . CP001312 . 1 ) as described previously using Bowtie 2 ( Langmead and Salzberg , 2012 ) . HTSeq-count ( Anders et al . , 2015 ) was used to count read numbers in each gene followed by differential expression analysis using DESeq2 package in R ( Love et al . , 2014 ) . Genes were considered to be significantly different if they had a p-adjusted value <0 . 01 . Total RNA that was extracted from the previous RNA-seq step was used for RNA ligase-mediated rapid amplification of 5’ cDNA ends ( RLM-RACE ) . GeneRacer Kit ( Invitrogen ) was used to generate RACE-ready cDNA , except that the calf intestinal alkaline phosphatase ( CIP ) treatment was replaced by Terminator exonuclease ( Epicentre ) treatment in order to select for primary transcripts . GeneRacer 5’ primer and an aerR specific primer were used in 5’ RACE PCR . RACE PCR product was gel purified and cloned into pCR-4 TOPO ( Invitrogen ) . 20 clones were selected for sequencing to validate the transcription start site of aerR . For biochemical analysis , AerR variant proteins were purified as described previously ( Cheng et al . , 2014 ) . E . coli strain BL21 ( DE3 ) with pSUMO-AerR or pSUMO-SAerR was grown in LB medium at 37˚C to an OD600 of 0 . 7 . E . coli cultures were then cooled and then 50 µM isopropyl-β-D-thiogalactopyranoside ( IPTG ) was added with cultivation continued at 16˚C for 16 hr . Collected cells was resuspended in a lysis buffer ( 20 mM Tris-HCl pH 8 . 0 , 150 mM NaCl , 5 mM imidazole , and 10% glycerol ) and disrupted using a French press cell three times at 18 , 000 psi . The lysate was clarified by centrifugation at 30 , 000 x g for 30 min at 4˚C . To bind hydroxyl-Cbl , 10 µM adenosyl-Cbl was added to the supernatant followed by illumination with white light for 5 min . The supernatant was then passed through a 0 . 45 µM membrane filter and applied to a 1 ml HisTrap column using ÄKTA chromatography system . The column was washed with wash buffer ( 20 mM Tris-HCl pH 8 . 0 , 150 mM NaCl , 20 mM imidazole , and 10% glycerol ) and SUMO-AerR was eluted with a gradient of 20 mM to 500 mM imidazole in the wash buffer over 15 column volume . Eluted SUMO-AerR was incubated with SUMO protease Ulp1 in presence of 1 mM DTT at RT for 16 hr followed by a desalting column against the wash buffer . Digested SUMO-tag was trapped by Ni-sepharose column with tag-less AerR further purified by Superose 12 size exclusion chromatography in 20 mM Tris-HCl ( pH 8 . 0 ) and 200 mM NaCl . Microscale thermophoresis ( MST ) analysis was performed using Monolith NT . 115 ( Nanotemper ) as described previously ( Cheng et al . , 2014 ) . CrtJ was labeled using RED-NHS protein labeling kit ( Nanotemper ) and the labeling efficiency was evaluated spectroscopically . For the MST experiment , concentration of labeled CrtJ was kept constant ( either 200 or 500 nM ) in 20 mM Tris-HCl ( pH 8 . 0 ) and 200 mM NaCl and with AerR concentration varied from µM to nM scale . Cobalamin binding assay was performed using cobalamin unbound SUMO-tagged AerR protein . Cobalamin unbound SUMO- ( S ) AerR was purified as same as mentioned above with the exception of cobalamin addition to cell lysates followed by incubation for 20 min at RT under dark conditions . Unbound cobalamin molecules were removed by desalting column . Cobalamin binding was evaluated by the spectrum of the SUMO- ( S ) AerR fraction . Total pigment was extracted from R . capsulatus cell by acetone/methanol ( v:v = 7:2 ) . Collected cells were dissolved in the acetone/methanol solution , followed by cell disruption using sonication . After the extract was clarified by centrifugation ( 13 , 300 rpm , 10 min , 4˚C ) , absorption spectrum was scanned from 350 nm to 900 nm with a HP 8453 UV-Visible Spectrometer . Relative amounts of bacgteriochlorophyll and carotenoid were calculated from absorbance at 768 nm , and 480 nm , respectively .	Some bacteria are able to use a process called photosynthesis to convert energy from sunlight into another form of energy they can use to grow . Within the bacteria , structures known as photosystems are responsible for absorbing light and transferring the energy to other molecules . The levels of light surrounding the bacteria continually fluctuate . To optimize the amount of light they absorb for photosynthesis , the bacteria have receptors that detect light and regulate the activities of the genes that produce photosystems . One group of bacteria that carry out photosynthesis are collectively known as purple bacteria . These bacteria contain a light receptor called AerR that interacts with a protein called CrtJ , which can directly bind to and alter the activity of genes involved in photosynthesis . AerR senses light by binding to a molecule called vitamin B12 , which can absorb blue light , but it was not clear how it affects the CrtJ protein . Fang , Yamamoto et al . used biochemical and genetic approaches to study AerR in a purple bacterium known as Rhodobacter capsulatus . The experiments show that R . capsulatus makes two different versions of AerR . The larger version only binds to vitamin B12 that is carrying light energy and stimulates CrtJ to activate genes involved in photosynthesis . On the other hand , the shorter version binds to vitamin B12 in the dark and causes CrtJ to repress genes that produce photosystems . Receptors similar to AerR are found in many bacteria and other single-celled organisms known as Archaea , including in species that do not perform photosynthesis . Therefore , these findings are likely to be useful to researchers studying how bacteria and Archaea sense light in a variety of situations . A next step will be to find out how the different forms of AerR can change the properties of CrtJ .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2018	Differing isoforms of the cobalamin binding photoreceptor AerR oppositely regulate photosystem expression
Red blood cells ( RBCs ) experience significant mechanical forces while recirculating , but the consequences of these forces are not fully understood . Recent work has shown that gain-of-function mutations in mechanically activated Piezo1 cation channels are associated with the dehydrating RBC disease xerocytosis , implicating a role of mechanotransduction in RBC volume regulation . However , the mechanisms by which these mutations result in RBC dehydration are unknown . In this study , we show that RBCs exhibit robust calcium entry in response to mechanical stretch and that this entry is dependent on Piezo1 expression . Furthermore , RBCs from blood-cell-specific Piezo1 conditional knockout mice are overhydrated and exhibit increased fragility both in vitro and in vivo . Finally , we show that Yoda1 , a chemical activator of Piezo1 , causes calcium influx and subsequent dehydration of RBCs via downstream activation of the KCa3 . 1 Gardos channel , directly implicating Piezo1 signaling in RBC volume control . Therefore , mechanically activated Piezo1 plays an essential role in RBC volume homeostasis . Mammalian red blood cells ( RBCs ) are rather unique in that they lack a nucleus and many organelles and that they traverse the circulatory system several hundred thousand times in their life cycle . RBCs experience significant mechanical forces while recirculating that influence their physiology in many ways , including changes in deformability ( Chien , 1987 ) , ATP release ( Sprague et al . , 2001 ) , NO release ( Yalcin et al . , 2008 ) , and Ca2+ influx ( Larsen et al . , 1981; Dyrda and et al . , 2010 ) , the latter of which can influence RBC volume . Changes in RBC volume can affect their membrane integrity and ability to travel through capillaries with diameters smaller than the RBCs themselves . The critical importance of RBC volume regulation is demonstrated by several pathologies resulting from either overhydration or dehydration of RBCs ( Gallagher , 2013 ) . However , the molecular mechanisms by which RBCs sense mechanical forces and the effects of these forces on volume homeostasis have remained unclear . Recent studies have identified a conserved family of mechanosensitive non-selective cation channels , Piezo1 and Piezo2 ( Coste et al . , 2010 , 2012 ) . Piezo1 responds to a wide array of mechanical forces , including poking , stretching , and shear stress , and is essential for proper vascular development in mice ( Nilius and Honore , 2012; Li et al . , 2014; Ranade et al . , 2014 ) . The potential role of Piezo1 in RBC physiology is most clearly demonstrated by many gain-of-function mutations in Piezo1 that have been identified in patients with the RBC disease xerocytosis , also called dehydrated hereditary stomatocytosis ( DHS ) ( Zarychanski et al . , 2012; Albuisson et al . , 2013; Andolfo et al . , 2013; Bae et al . , 2013 ) . In addition , whole body treatment of zebrafish with Piezo1 morpholino affects RBC volume homeostasis ( Faucherre et al . , 2014 ) . Finally , the Piezo1 locus has also been implicated in a genome-wide association screen for affecting the RBC mean corpuscular hemoglobin concentration ( MCHC ) in humans ( van der Harst et al . , 2012 ) . However , whether any of these effects are due to Piezo1 expression on the RBCs themselves is not understood nor is the normal role of Piezo1 in mammalian RBC physiology . We first investigated whether Piezo1 is expressed on mouse RBCs using Piezo1P1-tdTomato mice that express a Piezo1-tdTomato fusion protein from the Piezo1 locus ( Ranade et al . , 2014 ) . Both peripheral blood RBCs ( Figure 1A ) and developing bone marrow pro-RBCs ( Figure 1B ) from Piezo1P1-tdTomato mice exhibited increased tdTomato fluorescence by flow cytometry compared to those from Piezo1+/+ mice . Peripheral RBCs from Piezo1P1-tdTomato mice had clear expression of a ∼320 kDa Piezo1-tdTomato fusion protein by Western blot ( Figure 1A ) . To further investigate the role of Piezo1 in RBC physiology , we set out to genetically ablate it . Mice deficient in Piezo1 die in utero , so we deleted Piezo1 specifically in the hematopoietic system . We bred Vav1-iCre mice , which express Cre recombinase early in hematopoiesis ( Shimshek et al . , 2002 ) , to mice where exons 20–23 of Piezo1 are flanked by loxP sites ( P1f ) , thus generating viable , fertile Vav1-iCre P1f/f ( Vav1-P1cKO ) mice ( Figure 1—figure supplement 1A ) . Vav1-P1cKO lymphocytes exhibited >95% deletion of piezo1 transcript , demonstrating efficient Cre-mediated excision ( Figure 1—figure supplement 1C ) . Hematological analysis of blood from Vav1-P1cKO mice revealed significant changes in RBC physiology without significant anemia ( Table 1 ) . Notably , compared to WT mice , Vav1-P1cKO mice had elevated ( % of WT ± SEM ) mean corpuscular volume ( MCV , 109 . 51 ± 1 . 51 ) and mean corpuscular hemoglobin ( MCH , 103 . 14 ± 0 . 48 ) and reduced mean corpuscular hemoglobin concentration ( MCHC , 94 . 37 ± 1 . 08 ) , suggesting that Piezo1-deficient RBCs were overhydrated . Since increased MCV can also be observed in the dehydrated RBCs in xerocytosis , we further tested whether Piezo1-deficient RBCs were actually overhydrated . Overhydrated RBCs exhibit increased osmotic fragility and increased size as measured by forward scatter using flow cytometry . Blood from Vav1-P1cKO mice exhibited both of these characteristics ( Figure 1C and Figure 1—figure supplement 2A ) , demonstrating that Piezo1-deficient RBCs are overhydrated . While Vav1-P1cKO RBCs were overhydrated , scanning electron microscopy of WT and Vav1-P1cKO RBCs revealed that Vav1-P1cKO RBCs had relatively normal discoid morphology , unlike more severe overhydration pathologies such as spherocytosis ( Figure 1—figure supplement 2B ) . Regardless , these results suggest that Piezo1 expression on RBCs is a negative regulator of RBC volume . 10 . 7554/eLife . 07370 . 003Figure 1 . Deletion of Piezo1 in blood cells causes RBC fragility and splenic sequestration . ( A ) Left: flow cytometry histograms of tdTomato fluorescence on Ter-119+ peripheral blood red blood cells ( RBCs ) . Rightward shifts indicate increased fluorescence . Right: Western blot for tdTomato from lysates of packed RBCs . ( B ) Flow cytometry histograms of tdTomato fluorescence from less ( CD71+ FSC-AHi ) and more ( CD71+ FSC-ALo ) mature RBC progenitor cells . ( C ) Percent hemolysis of blood of WT and Vav1-P1cKO mice when exposed to hypotonic solutions of indicated relative tonicity . C50 values ( relative tonicity at half maximal lysis ) were calculated by fitting the data to a 4-parameter logistic sigmoidal curve . ( D ) Total number of Ter-119+ erythroid cells found in the spleens of WT and Vav1-P1cKO mice . ( E ) Plasma haptoglobin concentrations of both WT and Vav1-P1cKO mice as determined by ELISA . A and B are representative histograms and blots from three individual mice per genotype . Graphs in C and E result from individual experiments consisting of at least 3 WT and 3 Vav1-P1cKO mice each , with each experiment repeated 2 , 3 , and 3 times for C , D , and E , respectively . p < 0 . 05 , *p < 0 . 001 by unpaired Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 00310 . 7554/eLife . 07370 . 004Figure 1—figure supplement 1 . Generation and validation of Vav1-P1cKO mice . ( A ) Schematic for generation of Vav1-P1cKO mice . Deletion of exons 20–23 causes a frameshift in the piezo1 transcript . ( B ) Genotyping of Piezo1 mice with the given Piezo1 genotypes: 1: +/+ , 2: +/− , 3: f/+ , 4: f/f , 5: f/− , 6: P1f/f , 7: Vav1-iCre+ P1f/f . Arrow indicates the visible ( – ) band in Vav1-P1cKO mice . ( C ) piezo1 transcript expression ( mean + SEM , n = 4 ) of lymphocytes isolated from WT and Vav1-P1cKO mice . Transcript levels were calculated using the 2−ΔΔCT method using gapdh as a reference gene and were normalized to the average expression from WT samples . DOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 00410 . 7554/eLife . 07370 . 005Figure 1—figure supplement 2 . Histology of WT and Vav1-P1cKO spleens . ( A ) Forward scatter area ( size ) of different populations of splenic Ter-119+ RBCs from WT and Vav1-P1cKO mice . ( B ) Left—scanning electron micrographs of WT ( left ) and Vav1-P1cKO ( right ) RBCs . ( C ) Representative hematoxylin- and eosin-stained spleen sections from WT and Vav1-P1cKO mice . Right—quantification of red pulp area in WT and Vav1-P1cKO spleens . ( D ) Left—representative Prussian blue-stained spleen sections from WT and Vav1-P1cKO mice . Right—quantification of percent of spleen area stained with Prussian blue in WT and Vav1-P1cKO mice . Graph in A is representative from three experiments , with at least 3 mice per group per experiment . Graphs in C and D result from analysis of 3 spleen sections from each of 4 WT and 4 Vav1-P1cKO spleens . DOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 00510 . 7554/eLife . 07370 . 006Table 1 . Hematological indices from blood isolated from 8- to10-week-old WT and Vav1-P1cKO miceDOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 006WT ± SEM ( n = 19 ) Vav1-P1cKO ± SEM ( n = 18 ) RBC100 ± 0 . 5896 . 60 ± 1 . 10HGB100 ± 0 . 5499 . 50 ± 1 . 10HCT100 ± 0 . 51105 . 59 ± 1 . 13*MCV100 ± 0 . 23109 . 51 ± 1 . 51MCH100 ± 0 . 25103 . 14 ± 0 . 48MCHC100 ± 0 . 2694 . 37 ± 1 . 08RDW100 ± 0 . 92114 . 49 ± 2 . 64*Data are pooled from four individual experiments , each experiment consisting of at least three age- and sex-matched mice per genotype . RBC: red blood cell count per unit volume; HGB: hemoglobin content; HCT: hematocrit; MCV: mean corpuscular volume; MCH: mean corpuscular hemoglobin; MCHC: mean corpuscular hemoglobin concentration; RDW: red cell distribution width . Indices for each mouse were normalized to the average value of WT mice from the same experiment . Statistics were calculated by two-tailed Mann–Whitney test . p < 0 . 05 , **p < 0 . 001 . Because changes in RBC volume commonly result in pathology in the spleen , we compared Vav1-P1cKO spleens with those of WT littermates . Although they appeared visibly darker and redder following H&E staining , spleens from Vav1-P1cKO mice exhibited normal formation of both red and white pulp without an evident expansion of red pulp or increased iron deposition ( Figure 1—figure supplement 2C , D ) . However , flow cytometric analysis of splenic RBC subpopulations revealed an increased number of fully mature Ter119+ CD71− RBCs , but not immature Ter119+ CD71+ RBCs ( Figure 1D ) , suggesting that the darker splenic color is due in part to retention of overhydrated circulating mature RBCs . Consistent with this , immature splenic RBCs had similar forward scatter in WT and Vav1-P1cKO mice , indicating that they were of similar size , while fully mature RBCs in Vav1-P1cKO exhibited increased forward scatter indicative of increased size ( Figure 1—figure supplement 2A ) . We also found that Vav1-P1cKO mice exhibited significantly lower plasma haptoglobin concentrations , indicative of intravascular hemolysis in vivo ( Figure 1E ) . Thus , Piezo1-deficient RBCs have increased fragility and are aberrantly retained within the spleen , suggesting that Piezo1 helps maintain RBC integrity and normal recirculation . Piezo1 is a mechanically activated , calcium-permeable non-selective cation channel . RBCs experience significant mechanical forces during circulation; we , therefore , sought to determine whether acute application of mechanical force could cause Ca2+ influx and whether any Ca2+ influx observed was dependent on Piezo1 . However , many of the existing methods of mechanical stimulation available to us proved unsuitable for this purpose . Patch-clamp experiments proved impractical as conditions that allowed the formation of gigaohm seals in a cell-attached setting resulted in cell rupture upon application of negative pressure . Additionally , RBCs were also too small and fragile for indentation using a glass probe with or without simultaneous electrophysiological recording . Calcium imaging studies using shear stress in laminar flow chambers did not yield detectable increases in intracellular Ca2+ . We , therefore , developed a mechanical stimulation assay combining the advantage of calcium imaging ( no gigaohm seal necessary ) and patch clamp ( a pipette to capture the cell ) . We used micropipettes with a long , tapered tip and optimized their size and shape ( tip diameter ∼1–1 . 5 μm ) so that RBCs could only partially enter into the pipette ( Figure 2A and Figure 2—figure supplement 1 ) . In the presence of 0 . 05% BSA , mechanical stimulation of these cells was possible without causing cell rupture by applying negative pipette pressure using a high-speed pressure clamp device ( Figure 2A ) . This method allowed for quantitative and reproducible application of force while detecting changes in intracellular Ca2+ . Application of negative pressure induced a rapid rise in intracellular Ca2+ levels in Fluo-4-loaded WT RBCs with a threshold of ∼ −5 mmHg and a P50 of −9 . 72 ± 0 . 51 mmHg ( Figure 2B , C and Video 1 ) . As seen in Figure 2D , repeated applications of negative pressure induced a mild rundown in the calcium response . To account for this , we normalized all responses in Figure 2C to the average of two maximal ( −25 mmHg ) stimuli flanking each test pulse ( Figure 2C ) . Following removal of a half-maximal −10 mmHg stimulus , Ca2+ levels declined to baseline with an average t1/2 of 21 ± 1 . 8 s . This increase in intracellular Ca2+ concentration must be a result of influx rather than receptor-mediated store release as RBCs do not possess intracellular stores . To conclusively determine whether this Ca2+ influx was Piezo1-mediated , we subjected Vav1-P1cKO RBCs to similar negative pressure stimulation . We found that Ca2+ influx was not detectable in any Vav1-P1cKO RBCs , even at pressures up to −35 mmHg ( Figure 2D ) . Experiments using Vav1-P1cKO RBCs were conducted using the same pipettes that previously elicited Ca2+ influx in WT RBCs , allowing for equivalent mechanical stimulation between genotypes . These results demonstrate that RBCs can respond to mechanical force by allowing Ca2+ to enter the cell and that this Ca2+ influx is dependent on Piezo1 . 10 . 7554/eLife . 07370 . 007Figure 2 . RBCs exhibit Piezo1-dependent Ca2+ influx in response to mechanical stretch . ( A ) Left: cartoon representation of mechanical stretching of RBCs . Right: brightfield images of an individual RBC before ( top ) and during ( bottom ) application of −35 mmHg . Dotted line indicates starting location of RBC membrane prior to stretching . ( B ) Left: representative plot of background subtracted Fluo-4 fluorescence of an individual RBC following application of −25 mmHg for the time indicated by the gray shaded area . Right: images of the RBC plotted on left at the times indicated . ( C ) Left: representative plot of background subtracted Fluo-4 fluorescence from an individual RBC when subjected to pressure pulses of different magnitudes . Duration of the pulses is as indicated by shaded areas on plot; magnitude of pressure in mmHg is indicated above lines . Right: pressure-response curve ( mean ± SEM ) generated from 8 RBCs subjected to varying pressures . Responses to each different pressure were normalized to the average of the flanking −25 mmHg pulses , and the order of each different pressure applied was randomized for each separate RBC . ( D ) Representative plots from individual WT ( Left ) and Vav1-P1cKO ( Right ) RBCs subjected to pressure pulses as indicated in mmHg . Right graph represents mean ± SEM of fluorescence change in response to first −35 mmHg pulse from 5 WT and 5 Vav1-P1cKO RBCs subjected to mechanical stretching protocol as shown in plots . Numbers above graph indicate number of cells that had responses over 10 AFU out of total cells tested . DOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 00710 . 7554/eLife . 07370 . 008Figure 2—figure supplement 1 . Bright field image of representative pipette used to elicit Ca2+ influx in RBCs . DOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 00810 . 7554/eLife . 07370 . 009Video 1 . Video of Ca2+ influx into a RBC following mechanical stimulation . Time lapse video of a single Fluo-4-loaded RBC subjected to repeated 10-s pulses of −35 mmHg as indicated . Time scale is in min:s . DOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 009 The ability of mechanical force to cause Ca2+ entry through Piezo1 , coupled with the observed overhydration of Piezo1-deficient RBCs , suggests an important role for Piezo1-dependent Ca2+ influx in regulating RBC volume following mechanical stress . To directly test the effect that Piezo1 activation could have on RBC volume , we utilized the recently identified Piezo1-selective activator Yoda1 ( Syeda et al . , 2015 ) . WT or Vav1-P1cKO RBCs were loaded with the calcium-sensitive dye Fluo-4 and then treated with 15 μM Yoda1 . Yoda1 caused robust increases in fluorescence of WT , but not in Vav1-P1cKO RBCs . In contrast , both WT and Vav1-P1cKO RBCs exhibited similar Ca2+ influx in response to the Ca2+ ionophore A23187 ( Figure 3A ) . 10 . 7554/eLife . 07370 . 010Figure 3 . Piezo1 activation causes Ca2+ influx and KCa3 . 1-dependent RBC dehydration . ( A ) Flow cytometry histograms of Fluo-4 fluorescence of WT ( left ) or Vav1-P1cKO ( right ) RBCs treated with vehicle ( gray shaded ) , 15 μM Yoda1 ( solid line ) , or 10 μM A23187 ( dashed line ) for 1 min as indicated . Rightward shifts in fluorescence indicate increased intracellular Ca2+ . ( B ) Brightfield images from RBCs from WT ( top ) or Vav1-P1cKO ( bottom ) RBCs at the indicated times after superfusion with 15 μM Yoda1 . ( C ) Osmotic fragility ( ±SEM , n = 3 ) of blood from WT ( left ) or Vav1-P1cKO ( middle ) treated with 2 μM the KCa3 . 1 antagonist TRAM-34 and/or 5 μM Yoda1 as indicated . Blood was incubated with TRAM-34 or vehicle for 10 min , and then incubated with Yoda1 or vehicle for 30 min . Graph on right depicts C50 ± SEM for hemolysis for the genotypes and treatments in the left graphs . p values were calculated using one-way ANOVA . ( D ) Osmotic fragility ( ±SEM , n = 3 ) of WT and Vav1-P1cKO blood treated with 1 μM of the Ca2+ ionophore A23187 for 30 min . p < 0 . 05; ***p < 0 . 001 compared to genotype-matched , vehicle-treated blood by Student's t-test . ( E ) Working model for how Piezo1 activation regulates RBC volume . Experiments were repeated the following number of times: A: 3 , B: 3 , C: 2 , D: 2 , with results from an individual experiment being presented . DOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 01010 . 7554/eLife . 07370 . 011Figure 3—figure supplement 1 . Tram-34 does not block Piezo1 . Average maximal response ( mean ± SEM , vehicle , n = 194 , TRAM-34 , n = 191 ) of Fura2-loaded HEK293T cells transfected with mPiezo1-IRES-eGFP after superfusion with either vehicle or 2 μM TRAM-34 for 4 min prior to superfusion with 15 μM Yoda1 . DOI: http://dx . doi . org/10 . 7554/eLife . 07370 . 011 A rise in intracellular Ca2+ in response to Yoda1 would be expected to activate KCa3 . 1 , also known as the Gardos channel , which can mediate K+ efflux , H2O efflux , and RBC dehydration ( Maher and Kuchel , 2003 ) . In fact , RBCs from one strain of mice lacking KCa3 . 1 exhibit increased size and osmotic fragility similar to what is seen in Vav1-P1cKO mice ( Grgic et al . , 2009 ) . We tested whether Yoda1 could cause RBC dehydration , and whether any possible dehydration was mediated through KCa3 . 1 activation by Piezo1-dependent Ca2+ influx . Treatment of WT RBCs , but not Vav1-P1cKO RBCs , with 15 μM Yoda1 led to a rapid change in their shape , progressing from discocytes to echinocytes to spherocytes , similar to what has been demonstrated for treatment with A23187 ( Steffen et al . , 2011 ) ( Figure 3B ) . We further tested the osmotic fragility of blood following treatment with Yoda1 and/or the KCa3 . 1 antagonist TRAM-34 ( Wulff et al . , 2000 ) . Incubation of WT , but not Vav1-P1cKO blood , for 30 min with 5 μM Yoda1 caused a marked reduction in RBC osmotic fragility that was prevented by a 10-min pretreatment with 2 μM TRAM-34 ( Figure 3C ) . Both WT and Vav1-P1cKO RBCs exhibited reduced osmotic fragility when treated for 30 min with 1 μM A23187 , demonstrating normal functionality of KCa3 . 1 in Vav1-P1cKO RBCs ( Figure 3D ) . The lack of any Ca2+ influx , shape changes , or changes in osmotic fragility of Vav1-P1cKO RBCs in response to Yoda1 further demonstrates the specificity of Yoda1 on Piezo1 . Importantly , TRAM-34 did not block Piezo1 HEK293T cells transfected with Piezo1 ( Figure 3—figure supplement 1 ) . These data as a whole suggest a model shown in Figure 3E whereby activation of Piezo1 on RBCs leads to calcium influx , potassium efflux through KCa3 . 1 that is accompanied by water loss , resulting in RBC dehydration . Many of the effects exerted by physical forces on cellular physiology remain unclear . Here , we have shown that these forces have a significant effect on RBC physiology by activating the mechanosensitive ion channel Piezo1 . Our findings suggest a link between mechanical forces and RBC volume via Ca2+ influx through Piezo1 . The ability of RBCs to reduce their volume in response to mechanical forces could improve their ability to traverse through small-diameter capillaries and splenic sinusoids . Additionally , it is possible that this reduction in volume could aid in oxygen/CO2 exchange in the periphery by concentrating hemoglobin within RBCs , which may promote release of oxygen from hemoglobin; in fact , mechanical stimulation of RBCs via optical tweezers has been shown to cause such a release of oxygen ( Rao et al . , 2009 ) . It was not previously clear whether the cause of RBC dehydration in DHS patients is due to direct or indirect mechanisms . We have demonstrated that genetic deletion of Piezo1 in blood cells leads to overhydrated , fragile RBCs . We have also shown that mechanical force can cause calcium influx into RBCs that is dependent on Piezo1 expression . We cannot exclude the possibility that deletion of Piezo1 alters RBC membrane properties , resulting in decreased activity of a separate mechanosensitive ion channel rather than calcium entering the cell through Piezo1 itself . We find this unlikely given the relatively mild overhydration of Vav1-P1cKO RBCs and the complete absence of any mechanically induced calcium influx in these cells . Finally , using the first identified selective small molecule activator of Piezo1 , we have shown that calcium influx through Piezo1 dehydrates RBCs via the actions of KCa3 . 1 . These findings are consistent with the observation of dehydrated RBCs from DHS patients with gain-of-function Piezo1 mutations ( Zarychanski et al . , 2012; Albuisson et al . , 2013; Andolfo et al . , 2013; Bae et al . , 2013 ) , further supporting that the dehydration seen in DHS patients is due to increased Piezo1-mediated calcium influx in response to mechanical forces in RBCs themselves . This work also demonstrates the power of combining both genetic and chemical approaches . Experiments performed on Piezo1-deficient background demonstrate the specificity of Yoda1; meanwhile , the acute effect of Yoda1 on RBC dehydration allows us to conclude that Piezo1 activity can regulate cell volume regulation via KCa3 . 1 activation and suggests that changes in RBC volume in Vav1-P1cKO mice are not a consequence of developmental compensation or non-cell autonomous mechanisms . Piezo1 has been proposed by many to mediate a non-selective current observed in sickle cell disease ( SCD ) called PSickle , which acts upstream of KCa3 . 1 to mediate RBC sickling ( Lew et al . , 1997; Ranney , 1997; Ma et al . , 2012; Demolombe et al . , 2013 ) . While KCa3 . 1 inhibitors can improve many hematological indices of SCD patients , they have thus far been ineffective in preventing painful vasculo-occlusive crises ( Ataga et al . , 2008 ) . Generation of mice containing conditional deletion of Piezo1 in blood cells combined with full replacement of normal murine hemoglobin genes with sickling human hemoglobin genes could help determine whether Piezo1 mediates both PSickle and the pathology of SCD . Inhibiting Piezo1 may modulate other pathways in addition to potentially blocking KCa3 . 1-dependent dehydration that may have therapeutic benefits in patients with SCD . Piezo1P1-tdTomato fusion knockin mice were generated as described in Ranade et al . ( 2014 ) . P1f mice were generated using the Piezo1tm1a ( KOMP ) Wtsi Knockout First , promoter driven targeting construct from KOMP . This construct was electroporated into Bruce4 ( C57Bl/6-derived ) ES cells , yielding homologously recombined ES cells that were injected into B6 ( Cg ) -Tyr<c-2J>/J blastocysts at the TSRI Mouse Genetics Core . Chimeric mice from these injections were bred to C57Bl/6J mice , yielding germ line-transmitted mice . These mice were then bred to FLP-expressing mice to excise the neomycin resistance cassette , then once more to C57Bl/6J mice to remove FLP , yielding the P1f locus . Vav1-iCre mice on C57Bl/6J background ( Stock # 008610 ) were purchased from the Jackson Laboratory . Mice were genotyped using the following primers: P1 F: CTT GAC CTG TCC CCT TCC CCA TCA AG , P1 WT/fl R: CAG TCA CTG CTC TTA ACC ATT GAG CCA TCT C , P1 KO R: AGG TTG CAG GGT GGC ATG GCT CTT TTT using Phire II polymerase ( Thermo Scientific , Waltham , MA ) with the following cycling conditions: Initial denaturation 98°C for 30 s , followed by 31 cycles of 98°C for 5 s , 65°C for 5 s , 72°C for 5 s , followed by a final hold of 72°C for 2 min . Vav1-iCre was genotyped using protocols as described by Jackson Laboratory . Reactions were separated on 2% agarose gels yielding the following band sizes: P1+: 160 bp , P1f: 330 bp , P1−: 230 bp . It was noted that an obvious P1− band was found in most , but not all , Vav1-iCre+ P1f/f mice . The rare mice that were genotyped as Vav1-iCre+ P1f/f but did not exhibit a P1− band were found to have no changes in hematological indices and displayed Ca2+ influx in response to Yoda1 . Such mice were excluded from analysis , as they were likely a result of inefficient Cre-mediated deletion of Piezo1 . All mice were housed in 12-hr light/dark cycle room with food and water provided ad libitum . All animal procedures were approved by the TSRI Institutional Animal Care and Use Committee . Blood was isolated from mice from either the retro-orbital plexus of mice anesthetized with isoflurane or from the heart of euthanized mice . Hematological data were collected using either a CellDyn 3700 ( Abbott Laboratories , Abbott Park , IL ) or a Procyte Dx ( IDEXX Laboratories , Westbrook , ME ) hematology analyzer . Spleens from WT or Vav1-P1cKO were excised and fixed in 10% neutral buffer formalin for at least 24 hr , then processed and embedded in paraffin . 6-μm thick sections were cut , deparaffinized , and then were either stained with hematoxylin and eosin or with Prussian blue followed by Nuclear Fast Red . Red pulp and Prussian blue-stained areas were determined using Nikon Elements and were normalized to spleen area . For scanning electron micrographs , blood obtained from cardiac puncture was resuspended in 149 mM NaCl + 2 mM HEPES pH 7 . 4 and centrifuged at 500×g for 5 min . Cells were then resuspended in ice-cold fixative consisting of 2 . 5% glutaraldehyde in 0 . 1 M cacodylate buffer with 1 mM calcium . Aliquots of the fixed cells were allowed to settle on 12-mm coverslips previously coated with polylysine . The coverslips with adherent cells were subsequently washed in 0 . 1 M cacodylate buffer and postfixed in buffered 1% osmium tetroxide for 1 hr . The cells were washed extensively in distilled water , and then gradually dehydrated with addition of ethanol to the water . The coverslips were critical point dried ( tousimis Autosamdri 815 ) and the mounted onto SEM stubs with carbon tape . The stubs with attached coverslips were then sputter coated with 6 nm iridium ( EMS model 150T S ) for subsequent examination and documentation on a Hitachi S-4800 SEM ( Hitachi High Technologies America Inc . , Pleasanton CA ) operating at 5 kV . For analysis of Piezo1-tdTomato expression and splenic RBC composition , single cell suspensions of blood , bone marrow , or spleen were obtained in PBS containing 2% FBS . Cells were stained with fluorescently labeled antibodies , washed once , and data were acquired on a LSR-II flow cytometer ( BD ) and analyzed using FlowJo ( FlowJo Inc , Ashland , OR ) . The following antibodies were used for staining , all at 1:100 dilution: PE-Cy7 Ter-119 ( eBioscience , San Diego , CA ) , APC CD71 ( eBioscience ) , APC-Cy7 CD45 . 2 ( Biolegend , San Diego , CA ) . Absolute cell counts were determined using CountBright Absolute Counting Beads ( Life Technologies , Carlsbad , CA ) according to manufacturer's instructions . For analysis of Ca2+ influx by flow cytometry , cells were loaded with Fluo-4 ( Life Technologies ) for 30 min at 37°C in PBS containing 0 . 05% BSA , washed once , and resuspended in the same buffer containing 2 mM CaCl2 . Compounds were added from DMSO stock at 500× , quickly vortexed , and incubated at room temperature for 2 min prior to acquisition . For Western Blot of P1-tdTomato from erythrocytes , 20 μl of packed erythrocytes was lysed in ice-cold RIPA buffer ( G-Biosciences , St . Louis , MO ) containing 1:100 Protease Inhibitor cocktail ( Cell Signaling Technologies , Danvers , MA ) . Protein amounts were calculated by Pierce micro BCA assay ( Life Technologies ) , and then 20 μg of protein was loaded into 3–8% Novex Tris-Acetate polyacrylamide gels ( Life Technologies ) under denaturing conditions . Protein was then transferred to PVDF membranes using an iBlot transfer system ( Life Technologies ) . Blots were incubated for 1 hr with 5% ( wt/vol ) milk in TBST at room temperature and then incubated with a rat anti-mCherry antibody ( a gift from the laboratory of Hugh Rosen ) at a concentration of 1 μg/ml in 5% milk/TBST overnight at 4°C . Blots were then incubated with HRP-conjugated goat anti-rat ( Jackson Immunoresearch , West Grove , PA ) for 1 hr at room temperature at a concentration of 1:10 , 000 , and chemiluminescence was generated using Pierce ECL Plus ( Life Technologies ) reagent . Chemiluminescence was detected using a FluorChem Q ( ProteinSimple , San Jose , CA ) . For detecting plasma haptoglobin , plasma was isolated from blood by centrifugation at 2 , 000×g for 10 min at 4°C , and haptoglobin concentrations were calculated by ELISA ( Genway Biotech , San Diego , CA ) according to manufacturer's instructions . The identification and validation of Yoda1 is described ( Syeda et al . , 2015 ) . A23187 and TRAM-34 were both purchased from Tocris ( Bristol , United Kingdom ) and were dissolved in DMSO . To determine osmotic fragility , blood was first diluted in a ratio of 1:50 into normal saline ( NS , 149 mM NaCl , 2 mM HEPES , pH 7 . 4 ) , and then 10 μl of diluted blood was added to V-bottom 96-well plates . Solutions of varying tonicity were generated by mixing NS ( 100% ) with 2 mM HEPES , pH 7 . 4 ( 0% ) . 250 μl of these solutions was added to the diluted blood and incubated for 5 min at room temperature . Plates were then spun down and 200 μl of the supernatant was transferred to flat-bottom 96-well plates . Absorbance at 540 nm was measured using either an Enspire ( Perkin Elmer , Waltham , MA ) or Cytation3 ( BioTek , Winooski , VT ) plate reader . C50 values were determined by fitting the data to 4-parameter sigmoidal dose–response curves using Prism ( Graphpad , La Jolla , CA ) . For examining the effects of compounds on osmotic fragility , blood was incubated with Yoda1 or A23187 for 30 min prior to addition into V-bottom 96-well plates; blood was incubated with TRAM-34 for 10 min prior to agonist addition when used . When treating with compounds , 2 mM CaCl2 and 4 mM KCl were added to NS during incubation with compounds . Cell shape change was visualized by allowing blood diluted in normal extracellular medium containing ( in mM ) 137 NaCl , 3 . 5 KCl , 2 CaCl2 , 1 MgCl2 , 10 HEPES , 10 dextrose with the addition of 0 . 05% BSA ( NECM/BSA ) . Cells were allowed to settle onto uncoated glass coverslips . Non-adherent cells were washed away using whole-chamber perfusion , which was also used to deliver Yoda1 to the adherent erythrocytes . To monitor cell shape in response to this treatment , bright-field images were acquired every second for 2 min using a Axiovert S100 microscope ( Zeiss , Oberkochen , Germany ) at 40× magnification . Blood was diluted 1:1000 into NECM/BSA and incubated with 5 μM Fluo-4 AM ( Life Technologies ) while rotating for at least 1 hr at 4°C . Cells were then placed in an imaging chamber and washed via whole-chamber perfusion for removal of excess extracellular dye . Mechanical stimulation of erythrocytes was achieved by capturing individual RBCs in the ∼1 μm aperture of custom-made micropipettes ( as shown in Figure 2 ) and subsequently applying pulses of negative pressure to the pipette compartment using a High-Speed Pressure-Clamp ( HSPC ) device ( ALA scientific , Farmingdale , NY ) . Micropipettes were pulled using 1 . 5/0 . 85-mm ( OD/ID ) borosilicate glass capillaries ( Sutter Instruments , Novato , CA ) with a P-97 Flaming/Brown micropipette puller ( Sutter Instruments ) . The electrical resistance of such micropipettes varied in the range of 15–20 MΩ . Following expulsion of the erythrocyte after each measurement using positive pipette pressures , individual pipettes were reused for subsequent measurements ( approx . 6–10 measurements possible with single pipettes before clogging ) , allowing for reliable comparison of the applied mechanical stimuli across experimental groups . Fluorescent calcium measurements were performed using a Lambda DG4 fluorescent excitation source ( Sutter Instruments ) attached to a Zeiss Axiovert S100 microscope . Images were acquired at 1-s intervals using the Zen Pro acquisition suite ( Zeiss ) . Background subtracted average fluorescence intensity was computed using FIJI ( Schindelin et al . , 2012 ) and plotted as arbitrary fluorescence units as a function of time . Data analysis and plots were generated using Prism ( GraphPad ) . Osmotic fragility and pressure-response curves were generated by fitting the data to a 4-parameter sigmoidal dose–response curve . Student's t-test or one-way ANOVA was used to determine statistical significance where appropriate . Normal distribution of data was confirmed using the Shapiro–Wilk test .	Within our bodies , cells and tissues are constantly being pushed and pulled by their surrounding environment . These mechanical forces are then transformed into electrical or chemical signals by cells . This process is crucial for many biological structures , such as blood vessels , to develop correctly , and is also a key part of our senses of touch and hearing . In 2010 , researchers discovered a group of ion channels—proteins embedded in the membrane that surrounds a cell—that open up when a force is applied and allow calcium and other ions to enter the cell . This movement of ions generates the electrical response of the cell to the applied force . However , not much is known about the roles of these ‘Piezo’ ion channels . Red blood cells experience significant forces when they pass through narrow blood vessels . In a disease called xerocytosis , the red blood cells become severely dehydrated and shrink . In 2013 , researchers discovered that patients with this disease have mutations in the gene that codes for the Piezo1 protein: a Piezo protein that has also been linked to a role in blood vessel development in embryos . This suggested that Piezo1 may regulate the volume of red blood cells . Cahalan , Lukacs et al . —including some of the researchers who worked on the 2010 and 2013 studies—have now investigated the role of Piezo1 in red blood cells in more detail . Applying strong forces to red blood cells from mice caused calcium to rapidly enter cells through Piezo1 channels . Cahalan , Lukacs et al . then deleted the Piezo1 gene from red blood cells . This made the cells larger and more fragile than normal cells because they contained too much water . To investigate how Piezo1 regulates water content , the cells were treated with a chemical compound called Yoda1 . This compound was shown in a separate study by Syeda et al . to activate Piezo1 channels . Activating Piezo1 caused a second type of ion channel to open up as well , which allowed potassium ions and water molecules to leave the cell . This resulted in the cell becoming dehydrated . This work raises the possibility that Piezo proteins are involved in other diseases where red blood cell volume is altered . In particular , many believe that Piezo1 may be involved in sickle cell disease , a possibility that can now be tested using the tools described in this study .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "cell", "biology", "neuroscience" ]	2015	Piezo1 links mechanical forces to red blood cell volume
Sustainable biofuel production from renewable biomass will require the efficient and complete use of all abundant sugars in the plant cell wall . Using the cellulolytic fungus Neurospora crassa as a model , we identified a xylodextrin transport and consumption pathway required for its growth on hemicellulose . Reconstitution of this xylodextrin utilization pathway in Saccharomyces cerevisiae revealed that fungal xylose reductases act as xylodextrin reductases , producing xylosyl-xylitol oligomers as metabolic intermediates . These xylosyl-xylitol intermediates are generated by diverse fungi and bacteria , indicating that xylodextrin reduction is widespread in nature . Xylodextrins and xylosyl-xylitol oligomers are then hydrolyzed by two hydrolases to generate intracellular xylose and xylitol . Xylodextrin consumption using a xylodextrin transporter , xylodextrin reductases and tandem intracellular hydrolases in cofermentations with sucrose and glucose greatly expands the capacity of yeast to use plant cell wall-derived sugars and has the potential to increase the efficiency of both first-generation and next-generation biofuel production . The biological production of biofuels and renewable chemicals from plant biomass requires an economic way to convert complex carbohydrate polymers from the plant cell wall into simple sugars that can be fermented by microbes ( Carroll and Somerville , 2009; Chundawat et al . , 2011 ) . In current industrial methods , cellulose and hemicellulose , the two major polysaccharides found in the plant cell wall ( Somerville et al . , 2004 ) , are generally processed into monomers of glucose and xylose , respectively ( Chundawat et al . , 2011 ) . In addition to harsh pretreatment of biomass , large quantities of cellulase and hemicellulase enzyme cocktails are required to release monosaccharides from plant cell wall polymers , posing unsolved economic and logistical challenges ( Lynd et al . , 2002; Himmel et al . , 2007; Jarboe et al . , 2010; Chundawat et al . , 2011 ) . The bioethanol industry currently uses the yeast Saccharomyces cerevisiae to ferment sugars derived from cornstarch or sugarcane into ethanol ( Hong and Nielsen , 2012 ) , but S . cerevisiae requires substantial engineering to ferment sugars derived from plant cell walls such as cellobiose and xylose ( Kuyper et al . , 2005; Jeffries , 2006; van Maris et al . , 2007; Ha et al . , 2011; Hong and Nielsen , 2012; Young et al . , 2014 ) . In contrast to S . cerevisiae , many cellulolytic fungi including Neurospora crassa ( Tian et al . , 2009 ) naturally grow well on the cellulose and hemicellulose components of the plant cell wall . By using transcription profiling data ( Tian et al . , 2009 ) and by analyzing growth phenotypes of N . crassa knockout strains , we identified separate pathways used by N . crassa to consume cellodextrins ( Galazka et al . , 2010 ) and xylodextrins released by its secreted enzymes ( Figure 1A and Figure 1—figure supplement 1 ) . A strain carrying a deletion of a previously identified cellodextrin transporter ( CDT-2 , NCU08114 ) ( Galazka et al . , 2010 ) was unable to grow on xylan ( Figure 1—figure supplement 2 ) , and xylodextrins remained in the culture supernatant ( Figure 1—figure supplement 3 ) . As a direct test of transport function of CDT-2 , S . cerevisiae strains expressing cdt-2 were able to import xylobiose , xylotriose , and xylotetraose ( Figure 1—figure supplement 4 ) . Notably , N . crassa expresses a putative intracellular β-xylosidase , GH43-2 ( NCU01900 ) , when grown on xylan ( Sun et al . , 2012 ) . Purified GH43-2 displayed robust hydrolase activity towards xylodextrins with a degree of polymerization ( DP ) spanning from 2 to 8 , and with a pH optimum near 7 ( Figure 1—figure supplement 5 ) . The results with CDT-2 and GH43-2 confirm those obtained independently in Cai et al . ( 2014 ) . As with cdt-1 , orthologues of cdt-2 are widely distributed in the fungal kingdom ( Galazka et al . , 2010 ) , suggesting that many fungi consume xylodextrins derived from plant cell walls . Furthermore , as with intracellular β-glucosidases ( Galazka et al . , 2010 ) , intracellular β-xylosidases are also widespread in fungi ( Sun et al . , 2012 ) ( Figure 1—figure supplement 6 ) . 10 . 7554/eLife . 05896 . 003Figure 1 . Consumption of xylodextrins by engineered S . cerevisiae . ( A ) Two oligosaccharide components derived from the plant cell wall . Cellodextrins , derived from cellulose , are a major source of glucose . Xylodextrins , derived from hemicellulose , are a major source of xylose . The 6-methoxy group ( blue ) distinguishes glucose derivatives from xylose . R1 , R2 = H , cellobiose or xylobiose; R1 = β-1 , 4-linked glucose monomers in cellodextrins of larger degrees of polymerization; R2 = β-1 , 4-linked xylose monomers in xylodextrins of larger degrees of polymerization . ( B ) Xylose and xylodextrins remaining in a culture of S . cerevisiae grown on xylose and xylodextrins and expressing an XR/XDH xylose consumption pathway , CDT-2 , and GH43-2 , with a starting cell density of OD600 = 1 under aerobic conditions . ( C ) Xylose and xylodextrins in a culture as in ( B ) but with a starting cell density of OD600 = 20 . In both panels , the concentrations of xylose ( X1 ) and xylodextrins with higher DPs ( X2–X5 ) remaining in the culture broth after different periods of time are shown . All experiments were conducted in biological triplicate , with error bars representing standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 00310 . 7554/eLife . 05896 . 004Figure 1—figure supplement 1 . Transcriptional levels of transporters expressed in N . crassa grown on different carbon sources . Transcript levels reported in fragments per kilobase per million reads ( FPKM ) are derived from experiments published in Coradetti et al . ( 2012 ) ; Sun et al . ( 2012 ) . *CBT-1 transports cellobionic acid , the product of lytic polysaccharide monooxygenases ( LPMOs , or CaZy family AA9 and AA10 ) ( Xiong et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 00410 . 7554/eLife . 05896 . 005Figure 1—figure supplement 2 . Growth of N . crassa strains on different carbon sources . ( A ) Wild-type ( WT ) N . crassa , or N . crassa with deletions of transporters cdt-1 ( Δcdt-1 ) or cdt-2 ( Δcdt-2 ) , were grown on M . giganteus plant cell walls , or purified plant cell wall components . Avicel is a form of cellulose derived from plant cell walls . The black box shows the severe growth phenotype of the Δcdt-2 strain grown on xylan medium . ( B ) N . crassa biomass accumulation after 3 days of growth on xylan . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 00510 . 7554/eLife . 05896 . 006Figure 1—figure supplement 3 . Xylodextrins in the xylan culture supernatant of the N . crassa Δcdt-2 strain . 25 µl of 1:200 diluted N . crassa xylan culture supernantant was analyzed by HPAEC on a CarboPac PA200 column . While no detectable soluble sugars were found in the culture supernatant of the wild-type strain ( magenta line ) , the Δcdt-2 strain ( blue line ) left a high concentration of unmodified and modified xylodextrins in the culture supernatant . Little xylose was found , indicating xylose was transported by means of different transporters . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 00610 . 7554/eLife . 05896 . 007Figure 1—figure supplement 4 . Transport of xylodextrins into the cytoplasm of S . cerevisiae strains expressing N . crassa transporters . The starting xylodextrin concentration for each purified component was 100 µM . The remaining xylose ( X1 ) and xylodextrins in the culture media are shown for experiments with S . cerevisiae harboring an empty expression plasmid ( vector ) , or with S . cerevisiae individually expressing transporters CDT-1 or CDT-2 . Xylodextrins used include xylobiose ( X2 ) , xylotriose ( X3 ) , xylotetraose ( X4 ) , and xylopentaose ( X5 ) . Error bars indicate standard deviations of biological triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 00710 . 7554/eLife . 05896 . 008Figure 1—figure supplement 5 . Xylobiase activity of the predicted β-xylosidase GH43-2 . ( A ) GH43-2 hydrolysis of xylodextrins with degrees of polymerization from at least 2–8 ( X2–X8 ) . The 30 min chromatogram is offset for clarity . ( B ) The pH optimum of GH43-2 , determined by measuring the extent of hydrolysis of xylobiose to xylose . The HPAEC chromatogram peak area change for xylose is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 00810 . 7554/eLife . 05896 . 009Figure 1—figure supplement 6 . Phylogenetic distribution of predicted intracellular β-xylosidases GH43-2 in filamentous fungi . Homologs of GH43-2 ( NCU01900 ) were found with BLAST ( Altschul et al . , 1997 ) queries of respective sequence against NCBI protein database . Representative sequences from a diversified taxonomy were chosen and aligned with the MUSCLE algorithm ( Edgar 2004 ) . A maximum likelihood phylogenetic tree was calculated based on the alignment with the Jones-Taylor-Thornton model by using software MEGA v6 . 05 ( Tamura et al . , 2013 ) . Xylan-induced extracellular GH43-3 ( NCU05965 ) was used as an outgroup . The NCBI GI numbers of the sequences used to build the phylogenetic tree were indicated besides the species names . 1000 bootstrap replicates were performed to calculate the supporting values shown on the branches . The scale bar indicates 0 . 2 substitutions per amino acid residue . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 00910 . 7554/eLife . 05896 . 010Figure 1—figure supplement 7 . Xylodextrin consumption profiles of S . cerevisiae strains lacking the xylodextrin pathway . Shown are the concentrations of the remaining sugars in the culture broth after different periods of time of ( A ) the WT D452-2 strain with starting cell density at OD600 = 1 , ( B ) D452-2 with a S . stipitis xylose utilization pathway ( plasmid pLNL78 , Table 1 ) with a starting cell density at OD600 = 1 , ( C ) WT D452-2 strain with a starting cell density at OD600 = 20 , and ( D ) D452-2 with a S . stipitis xylose utilization pathway ( plasmid pLNL78 ) with a starting cell density at OD600 = 20 . In all panels , xylose ( X1 ) and xylodextrins of higher DPs ( X2–X5 ) are shown . Error bars represent standard deviations of biological triplicates ( panels A–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 010 Cellodextrins and xylodextrins derived from plant cell walls are not catabolized by wild-type S . cerevisiae ( Matsushika et al . , 2009; Galazka et al . , 2010; Young et al . , 2010 ) . Reconstitution of a cellodextrin transport and consumption pathway from N . crassa in S . cerevisiae enabled this yeast to ferment cellobiose ( Galazka et al . , 2010 ) . We therefore reasoned that expression of a functional xylodextrin transport and consumption system from N . crassa might further expand the capabilities of S . cerevisiae to utilize plant-derived xylodextrins . Previously , S . cerevisiae was engineered to consume xylose by introducing xylose isomerase ( XI ) , or by introducing xylose reductase ( XR ) and xylitol dehydrogenase ( XDH ) ( Jeffries , 2006; van Maris et al . , 2007; Matsushika et al . , 2009 ) . To test whether S . cerevisiae could utilize xylodextrins , a S . cerevisiae strain was engineered with the XR/XDH pathway derived from Scheffersomyces stipitis—similar to that in N . crassa ( Sun et al . , 2012 ) —and a xylodextrin transport ( CDT-2 ) and consumption ( GH43-2 ) pathway from N . crassa . The xylose utilizing yeast expressing CDT-2 along with the intracellular β-xylosidase GH43-2 was able to directly utilize xylodextrins with DPs of 2 or 3 ( Figure 1B and Figure 1—figure supplement 7 ) . Notably , although high cell density cultures of the engineered yeast were capable of consuming xylodextrins with DPs up to 5 , xylose levels remained high ( Figure 1C ) , suggesting the existence of severe bottlenecks in the engineered yeast . These results mirror those of a previous attempt to engineer S . cerevisiae for xylodextrin consumption , in which xylose was reported to accumulate in the culture medium ( Fujii et al . , 2011 ) . Analyses of the supernatants from cultures of the yeast strains expressing CDT-2 , GH43-2 and the S . stipitis XR/XDH pathway surprisingly revealed that the xylodextrins were converted into xylosyl-xylitol oligomers , a set of previously unknown compounds rather than hydrolyzed to xylose and consumed ( Figure 2A and Figure 2—figure supplement 1 ) . The resulting xylosyl-xylitol oligomers were effectively dead-end products that could not be metabolized further . 10 . 7554/eLife . 05896 . 011Figure 2 . Production and enzymatic breakdown of xylosyl-xylitol . ( A ) Structures of xylosyl-xylitol and xylosyl-xylosyl-xylitol . ( B ) Computational docking model of xylobiose to CtXR , with xylobiose in yellow , NADH cofactor in magenta , protein secondary structure in dark green , active site residues in bright green and showing side-chains . Part of the CtXR surface is shown to depict the shape of the active site pocket . Black dotted lines show predicted hydrogen bonds between CtXR and the non-reducing end residue of xylobiose . ( C ) Production of xylosyl-xylitol oligomers by N . crassa xylose reductase , XYR-1 . Xylose , xylodextrins with DP of 2–4 , and their reduced products are labeled X1–X4 and xlt1–xlt4 , respectively . ( D ) Hydrolysis of xylosyl-xylitol by GH43-7 . A mixture of 0 . 5 mM xylobiose and xylosyl-xylitol was used as substrates . Concentration of the products and the remaining substrates are shown after hydrolysis . ( E ) Phylogeny of GH43-7 . N . crassa GH43-2 was used as an outgroup . 1000 bootstrap replicates were performed to calculate the supporting values shown on the branches . The scale bar indicates 0 . 1 substitutions per amino acid residue . The NCBI GI numbers of the sequences used to build the phylogenetic tree are indicated beside the species names . ( F ) Activity of two bacterial GH43-7 enzymes from B . subtilis ( BsGH43-7 ) and E . coli ( EcGH43-7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 01110 . 7554/eLife . 05896 . 012Figure 2—figure supplement 1 . Xylosyl-xylitol oligomers generated in yeast cultures with xylodextrins as the sole carbon source . ( A ) Carbohydrates from culture supernatants of strain SR8U expressing CDT-2 and GH43-2 ( plasmid pXD8 . 4 ) , resolved by HPAEC , abbreviated as follows: X1 , xylose; X2 , xylobiose; X3 , xylotriose; X4 , xylotetraose; xlt , xylitol; xlt2 , xylosyl-xylitol; xlt3 , xylosyl-xylosyl-xylitol . ( B ) LC-MS and LC-MS/MS spectra for xylosyl-xylitol . High-resolution MS spectra show m/z ratios for the negative ion mode . The deprotonated and formate adduct ions were determined with an accuracy of 0 . 32 and 0 . 33 ppm , respectively . The MS/MS spectrum in the lower panel shows the product ion matching the predicted fragment . The parental ion , [xylosyl-xylitol + H]− , is denoted with the black diamond mark . ( C ) LC-MS and LC-MS/MS spectra for xylosyl-xylosyl-xylitol . The deprotonated and formate adduct ions were determined with an accuracy of 0 . 51 and 0 . 37 ppm , respectively . The MS/MS spectrum in the lower panel shows the product ions matching the predicted fragments . The parental ion , [xylosyl-xylosyl-xylitol + H]− , is denoted with the black diamond mark . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 01210 . 7554/eLife . 05896 . 013Figure 2—figure supplement 2 . Xylodextrin metabolism by a co-culture of yeast strains to identify enzymatic source of xylosyl-xylitol . A mixture of a xylose utilizing strain ( SR8 ) with a cell density at OD600 = 1 . 0 and a xylodextrin hydrolyzing strain ( D452-2 expressing CDT-2 and GH43-2 from plasmid pXD8 . 4 ) with a cell density at OD600 = 20 was co-cultured in a medium containing 2% xylodextrin . Xylobiose ( X2 ) and xylotriose ( X3 ) decreased , whereas xylose ( X1 ) initially increased . Subsequent X1 consumption correlated with production of xylitol . Notably , xylosyl-xylitol oligomers were not detected , suggesting that the xylodextrin reductase activity was present only in the xylose-fermenting strain expressing XR . Error bars represent standard deviations of biological triplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 01310 . 7554/eLife . 05896 . 014Figure 2—figure supplement 3 . Chromatogram of xylosyl-xylitol hydrolysis products generated by β-xylosidases . Reaction products from the enzymatic assays in Figure 2D were resolved by ion-exclusion HPLC . Peak areas were used to quantify the concentration of substrates and products at the end of the reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 014 Since the production of xylosyl-xylitol oligomers as intermediate metabolites has not been reported , the molecular components involved in their generation were examined . To test whether the xylosyl-xylitol oligomers resulted from side reactions of xylodextrins with endogenous S . cerevisiae enzymes , we used two separate yeast strains in a combined culture , one containing the xylodextrin hydrolysis pathway composed of CDT-2 and GH43-2 , and the second with the XR/XDH xylose consumption pathway . The strain expressing CDT-2 and GH43-2 would cleave xylodextrins to xylose , which could then be secreted via endogenous transporters ( Hamacher et al . , 2002 ) and serve as a carbon source for the strain expressing the xylose consumption pathway ( XR and XDH ) . The engineered yeast expressing XR and XDH is only capable of consuming xylose ( Figure 1B ) . When co-cultured , these strains consumed xylodextrins without producing the xylosyl-xylitol byproduct ( Figure 2—figure supplement 2 ) . These results indicate that endogenous yeast enzymes and GH43-2 transglycolysis activity are not responsible for generating the xylosyl-xylitol byproducts , that is , that they must be generated by the XR from S . stipitis ( SsXR ) . Fungal xylose reductases such as SsXR have been widely used in industry for xylose fermentation . However , the structural details of substrate binding to the XR active site have not been established . To explore the molecular basis for XR reduction of oligomeric xylodextrins , the structure of Candida tenuis xylose reductase ( CtXR ) ( Kavanagh et al . , 2002 ) , a close homologue of SsXR , was analyzed . CtXR contains an open active site cavity where xylose could bind , located near the binding site for the NADH co-factor ( Kavanagh et al . , 2002; Kratzer et al . , 2006 ) . Notably , the open shape of the active site can readily accommodate the binding of longer xylodextrin substrates ( Figure 2B ) . Using computational docking algorithms ( Trott and Olson , 2010 ) , xylobiose was found to fit well in the pocket . Furthermore , there are no obstructions in the protein that would prevent longer xylodextrin oligomers from binding ( Figure 2B ) . We reasoned that if the xylosyl-xylitol byproducts are generated by fungal XRs like that from S . stipitis , similar side products should be generated in N . crassa , thereby requiring an additional pathway for their consumption . Consistent with this hypothesis , xylose reductase XYR-1 ( NCU08384 ) from N . crassa also generated xylosyl-xylitol products from xylodextrins ( Figure 2C ) . However , when N . crassa was grown on xylan , no xylosyl-xylitol byproduct accumulated in the culture medium ( Figure 1—figure supplement 3 ) . Thus , N . crassa presumably expresses an additional enzymatic activity to consume xylosyl-xylitol oligomers . Consistent with this hypothesis , a second putative intracellular β-xylosidase upregulated when N . crassa was grown on xylan , GH43-7 ( NCU09625 ) ( Sun et al . , 2012 ) , had weak β-xylosidase activity but rapidly hydrolyzed xylosyl-xylitol into xylose and xylitol ( Figure 2D and Figure 2—figure supplement 3 ) . The newly identified xylosyl-xylitol-specific β-xylosidase GH43-7 is widely distributed in fungi and bacteria ( Figure 2E ) , suggesting that it is used by a variety of microbes in the consumption of xylodextrins . Indeed , GH43-7 enzymes from the bacteria Bacillus subtilis and Escherichia coli cleave both xylodextrin and xylosyl-xylitol ( Figure 2F ) . To test whether xylosyl-xylitol is produced generally by microbes as an intermediary metabolite during their growth on hemicellulose , we extracted and analyzed the metabolites from a number of ascomycetes species and B . subtilis grown on xylodextrins . Notably , these widely divergent fungi and B . subtilis all produce xylosyl-xylitols when grown on xylodextrins ( Figure 3A and Figure 3—figure supplement 1 ) . These organisms span over 1 billion years of evolution ( Figure 3B ) , indicating that the use of xylodextrin reductases to consume plant hemicellulose is widespread . 10 . 7554/eLife . 05896 . 015Figure 3 . Xylosyl-xylitol and xylosyl-xylosyl-xylitol production by a range of microbes . ( A ) Xylodextrin-derived carbohydrate levels seen in chromatograms of intracellular metabolites for N . crassa , T . reesei , A . nidulans and B . subtilis grown on xylodextrins . Compounds are abbreviated as follows: X1 , xylose; X2 , xylobiose; X3 , xylotriose; X4 , xylotetraose; xlt , xylitol; xlt2 , xylosyl-xylitol; xlt3 , xylosyl-xylosyl-xylitol . ( B ) Phylogenetic tree of the organisms shown to produce xylosyl-xylitols during growth on xylodextrins . Ages taken from Wellman et al . ( 2003 ) ; Galagan et al . ( 2005 ) ; Hedges et al . ( 2006 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 01510 . 7554/eLife . 05896 . 016Figure 3—figure supplement 1 . LC-MS/MS multiple reaction monitoring chromatograms of xylosyl-xylitols from cultures of microbes grown on xylodextrins . Shown are MS/MS transitions for xylosyl-xylitol ( in red , m/z 283 . 1035 → 151 . 0612 transition ) and xylosyl-xylosyl-xylitol ( in green , m/z 415 . 1457 → 151 . 0612 transition ) analyzed from intracellular metabolites of N . crassa , T . reesei , A . nidulans and B . subtilis grown on xylodextrins , after separation by liquid chromatography . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 016 We next tested whether integration of the complete xylodextrin consumption pathway would overcome the poor xylodextrin utilization by S . cerevisiae ( Figure 1 ) ( Fujii et al . , 2011 ) . When combined with the original xylodextrin pathway ( CDT-2 plus GH43-2 ) , GH43-7 enabled S . cerevisiae to grow more rapidly on xylodextrin ( Figure 4A ) and eliminated accumulation of xylosyl-xylitol intermediates ( Figure 4B–D and Figure 4—figure supplement 1 ) . The presence of xylose and glucose greatly improved anaerobic fermentation of xylodextrins ( Figure 5 and Figure 5—figure supplement 1 and Figure 5—figure supplement 2 ) , indicating that metabolic sensing in S . cerevisiae with the complete xylodextrin pathway may require additional tuning ( Youk and van Oudenaarden , 2009 ) for optimal xylodextrin fermentation . Notably , we observed that the XR/XDH pathway produced much less xylitol when xylodextrins were used in fermentations than from xylose ( Figure 5 and Figure 5—figure supplement 2B ) . Taken together , these results reveal that the XR/XDH pathway widely used in engineered S . cerevisiae naturally has broad substrate specificity for xylodextrins , and complete reconstitution of the naturally occurring xylodextrin pathway is necessary to enable S . cerevisiae to efficiently consume xylodextrins . 10 . 7554/eLife . 05896 . 017Figure 4 . Aerobic consumption of xylodextrins with the complete xylodextrin pathway . ( A ) Yeast growth curves with xylodextrin as the sole carbon source under aerobic conditions with a cell density at OD600 = 1 . Yeast strain SR8U without plasmids , or transformed with plasmid expressing CDT-2 and GH43-2 ( pXD8 . 4 ) , CDT-2 and GH43-7 ( pXD8 . 6 ) or all three genes ( pXD8 . 7 ) are shown . ( B–D ) Xylobiose consumption with xylodextrin as the sole carbon source under aerobic conditions with a cell density of OD600 = 20 . Xylosyl-xylitol ( xlt2 ) accumulation was only observed in the SR8U strain bearing plasmid pXD8 . 4 , that is , lacking GH43-7 . Error bars represent standard deviations of biological triplicates ( panels A–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 01710 . 7554/eLife . 05896 . 018Figure 4—figure supplement 1 . Culture media composition during yeast growth on xylodextrin . Yeast growth with xylodextrin as the sole carbon source ( concentration g/l ) under aerobic conditions with a cell density at OD600 = 20 . Yeast strain SR8 transformed with plasmid expressing CDT-2 and GH43-2 ( pXD8 . 4 ) , CDT-2 and GH43-7 ( pXD8 . 6 ) , or all three genes ( pXD8 . 7 ) . All growth experiments were performed in biological triplicate , and error bars indicate the standard deviation between experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 01810 . 7554/eLife . 05896 . 019Figure 5 . Anaerobic fermentation of xylodextrins in co-fermentations with xylose or glucose . ( A ) Anaerobic fermentation of xylodextrins and xylose , in a fed-batch reactor . Strain SR8U expressing CDT-2 , GH43-2 , and GH43-7 ( plasmid pXD8 . 7 ) was used at an initial OD600 of 20 . Solid lines represent concentrations of compounds in the media . Blue dotted line shows the total amount of xylose added to the culture over time . Error bars represent standard deviations of biological duplicates . ( B ) Anaerobic fermentation of xylodextrins and glucose , in a fed-batch reactor . Glucose was not detected in the fermentation broth . Error bars represent standard deviations of biological duplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 01910 . 7554/eLife . 05896 . 020Figure 5—figure supplement 1 . Anaerobic xylodextrin utilization in the presence of xylose . Strain carrying the complete xylodextrin pathway ( CDT-2 , GH43-2 , GH43-7 , XR/XDH ) grown under anaerobic conditions in oMM media ( Lin et al . , 2014 ) containing 4% xylose and 3% xylodextrin . The consumption of xylobiose ( X2 ) and xylotriose ( X3 ) stalled when xylose ( X1 ) was depleted and resumed after supplying additional xylose at hour 48 . This experiment is representative of those carried out with different xylose to xylodextrin ratios . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 02010 . 7554/eLife . 05896 . 021Figure 5—figure supplement 2 . Control anaerobic fermentations with S . cerevisiae strain expressing the complete xylodextrin utilization pathway . Strain SR8U with plasmid pXD8 . 7 expressing CDT-2 , GH43-2 , and GH43-7 was used at an initial OD600 of 20 . Solid lines represent concentrations of compounds in the media . Blue dotted line shows the total amount of xylose added to the culture over time . ( A ) Fermentation profile of the strain in oMM medium containing 4% xylodextrin in the reactor without feeding xylose . ( B ) Fermentation profile of the strain in oMM medium without xylodextrin in the reactor but with continuous xylose feeding . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 021 The observation that xylodextrin fermentation was stimulated by glucose ( Figure 5B ) suggested that the xylodextrin pathway could serve more generally for cofermentations to enhance biofuel production . We therefore tested whether xylodextrin fermentation could be carried out simultaneously with sucrose fermentation , as a means to augment ethanol yield from sugarcane . In this scenario , xylodextrins released by hot water treatment ( Hendriks and Zeeman , 2009; Agbor et al . , 2011; Vallejos et al . , 2012 ) could be added to sucrose fermentations using yeast engineered with the xylodextrin consumption pathway . To test this idea , we used strain SR8U engineered with the xylodextrin pathway ( CDT-2 , GH43-2 , and GH43-7 ) in fermentations combining sucrose and xylodextrins . We observe simultaneous fermentation of sucrose and xylodextrins , with increased ethanol yields ( Figure 6 ) . Notably , the levels of xylitol production were found to be low ( Figure 6 ) , as observed in cofermentations with glucose ( Figure 5B ) . 10 . 7554/eLife . 05896 . 022Figure 6 . Xylodextrin and sucrose co-fermentations . ( A ) Sucrose fermentation . Vertical axis , g/l; horizontal axis , time in hours . ( B ) Xylodextrin and sucrose batch co-fermentation using strain SR8U expressing CDT-2 , GH43-2 , and GH43-7 ( plasmid pXD8 . 7 ) . Vertical axis , g/l; horizontal axis , time in hours . The xylodextrins were supplied at 10 g/l which containing xylobiose ( 4 . 2 g/l ) and xylotriose ( 2 . 3 g/l ) . Not fermented in the timeframe of this experiment , the xylodextrin sample also included xylotetraose and xylopentaose , in addition to hemicellulose modifiers such as acetate . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 022 Using yeast as a test platform , we identified a xylodextrin consumption pathway in N . crassa ( Figure 7 ) that surprisingly involves a new metabolic intermediate widely produced in nature by many fungi and bacteria . In bacteria such as B . subtilis , xylosyl-xylitol may be generated by aldo-keto reductases known to possess broad substrate specificity ( Barski et al . , 2008 ) . The discovery of the xylodextrin consumption pathway along with cellodextrin consumption ( Galazka et al . , 2010 ) in cellulolytic fungi for the two major sugar components of the plant cell wall now provides many modes of engineering yeast to ferment plant biomass-derived sugars ( Figure 7 ) . An alternative xylose consumption pathway using xylose isomerase could also be used with the xylodextrin transporter and xylodextrin hydrolase GH43-2 ( van Maris et al . , 2007 ) . However , the XR/XDH pathway may provide significant advantages in realistic fermentation conditions with sugars derived from hemicellulose . The breakdown of hemicellulose , which is acetylated ( Sun et al . , 2012 ) , releases highly toxic acetate , degrading the performance of S . cerevisiae fermentations ( Bellissimi et al . , 2009; Sun et al . , 2012 ) . The cofactor imbalance problem of the XR/XDH pathway , which can lead to accumulation of reduced byproducts ( xylitol and glycerol ) and therefore was deemed a problem , can be exploited to drive acetate reduction , thereby detoxifying the fermentation medium and increasing ethanol production ( Wei et al . , 2013 ) . 10 . 7554/eLife . 05896 . 023Figure 7 . Two pathways of oligosaccharide consumption in N . crassa reconstituted in S . cerevisiae . Intracellular cellobiose utilization requires CDT-1 or CDT-2 along with β-glucosidase GH1-1 ( Galazka et al . , 2010 ) and enters glycolysis after phosphorylation by hexokinases ( HXK ) to form glucose-6-phosphate ( Glc-6-P ) . Intracellular xylodextrin utilization also uses CDT-2 and requires the intracellular β-xylosidases GH43-2 and GH43-7 . The resulting xylose can be assimilated through the pentose phosphate pathway consisting of xylose/xylodextrin reductase ( XR ) , xylitol dehydrogenase ( XDH ) , and xylulokinase ( XK ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 023 With optimization , that is , through improvements to xylodextrin transporter performance and chromosomal integration ( Ryan et al . , 2014 ) , the newly identified xylodextrin consumption pathway provides new opportunities to expand first-generation bioethanol production from cornstarch or sugarcane to include hemicellulose from the plant cell wall . For example , we propose that xylodextrins released from the hemicellulose in sugarcane bagasse by using compressed hot water treatment ( Hendriks and Zeeman , 2009; Agbor et al . , 2011; Vallejos et al . , 2012 ) could be directly fermented by yeast engineered to consume xylodextrins , as we have shown in proof-of-principle experiments ( Figure 6 ) . Xylodextrin consumption combined with glucose or cellodextrin consumption ( Figure 7 ) could also improve next-generation biofuel production from lignocellulosic feedstocks under a number of pretreatment scenarios ( Hendriks and Zeeman , 2009; Vallejos et al . , 2012 ) . These pathways could find widespread use to overcome remaining bottlenecks to fermentation of lignocellulosic feedstocks as a sustainable and economical source of biofuels and renewable chemicals . N . crassa strains obtained from the Fungal Genetics Stock Center ( FGSC ) ( McCluskey et al . , 2010 ) include the WT ( FGSC 2489 ) , and deletion strains for the two oligosaccharide transporters: NCU00801 ( FGSC 16575 ) and NCU08114 ( FGSC 17868 ) ( Colot et al . , 2006 ) . Conidia were inoculated at a concentration equal to 106 conidia per ml in 3 ml Vogel's media ( Vogel , 1956 ) with 2% wt/vol powdered Miscanthus giganteus ( Energy Bioscience Institute , UC-Berkeley ) , Avicel PH 101 ( Sigma-Aldrich , St . Louis , MO ) , beechwood xylan ( Sigma-Aldrich ) , or pectin ( Sigma-Aldrich ) in a 24-well deep-well plate . The plate was sealed with Corning breathable sealing tape and incubated at 25°C in constant light and with shaking ( 200 rpm ) . Images were taken at 48 hr . Culture supernatants were diluted 200 times with 0 . 1 M NaOH before Dionex high-performance anion exchange chromatographic ( HPAEC ) analysis , as described below . N . crassa growth on xylan was also determined by measuring N . crassa biomass accumulation . N . crassa grown on xylan for 3 days was harvested by filtration over a Whatman glass microfiber filter ( GF/F ) on a Büchner funnel and washed with 50 ml water . Biomass was then collected from the filter , dried in a 70°C oven , and weighed . Template gDNA from the N . crassa WT strain ( FGSC 2489 ) and from the S . cerevisiae S288C strain was extracted as described in http://www . fgsc . net/fgn35/lee35 . pdf ( McCluskey et al . , 2010 ) . Open reading frames ( ORFs ) of the β-xylosidase genes NCU01900 and NCU09652 ( GH43-2 and GH43-7 ) were amplified from the N . crassa gDNA template . For biochemical assays , each ORF was fused with a C-terminal His6-tag and flanked with the S . cerevisiae PTEF1 promoter and CYC1 transcriptional terminator in the 2µ yeast plasmid pRS423 backbone . Plasmid pRS426_NCU08114 was described previously ( Galazka et al . , 2010 ) . Plasmid pLNL78 containing the xylose utilization pathway ( xylose reductase , xylitol dehydrogenase , and xylulose kinase ) from S . stipitis was obtained from the lab of John Dueber ( Latimer et al . , 2014 ) . Plasmid pXD2 , a single-plasmid form of the xylodextrin pathway , was constructed by integrating NCU08114 ( CDT-2 ) and NCU01900 ( GH43-2 ) expression cassettes into pLNL78 , using the In-Fusion Cloning Kit ( Clontech ) . Plasmid pXD8 . 4 derived from plasmid pRS316 ( Sikorski and Hieter , 1989 ) was used to express CDT-2 and GH43-2 , each from the PCCW12 promoter . Plasmid pXD8 . 6 was derived from pXD8 . 4 by replacing the GH43-2 ORF with the ORF for GH43-7 . pXD8 . 7 contained all three expression cassettes ( CDT-2 , GH43-2 , and GH43-7 ) using the PCCW12 promoter for each . S . cerevisiae strain D452-2 ( MATa leu2 his3 ura3 can1 ) ( Kurtzman , 1994 ) and SR8U ( the uracil autotrophic version of the evolved xylose fast utilization strain SR8 ) ( Kim et al . , 2013 ) were used as recipient strains for the yeast experiments . The ORF for N . crassa xylose reductase ( xyr-1 , NcXR ) was amplified from N . crassa gDNA and the introns were removed by overlapping PCR . XR ORF was fused to a C-terminal His6-tag and flanked with the S . cerevisiae PCCW12 promoter and CYC1 transcriptional terminator and inserted into plasmid pRS313 . A list of the plasmids used in this study can be found in Table 1 . 10 . 7554/eLife . 05896 . 024Table 1 . A list of plasmids used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 05896 . 024PlasmidGenotype and useUseRef . pRS426_NCU08114PPGK1-CDT-2transport assay ( Galazka et al . , 2010 ) pRS423_GH43-2PTEF1-GH43-2enzyme purificationthis studypRS423_GH43-7PTEF1-GH43-7enzyme purificationthis studypRS313_NcXRPCCW12-NcXRenzyme purificationthis studypET302_EcGH43-7EcGH43-7enzyme purificationthis studypET302_BsGH43-7BsGH43-7enzyme purificationthis studypLNL78PRNR2-SsXK::PTEF1-SsXR::PTEF1-SsXDHfermentation ( Galazka et al . , 2010 ) pXD2PRNR2-SsXK::PTEF1-SsXR::PTEF1-SsXDH::PPGK1-CDT-2::PTEF1-GH43-2fermentationthis studypXD8 . 4PCCW12-CDT-2::PCCW12-GH43-2fermentationthis studypXD8 . 6PCCW12-CDT-2::PCCW12-GH43-7fermentationthis studypXD8 . 7PCCW12-CDT-2::PCCW12-GH43-7::PCCW12-GH43-7fermentationthis study S . cerevisiae was grown in an optimized minimum medium ( oMM ) lacking uracil into late log phase . The oMM contained 1 . 7 g/l YNB ( Sigma-Aldrich , Y1251 ) , twofold appropriate CSM dropout mixture , 10 g/l ( NH4 ) 2SO4 , 1 g/l MgSO4 . 7H2O , 6 g/l KH2PO4 , 100 mg/l adenine hemisulfate , 10 mg/l inositol , 100 mg/l glutamic acid , 20 mg/l lysine , 375 mg/l serine , and 100 mM 4-morpholineethanesulfonic acid ( MES ) , pH 6 . 0 ( Lin et al . , 2014 ) . Cells were then harvested and washed three times with assay buffer ( 5 mM MES , 100 mM NaCl , pH 6 . 0 ) and resuspended to a final OD600 of 40 . Substrate stocks were prepared in the same assay buffer at a concentration of 200 μM . Transport assays were initiated by mixing equal volumes of the cell suspension and the substrate stock . Reactions were incubated at 30°C with continuous shaking for 30 min . Samples were centrifuged at 14 , 000 rpm at 4°C for 5 min to remove yeast cells . 400 μl of each sample supernatant was transferred to an HPLC vial containing 100 μl 0 . 5 M NaOH , and the concentration of the remaining substrate was measured by HPAEC as described below . S . cerevisiae strains transformed with pRS423_GH43-2 , pRS423_GH43-7 , or pRS313_NcXR were grown in oMM lacking histidine with 2% glucose until late log phase before harvesting by centrifugation . E . coli strains BL21DE3 transformed with pET302_BsGH43-7 or pET302_EcGH43-7 were grown in TB medium , induced with 0 . 2 mM IPTG at OD600 of 0 . 8 , and harvested by centrifugation 12 hr after induction . Yeast or E . coli cell pellets were resuspended in a buffer containing 50 mM Tris–HCl , 100 mM NaCl , 0 . 5 mM DTT , pH 7 . 4 and protease inhibitor cocktail ( Pierce Biotechnology , Rockford , IL ) . Cells were lysed with an Avestin homogenizer , and the clarified supernatant was loaded onto a HisTrap column ( GE Healthcare , Sweden ) . His-tagged enzymes were purified with an imidazole gradient , buffer-exchanged into 20 mM Tris–HCl , 100 mM NaCl , pH 7 . 4 , and concentrated to 5 mg/ml . For the β-xylosidase assay of GH43-2 with xylodextrins , 0 . 5 μM of purified enzyme was incubated with 0 . 1% in-house prepared xylodextrin or 1 mM xylobiose ( Megazyme , Ireland ) in 1× PBS at 30°C . Reactions were sampled at 30 min and quenched by adding 5 vol of 0 . 1 M NaOH . The products were analyzed by HPAEC as described below . For pH profiling , acetate buffer at pH 4 . 0 , 4 . 5 , 5 . 0 , 5 . 5 , 6 . 0 , and phosphate buffer at 6 . 5 , 7 . 0 , 7 . 5 , 8 were added at a concentration of 0 . 1 M . For the β-xylosidase assay of GH43-2 and GH43-7 with xylosyl-xylitol , 10 µM of purified enzyme was incubated with 4 . 5 mM xylosyl-xylitol and 0 . 5 mM xylobiose in 20 mM MES buffer , pH = 7 . 0 , and 1 mM CaCl2 at 30°C . Reactions were sampled at 3 hr and quenched by heating at 99°C for 10 min . The products were analyzed by ion-exclusion HPLC as described below . For the xylose reductase assays of NcXR , 1 μM of purified enzyme was incubated with 0 . 06% xylodextrin and 2 mM NADPH in 1× PBS at 30°C . Reactions were sampled at 30 min and quenched by heating at 99°C for 10 min . The products were analyzed by LC-QToF as described below . Xylodextrin was purchased from Cascade Analytical Reagents and Biochemicals or prepared according to published procedures ( Akpinar et al . , 2009 ) with slight modifications . In brief , 20 g beechwood xylan ( Sigma–Aldrich ) was fully suspended in 1000 ml water , to which 13 . 6 ml 18 . 4 M H2SO4 was added . The mixture was incubated in a 150°C oil bath with continuous stirring . After 30 min , the reaction was poured into a 2-L plastic container on ice , with stirring to allow it to cool . Then 0 . 25 mol CaCO3 was slowly added to neutralize the pH and precipitate sulfate . The supernatant was filtered and concentrated on a rotary evaporator at 50°C to dryness . The in-house prepared xylodextrin contained about 30% xylose monomers and 70% oligomers . To obtain a larger fraction of short chain xylodextrin , the commercial xylodextrin was dissolved to 20% wt/vol and incubated with 2 mg/ml xylanase at 37°C for 48 hr . Heat deactivation and filtration were performed before use . Xylosyl-xylitol was purified from the culture broth of strain SR8-containing plasmids pXD8 . 4 in xylodextrin medium . 50 ml of culture supernatant was concentrated on a rotary evaporator at 50°C to about 5 ml . The filtered sample was loaded on an XK 16/70 column ( GE Healthcare ) packed with Supelclean ENVI-Carb ( Sigma–Aldrich ) mounted on an ÄKTA Purifier ( GE Healthcare ) . The column was eluted with a gradient of acetonitrile at a flow rate of 3 . 0 ml/min at room temperature . Purified fractions , verified by LC-MS , were pooled and concentrated . The final product , containing 90% of xylosyl-xylitol and 10% xylobiose , was used as the substrate for enzyme assays and as an HPLC calibration standard . N . crassa strain ( FGSC 2489 ) and Aspergillus nidulans were stored and conidiated on agar slants of Volgel's medium ( Vogel , 1956 ) with 2% glucose . Trichoderma reesei ( strain QM6a ) was conidiated on potato dextrose agar ( PDA ) plates . Condia from each fungi were collected by resuspending in water and used for inoculation at a concentration of 106 cells per ml . N . crassa and A . nidulans were inoculated into Volgel's medium with 2% xylodextrin . T . reesei was inoculated into Trichoderma minimal medium ( Penttilä et al . , 1987 ) with 2% xylodextrin . N . crassa , A . nidulans , and T . reesei were grown in shaking flasks at 25°C , 37°C , and 30°C respectively . After 40 hr , mycelia from 2 ml of culture were harvested and washed with water on a glass fiber filter and transferred to a pre-chilled screw-capped 2 ml tube containing 0 . 5 ml Zirconia beads ( 0 . 5 mm ) and 1 . 2 ml acidic acetonitrile extraction solution ( 80% Acetonitrile , 20% H2O , and 0 . 1 M formic acid , [Rabinowitz and Kimball , 2007] ) . The tubes were then plunged into liquid nitrogen . The harvest process was controlled within 30 s . Samples were kept at −80°C until extraction , as described below . B . sublitis was stored on 0 . 5× LB ( 1% tryptone , 0 . 5% yeast extract , and 0 . 5% NaCl ) agar plates . A single colony was inoculated into 0 . 5× LB liquid medium with 1% glucose and allowed to grow in a 37°C shaker overnight . An inoculum from the overnight culture was transferred to fresh 0 . 5× LB liquid medium with 1% xylodextrin at an initial OD600 of 0 . 2 . After 40 hr , 2 ml of the culture was spun down and washed with cold PBS solution . Zirconia beads and acidic acetonitrile extraction solution were added to the cell pellet . The tubes were then flash frozen immediately and kept at −80°C until extraction . For extraction , all samples were allowed to thaw at 4°C for 10 min , bead beat for 2 min , and vortexed at 4°C for 20 min . 50 µl of the supernatant from each sample was analyzed by LC-MS/MS ( see ‘Mass spectrometric analyses’ section ) . Yeast strains were pre-grown aerobically overnight in oMM medium containing 2% glucose , washed three times with water , and resuspended in oMM medium . For aerobic growth , strains were inoculated at a starting OD600 of 1 . 0 or 20 in 50 ml oMM medium with 3% wt/vol xylodextrins and cultivated in 250 ml Erlenmeyer flasks covered with four layers of miracle cloth , shaking at 220 rpm . At the indicated time points , 0 . 8 ml samples were removed and pelleted . 20 μl supernatants were analyzed by ion-exclusion HPLC to determine xylose , xylitol , glycerol , and ethanol concentrations . 25 μl of 1:200 diluted or 2 μl of 1:100 diluted supernatant was analyzed by HPAEC or LC-QToF , respectively , to determine xylodextrin concentrations . Anaerobic fermentation experiments were performed in a 1-L stirred tank bioreactor ( DASGIP Bioreactor system , Type DGCS4 , Eppendorf AG , Germany ) , containing oMM medium with 3% wt/vol xylodextrins inoculated with an initial cell concentration of OD600 = 20 . The runs were performed at 30°C for 107 hr . The culture was agitated at 200 rpm and purged constantly with 6 l/hr of nitrogen . For xylose plus xylodextrin co-fermentations , xylose was fed continuously at 0 . 8 ml/hr from a 25% stock . During the fermentation , 3 ml cell-free samples were taken each 4 hr with an autosampler through a ceramic sampling probe ( Seg-Flow Sampling System , Flownamics , Madison , WI ) . 20 μl of the supernatant fraction were analyzed by ion-exclusion HPLC to determine xylose , xylitol , glycerol , acetate , and ethanol concentrations . 2 μl of 1:100 diluted supernatant was analyzed by LC-QToF to determine xylodextrin concentrations . For glucose plus xylodextrin co-fermentations , glucose was fed continuously at 2 ml/hr from a 10% stock . Analytes were detected as described for xylose plus xylodextrin co-fermentations , with the addition of the measurement of glucose concentrations in the culture broth . Yeast strain SR8U with plasmid pXD8 . 7 was pre-grown aerobically to late-log phase in oMM medium lacking uracil and containing 2% glucose , washed with water , and resuspended in oMM medium . Media containing 75 g/l sucrose plus or minus 15 g/l xylodextrins were inoculated with 20 OD of the washed yeast seed culture and purged with N2 . Fermentations were carried out in 50 ml of oMM medium in 125 ml serum bottles shaking at 220 rpm in a 30°C shaker . At the indicated time points , 1 ml samples were removed and pelleted . 5 μl supernatants were analyzed by ion-exclusion HPLC to determine sucrose , glucose , fructose , xylose , xylitol , glycerol , and ethanol concentrations . 2 μl of 1:100 diluted supernatant was analyzed by LC-QToF , as described below , to determine xylodextrin concentrations . Ion-exclusion HPLC was performed on a Prominence HPLC ( Shimadzu , Japan ) equipped with a refractive index detector . Xylose fermentation samples were resolved on a Rezex RFQ-Fast Fruit H+ 8% column ( 100 × 7 . 8 mm , Phenomenex , Torrance , CA ) using a flow rate of 1 ml/min at 50°C . Xylodextrin fermentation samples were resolved on Aminex HPX-87H Column ( 300 × 7 . 8 mm , Bio-Rad , Hercules , CA ) at a flow rate of 0 . 6 ml/min at 40°C . Both columns used a mobile phase of 0 . 01 N H2SO4 . HPAEC analysis was performed on a ICS-3000 HPLC ( Thermo Fisher , Sunnyvale , CA ) using a CarboPac PA200 analytical column ( 150 × 3 mm ) and a CarboPac PA200 guard column ( 3 × 30 mm ) at 30°C . Following injection of 25 μl of diluted samples , elution was performed at 0 . 4 ml/min using 0 . 1 M NaOH in the mobile phase with sodium acetate gradients . For xylodextrin and xylosyl-xylitol separation , the acetate gradients were 0 mM for 1 min , increasing to 80 mM in 8 min , increasing to 300 mM in 1 min , keeping at 30 mM for 2 min , followed by re-equilibration at 0 mM for 3 min . Carbohydrates were detected using pulsed amperometric detection ( PAD ) and peaks were analyzed and quantified using the Chromeleon software package . All mass spectrometric analyses were performed on an Agilent 6520 Accurate-Mass Q-TOF coupled with an Agilent 1200 LC ( Agilent Technologies , Santa Clara , CA ) . Samples were resolved on a 100 × 7 . 8 mm Rezex RFQ-Fast Fruit H+ 8% column ( Phenomenex ) using a mobile phase of 0 . 5% formic acid at a flow rate of 0 . 3 ml/min at 55°C . To determine the accurate masses of the unknown metabolites , 2 µl of 1:100 diluted yeast culture supernatant was analyzed by LC-QToF . Nitrogen was used as the instrument gas . The source voltage ( Vcap ) was 3000 V in negative ion mode , and the fragmentor was set to 100 V . The drying gas temperature was 300°C; drying gas flow was 7 l/min; and nebulizer pressure was 45 psi . The ESI source used a separate nebulizer for the continuous , low-level introduction of reference mass compounds ( 112 . 985587 , 1033 . 988109 ) to maintain mass axis calibration . Data were collected at an acquisition rate of 1 Hz from m/z 50 to 1100 and stored in centroid mode . LC-MS/MS was performed to confirm the identity of xylosyl-xylitol and xylosyl-xylosyl-xylitol . The compound with a retention time ( RT ) of 5 . 8 min and m/z ratio of 283 . 103 and the compound with an RT of 4 . 7 min and m/z ratio of 415 . 15 were fragmented with collision energies of 10 , 20 , and 40 eV . MS/MS spectra were acquired , and the product ions were compared and matched to the calculated fragment ions generated by the Fragmentation Tools in ChemBioDraw Ultra v13 . To quantify the carbohydrates and carbohydrate derivatives in the culture , culture supernatants were diluted 100-fold in water and 2 µl was analyzed by LC-QToF . Spectra were imported to Qualtitative Analysis module of Agilent MassHunter Workstation software using m/z and retention time values obtained from the calibration samples to search for the targeted ions in the data . These searches generated extracted ion chromatograms ( EICs ) based on the list of target compounds . Peaks were integrated and compared to the calibration curves to calculate the concentration . Calibration curves were calculated from the calibration samples , prepared in the same oMM medium as all the samples , and curve fitting for each compound resulted in fits with R2 values of 0 . 999 . 4-morpholineethanesulfonic acid ( MES ) , the buffer compound in the oMM medium with constant concentration and not utilized by yeast , was used as an internal standard ( IS ) for concentration normalization .	Plants can be used to make ‘biofuels’ , which are more sustainable alternatives to traditional fuels made from petroleum . Unfortunately , most biofuels are currently made from simple sugars or starch extracted from parts of plants that we also use for food , such as the grains of cereal crops . Making biofuels from the parts of the plant that are not used for food—for example , the stems or leaves—would enable us to avoid a trade-off between food and fuel production . However , most of the sugars in these parts of the plant are locked away in the form of large , complex carbohydrates called cellulose and hemicellulose , which form the rigid cell wall surrounding each plant cell . Currently , the industrial processes that can be used to make biofuels from plant cell walls are expensive and use a lot of energy . They involve heating or chemically treating the plant material to release the cellulose and hemicellulose . Then , large quantities of enzymes are added to break these carbohydrates down into simple sugars that can then be converted into alcohol ( a biofuel ) by yeast . Fungi may be able to provide us with a better solution . Many species are able to grow on plants because they can break down cellulose and hemicellulose into simple sugars they can use for energy . If the genes involved in this process could be identified and inserted into yeast it may provide a new , cheaper method to make biofuels from plant cell walls . To address this challenge , Li et al . studied how the fungus Neurospora crassa breaks down hemicellulose . This study identified a protein that can transport molecules of xylodextrin—which is found in hemicellulose—into the cells of the fungus , and two enzymes that break down the xylodextrin to make simple sugars , using a previously unknown chemical intermediate . When Li et al . inserted the genes that make the transport protein and the enzymes into yeast , the yeast were able to use plant cell wall material to make simple sugars and convert these to alcohol . The yeast used more of the xylodextrin when they were grown with an additional source of energy , such as the sugars glucose or sucrose . Li et al . 's findings suggest that giving yeast the ability to break down hemicellulose has the potential to improve the efficiency of biofuel production . The next challenge will be to improve the process so that the yeast can convert the xylodextrin and simple sugars more rapidly .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "computational", "and", "systems", "biology" ]	2015	Expanding xylose metabolism in yeast for plant cell wall conversion to biofuels
Animals respond to mitochondrial stress with the induction of mitochondrial unfolded protein response ( UPRmt ) . A cascade of events occur upon UPRmt activation , ultimately triggering a transcriptional response governed by two transcription factors: DVE-1 and ATFS-1 . Here we identify SUMO-specific peptidase ULP-4 as a positive regulator of C . elegans UPRmt to control SUMOylation status of DVE-1 and ATFS-1 . SUMOylation affects these two axes in the transcriptional program of UPRmt with distinct mechanisms: change of DVE-1 subcellular localization vs . change of ATFS-1 stability and activity . Our findings reveal a post-translational modification that promotes immune response and lifespan extension during mitochondrial stress . The ability of an organism to cope with an ever-changing and challenging environment lies in its ability to activate stress responses . Failure to appropriately respond to different stresses and maintain cellular and organismal homeostasis could result in multiple diseases including metabolic and neurodegenerative disorders ( Jovaisaite et al . , 2014; Lee and Ozcan , 2014; Wang and Kaufman , 2012 ) . Animals respond to mitochondrial stress with the induction of mitochondrial unfolded protein response ( UPRmt ) , a surveillance program that monitors mitochondrial function and initiates mitochondria-to-nucleus crosstalk to maintain mitochondrial protein-folding homeostasis ( Benedetti et al . , 2006; Yoneda et al . , 2004 ) and coordinate the expression of electron transport chain ( ETC ) components in mitochondrial and nuclear genomes ( Houtkooper et al . , 2013 ) . UPRmt also elicits global changes to reprogram metabolism ( Nargund et al . , 2015; Nargund et al . , 2012 ) , activate immune responses ( Liu et al . , 2014; Melo and Ruvkun , 2012; Pellegrino et al . , 2014 ) and extend lifespan ( Durieux et al . , 2011; Merkwirth et al . , 2016; Tian et al . , 2016 ) . UPRmt signaling ultimately activates a transcriptional response governed by two transcription factors: ATFS-1 and DVE-1 . ATFS-1 contains an N-terminal mitochondrial targeting sequence and a C-terminal nuclear localization sequence . Under normal condition , ATFS-1 is imported into mitochondria , where it is degraded by mitochondrial protease LON . During mitochondrial stress , mitochondrial import efficiency is impaired , resulting in nuclear accumulation of ATFS-1 ( Nargund et al . , 2012 ) . ATFS-1 controls approximately half of the mitochondrial stress response genes , including those encoding mitochondrial-specific chaperones , proteases and immune response genes ( Nargund et al . , 2012 ) . ATFS-1 also regulates genes involved in metabolic reprogramming , such as those functioning in glycolysis ( Nargund et al . , 2015 ) . Another axis of the UPRmt transcriptional program relies on DVE-1 , a homeobox transcription factor homologous to human SATB1/SATB2 . Upon mitochondrial perturbation , DVE-1 translocates from cytosol to nucleus , binds to the open-up chromatins devoid of H3K9me2 , and initiates the transcription of mitochondrial stress response genes ( Haynes et al . , 2007; Tian et al . , 2016 ) . While several core components of UPRmt have been identified , the regulation , especially post-translational regulation of these components has not been reported . The Small Ubiquitin-like Modifier ( SUMO ) post-translational modifies a large number of proteins that function in diverse biological processes , including transcription , chromatin remodeling , DNA repair and mitochondrial dynamics ( Gill , 2004; Hay , 2005; Hendriks et al . , 2014; Prudent et al . , 2015; Wasiak et al . , 2007; Yeh et al . , 2000 ) . Growing evidence suggests that rather than modifying a single protein , SUMO often targets multiple proteins within a complex , or within a pathway ( Chymkowitch et al . , 2015; Hendriks et al . , 2014 ) . Similar to ubiquitination , conjugation of SUMO to its substrates involves an enzymatic cascade including an E1 activating enzyme , an E2 conjugating enzyme and E3 ligases that determine the specificity ( Flotho and Melchior , 2013 ) . SUMOylation is also a dynamic process , which can be reversed by a family of conserved Sentrin/SUMO-specific proteases ( SENPs ) ( Mukhopadhyay and Dasso , 2007 ) . In C . elegans , the SENP family consists of four SUMO proteases ( ubiquitin-like proteases , ULPs ) ULP-1 , ULP-2 , ULP-4 and ULP-5 . Among them , ULP-2 has been reported to deSUMOylate E-cadherin and promotes its recruitment to adherens junctions ( Tsur et al . , 2015 ) . Moreover , ULP-4 has been reported to deSUMOylate HMGS-1 to control mevalonate pathway activity during aging ( Sapir et al . , 2014 ) . Aberrant activity of SUMOylation drastically affects cellular homeostasis and has been linked with many diseases ( Flotho and Melchior , 2013; Mo et al . , 2005; Sarge and Park-Sarge , 2009; Seeler et al . , 2007 ) . It has been reported that SUMO could covalently modify Drp1 , a protein essential for mitochondrial dynamics ( Prudent et al . , 2015 ) . In addition , SUMOylation of a pathogenic fragment of Huntingtin , a PolyQ-repeats protein that specifically binds to the outer membrane of mitochondria and impairs mitochondrial function , has been reported to exacerbate neurodegeneration in a Drosophila Huntington’s disease model ( Costa and Scorrano , 2012; Panov et al . , 2002; Steffan et al . , 2004 ) . In the present study , we find that under mitochondrial stress , SUMO-specific peptidase ULP-4 is required to deSUMOylate DVE-1 and ATFS-1 to activate UPRmt in C . elegans . ULP-4 is also required to promote UPRmt-mediated innate immunity and lifespan extension . Our results reveal an essential and unexplored function of post-translational regulation in UPRmt signaling . Previously , we have performed a genome-wide RNAi screen to identify genes that are required for the activation of mitochondrial unfolded protein response ( UPRmt ) in C . elegans ( Liu et al . , 2014 ) . ulp-4 , a gene encoding ortholog of SUMO-specific peptidase in C . elegans , is one of the hits from our primary screen . RNAi of ulp-4 impaired the activation of UPRmt that is induced by mitochondrial inhibitor antimycin A , or RNAi of nuclear encoded mitochondrial gene spg-7 ( mitochondrial metalloprotease ) ( Figure 1A–B and Figure 1—figure supplement 1A ) . RNAi of cco-1 ( nuclear-encoded cytochrome c oxidase-1 subunit ) is also widely used to disrupt mitochondrial function and activate UPRmt ( Durieux et al . , 2011; Nargund et al . , 2012; Pellegrino et al . , 2014 ) . Consistently , deficiency of ulp-4 also suppressed the induction of endogenous mitochondrial chaperone genes hsp-6 and hsp-60 under cco-1 RNAi ( Figure 1C and Figure 1—figure supplement 1B ) . Notably , transcript level of ulp-4 was also elevated during mitochondrial stress ( Figure 1—figure supplement 1C ) . In contrast , ulp-4 RNAi did not affect the induction of endoplasmic reticulum ( ER ) stress reporter hsp-4p::gfp nor heat shock stress reporter hsp-16 . 2p::gfp ( Figure 1D–E ) . In addition , worms treated with ulp-4 RNAi were still able to induce the expression of endogenous hsp-4 or hsp-16 . 2 during ER or heat shock stress ( Figure 1F ) . RNAi of ulp-4 from L1 stage only delayed worm development a bit ( Figure 1—figure supplement 1D ) . To further exclude the possibility that the suppression of UPRmt by ulp-4 RNAi is due to developmental delay , we also treated worms with ulp-4 RNAi starting at L4 stage and observed the reduction of UPRmt as well ( Figure 1—figure supplement 1E ) . Overexpression of ULP-4 in ulp-4 RNAi worms rescued UPRmt activation ( Figure 1G and Figure 1—figure supplement 1F ) . Lastly , we crossed an ulp-4 ( tm1597 ) mutant allele that lacks 404nt in the promoter region of ulp-4 with hsp-6p::gfp reporter , and showed that the induction of UPRmt was also impaired in ulp-4 mutants ( Figure 1H–I and Figure 1—figure supplement 1G ) . To see if other SUMO peptidases have similar effects to regulate UPRmt , we treated C . elegans with ulp-1 , 2 , or 5 RNAi ( Figure 1J ) and tested for their abilities to induce UPRmt . Deficiency of ulp-1 , 2 , or 5 failed to suppress antimycin- or spg-7 RNAi-induced UPRmt ( Figure 1K ) , suggesting a specific role of ulp-4 in mediating UPRmt . Sequence alignment ( Katoh and Standley , 2013 ) of ulp-1 , 2 , 4 and 5 revealed that the only conserved region among them is the catalytic domain ( Figure 1—figure supplement 2 ) . Thus , the specificity of ULP-4 in UPRmt signaling might be due to its ability to specifically interact with other protein components in UPRmt pathway . Conversely , RNAi of the E1 SUMO activating enzyme aos-1 or the E2 SUMO conjugating enzyme ubc-9 in C . elegans induced UPRmt more potently ( Figure 1—figure supplement 3A–C ) . Moreover , RNAi of smo-1 , the only SUMO ortholog gene in C . elegans , induced only weak UPRmt under unstressed condition ( Figure 1—figure supplement 3D–E ) . However , upon mitochondrial stress , smo-1 RNAi further activated UPRmt ( Figure 1—figure supplement 3F–G ) . More importantly , smo-1 RNAi rescued ulp-4 deficiency-suppressed UPRmt ( Figure 1—figure supplement 3F–G ) . Taken together , these results suggest that ULP-4 plays a specific role to mediate mitochondrial stress response through its SUMO peptidase activity . To understand the molecular mechanism of ULP-4 in mediating UPRmt , we sought to identify its protein targets . We first performed a cherry-picked yeast two-hybrid screen to test if ULP-4 could interact with known UPRmt pathway components . We found that DVE-1 , a homeodomain-containing transcription factor in UPRmt ( Haynes et al . , 2007 ) , interacted with ULP-4 ( Figure 2A ) . Notably , DVE-1 could specifically interact with ULP-4 , but not ULP-2 or ULP-5 ( Note: overexpression of ULP-1 in yeast is lethal ) ( Figure 2—figure supplement 1A ) . Consistently , smo-1 is one of the top hits from our yeast two-hybrid screen with DVE-1 as bait ( Figure 2—figure supplement 1B ) . Four possible mechanisms may explain the interaction between SMO-1 and a prey protein in the yeast two-hybrid experiment ( Figure 2B ) : ( I ) SMO-1 covalently modifies the prey; ( II ) SMO-1 modifies an adaptor protein , which interacts with the prey; ( III ) SMO-1 non-covalently interacts with the prey; ( IV ) SMO-1 non-covalently interacts with an adaptor protein , which associates with the prey . To identify which mechanism explains our result ( Figure 2A ) , we deleted C-terminal tail of SMO-1 to expose a conserved di-glycine motif ( GG’: active form ) or deleted the di-glycine motif to inactivate SMO-1 ( ΔGG: inactive form ) . Only wild type SMO-1 or SMO-1 GG’ , but not SMO-1 ΔGG , interacted with DVE-1 ( Figure 2C and Figure 2—figure supplement 1C ) , indicating that SMO-1 either covalently modifies DVE-1 , or covalently modifies an adaptor protein that associates with DVE-1 . SUMOylated DVE-1 could be detected in worms , the level of which was elevated under ulp-4 RNAi ( Figure 2—figure supplement 1D ) . We therefore used GPS-SUMO web service , which predicts SUMOyaltion sites of a protein , to identify several lysine residues of DVE-1 that could potentially be SUMOylated ( Figure 2D ) ( Ren et al . , 2009; Zhao et al . , 2014 ) . To further map the SUMOylation site of DVE-1 , we first tested the interaction between SMO-1 and fragments of DVE-1 in the yeast two-hybrid experiment . We found that DVE-1 301–468 associated with SMO-1 ( Figure 2—figure supplement 1E ) , suggesting that SUMOylation site may resides in 301–468 amino acids of DVE-1 . We then employed site-direct mutagenesis to mutate K327 , K461 or K465 residue of DVE-1 to arginine , and tested for its ability to associate with SMO-1 . K327R , but not other mutations , abolished SMO-1-DVE-1 interaction ( Figure 2C ) . Furthermore , SUMOylation of DVE-1 could be detected when we expressed wild type , but not DVE-1 K327R , with SMO-1 and E2 conjugating enzyme UBC9 in 293 T cells ( Figure 2E ) . We also noted that the size shift of SUMOylated DVE-1 was about the molecular weight of two EGFP-SMO-1 . To exclude the possibility that DVE-1 has another SUMOylation sites in addition of K327 , we expressed a fragment of DVE-1 ( 295–354 ) , which contains only one lysine residue ( K327 ) in this polypeptide in 293 T cells . We found that the size shift of DVE-1 295–354 is corresponding to di-SMO-1 ( Figure 2—figure supplement 1F ) , suggesting that DVE-1 only contains one SUMOylation site . Consistently , SUMOylation of DVE-1 on K327 residue was also observed in C . elegans ( Figure 2F ) . Taken together , these results suggested that SMO-1 covalently modifies K327 residue of DVE-1 . To test if ULP-4 could directly deSUMOylate DVE-1 , we employed a yeast three-hybrid experiment to induce the expression of ULP-4 or ULP-5 in yeasts . The expression of ULP-4 or ULP-5 was driven by a met17 promoter , which could be induced when growth media is deficient for methionine . Induction of ULP-4 , but not ULP-5 , prevented yeast growth caused by SMO-1-DVE-1 interaction ( Figure 2G ) , suggesting that ULP-4 may removes SUMO moiety from DVE-1 . Overexpression of ULP-4 in worms decreased SUMOylation level of DVE-1 ( Figure 2H ) . In addition , expression of ULP-4 C48 , the catalytic domain of ULP-4 ( Letunic and Bork , 2018 ) in 293 T cells could deSUMOylate DVE-1 in mammalian cells ( Figure 2I ) . Lastly , if during mitochondrial stress , ULP-4 is indeed required to deSUMOylate DVE-1 , mutation of K327 to arginine that prevents DVE-1 SUMOylation would bypass the requirement of ULP-4 for UPRmt induction . Indeed , we found that a CRISPR/Cas9 knock-in strain with DVE-1 K327R mutation bypassed the requirement of ulp-4 and was capable to activate UPRmt under ulp-4 RNAi ( Figure 2J and Figure 2—figure supplement 1G ) . We next aimed to understand how SUMOylation affects DVE-1 function . During mitochondrial stress , DVE-1 translocates from cytosol to nucleus ( Haynes et al . , 2007; Tian et al . , 2016 ) ( Figure 3A and B ) . Inactivation of ulp-4 by RNAi abolished the nuclear accumulation of DVE-1 in cco-1 RNAi-treated worms ( Figure 3A–C ) . Conversely , overexpression of ulp-4 increased the nuclear accumulation of DVE-1 ( Figure 3D ) . More importantly , we found that the induction of UPRmt correlated well with DVE-1 subcellular localization . When we fed worms with ulp-4 RNAi for one generation and treated their progeny with ulp-4 RNAi for twenty-four hours to allow efficient ulp-4 knockdown , and then fed animals with cco-1 RNAi , we observed cytosolic accumulation of DVE-1 and suppression of UPRmt ( Figure 3E ) . However , when we treated progeny with a mixture of ulp-4 and cco-1 RNAi , which induced mitochondrial stress before ulp-4 was efficiently knocked down , DVE-1 was still able to translocate to the nucleus and induce UPRmt ( Figure 3E ) . We also tested the subcellular localization of DVE-1 K327R under ulp-4 RNAi . Different from wild type proteins , DVE-1 K327R constitutively localized in the nucleus of C . elegans , even if ulp-4 was knocked down by RNAi ( Figure 3F ) . Conversely , SUMO-mimetic DVE-1 constitutively localized in the cytosol ( Figure 3G ) . Taken together , during mitochondrial stress , ULP-4 deSUMOylates DVE-1 at K327 residue to allow its nuclear accumulation to initiate UPRmt . Aside from DVE-1 , we found that a fragment of ATFS-1 ( 372–472 ) , another transcription factor in UPRmt , was also able to interact with ULP-4 ( Figure 4A , note: expression of full-length ATFS-1 is toxic in yeast ) . Similar as DVE-1 , ATFS-1 could only interact with ULP-4 , but not ULP-2 and ULP-5 ( Figure 4B ) . Four lysine residues were predicted by GPS-SUMO tool to be potential SUMOylation sites in ATFS-1 ( Figure 4C ) ( Zhao et al . , 2014 ) . In vivo SUMOylation assay in 293 T cells identified K326 residue as the bona fide ATFS-1 SUMOylation site ( Figure 4D ) . SUMOylation of ATFS-1 was also observed in C . elegans , which could be abolished by mutating K326 residue to arginine ( Figure 4E ) . Expression of ULP-4 C48 , the catalytic domain of ULP-4 , in 293 T cells deSUMOylated ATFS-1 ( Figure 4F ) . Lastly , expression of ATFS-1 K326R in atfs-1 ( tm4525 ) hypomorphic allele bypassed the requirement of ulp-4 in UPRmt activation ( Figure 4G ) . To see if SUMO also affects ATFS-1 localization , we used hsp-16 . 2 heat shock promoter to drive the expression of GFP-tagged full-length ATFS-1 . Upon mitochondrial perturbation , ATFS-1 was still able to translocate to the nucleus when ULP-4 expression is diminished ( Figure 5A ) . It should be noted that it is very difficult to express full-length ATFS-1 in worms , probably due to toxicity ( expression of full-length ATFS-1 in yeast is lethal ) . Therefore , we made truncations of ATFS-1 , and found that ATFS-1Δ1-184 expressed well and localized in the nucleus ( Figure 5—figure supplement 1A ) . Abolishing SUMOylation site of ATFS-1 ( K326R ) did not affect subcellular localization of ATFS-1 Δ 1-184 either ( Figure 5—figure supplement 1B ) . Interestingly , we noticed that ulp-4 RNAi greatly reduced the protein level of ATFS-1Δ 1-184 ( Figure 5—figure supplement 1C ) , suggesting that ulp-4 RNAi may affect ATFS-1 expression , or stability . Seven hours after heat induction of ATFS-1 expression , levels of ATFS-1Δ 1-184 were comparable in control or ulp-4 RNAi animals ( Figure 5B ) . However , after 24 hours , ATFS-1Δ 1-184 level in ulp-4 RNAi treated animals significantly decreased ( Figure 5B ) . Treating worms with proteasome inhibitor MG132 partially rescued the reduction of ATFS-1 Δ 1-184 under ulp-4 RNAi ( Figure 5—figure supplement 1C ) , suggesting that ulp-4 RNAi affects ATFS-1 protein stability . Full-length ATFS-1 proteins could be detected when mitochondrial protease LON-1 was inhibited ( Nargund et al . , 2012 ) . Treating worms with lon-1 RNAi , we showed that full-length ATFS-1 levels were also reduced under ulp-4 RNAi ( Figure 5C ) . Mutation ( K326R ) that abolished ATFS-1 SUMOylation partially restored its protein level under ulp-4 RNAi ( Figure 5D–E ) . In contrary , fusion of SMO-1 to mimic SUMOylated form of ATFS-1 significantly reduced its protein level ( Figure 5D–E ) . Taken together , these results suggest that SUMOylation reduces the stability of ATFS-1 . It has been shown that ATFS-1Δ1-32 , with impaired mitochondrial targeting sequence , could be expressed in the nucleus of HeLa cells ( Nargund et al . , 2012 ) . Therefore , it might be possible to directly test the transcriptional activity of ATFS-1Δ1-32 in mammalian system . We employed a luciferase reporter assay , in which transcription factor of interest ( e . g . wild type ATFS-1Δ1-32 , ATFS-1Δ1-32 K326R or SMO-1-ATFS-1Δ1-32 ) is fused with Gal4 binding domain ( BD ) to drive the expression of luciferase ( Figure 5F and Figure 5—figure supplement 1D ) . We found that mutation of ATFS-1 SUMOylation site greatly enhanced its transcriptional activity , whereas SUMO-mimetic significantly impaired the activity ( Figure 5G ) . Lastly , transcriptions of genes known to be driven by ATFS-1 ( eg . gpd-2 and gst-14 ) ( Nargund et al . , 2012; Nargund et al . , 2015 ) were blocked by ulp-4 RNAi , which could be partially rescued with ATFS-1 K326R mutation ( Figure 5H ) . Thus , SUMOylation also impairs the transcriptional activity of ATFS-1 . During mitochondrial stress , UPRmt not only initiates mitochondrial protective responses , but also activates immune responses and extends worm lifespan ( Liu et al . , 2014; Merkwirth et al . , 2016; Pellegrino et al . , 2014; Tian et al . , 2016 ) . The essential function of ulp-4 in signaling UPRmt makes it likely to be crucial for animal fitness during mitochondrial stress . Indeed , worms treated with cco-1 RNAi had a severe synthetic growth defect on ulp-4 RNAi ( Figure 6A ) . Consistently , ulp-4 ( tm1597 ) mutants revealed a more severe developmental delay when grown on spg-7 RNAi ( Figure 6—figure supplement 1A–B ) . Mutation of the SUMOylation sites of ATFS-1 and DVE-1 in C . elegans partially rescued the developmental delay of spg-7 mutants ( Figure 6—figure supplement 1C–E ) . The survival rate of worms exposed to high dosage of antimycin was also significantly reduced in ulp-4 RNAi ( Figure 6—figure supplement 1F–G ) . A broad range of microbes isolated from natural habitats harboring wild C . elegans populations could perturb mitochondrial function and induce the expression of hsp-6p::gfp ( Liu et al . , 2014 ) . ulp-4 RNAi also impaired UPRmt activation when we challenged worms with a Pseudomonas strain , a mitochondrial insult isolated from the natural habitat of C . elegans ( Liu et al . , 2014 ) ( Figure 6B ) . Moreover , ulp-4 RNAi suppressed the activation of immune response and xenobiotic detoxification response ( Figure 6C ) . To further validate the requirement of ulp-4 in UPRmt-mediated immune response , we treated irg-1p::gfp transgenic worms , a reporter strain for pathogen-infected response ( Estes et al . , 2010 ) , with control or ulp-4 RNAi and then challenged them with Pseudomonas . We showed that deficiency of ulp-4 greatly suppressed the induction of irg-1 ( Figure 6D ) . Deficiency of ulp-4 also impaired worm development and survival rate when they were infected with Pseudomonas ( Figure 6E–F ) ( Kirienko et al . , 2014 ) . The reduced survival rate of ulp-4-deficient worms could be rescued with atfs-1 K326R; dve-1 K327R mutations , further demonstrating that deSUMOylation of ATFS-1 and DVE-1 is the major function of ULP-4 during mitochondrial stress ( Figure 6G ) . Finally , we analyzed the lifespans of control or ulp-4 RNAi worms . Under unstressed condition , ulp-4 RNAi did not affect worm lifespan . However , ulp-4 RNAi greatly suppressed the lifespan extension in cco-1 RNAi-treated worms , which could be rescued by mutations of SUMOylation site within ATFS-1 and DVE-1 ( Figure 6H ) . Overexpression of ulp-4 neither affects the basal level of UPRmt under unstressed condition ( Figure 1G ) , nor affects worm lifespans with or without mitochondrial stress ( Figure 6—figure supplement 1H ) . In contrary , smo-1 RNAi shortened worm lifespan , but greatly extended the lifespan of spg-7 mutants ( Figure 6—figure supplement 1I ) . Mutations of SUMOylation site within ATFS-1 and DVE-1 extended lifespan of spg-7 mutants as well ( Figure 6—figure supplement 1J ) . Taken together , these results suggest that ulp-4 is required for mitochondrial stress-induced lifespan extension . In summary , our studies indicate that mitochondrial stress signals through ULP-4 , which deSUMOylates DVE-1 and ATFS-1 to modulate their localization , stability and transcriptional activity . Consequences of these events are elevated innate immunity and prolonged lifespan ( Figure 7 ) . We have identified a SUMO-specific peptidase ULP-4 that participates in C . elegans UPRmt . ULP-4 regulates the entire transcriptional program of UPRmt , underscoring the importance of ULP-4-mediated deSUMOylation in UPRmt signaling . However , how mitochondrial stress signals to ULP-4 warrants future analysis . SUMOylation affects protein function through several mechanisms , including changes of protein conformation , protein–protein interaction , protein stability and subcellular localization ( Chymkowitch et al . , 2015 ) . Interestingly , we find that SUMOylation affects DVE-1 and ATFS-1 through two distinct mechanisms: change of DVE-1 subcellular localization vs . changes of ATFS-1 stability and transcriptional activity . DVE-1 and ATFS-1 constitute the two axes in the transcriptional program of UPRmt , each might regulate a different subset of downstream genes . For instance , ATFS-1 has been shown to be the primary factor that controls the expression of genes involved in mitochondrial protein folding , glycolysis , xenobiotic detoxification and immune response ( Nargund et al . , 2012; Pellegrino et al . , 2014 ) . A detailed analysis of DVE-1 and ATFS-1 substrate selection may facilitate the understanding of why cells employ such intricate regulation of transcriptional response during mitochondrial stress . DVE-1 is homologous to mammalian SATB class of proteins that function in chromatin remodeling and transcription . Interestingly , it is reported that SATB1 and SATB2 could also be SUMOylated . For example , SUMOylation of SATB2 targets it to the nuclear periphery , where it regulates immunoglobulin μ gene expression ( Dobreva et al . , 2003 ) . SUMOylation of SATB1 targets it to the promyelocytic leukemia ( PML ) nuclear bodies where it undergoes caspase-mediated cleavage ( Tan et al . , 2008 ) . SATB1 has also been shown to form a ‘cage’-like distribution and anchors specialized DNA sequences onto its network ( Cai et al . , 2003 ) . Histone H3K9 and H3K14 acetylation mark the binding sites of SATB1 , whereas in SATB1 deficient cells , these sites are marked by H3K9 methylation ( Cai et al . , 2003 ) . Similarly , studies in C . elegans reported that during mitochondrial perturbation , H3K9 di-methylation globally marks chromatin , leaving portions of chromatin open-up where binding of DVE-1 occurs ( Tian et al . , 2016 ) . All these findings point to the conserved function and regulatory mechanisms of DVE-1 and SATB1 . Therefore , it will be interesting in the future to directly test if SATB1 functions as mammalian DVE-1 to signal UPRmt . Furthermore , ATF5 has been reported to constitute mammalian homolog of ATFS-1 ( Fiorese et al . , 2016 ) . It will also be interesting to see if SUMOylation can affect the stability and activity of ATF5 . Several mitochondrial quality control processes have evolved to maintain and restore proper mitochondrial function , including mitochondrial unfolded protein response ( UPRmt ) , mitochondrial dynamics , and mitophagy ( Andreux et al . , 2013 ) . Cells selectively activate each quality control pathway , depending on the stress level of mitochondria ( Andreux et al . , 2013; Pellegrino et al . , 2013 ) . Mild mitochondrial inhibition is often associated with the activation of UPRmt to maintain and restore proteostasis . As stress exceeds the protective capacity of UPRmt , cells may employ mitochondrial fusion to dilute damaged materials , and activate mitochondrial fission to isolate severely damaged mitochondria for removal through mitophagy . Drp1 , the central protein that controls mitochondrial fission , could be SUMOylated ( Prudent et al . , 2015 ) . A RING-finger containing protein MAPL functions as the E3 ligase to promote Drp1 SUMOylation on the mitochondria . SUMOylated Drp1 facilitates cristae remodeling , calcium flux and release of cytochrome c , and stabilizes ER/mitochondrial contact sites . Whether SUMOylation affects other proteins in the mitochondrial quality control processes , such as those govern mitophagy , are worth to explore . The discovery of SUMOylation in modulating UPRmt opens up a new research direction to study post-translational regulation of UPRmt , and UPRmt-mediated immunity and longevity . SJ4100 ( zcIs13[hsp-6p::gfp] ) , CL2070 ( dvIs70[hsp-16 . 2p::gfp] ) , SJ4005 ( zcIs4[hsp-4p::gfp] ) , SJ4058 ( zcls9[hsp-60p::gfp] ) , SJ4198 ( zcls39[dve-1p::dve-1::gfp] ) , ulp-4 ( tm1597 ) , spg-7 ( ad2249 ) and N2 wild-type worms were obtained from Caenorhabditis Genetics Center . hsp-60p::gfp;atfs-1 ( tm4525 ) is a generous gift from Dr . Cole Haynes . dve-1p::dve-1::gfp plasmid is a gift from Dr . Cole Haynes . We used site-direct mutagenesis to generate dve-1p::dve-1 K327R::gfp plasmid and micro-injected into worms . For generation of hsp-16 . 2p::dve-1::smo-1::gfp worms , hsp-16 . 2 promoter , dve-1::smo-1 and gfp sequences were sub-cloned into pDD49 . 26 vector . Three glycine residues were placed as linker between smo-1 and dve-1 . HEK293T cell was obtained from ATCC , which was authenticated by ATCC . Cells were validated to be free of mycoplasma contamination . No commonly misidentified cell lines were used . C . elegans were cultured at 20°C and fed with E . coli OP50 on Nematode Growth Media unless otherwise indicated . Yeast strain AH109 for yeast two-hybrid and three-hybrid assays was cultured with YPDA media at 30°C unless otherwise indicated . 293 T cells were cultured with DMEM ( 10% fetal bovine serum ) at 37°C . For RNAi-induced UPRmt , RNAi bacteria were grown in LB containing 50 µg/ml carbenicillin at 37°C overnight . 200 µl of RNAi bacteria was seeded onto 6 cm worm plates with 5 mM IPTG . Dried plates were kept at room temperature overnight to allow IPTG induction of dsRNA expression . Synchronized L1 worms were raised on the RNAi plates at 20°C . After 24 hr , 200 µl 10X concentrated RNAi ( atp-2 , cco-1 or spg-7 ) bacteria were provided . GFP expressions were imaged after 48 hr . For antimycin A induced UPRmt , synchronized L1 worms were raised on 6 cm worm plates for 48 hr . 200 µl of 20 µg/ml antimycin were then provided . Fluorescent images were taken 24 hr after the addition of antimycin . For P . aeruginosa induced UPRmt , synchronized L1 worms were raised on 6 cm worm plates for 24 hr before exposure to P . aeruginosa . Worms were imaged at adulthood day 1 . Worms with each indicated fluorescent reporter were dropped in 100 mM NaN3 droplet on 2% agarose pads and imaged with a Zeiss Imager M2 microscopy . Worms raised under each described condition were washed off plates with M9 buffer and then washed several times with M9 until supernatant was clear . 2X SDS Laemmli buffer ( 4% SDS , 20% glycerol , 10% 2-mercaptoethanol , 0 . 004% bromophenol blue , 0 . 125M tris-HCl , pH 6 . 8 ) was used to re-suspend the worm pellet . Samples were boiled at 95°C for 5 min . Lysates containing the same amount of protein were loaded onto SDS-PAGE and transferred onto PVDF membranes ( Bio-Rad ) . After blocked with 5% non-fat milk , the membrane was probed with the designated first and second antibodies ( mouse monoclonal anti-GFP , sungen biotech #KM8009; rabbit polyclonal anti-GFP , abcam #ab290; anti-Myc , CST #2276; anti-tubulin , abcam #ab6161; anti-UBC9 , abcam #ab75854; anti-ATFS-1 , anti-DVE-1 and anti-SMO-1 antibodies were developed by abclonal ) , developed with the enhanced chemiluminescence method ( Pierce , CAT#32106 ) , and visualized by Tanon 5200 chemical luminescence imaging system . The result analysis was performed by ImageCal ( Tanon ) . 293 T cells in 10 cm plate were transfected with 3 µg EGFP-SMO-1 , 2 µg UBC9 , 5 µg MYC-DVE-1 or MYC-ATFS-1△1-32 plasmids via lipofectamine 3000 ( Invitrogen , CAT#L3000015 ) or PEI . For ULP-4 deSUMOylation assay , additional 5 µg MYC-ULP-4 C48 plasmid was co-transfected . 24 hr after transfection , cells were washed with 1X PBS buffer , scraped off the plates and pelleted by centrifugation at 1000 rpm for 1 min . Immunoprecipitation was performed at 4°C or on ice . Cell pellet was resuspended in 1 ml lysis buffer ( 50 mM tris-HCl pH 7 . 5 , 150 mM NaCl , 1% NP40 , 1 mM EDTA , 1 mM EGTA , 20 mM N-Ethylmaleimide , proteinase inhibitor cocktail ) and sonicated . Samples were spun down at 21 , 000 g for 15 min . Supernatants were transferred into new tubes . Protein concentration was quantitated by BCA assay ( Thermo , CAT#23225 ) . Same amount of protein wasused for immunoprecipitation with appropriate antibodies . Samples were incubated with agitation at 4°C for 4 hr . 20 µl pre-washed protein G beads ( Invitrogen , CAT#10004D ) were added to each sample and rotated for additional 2 hr at 4°C . Protein G beads were washed four times with lysis buffer . 50 ul 2X SDS Laemmli ( 4% SDS , 20% glycerol , 10% 2-mercaptoenthnol , 0 . 004 bromophenol blue , 0 . 125M Tri-HCL , pH 6 . 8 ) buffer was added to the beads and boiled for 5 min at 95°C . Worms were fed with each indicated RNAi or exposed to pathogen . Adulthood day one worms were washed off plates and washed several times with M9 buffer until supernatant was clear . Worm pellets were re-suspended with TRIzol reagent ( cwbiol , CAT#cw0580A ) . Samples were frozen and thawed six times to crack worms . Total RNA was isolated by chloroform extraction , followed by ethanol precipitation and DNase treatment . cDNA was then synthesized by reverse transcription ( transgene biotech , CAT#AT311-03 ) . Quantitative real-time PCR was carried out using SYBR GREEN PCR Master Mix ( Bio-Rad , CAT#1725121 ) . Quantification of transcripts was normalized to act-3 . Each indicated gene was cloned into pGADT7 or pGBDT7 plasmid ( Clontech ) . Plasmids were then transformed into AH109 yeast strain . We first seeded yeast on –Trp and -Leu double dropout solid culture medium . After yeast colonies were formed , we picked individual yeast colony into sterile water , adjusted it to the same OD and dropped onto -Trp , -Leu , -His and -Ade four dropout solid culture medium . Images of yeast were taken after culturing at 30°C for 2–4 days . Full-length dve-1 was cloned into pGBDT7 , and transformed into AH109 yeast strain to test for auto-activation and protein expression . We then transformed C . elegans AD library plasmids into yeast that already contains dve-1 BD plasmid . Yeasts were seeded onto four dropout solid culture medium and grown at 30°C for a week . Each colony was picked , followed by mini-prep to extract AD plasmid for PCR and sequencing . Yeast three-hybrid for DVE-1 deSUMOylation were carried out using pBRIDGE and pGADT7 vectors . smo-1 was cloned into pGADT7 . dve-1 alone , or together with ulp-4/5 , were cloned into pBRIDGE . pGADT7 and pBRIDGE plasmids were transformed into AH109 strain and cultured on -Leu and -Trp double dropout solid culture medium for yeast growth , on -Leu , -Trp and -His dropout medium for SMO-1 and DVE-1 interaction assay , on -Met , -Leu , -Trp and -His dropout medium to induce the expression of ULP-4 or ULP-5 and test for deSUMOylation activity . Worms were grown on control or ulp-4 RNAi plates for 48 hr before heat shock at 37°C for 1 hr . The worms were then transferred in M9 containing indicated RNAi bacteria and 100uM MG132 . Images were taken after 24 hr . Each indicated gene was cloned into pCMV-BD vector . We then transfected 0 . 5 µg pCMV-BD , 0 . 5 µg pFR-luci ( Photinus pyralis ) and 0 . 1 µg pActin-luci ( Renilla reniformis ) into one 24-well of 293 T cells . 24 hr after transfection , we assayed luciferase activity with dual luciferase reporter assay system ( Promega , CAT#1910 ) . To assay developmental delay induced by P . aeruginosa , worms were pre-treated with control or ulp-4 RNAi at 20°C for 16 hr . P . aeruginosa was then dropped onto worm plates . Images were taken two days later . To assay survival rate of worms under antimycin treatment , worms were grown at 20°C on 6-well RNAi plates seeded with control or ulp-4 RNAi . After 48 hr , 20 µg antimycin A were provided . Lifespan analyses were conducted on RNAi plates at 20°C . More than 100 synchronized L1 were seeded onto 6 cm worm plates with control or ulp-4 RNAi . Animals that did not move when gently touched were scored as dead . Worms were transferred every 2 days to new plates during the first 10 days and were transferred every 3–5 days afterwards . Lifespan experiments were performed twice . PA14 survival assay was carried out as described previously ( Kirienko et al . , 2014 ) . PA14 was freshly streaked from frozen stock and cultured 37°C for 16 hr . 10 µL PA14 were then spread onto 3 . 5 cm slow killing agar plates ( 3 . 5 mg/mL peptone , 3 mg/mL NaCl , 17 mg/mL Agar , 1 mM MgSO4 , 25 mM 1MKH2PO4 , pH 6 , 1 mM CaCla , 5 µg/mL cholesterol ) . The slow killing plates seeded with PA14 were cultured at 37°C for 24 hr . FUDR was spread on the plates and L4 worms were picked onto plates and cultured at 25°C . Worms were scored at least three times per day after16 hours , until all worms were dead . PA14 survival assay was performed 2 times with three replications each time . The pDD162 CRISPR/Cas9 expression plasmid was obtained from Addgene ( #47549 ) . C . elegans Cas9 target prediction tool ( https://crispr . cos . uni-heidelberg . de ) was used to design target sequences . Templates for recombination were cloned into pDD49 . 26 vector with ~500 bp overhang upstream and downstream of the target . Cas9 pam sequences were mutated in the templates . The plasmid was injected into worms , with Cas9-rol-6 as co-injection marker . Rolling worms were singled and validated by PCR and sequencing . Primers: dve-1 K327R F: TTTCCATCAATATTCGACCAGAACCGG dve-1 K327R R: CGAATATTGATGGAAAACAAAGTATCTTGAATAGTTTC . atfs-1 K326R F: TTTTAAGCGTCCAGAAGCATTTTTCCGGGAAGAACCCATG atfs-1 K326R R: CGGAAAAATGCTTCTGGACGCTTAAAAACGTC All experiments in this paper , if not specifically indicated , have been repeated for at least three times . Statistical analysis was performed with GraphPad . DVE-1 subcellular localization , transcriptional activity assay , QPCR and worm length were analyzed by Student’s t-test . PA14 survival and lifespan assays were analyzed using Log-Rank method . p<0 . 05 , p<0 . 002 and **p<0 . 0002 . Worm lengths , relative gfp expression and immunoblot quantification were analyzed by Image J . DVE-1 subcellular localization was counted in more than 40 worms per plate and three independent replicates were analyzed for each condition . Transcriptional activity assay wasanalyzed with three independent replicates for each condition . PA14 survival assay was performed with more than 35 worms per plate and at least two independent replicates were analyzed for each condition . The experiment was repeated for at least two times . Lifespan assay was performed with two biological replicates .	Most animal cells carry compartments called mitochondria . These tiny powerhouses produce the energy that fuels many life processes , but they also store important compounds and can even cause an infected or defective cell to kill itself . For a cell , keeping its mitochondria healthy is often a matter of life and death: failure to do so is linked with aging , cancer or diseases such as Alzheimer’s . The cell uses a surveillance program called the mitochondrial unfolded protein response to assess the health of its mitochondria . If something is amiss , the cell activates specific mechanisms to fix the problem , which involves turning on specific genes in its genome . A protein named ULP-4 , which is found in the worm Caenorhabditis elegans but also in humans , participates in this process . This enzyme cuts off chemical ‘tags’ known as SUMO from proteins . Adding and removing these labels changes the place and role of a protein in the cell . However , it was still unclear how ULP-4 played a role in the mitochondrial unfolded protein response . Here , Gao et al . show that when mitochondria are in distress , ULP-4 removes SUMO from DVE-1 and ATFS-1 , two proteins that control separate arms of the mitochondrial unfolded protein response . Without SUMO tags , DVE-1 can relocate to the area in the cell where it can turn on genes that protect and repair mitochondria; meanwhile SUMO-free ATFS-1 becomes more stable and can start acting on the genome . Finally , the experiments show that removing SUMO on DVE-1 and ATFS-1 is essential to keep the worms healthy and with a long lifespan under mitochondrial stress . The experiments by Gao et al . show that the mitochondrial unfolded protein response relies , at least in part , on SUMO tags . This knowledge opens new avenues of research , and could help fight diseases that emerge when mitochondria fail .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2019	SUMO peptidase ULP-4 regulates mitochondrial UPR-mediated innate immunity and lifespan extension
Most vertebrate and plant RNA and small DNA viruses suppress genomic CpG and UpA dinucleotide frequencies , apparently mimicking host mRNA composition . Artificially increasing CpG/UpA dinucleotides attenuates viruses through an entirely unknown mechanism . Using the echovirus 7 ( E7 ) model in several cell types , we show that the restriction in E7 replication in mutants with increased CpG/UpA dinucleotides occurred immediately after viral entry , with incoming virions failing to form replication complexes . Sequences of CpG/UpA-high virus stocks showed no evidence of increased mutational errors that would render them replication defective , these viral RNAs were not differentially sequestered in cytoplasmic stress granules nor did they induce a systemic antiviral state . Importantly , restriction was not mediated through effects on translation efficiency since replicons with high CpG/UpA sequences inserted into a non-coding region were similarly replication defective . Host-cells thus possess intrinsic defence pathways that prevent replication of viruses with increased CpG/UpA frequencies independently of codon usage . A primary function of the genomes of RNA viruses is to code for viral genes that replicate and package genomic RNA for new rounds of infection . It is increasingly recognised , however , that RNA virus genomes possess a range of other organisational features , such as formation of RNA secondary and tertiary structures that interact with host cell elements , such as ribosomal proteins in viral internal ribosomal entry sites and in the encoding of replication structures such as the cis- active replication element of picornaviruses ( Martínez-Salas et al . , 2015; Goodfellow et al . , 2003; Steil and Barton , 2009 ) . The genomes of RNA viruses are also subject to a range of poorly understood mutational and compositional constraints , with substantial variability in G + C content and the apparent avoidance of certain dinucleotides ( the use of two adjacent nucleotides in a linear sequence ) , such as CpG and UpA ( Simmonds et al . , 2013; Karlin et al . , 1994; Rima and McFerran , 1997 ) . At least in part , this pattern of under-representation may be shared by the hosts they infect where suppression of CpG and UpA is widespread . In coding sequences of most organisms , TpA ( UpA in RNA ) is under-represented while vertebrate and plant genomes additionally show strong suppression of CpG dinucleotides ( Josse et al . , 1961; Russell et al . , 1976 ) . UpA dinucleotides in cytoplasmic mRNA and likely also viral RNA are under direct selection as the dinucleotide is recognised by RNA-degrading enzymes in the cytoplasm . The degree of UpA dinucleotides in a RNA molecule has therefore been hypothesized to control cellular RNA turn-over ( Duan and Antezana , 2003; Beutler et al . , 1989 ) . A different , enzymatic mechanism underlies the suppression of CpG in host genomes; the cytosine in a CpG dinucleotide can be methylated , making it more likely to deaminate into a thymine . This selectively reduces CpG dinucleotide frequencies in both plant and vertebrate genomes where DNA methylation is extensive ( Coulondre et al . , 1978; Bird , 1980 ) . Most small DNA viruses and viruses with single stranded RNA genomes appear to mimic host-cell mRNA dinucleotide frequencies , with a strong bias in both UpA and CpG dinucleotide frequencies in viruses of plants and vertebrates ( Simmonds et al . , 2013; Karlin et al . , 1994; Rima and McFerran , 1997 ) . In contrast , the genome of many invertebrates lack methylation and consequently show little if any suppression of CpG dinucleotide frequencies . Consistent with the hypothesis for virus mimicry , the genomes of viruses that infect invertebrates show little suppression of CpG ( Lobo et al . , 2009; Simmonds et al . , 2013 ) . There is abundant evidence that modifying dinucleotide frequencies has a direct functional effect on virus replication . For example , increasing CpG or UpA dinucleotides in coding regions of echovirus7 ( E7 , enterovirus , Picornaviridae ) while keeping protein coding identical strongly attenuated E7 independently of classical antiviral signalling pathways , whereas removing CpG and UpA dinucleotides increased viral replication rates in vitro beyond those of WT virus ( Atkinson et al . , 2014; Witteveldt et al . , 2016 ) . In mice , infections with influenza virus ( Influenzavirus A , Orthomyxoviridae ) with increased CpG and UpA dinucleotide frequencies were entirely non-pathogenic and viral loads were markedly reduced . Such infections did however elicit inflammatory cytokine production and T-cell and antibody responses equal to or exceeding that of wild type virus and conferred complete protection from a lethal challenge dose of wild type influenza virus ( Gaunt et al . , 2016 ) . Because of the compact nature of RNA virus genomes , alteration of CpG and UpA frequencies in viruses in these previous studies inevitably involved extensive modification of coding regions . Such changes may have the secondary effect of introducing normally unfavoured codons or codon pairs in viral genomes , reducing translation rates that may cause further virus attenuation ( Martínez et al . , 2016 ) . It is possible , however , to modify sequences in such a way that limited changes in CpG/UpA dinucleotide frequencies can be introduced while keeping codon pair bias ( as measured by the summary metric , codon pair bias ) constant , and vice versa . When done in the E7 picornavirus model , we found it was dinucleotide frequency changes that were the primary factor behind attenuation ( Tulloch et al . , 2014 ) , although this conclusion has been disputed ( Futcher et al . , 2015 ) . To resolve this issue , in the current study we have developed a new replicon for E7 which allows additional sequence of varied dinucleotide composition to be placed in a non-translated context; any changes in the replication ability of mutants with different dinucleotide compositions therefore cannot be attributed to effects on translational efficiency that have been advocated by other groups . The replication cycle of E7 , a typical enterovirus , is relatively well understood and allows functional dissection of the various replication steps in which attenuation of high CpG and UpA mutants of the virus may occur . Upon entry into the host cell the viral genomic RNA is quickly released into the cytoplasm . Through a cap-independent mechanism viral RNA is translated into a single polyprotein that is further processed . Changes in dinucleotides frequencies may influence the stability of RNA post entry or the efficiency of translation initiation or processivity . RNA replication occurs in endoplasmic reticulum ( ER ) associated vesicles , where from a negative-sense RNA intermediate progeny positive-sense RNA genomes are transcribed . The dsRNA replication intermediate is a known activator of cytosolic pattern recognition receptors , such as retinoic acid-inducible gene I ( RIG-I ) and melanoma differentiation-associated gene 5 ( MDA5 ) ( Baum et al . , 2010; Peisley et al . , 2011 ) . The consequent induction and secretion of interferon induces expression of a large number of cellular genes whose expression leads to the induction of an antiviral state within the cell ( Randall and Goodbourn , 2008 ) . Viral RNAs with modified dinucleotide frequencies may be differentially susceptible to the effects of the interferon response . Finally , the generation and packaging of viral RNAs may be influenced by post-transcriptional modifications of viral RNA , such as deamination by the interferon-inducible ADAR and APOBEC proteins and lead to progeny virus that is intrinsically replication defective . Using a wide range of pathway knockouts and inhibitors and direct observation of the fate of viral RNA post-entry , we were able to determine where in the viral replication cycle attenuation by unfavoured dinucleotides occurred and what components of the cellular antiviral response were responsible for virus attenuation . We previously demonstrated the marked inhibition in replication of E7 mutants in which CpG and UpA frequencies were artificially increased in one or two regions of the genome ( R1 , R2 , Figure 1A ) . These regions were selected for the absence of secondary RNA structures or specific RNA sequences important for enterovirus replication . This was supported by a high synonymous site variability and low mean folding energies ( Atkinson et al . , 2014 ) . In these experiments a permutated mutant of E7 , with the native sequences of R1 or R2 scrambled but retaining coding and native dinucleotide frequencies , showed WT levels of replication , indicative that these genome region can be safely modified without consequences for viral replication ( Atkinson et al . , 2014; Tulloch et al . , 2014 ) . Attenuation of viruses containing UpA-high and CpG-high sequences was evident in RD and A549 cell lines , but whether the restriction in replication extended to cell lines of different tissue origins was not determined , nor whether the restriction was related to host-cell susceptibility to E7 infection . To investigate this , we infected a range of different cells at low multiplicity of infection ( MOI ) with wild type ( WT ) E7 and mutants with modified R2 sequences with elevated CpG ( C ) or UpA ( U ) frequencies ( Figure 1B and Table 1 ) . R2 mutants were used in preference to the R1/R2 double mutants to ensure that replication kinetics for increasingly attenuated mutants could still be measured to an acceptable accuracy in some of the less permissive cell lines . Twenty-four hours post infection with the E7 R2 mutant viruses , infectious progeny virus was measured with an end point dilution assay ( EPDA ) . Across the different cell types , the viral titres of E7 with R2_U or R2_C were consistently lower compared to WT E7 ( Figure 1B , one-way ANOVA , p<0 . 01 , ) . The relative replication rates ( RRRs; ratio of TCID50s of mutant/WT virus ) of E7 mutants differed between cell types . In RD cells ( human muscle ) and A549 cells ( human lung epithelium ) R2_U E7 displayed an RRR of 0 . 1 and 0 . 4 respectively . The RRR of R2_C was further suppressed to 0 . 01 in these cells . Cells that originated from the kidney displayed a more moderate E7 R2_C RRR of approximately 0 . 09 ( Figure 1B ) . The viral titres of WT E7 at 24 hr post infection varied strongly between cell types and may potentially be the cause of the differing RRR of UpA and CpG-high viruses . However , there was no relationship between RRR ( Figure 1C , x-axis ) and virus titre of WT virus at 24 hr ( Figure 1C , y-axis , representing cell susceptibility; Pearson correlation coefficient CpG −0 . 57 and UpA −0 . 63 ) . Relative replication rates were similarly determined in cells with impaired innate responses to RNA virus infections . These included A549 cells expressing the bovine viral diarrhoea virus ( BVDV , genus Pestivirus ) N-terminal protease fragment ( NPro ) , which blocks the activity of interferon regulatory factor 3 ( IRF-3 ) ( Hilton et al . , 2006 ) and IRF7 ( Fiebach et al . , 2011 ) or the hepatitis C virus ( HCV , genus Hepacivirus ) protein NS3/4A that inhibits cytokine gene expression by cleavage of IPS-1/MAVS/VISA/Cardif ( Kaukinen et al . , 2006 ) . Both cell lines showed similar or greater restriction in replication for both CpG- and UpA-high mutants compared to the parental cell line . The monkey kidney fibroblast cell line , Vero , which has intact IFN signalling pathways but cannot produce type 1 IFNs ( Desmyter et al . , 1968 ) , restricted mutated E7 viruses comparable to that of other cultured kidney cells tested in this study with a more moderate attenuation between of the CpG high virus ( Figure 1BC ) . Together this indicates that reduced replication of E7 with increased UpA and CpG dinucleotide frequencies occurs in all the tested cell types and restriction cannot be lifted by inhibiting some of the most potent antiviral signalling cascades . There was additionally some intrinsic variability between kidney and other cell lines in the extent to which replication inhibition occurred . It has been hypothesised that the observed attenuation of CpG- and UpA-high mutants arises through the effects of dinucleotide choice on translation efficiency , either through selection of disfavoured codon usage or codon pairs that are translated less efficiently than native sequences ( Mueller et al . , 2010; Coleman et al . , 2008; Burns et al . , 2006 ) . To investigate this possibility directly , we compared the attenuating effects of CpG and UpA dinucleotides added to either the coding ( in R2 ) or non-coding region of the E7 replicon . To achieve the latter , the E7 replicon was modified by insertion of region 1 ( R1 , Figure 1A and Table 1 ) compositional variants as additional non-coding regions ( ncR1 ) after the stop codon ( nt 7325 ) , but before replication structures in the viral 3’-untranslated region ( UTR ) ( Figure 2A ) . Cells were transfected with in vitro transcribed RNA of the E7 replicon with ncR1 variants and assayed for firefly luciferase expression at indicated times post transfection . Being of insect origin , the wild type firefly luciferase gene ( luc_WT ) contains a relatively high ratio of CpG dinucleotides ( CpG 1 . 210 and UpA 0 . 695 observed/expected ) . Synonymous removal of all possible CpG and UpA dinucleotides from firefly luciferase ( luc_cu , ratio CpG 0 . 013 and UpA 0 . 154 O/E ) strongly increased the replication rate of the E7luc replicon ( Figure 2—figure supplement 1 ) . To reliably measure potential reductions in E7 replicon RNA replication the E7luc_cu replicon was used in the remainder of this study . Addition of the 1242 nucleotide long ncR1 of WT composition to the E7 replicon RNA had no effect on replication compared to that of the E7 replicon without ncR1 ( Figure 2—figure supplement 1 ) . Similar replication was observed in the replicon with the R1 permuted control ( P ) that retained identical CpG and UpA frequencies to the WT control . This indicates that the replication structures in the 3’-UTR are not affected by the addition of 1242 nucleotides to the 5’-end of the 3’-UTR . However , RNA replication rates of E7 replicons with a high UpA ncR1 tail ( ncR1_U ) in RD and A549 cells showed mildly reduced relative replication rates ( to 0 . 5 ) , whereas the high CpG sequence , ncR1_C , reduced replication rates by approximately 10-fold at 6 and 9 hr post RNA transfection ( hpt ) ( Figure 2B ) . Similar degrees of attenuation were observed in replicons where CpG and UpA frequencies were increased through mutation of the coding region in R2 ( Figure 2B ) . In contrast , the RRRs of both coding and non-coding CpG- and UpA-mutants approached that of WT in kidney-derived BHK cells ( Figure 2—figure supplement 2 ) . No effects on replication were observed in replicons with ncR1 ( non-coding ) or R2 ( coding ) sequences possessing lower ( cu ) CpG/UpA frequencies than WT ( Figure 2 ) . We previously reported that kinase inhibitor C16 was able to largely reverse the attenuation of CpG- and UpA-high mutants of E7 virus , independently of PKR ( Atkinson et al . , 2014 ) . To further investigate whether the replication of infectious virus and replicons with ncR1 extensions were similarly influenced by C16 . RD cells were either transfected with replicon RNA of E7 ncR1 variants or infected with infectious E7 virus with mutated R2 , both in the presence or absence of C16 . The relative luciferase expression of E7 replicons with increased UpA or CpG dinucleotides in their ncR1 were increased by 1 . 8 and 6 fold respectively in the presence of C16 while the replication of WT and CDLR was unaffected ( Figure 2—figure supplement 3 , p-values 0 . 069 and 0 . 01 , T-test ) . The observed reversal of attenuation of the E7 ncR1 variants closely resembled that of viruses with modifications in the coding region ( Figure 2—figure supplement 3 ) and provides evidence that the mechanisms of attenuation in the two systems were similar . We previously demonstrated that relative replication rates of infectious viruses with increased CpG- and UpA dinucleotide frequencies in R2 varied between cell lines ( Figure 1B ) . Transfecting replicons with CpG and UpA dinucleotides incorporated into the ncR1 tail into the same cell lines provided an opportunity to compare the extent of attenuation ( Figure 3A and B ) . The replicon with ncR1_C showed an RRR of approximately 0 . 1 in A549 and RD cells , a greater degree of attenuation than observed in the kidney-derived 293 , 293T and Vero cells , while attenuation in the pathway knockout cells , A549-V and A549 NS3/4A was greater ( Figure 3A ) . Although the overall degree of attenuation in the replicon format was lower than in the low MOI infection experiments performed for infectious virus ( Figure 1B ) , the ranking of cell lines for attenuation of both CpG and UpA-elevated mutants was similar between the two challenge systems ( Figure 3B ) . Replicons of E7 can also replicate in murine cells ( Zhang and Racaniello , 1997 ) and this provided an opportunity to investigate a wider range of cell lines on the attenuated phenotype of CpG and UpA-high mutants of E7 ( Figure 3C ) . These included mouse embryonic fibroblasts ( MEF and 3T3 ) , a B lymphoblast cell line ( A9 ) and two neuronal cell types , neuroblasts and microglia cells ( dN2a and BV2 , respectively ) . At six hpt , samples were lysed and assayed for luciferase expression . In all cell types replication rates of replicons with increased UpA or CpG dinucleotide frequencies in the non-coding region were consistently lower than WT ( RRRs ranging from 0 . 01 to 0 . 3 for ncR1_C and 0 . 2–0 . 7 for ncR1_U ) . In contrast , replicon mutants with an ncR1 of P or cu composition invariably showed equivalent replication to WT ( Figure 3 ) . To further disentangle the effects of CpG and UpA dinucleotide frequencies on virus attenuation and translation efficiency , non-replicating E7 replicons with various dinucleotide frequencies in their non-coding R1 were transfected into cells and the subsequent translation of luciferase was measured . Replication was inhibited by treatment of cells with a replication inhibitor , guanidine hydrochloride ( GuHCl ) . It is well established that GuHCl effectively inhibits picornavirus RNA replication through inhibiting initiation of negative strand RNA synthesis and RNA strand elongation ( Barton and Flanegan , 1997; Pfister and Wimmer , 1999 ) . In addition , replication-defective mutants of the E7 luciferase replicon with ncR1_WT and ncR1_C were made by site-directed mutagenesis of the viral polymerase . In these non-replicating formats , any differences in luciferase expression arise from effects on translation efficiency and/or stability of the replicon RNA and are not compounded by effects of CpG or UpA modification on replicon replication . RD cells were transfected with E7 replicon RNA and either treated with GuHCl or left untreated ( Figure 4A and B ) . In a parallel experiment cells were transfected with the mutated replication-defective replicons ( Figure 4C ) . GuHCl treatment or mutation of the viral polymerase similarly reduced luciferase readings throughout the time-course experiment , with no sign of replicon RNA replication . Importantly , E7 replicons with CpG or UpA high ncR1 showed comparable luciferase expression as replicons with WT ncR1 sequences when replication was inhibited by either GuHCl or mutation of the viral polymerase ( Figure 4B and C ) . Untreated , replication competent replicons displayed the typical attenuation of luciferase expression in CpG and UpA mutants ( Figure 4A ) . There was a modest , approximate 5-fold reduction in luciferase expression over the 24 hr course of the experiment observed in non-replicating replicons that may reflect partial degradation of the transfected RNA . Reductions in luciferase expression were comparable between WT and CpG/UpA modified replicons ( Figure 4A–C ) . In an additional experiment , translation rates of replicating and non-replicating E7 variants were compared between different cell types ( Figure 4D ) . At six hours post transfection , replicating E7 replicon RNA with a CpG-high ncR1 displayed an approximate 10-fold reduction in luciferase expression when compared to WT ( Figure 4D , gray bars ) . In contrast , when replication was inhibited by GuHCl , translation from the CpG-high replicon RNA was comparable to that of the WT in all tested cell types ( Figure 4D , black bars ) . Replicons with UpA-modified ncR1 also expressed similar levels of luciferase compared to WT in a non-replicating context with ratios ranging from 0 . 5 to 2 . 9 ( Figure 4—figure supplement 1 ) . Combined , these results demonstrate that the observed attenuation of UpA or CpG high replicon mutants is not dependent on differences in translation rates or instability of the coding region RNA sequences in the cell . To investigate whether increasing the absolute number of unfavourable dinucleotides strengthens the observed restricted phenotype of UpA and CpG-high mutants of E7 , different lengths of nucleotide sequence were inserted in the 3’-non-coding region of the E7 luciferase replicon system , either as single ( 800 nt ) or double ( 1600 nt ) blocks . The initial sequence was normalised to WT E7 nucleotide composition , but with a completely random and non-coding sequence order ( Norm ) . Subsequently , this sequence was altered to contain either increased UpA or CpG dinucleotides or to contain no unfavourable dinucleotides at all ( U , C and cu respectively ) . These blocks were cloned in the non-coding region and referred to as single or double non-coding synthetic sequences ( ncS , Figure 5A and Table 1 ) . Transfection of these replicon RNAs with various compositions in their ncS regions into RD cells resulted in significantly reduced E7 RNA replication for the single CpG-high ncS ( ncS_C ) but not the single UpA-high ncS_U ( Figure 5B ) . However replicons with the double ncS of both U and C composition showed significantly reduced replication compared to their respective single ncS replicons ( Figure 5B ) . Indeed , luciferase expression of the CpG-high double region mutant was little different from that of the non-replicative guanidine treated control ( Figure 5B; dotted red line ) , indicating that this longer insertion almost entirely abrogated replication of this construct . Together the observations that cumulative CpG , and to a lesser extent UpA dinucleotides in non-coding regions of E7 replicons have restricted RRRs , but translate equally efficiently in a non-replicating setting indicates that codon bias , codon pair bias or other direct effects on translational efficiency are not at the heart of the observed restriction . Rather it implies a cell-mediated restriction of viral RNA replication and/or an alternative fate of the compositionally altered ( viral ) RNA . Making use of both infectious virus and replicon systems , we investigated the cellular basis of the observed restriction during the E7 infection cycle . To investigate when in the replication cycle , the restriction of CpG- and UpA-high mutants of E7 occurred , RD cells were infected with equal RNA copy numbers ( 1000 RNA copies/cell ) or equal infectivity ( MOI 0 . 01 ) of E7 with various R2 compositions and replication monitored at early time points post-infection . Infection with equal RNA resulted in reduced infectivity of E7 R2_C virus as early as two hours post infection and a prolonged delay in the production of progeny virus and RNA levels ( Figure 6 ) . In this experiment there was no difference in initial receptor binding/viral entry between the E7 variant R2_C and other mutants as the genomic RNA copies detected after washing off the inoculum were identical to the mean of all samples at 1 hr post infection ( mean 1 . 00 , R2_C 1 . 00 , R2_U 1 . 03 SD 0 . 04 ) . Interestingly , after the initial delay in RNA replication , E7 with increased CpG dinucleotide frequencies replicated at a rate similar to that of other E7 variants ( Figure 6B ) . In comparison , infections with equal infectivity resulted in the suppression of progeny E7 viruses with increased CpG and UpA dinucleotide frequencies after the first round of replication ( Figure 6—figure supplement 1A ) . The restriction of R2_C viral RNA replication became more apparent during subsequent replication cycles ( Figure 6—figure supplement 1B ) . To further investigate the fate of viral RNA post-entry , the intracellular localization of E7 genomic RNA was visualized by RNA fluorescent in situ hybridisation ( FISH ) . RD cells were infected with equal RNA copies of either R2_WT , U , or C mutant viruses . Small localized foci with intracellular genomic E7 RNA were observed at two hpi ( Figure 7A , white arrowheads ) in all infected cells at frequencies that were comparable between WT and mutant viruses . At four hours post infection , WT and R2_U mutants developed clear pockets of replication in a large percentage of cells , whereas detection of E7 RNA with increased CpG dinucleotides became less frequent . Those cells that did display E7 with R2_C emitted a more variable fluorescent intensity ranging from small RNA foci ( Figure 7A ) to areas of RNA replication similar to the other viruses . At six hours , RNA replication of E7 R2_C had progressed to display similar fluorescent intensities compared to the other E7 variants , but only in 3 . 5% of cells , compared to 49% and 36% of cells infected with R2_WT or R2_U viruses , respectively ( Figure 7BC ) . The striking observation that dinucleotide frequency changes influenced the frequency of infected cells , but not necessarily their subsequent replication ability was verified by flow cytometry of a further set of E7 replicon constructs in which the luciferase gene was replaced by enhanced green fluorescent protein ( EGFP ) . A series of mutants were created with altered nucleotide compositions in their coding R2 region and/or ncR1 tails . In vitro transcribed RNA was subsequently transfected into either RD or BHK cells and scored for fluorescent intensity . Similar to the above RNA FISH experiments , FACS analysis showed that the percentage of EGFP expressing cells at 6 hpt differed dramatically between WT and CpG-high mutant replicons while the mean fluorescent intensities of the cells that were fluorescent was comparable ( Figure 8AB , Figure 8—figure supplement 1 and Table 2 ) . Attenuation was enhanced for CpG- and UpA-high mutants when both coding R2 and ncR1 were modified ( Figure 8 and Table 2 ) . Similar to E7 replicons that expressed luciferase , transfection of replicons with either R2 or ncR1 of UpA and CpG high composition into BHK cells resulted in EGFP expression more similar to WT replicons , though replicons with both regions mutated did show a marked reduction in EGFP expression ( Figure 8CD , Figure 8—figure supplement 2 and Table 2 ) . Together , this indicates that while entry of E7 WT , UpA- and CpG-high mutants was unaffected , replication of these mutants was profoundly restricted at early time points during the replication cycle . However , in a small number of cells , E7 CpG-high and UpA-high mutants were able to overcome restriction , suggesting that increasing CpG and UpA dinucleotides renders a large proportion of E7 RNA replication defective or E7 replication is able to counteract or saturate supposed intracellular restriction factors . The inability of the majority of CpG-high E7 virions and a measurable proportion of those with high UpA sequences to initiate replication post-entry may originate from mutational defects in their genome sequences which is possibly mediated through ADAR-2 or APOBEC activity in cells that produced them . To investigate this possibility , frequencies of defective viral genome sequences were determined by extensive sequencing of virus stocks of E7 with R2s of WT , P , cu , U and C composition in amplicons derived from the variable R2 region and in a non-mutated region ( positions 2312–3083 in the E7 genome ) . Multiple clones from each PCR product were sequenced and differences from the cloned mutated region recorded ( Table 3 ) . For each , RNA from three biological replicates of WT and all four mutant virus stocks showed comparable misincorporation frequencies in the R2 variant viruses . The average mutation rate of approximately 10−4mutations/nucleotide closely resembled that of previously published mutation frequency for poliovirus ( 9−5[Sanjuán et al . , 2010] ) . These results indicate that these viral RNAs are not intrinsically replication defective . The observation that fitness defects in both UpA and CpG-high mutant viruses and replicons can be reversed by kinase inhibitor C16 ( Figure 2—figure supplement 3 ) provides further evidence that the nature of the replication defect is not caused by increased mutation rates . Together , this suggests that the inhibition in replication of incoming RNA with increased UpA or CpG dinucleotide frequencies is the result of host-cell restriction factors . An alternative explanation for the failure of CpG- and UpA-high mutants to initiate replication as effectively as WT virus is that their compositional differences may lead to differential sequestration of incoming viral RNA sequences into cytoplasmic stress granules ( SGs ) and/or induce a greater stress response that prevents initial translation of the genomic RNA . SGs are cytoplasmic foci containing RNA binding proteins , RNAs and translation initiation factors . SGs are rapidly formed in response to translation attenuation and environmental stress including that induced by viral infections ( Buchan and Parker , 2009; Kedersha and Anderson , 2002 ) . This intrinsic response pathway contributes to the cellular antiviral response ( Reineke and Lloyd , 2015 ) and multiple diverse viruses have shown to inhibit SG function ( Fros et al . , 2012; Emara and Brinton , 2007; Borghese and Michiels , 2011 ) . However , picornaviruses have been shown to inhibit the formation of SGs by expressing a viral protease that cleaves Ras-GAP SH3 domain-binding protein ( G3BP ) , which plays a central role in the formation of SGs ( Fung et al . , 2013; White et al . , 2007 ) . To investigate whether E7 mutants with increased CpG or UpA dinucleotide frequencies are differentially sequestered into SGs and therefore prevented from replication initiation ( Figures 6–8 ) , RD cells were infected with 1000 RNA copies/cell and after 4–6 hours cells were fixed and stained for viral RNA and G3BP1 . Regardless of the nucleotide composition of the incoming viruses ( WT , C , U ) , no typical G3BP positive cytoplasmic granules corresponding to those found in uninfected sodium arsenite treated samples were observed during infection with E7 , nor did G3BP localize with E7 genomic RNA ( Figure 9 ) . Further evidence that such cytoplasmic response pathways are not involved directly in the control of E7 replication was provided by measurement of phosphorylation on serine 51 of the eukaryotic translation initiation factor eIF2α . This is a central mediator of cellular responses to environmental stress including that induced by virus infection , that acts to inhibit general translation ( Holcik and Sonenberg , 2005 ) and phosphorylation is generally associated with SG formation . Infection with the E7 UpA and CpG high variants at either equal infectivity or equal RNA did not result in a considerable increase ineIF2α phosphorylation compared to infection with WT E7 or the permutated control ( Figure 9—figure supplement 1A ) . Next we investigated whether RNA silencing through the RNA-induced silencing complex ( RISC ) may differentially affect replication of the E7 mutants . Loading of small molecules to the RISC complex is effectively inhibited by acriflavine ( ACF ) ( Madsen et al . , 2014 ) . However , treatment of cells with ACF did not change the RRR of E7luc replicons with ncR1 variants from their respective DMSO treated controls ( Figure 9—figure supplement 1B ) . Finally , we investigated whether mutants with higher frequencies of CpG or UpA dinucleotides were more potent inducers of apoptosis . Caspase 3/7 activity of cells infected with WT , P , cu , C , or U R2 mutants of E7 at equal MOI was determined at 24 hr post infection . Levels correlated with viral replication rates rather than with CpG or UpA dinucleotide frequencies , providing strong evidence that the reduced replication of viruses with increased genomic UpA or CpG dinucleotide frequencies was not caused by an increased induction of programmed cell death ( Figure 9—figure supplement 1C ) . Together , this suggests that viruses with increased CpG or UpA dinucleotide frequencies do not differentially induce stress , interferon-coupled or apoptosis-associated antiviral pathways and that none of these can be plausibly attributed as mediations of their restricted replication phenotypes . While conventional antiviral pathways or effects of siRNA induction could not be implicated in the observed restriction of replication of CpG- or UpA-high mutants of E7 , it is possible that their attenuation was mediated through alternative pathways that induced an antiviral state within the infected cells . To investigate this , we determined whether the replication of WT E7 could be suppressed in trans through the effects of co-transfection or expression of RNA sequences in the cell with elevated CpG or UpA dinucleotide frequencies . In the first experiment , high CpG/UpA and permutated control R2 region RNA were transiently expressed from a plasmid vector transfected into HEK293 cells off a CMV promoter . After 24 hr , during peak expression of the transcripts , cells were infected with WT E7 at a MOI of 0 . 01 for 48 hr and effects of the RNA transcripts monitored by measurement of viral titres in an EPDA ( Figure 10A ) . The expression levels of the R2 mRNA transcripts were comparable between mutant R2 sequences ( data not shown ) . However , titres of the superinfecting WT virus were similar between all samples and therefore entirely unaffected by the nucleotide composition of the R2 RNA co-expressed in these cells ( Figure 10A ) . In an alternative experimental format , the above described non-coding synthetic sequences ( ncS , Table 1 ) , 800 bases in length , with variable dinucleotide frequencies ( Norm , cu , U , and C ) were transcribed in vitro and RNA co-transfected into RD cells with the E7 luciferase replicon extended by an ncR1 of WT composition ( Figure 10B ) . At six hpt , cells were lysed and luciferase expression was measured . Luciferase expression of the E7 replicon was not affected by co-transfected ncS RNAs , regardless of their dinucleotide frequencies ( Figure 10B ) . In these experiments , the high CpG/UpA RNA sequences were expressed in cells in a non-replicating context . To determine whether effects of high CpG and UpA RNAs required replication in order to exert their inhibitory effects , we co-transfected cells with reporter and interfering replicons . The reporter replicon contained a luciferase reporter gene and an E7 WT 3’ ncR1 sequence . Interfering replicons contained a GFP reporter gene of varying composition ( WT , P , cu , U , or C ) . Equimolar amounts of reporter and interfering replicons were co-transfected into RD cells . EGFP expression was detectable in all samples , but with far fewer cells expressing EGFP from the replicon with ncR1_C sequences ( Figure 10—figure supplement 1 ) . In contrast to the effects dinucleotide frequencies have on viral RNA replication in cis , luciferase expression of the co-transfected reporter replicon was unaffected ( Figure 10C ) . Together , these findings indicate that neither replicating viral RNA nor non-replicating RNA sequences in the cytoplasm with elevated CpG or UpA dinucleotide frequencies had any detectable trans effect on E7 replication . The strict restriction on replication in cis demonstrates that expression of RNAs with elevated CpG or UpA frequencies mediates a quite different form of replication inhibition than the antiviral state induced by stress pathways or IFN-β induction through activation of conventional PRRs . Most genomic sequences of ssRNA viruses show marked suppression of UpA and CpG dinucleotides ( Karlin et al . , 1994; Rima and McFerran , 1997; Simmonds et al . , 2013 ) . However , the suppression of CpG dinucleotides is composition dependent . Higher G + C content generally allows for a higher frequency of CpG dinucleotides in naturally occurring sequences ( Fryxell and Moon , 2005; Simmonds et al . , 2013 ) , including isolates of the enterovirus genus ( Figure 11A ) . To investigate whether this striking correlation is the result of functional constraints that also shapes the direct context surrounding a CpG dinucleotide , synthetic sequences were designed to have an identical G + C content and equal amounts of CpG and UpA dinucleotides as WT R1 ( Table 1 ) , but with variable positioning of A and U bases that may create more potent motifs restricting replication than CpG alone . Specifically , sequences were generated in which A and U bases were positioned in either AACGAA or UUCGUU contexts . These novel sequences were cloned into the non-coding region of the E7 luciferase replicon system creating E7 ncR1_AACGAA and ncR1_UUCGUU . Despite these mutants possessing the WT number of CpG dinucleotides , their replication was profoundly impaired ( Figure 11B ) ; the ncR1_AACGAA showed an RRR comparable to that of the CpG-high sequence ( containing 181 CpG dinucleotides ) . Remarkably , the replication of the ncR1_UUCGUU was further impaired with an RRR 30-fold lower than the WT control of identical CpG content . The context of the bases surrounding the CpG dinucleotide has a potent effect on replication attenuation . A further range of sequences require to be tested in this experimental paradigm to better characterise the minimal motif associated with CpG recognition . The RRRs of both viruses and replicons with increased CpG and UpA dinucleotides were lower than WT E7 in all cell types tested . The similarity in pattern of RRR observed in different cell lines between E7 virus and replicons , and their shared responsiveness to C16 corroborates the use of this replicon system with non-coding sequence variants . For both replication systems , cells originating from the kidney and especially BHK cells displayed a smaller restrictive phenotype to CpG-high E7 ( Figures 1–3 and 10 ) , although increasing CpG or UpA dinucleotides in multiple regions further reduced the RRR in BHK cells , indicating that these cells do share the ability to inhibit E7 replication ( Figure 8 ) . These findings are consistent with the reduced attenuation in BHK cells of dengue virus mutants with an increased frequency of unfavoured codon pairs ( Shen et al . , 2015 ) , a process that increases frequencies of CpG and UpA dinucleotides in the sequence ( Simmonds et al . , 2015; Tulloch et al . , 2014 ) . More broadly , the numerous studies that used codon pair bias to attenuate viruses ( Coleman et al . , 2008; Mueller et al . , 2010; Ni et al . , 2014; Martrus et al . , 2013; Le Nouën et al . , 2014 ) consistently report the same type of restricted replication phenotypes that we have observed in high CpG and UpA mutants of E7 . We and others have proposed on bioinformatic and experimental grounds that attenuation associated with unfavoured codon pairs originates from unintentional increase in CpG and UpA dinucleotides ( Tulloch et al . , 2014; Kunec and Osterrieder , 2016 ) . The data reported in the current study reinforces this conclusion , at least for the echovirus seven model we used , by demonstrating that replication rates were similarly affected by addition of CpG and UpA dinucleotides in the 3’ non-coding region as they were in the coding part of the genome ( Figures 2 , 5 and 8 ) . CpG- and UpA-induced attenuation therefore must be mainly mediated in a manner that is independent of codon usage or codon pair bias . Furthermore , transfection of non-replicating RNAs showed no influence of CpG or UpA addition on translation of luciferase or RNA stability ( Figure 4 ) . These observations were however restricted to the echovirus model , and while they may also underlie the observed attenuation of other viruses with codon or codon pair de-optimised coding regions , translation efficiency is very clearly a potential additional factor that may influence virus replication rates , and does not exclude the existence of additional mechanisms that may attenuate viruses based on codon or codon pair choice , particularly for viruses that have different replication mechanisms to E7 . For example , the expression of viral proteins from conventionally processed mRNAs by most DNA viruses , nuclear replicating RNA viruses such as influenza A virus and retroviruses clearly places their replication kinetics at the mercy of how effectively these are translated and how well this is coordinated for virus assembly and release . These factors are less relevant for positive sense RNA viruses , such as E7 that replicate in the cytoplasm . The findings do suggest , however , that the apparent necessity for under-representation of UpA and CpG dinucleotides in viral RNA is one element that has contributed to biases in nucleotide composition and as a result the choice of codons and codon pairs in native viral sequences . The importance of the current study is that we can at least for the E7 model , entirely disentangle the effects of dinucleotide frequency modification from translation efficiency and produce experimental findings that complement conclusions reached previously using other viruses and other experimental approaches ( Burns et al . , 2006; Burns et al . , 2009; Tulloch et al . , 2014; Kunec and Osterrieder , 2016 ) . Since the attenuation of replication of E7 with elevated CpG and UpA frequencies cannot be the result of changes in translation efficiency , we investigated the cellular basis of the observed restriction by following the outcome of infection of cells with E7 . Shortly after entry , mutants with increased CpG dinucleotide frequencies showed a substantial delay and reduced formation of replication complexes . However , this difference was not the result of reduced infectivity of CpG-high virions , as equal RNA copies were detected by PCR and RNA-FISH revealed similar frequencies of RNA genomes post-entry ( Figures 6 and 7 ) . The marked phenotypic effect arose because the initial replication entities of CpG-high viruses failed to progress to form the larger replication complexes observed in WT and also most UpA-high infected cells ( Figure 7 ) . Once formed , replication complexes from CpG-high viruses showed comparable levels of viral RNA by RNA-FISH . Infection outcomes with a parallel set of replicons in which the luc reporter gene was replaced by EGFP conformed these observations; FACS analysis provided quantitative evidence for a reduced frequency but equivalent fluorescent intensity of productively infected cells by the CpG-high mutant ( Figure 8 ) . This reduction in replication competent particles could be the result of post-transcriptional RNA-editing of the viral genome that rendered them replication defective . It is possible , for example , that ADAR-2 or APOBEC may be differentially upregulated in cells infected with compositionally altered mutants and create progeny viruses with high frequencies of hypermutated , replicative-defective genomes ( Powell et al . , 1987; Yang et al . , 1995; Rueter et al . , 1995; Tomaselli et al . , 2015 ) . In RNA these proteins create specific A-G and C-U transitions , respectively ( Smith et al . , 1997 ) . APOBEC editing substantially limits the replication ability of HIV-1 and other retroviruses; so powerfully indeed that many retroviruses have developed specific antagonism pathways , such as Vif in HIV-1 that limits APOBEC’s effect ( Hultquist et al . , 2011 ) . However , frequencies of mutations , both non-synonymous ( and potentially inactivating ) and synonymous , were comparable in E7 viral stocks of all compositions , and indeed were similar to those reported previously in WT E7 and poliovirus ( Sanjuán , 2010; Atkinson et al . , 2014 ) . There was also no evidence of a specific RNA editing signature among the identified mutations ( Table 3 ) . The observed failure of CpG-high viruses to initiate replication is therefore not the result of genomic sequences being defective . Viral replication may be directly inhibited in some other way by increased UpA and CpG dinucleotides . UpA and CpG dinucleotides are self-complementary and may increase the likelihood of intramolecular base pairing , which could impact the efficiency of the viral RNA dependent RNA polymerase ( Lai , 2005 ) . However , the reversal of the restricted replication phenotypes of CpG- and UpA-high mutants by the kinase inhibitor C16 ( Figure 2—figure supplement 3 ) and the reduction in their attenuation in BHK and other kidney cells lines ( Figures 1–3 and 8 ) argues strongly against the existence of an intrinsic replication defect in the mutant viruses . This conclusion is reinforced by the observation that CpG- or UpA- elevation in a non-functional genome region ( the 3’ ncR1 or ncS ) led to a similar attenuation as observed in coding region mutants ( Figures 2 , 3 and 5 ) . As there was no evidence that CpG- and UpA-high viruses were translationally or otherwise compromised in their replication abilities , we therefore sought evidence that their attenuation provoked a qualitatively or quantitatively different innate cellular response . Perhaps it was their greater visibility to cell defences or susceptibility to antiviral pathways that limited their replication . It is known that the structure and configuration of viral genomic RNA may serve as PAMPs for interferon pathway-coupled PRRs ( Yoneyama et al . , 2016; Oshiumi et al . , 2016 ) . UpA and UpU dinucleotides in viral RNAs can be cleaved by RNase L ( Cooper et al . , 2015 , 2014 ) and produce 5’ phosphorylated RNA termini that activate RIG-I mediated IFN-β expression ( Malathi et al . , 2007; Malathi et al . , 2010 ) . This may potentially account for the suppression of UpA dinucleotide frequencies in viral RNA sequences . However , this does not explain why viruses have not evolved to under-represent UpU dinucleotides as well as UpA - for E7 the UpU O/E ratio is 1 . 04 . We have previously shown that inhibition of IRF3 , a pathway intermediate that couples PRR recognition to interferon production , had no effect on the attenuation of CpG- or UpA-high mutants of E7 , nor was their replication any more affected by the addition of exogenous IFN than WT virus ( Atkinson et al . , 2014 ) . Several findings in the current study reinforce the conclusion that the IFN pathway is not involved , directly or indirectly in the attenuation of CpG-/UpA-high viruses . These include the reduced replication rates of UpA and CpG high viruses in Vero cells and the absence of any effect on MAVS inhibition ( mediated by expression of the HCV protease NS3/4A; Suppl . Figure 1 ) . Similarly , viral RNA was not sequestered in stress granules and eIF2α phosphorylation was not upregulated in cells infected with CpG- or UpA-high mutants of E7 ( Figure 9—figure supplement 1 ) . None of the currently described PRRs are known to recognise CpGs in RNA sequences , although CpG dinucleotides in oligodeoxynucleotides or other DNA sequences can activate antiviral gene expression through TLR9 ( Hemmi et al . , 2000; Dorn and Kippenberger , 2008 ) . In addition , removing CpGs from a transgene drastically improved its expression in vivo while at the same time reducing cytokine expression and increasing the persistence of the transgene ( Yew et al . , 2002; Hodges et al . , 2004; Hyde et al . , 2008 ) . Although also in this system , TLR9 was shown to be indispensable as persistent transgene expression in TLR9 knock out mice still required complete removal of CpG dinucleotides ( Bazzani et al . , 2016 ) . Together this indicates the existence of a TLR9 independent mechanism that detects CpG dinucleotides in transcripts . Alternatively , ssRNA sequences may be recognised in endosomes by TLR7 and TLR8 . GU-rich and AU-rich sequences have been identified as preferential TLR7/8 activator , enhancing cytokine expression of IFN-α and TNFα ( Lund et al . , 2004; Heil et al . , 2004; Diebold et al . , 2004 ) . Interestingly , an additional CpG dinucleotide in an AU-rich context enhances TLR7 mediated cytokine expression ( Jimenez-Baranda et al . , 2011 ) . However , the cell types used in this study are not known to express detectable levels of TLR7/8 , nor do increased CpG dinucleotide frequencies in either E7 or influenza viruses upregulate TLR7/8 specific cytokines ( Gaunt et al . , 2016; Atkinson et al . , 2014 ) . It is therefore unlikely that TLR7/8 signalling is at the heart of the observed restriction . The TLR protein family contains a number of additional molecules , however for some their function in pathogen recognition and especially subsequent induction of protein expression remain unknown ( reviewed in [Satoh and Akira , 2016] ) Furthermore , the absence of any differences in ISG expression between WT and mutant viruses ( Gaunt et al . , 2016; Atkinson et al . , 2014 ) together with the complete lack of any inducible antiviral response to UpA and CpG containing RNA ( Figure 10 ) makes a controlling role for interferon-coupled responses to such mutants highly unlikely . Picornaviruses furthermore are particularly well armoured against IFN-mediated cellular responses that reduces the likelihood of any role in attenuating CpG- or UpA-high mutants . The enterovirus-encoded 2C protease cleaves a variety of host factors ( i . e . MAVS , MDA5 , RIG-I ) ( Feng et al . , 2014; Wang et al . , 2013; Barral et al . , 2007 , 2009 ) and renders this PRR-linked pathway non-functional during virus replication . E7 replication was relatively insensitive to effects of pre-treating cells with high concentrations of exogenous IFN-α , nor were CpG- or UpA-high mutant differentially inhibited ( Atkinson et al . , 2014 ) . Several other aspects of the attenuation phenomenon similarly argue against a role of standard RNA virus recognition pathways in mediating the attenuation of CpG- or UpA-high viruses . Firstly , inhibition of replication was observed immediately after infection of cells ( Figures 6–7 ) before significant induction of ISGs , and the restriction mechanism appeared to prevent the establishment of replication complexes rather than inhibit virus expression once formed ( Figures 7–8 ) . Furthermore , using several experimental formats ( Figure 10 ) , the absence of a trans-acting effect of high CpG- or UpA- co-expressed RNAs or co-transfected replicons on the replication of a compositionally normal reporter replicon demonstrates that these compositionally altered RNAs do not induce a cellular antiviral state that makes cells generally non-permissive for virus replication . Such cellular responses would be expected if inhibition was mediated through ISGs or stress-induced translational arrest . Remarkably , it additionally appears that CpG is not the necessary and sufficient requirement for virus attenuation since bases upstream and downstream appear critical for attenuation ( Figure 11 ) . The 3’ extension provided complete freedom to insert RNA sequences of any composition into the replicon , freed from coding constraint . Our initial experiments placing U or A residues either side of CpG while keeping total CpGs the same as WT and overall base composition constant provides an example of the utility of this construct in the future dissection of recognition motifs . In these initial experiments , simply placing A/U on either side of CpG produced levels of attenuation of replicons that approached that of the CpG-high R1 mutant , despite possessing identical numbers of CpG residues to WT sequence . This context dependence is consistent with previous observation of suppression of CpG in an AU context in genome sequences of influenza A virus ( Jimenez-Baranda et al . , 2011 ) , reflecting perhaps the avoidance of more potent recognition motifs than CpG alone . Finally , the observation that plant viruses suppress CpG ( and UpA ) dinucleotide frequencies as much or even more intensively than vertebrate viruses leads to the tantalising possibility that the attenuation mechanism might be fundamental to eukaryotic virus defence and evolutionarily conserved and constantly active over the many hundreds of millions of years of eukaryote evolution . Animal and plant cells however differ almost entirely in their use and mechanism of action of PRRs , the latter depending more on siRNA-mediated silencing of viruses and some elements of the stress response observed in vertebrate cells . Enterovirus 71 protein 3A was recently shown to be a viral suppressor of RNA interference ( VSR ) in vertebrate cells . Without a functional VSR Echovirus 71 replication was reduced through activation of the RNA interference pathway ( Qiu et al . , 2017 ) . However , if such pathways were indeed shared between animals and plants , we obtained little evidence for stress response pathways nor siRNA-induced viral silencing being involved as mediators of E7 attenuation ( Figure 9—figure supplement 1B ) . Concerning the latter pathway , there is nothing in the mode of action of siRNAs in either recognition or effector pathways that would seem capable of causing the dinucleotide composition related differences in replication efficacy . Extension of the experimental approach used in the current study to plant viruses and other virus/host interactions may contribute to the identification of what may ultimately represent a novel and undocumented mechanism of eukaryotic virus control . In summary , the recognition and restriction mechanisms that attenuate the replication of CpG- and UpA-high mutant appear to lie outside the conventional paradigm of virus control by innate cellular immune pathways . Although mechanistically unclear , the restriction mechanism exerts a powerful evolutionary constraint on vertebrate RNA viruses to judge from the widespread suppression of CpG and UpA frequencies in viruses with major differences in replication strategies . Design , construction and recovery of E7 viruses with various nucleotide compositions in R2 were described previously ( Atkinson et al . , 2014 ) . Viral titres were verified by end point dilution assay ( EPDA ) on RD cells and RNA copy numbers were determined by quantitative RT-PCR using primer pair E7 5’UTR ( Supplementary file 1A ) , with a PCR amplicon as standard curve . Cells from a variety of tissues: RD ( ATCC: CCL-136 , RRID:CVCL_1649 , Homo sapiens , muscle rhabdomyosarcoma ) , A549 ( ATCC: CRM-CCL-185 , RRID:CVCL_0023 , Homo sapiens , lung epithelial ) , HEK-293 and HEK-293T ( ATCC: CRL-1576 , RRID:CVCL_6342 and ATCC: CRL-3216 , RRID:CVCL_0063 , Homo sapiens , embryonic Kidney ) , Nb324K ( RRID:CVCL_U409 , Homo sapiens , kidney ) , Vero E6 ( ATCC: CRL-1586 , RRID:CVCL_0574 , Cercopithecus aethiops , Kidney ) , BHK-21 ( ATCC: CCL-10 , RRID:CVCL_1915 , Mesocricetus auratus , kidney ) , MEF ( Mouse embryonic fibroblasts and knock outs thereof provided by Prof . Jan Rehwinkel and generated as described [Glück et al . , 2017] ) , NIH/3T3 ( ATCC: CRL-1658 , RRID:CVCL_0594 , Mus musculus , embryonic fibroblasts ) , A9 ( ATCC: CRL-1811 , RRID:CVCL_9094 , Mus musculus , B lymphocyte ) , differentiated Neuro-2a ( dN2a , ATCC CCL-131 , RRID:CVCL_0470 , Mus musculus , neuroblast ) and BV2 ( RRID:CVCL_0182 , Mus musculus , Microglia brain cells ) ] were cultured in Dulbecco modified Eagle medium ( DMEM ) with 10% foetal calf serum ( FCS ) , penicillin ( 100 U/ml ) and streptomycin ( 100 μg/ml ) and maintained at 37°C with 5% CO2 . The cell lines used in this study are not listed in the ICLA Database of Cross-Contaminated or Misidentified Cell Lines . All cell lines were derived from accredited sources in the Roslin Institute , University of Edinburgh . Initial cultures of each cell line was aliquoted and frozen after minimum passaging and cells used from experiments described in the study were derived from these . All cell lines are screened on a regular 6 month schedule for mycoplasma contamination with the PCR-based protocol as described in ( Young et al . , 2010 ) . No contamination was detected in any of the cell lines used over the period the study described in the manuscript . The plasmid pRiboE7luc contains the E7 genome in which the structural genes of E7 have been replaced by a firefly luciferase gene . The original firefly luciferase gene ( Observed/Expected ratio ( O/E ) CpG 1 . 242 and UpA 0 . 699 ) was replaced by a synthetic version of the luciferase gene with its CpG and UpA dinucleotides removed ( O/E ratio CpG 0 . 013 and UpA 0 . 145 ) while maintaining its coding sequence , referred to as E7 . To introduce dinucleotide variations in the non-coding region of E7 amplicons were created by amplification of pRiboE7luc with PCR_7146 s and PCR_7358as and PCR_7315 s and PCR_749 as primers . Both amplicons were fused together with a second PCR reaction using PCR_7146 s and PCR_749as primers ( Supplementary file 1A ) ( Phusion high-fidelity DNA polymerase , New England Biolabs , M0530S ) . The linker sequence was inserted with the existing PmlI and NotI restriction sites . The result was an E7 replicon with unchanged coding sequence , but SalI , SbfI and HpaI restriction sites immediately after the stop codon ( nt 7325 ) and before the original E7 3'UTR . Previously , two regions ( R1 , nts 1878–3119 and R2 , nts 5403–6462 ) of the full length E7 cDNA clone pT7:E7 , were synonymously altered creating viruses with variations in the nucleotide composition of their coding region ( Table 1 ) . In short; both R1 and R2 sequences included a wild type ( WT ) , permutated control ( P ) , CpG and UpA low ( cu ) , UpA high ( U ) and CpG high ( C ) variant . Using the restriction sites SalI and HpaI for R1 and EcoRI and BglII for R2 the various R1 sequences were introduced into the 3’ non-coding region of E7 , creating E7 with 3’-ncR1 variants and the R2 sequences into the original coding sequence of E7 with 3’-ncR1_WT . The R1 sequence was further altered such that it contained an equal amount of CpG ( 51 ) and UpA ( 62 ) dinucleotides to WT R1 , but that the context of all CpG dinucleotides was altered , surrounding CpG dinucleotides either with two adenines or two thymines on either side of the dinucleotide , resulting in 51 AACGAA or UUCGUU motifs ( Table 1 ) . These sequences were then cloned into the 3’ noncoding region of E7 via the described SalI and HpaI restriction sites , creating E7 with ncR1_AACGAA or ncR1_UUCGUU . Replication defective mutants of the E7 replicons with ncR1 of WT and CpG high composition were created by site-directed mutagenesis of a highly conserved motif within the viral polymerase GDD into GND . The mutation was introduced using the QuikChange II XL Site-Directed Mutagenesis kit from Agilent with E7-GND F and E7-GND R primers ( Supplementary file 1A ) . The 800 nucleotides 3201–4000 from the E7 genome were selected for being of average E7 nucleotide composition . The nucleotide order was non-synonymously scrambled with the NOR function in SSE . Subsequently the CpG and UpA dinucleotide frequencies were restored to wild type ratios ( O/E 0 . 546 and 0 . 649 , respectively ) , creating a normalised synthetic sequence ( S_Norm ) . Variants of S_Norm were created by removing all CpG and UpA dinucleotides ( S_cu ) , or increasing UpA ( S_U ) or CpG ( S_C ) dinucleotides ( Table 1 ) . Sequences were synthesised ( Geneart ) with 5’ HpaI - AscI and 3’ MluI –SalI restriction sites and ligated into the above described pRiboE7luc vector via the introduced SalI and HpaI restriction sites , creating E7 with 3’-non-coding ( ncS ) variants . Next the newly introduced regions were digested with AscI and SalI and ligated into the MluI and SalI sites of the E7 with 3’ ncS variants to generate single ( 800 nt ) and double ( 1600 nt ) linear repeats of the ncS variants in the noncoding region of the E7 replicon . The E7 EGFP replicon was constructed by replacing the firefly luciferase from the luciferase expressing E7 replicon with 3’-ncR1 of WT composition for EGFP using KasI and KflI restriction sites . Subsequently the ncR1 of WT composition was replaced by ncR1 of P , cu , U and C composition as described above . R2 sequences of various nucleotide compositions were amplified by PCR while adding 5’ -EcoRI and 3’-ApaI restriction sites and additional nucleotides that enabled transcription of the R2 sequences but prevented their translation ( R2_EcoRI Fwd and R2_ApaI Rev primers for each R2 mutant sequence , Supplementary file 1A ) . The same restriction sites were used to clone the sequences into a previously published pcDNA/DEST40 backbone downstream of the CMV promoter ( Fros et al . , 2012 ) . RD cells were seeded at 1 × 10∧5 cells per well in 24-well plates and subsequently infected with the wild-type ( WT ) E7 or E7 R2 mutants at a multiplicity of infection ( MOI ) of 0 . 01 or 1000 E7 RNA copies per cell . One hour post infection the inoculum was removed and cells were washed with phosphate buffered saline ( PBS ) before adding 500 μl cell culture medium . At the indicated times post infection the cell culture medium was aspirated and stored at −80°C . Where applicable , cells were lysed in 300 μl RLT lysis buffer and stored at −80°C before RNA isolation with the RNease kit ( Qiagen ) . Viral titres were determined in an end point dilution assay ( EPDA ) by determining the tissue culture infectious dose 50% ( TCID50 ) in RD cells . Total RNA was harvested according to the manufacturer’s protocol ( RNeasy , Qiagen ) . In a one-step reaction Quantifast Sybr green ( Qiagen ) total RNA was reverse transcribed with gene specific primers amplifying either E7 ( primer pair E7 5’-UTR ) or the internal control GAPDH ( Supplementary file 1A ) using the Stepone plus cycler ( Applied Biosystems ) . Custom Stellaris FISH Probes were designed against an unaltered WT portion of the E7 genomic RNA ( nt 3200–4200 ) by utilizing the Stellaris RNA FISH Probe Designer ( Biosearch Technologies , Inc . , Petaluma , CA ) available online at www . biosearchtech . com/stellarisdesigner . The resulting 32 E7 genomic RNA probes ( Supplementary file 1B ) were hybridized with CAL Fluor 590 red . RD cells were infected with 1000 RNA copies/cell of either WT , R2_U or R2_C viruses . At the end of infection cells were washed with PBS and fixed with 3 . 7% paraformaldehyde in PBS for 10 min . Cells were permealized by 70% ethanol for 2 hr at 4°C . Samples were stained for E7 RNA using the RNA FISH probe set , following the manufacturer's instructions , using the protocol for adherent cells or in case of co-staining with cellular proteins the sequential IF protocol , both available online at www . biosearchtech . com/stellarisprotocols . The primary G3BP1 antibody ( G6046; Sigma , RRID:AB_1840864 ) was diluted 1:500 in PBS containing 3% FCS . Cells were stained at room temperature for one hours , washed three times with PBS and stained with the secondary antibody Alexa Fluor 488 ( RRID:AB_2633280 , 1:2000 ) for one hour . Nuclei were stained with Hoechst 33342 . Samples were analysed using a Zippy API Deltavision core inverted microscope and Z-stacks were deconvolved with SoftWorx Deltavision software . Replicon plasmid DNA was linearized using NotI and isolated from agarose gel . Uncapped RNA transcripts were synthesized in vitro using T7 RNA polymerase ( MEGAscript T7 , Invitrogen , Carlsbad , CA ) for 3–6 hr . RNA integrity was confirmed on agarose gel and the concentration determined with Qubit Fluorometric Quantitation ( ThermoFisher Scientific ) . In a 96-well format , cells were transfected with 10 ng of RNA/well using 0 . 4 μl of lipofectamine 2000 ( ThermoFisher Scientific ) , according to the manufacturers protocol . At the indicated time post transfection cells were lysed in 60 μl passive lysis buffer ( Promega ) for 20 min . In a white F-bottom plate 50 μl cell lysate and 50 μl of firefly luciferase substrate ( Promega ) were mixed and measured in a GloMax 96-microplate luminometer ( Promega ) . RD cells either infected with E7 R2 variants or uninfected cells were washed once with ice cold PBS and lysed in laemmli buffer containing 2-mercaptoethanol . Lysate was heated to 100°C for five minutes and clarified by centrifugation at 13 , 000 rpm for one minute in an Eppendorf table top centrifuge . Protein samples were separated on a 4–12% SDS gel ( Biorad ) and transferred to an Immobilon membrane ( Millipore ) for analysis by Western blotting . Membranes were blocked in PBS with 0 . 05% Tween 60 ( PBST ) containing 3% skimmed milk in for 1 hr at room temperature . Membranes were washed three times for 5 min each with PBST and incubated for one hour at room temperature with anti-P-eIF2α ( diluted 1: 500; Abcam , RRID:AB_732117 ) or anti-HPRT ( diluted 1: 4000; Abcam , RRID:AB_297217 ) in PBST . Membranes were washed and treated with HRP-conjugated goat anti-rabbit IgG mAb , diluted 1: 3000 in PBST , for 45 min at room temperature . Membranes were washed three times with PBST . Proteins were detected by chemiluminescence using ECL prime Western blotting detection reagent ( GE Healthcare ) . RD cells were seeded at 1 . 5 × 10∧4 cells per well in 96-well plates and the next day cells were pre-treated for two hours with 2 . 5 μM acriflavine ( ACF ) or DMSO . Cells were transfected with 10 ng / well of E7luc replicon RNA containing ncR1 variants in the presence of ACF or DMSO respectively . Six hpt cells were lysed and luciferase measured . RD cells were seeded at 1 . 5 × 10∧4 cells per well in 96-well plates and subsequently infected with the wild-type ( WT ) or R2 mutants at a MOI of 0 . 01 . One hour post infection the inoculum was removed and cells were washed with PBS before adding 100 μl cell culture medium . At 24 hr post-infection caspase activity was measured using the Caspase Glo 3/7 Assay kit ( Promega ) according to manufacturer’s instructions . RD cells or BHK cells were seeded at 2 × 10∧5 cells per well in 12-well plates and left to adhere overnight . Cells were transfected with EGFP expressing E7 replicon RNA with various ncR1 and coding R2 mutants at 594 ng RNA/well and 4 . 75 µl lipofectamine 2000 ( ThermoFisher Scientific ) . Six hpt the cell culture media was removed , cells were washed in PBS , trypsinized and pelleted at 1 , 500 rpm for 5 min . Pellets were washed once in PBS and cells were fixed in 4% paraformaldehyde/PBS for 10 min . Cells were pelleted and resuspended in 100 µl PBS . EGFP expression was quantified on a MACSQuant Flow Cytometer . Data was analyzed using FlowJo software ( LCC ) . Interference assays were performed using three different methods . ( i ) 293 cells were seeded at 1 × 10∧5 cells per well in 24-well plates and left to adhere overnight . Cells were transfected with 250 ng of R2_pcDNA/DEST40 vectors/well using 2 µl lipofectamine 2000 ( ThermoFisher Scientific ) . 24 hpt cells were infected with E7 WT virus at MOI 0 . 01 for 48 hr . Virus-containing supernatants were titrated by EPDA on RD cells . ( ii ) RD cells were seeded at 1 . 5 × 10∧4 cells / well in 96-well plates and left to adhere overnight . Cells were co-transfected with 6 . 3 ng ncS variant RNA and 50 ng E7 luc replicon with ncR1_WT per well using 0 . 4 µl lipofectamine 2000 ( ThermoFisher Scientific ) . Six hpt luciferase activity was quantified as described above . ( iii ) RD cells were seeded at 1 . 5 × 10∧4 cells/well in 96-well plates and left to adhere overnight . Cells were co-transfected with 42 ng EGFP expressing E7 replicon with the indicated R2 and ncR1 variants and 50 ng of luciferase expressing E7 with ncR1 of WT composition using 0 . 4 µl lipofectamine 2000 ( ThermoFisher Scientific ) . Six hpt luciferase activity was quantified as described above . Biological replicates are defined as repeats of the same experiment . Each experimental replicate used cells from separate batches and virus dilutions , RNA transfections and possible additional treatments were separate suspensions . Significance for the described analyses was calculated using either the Microsoft Excel 2016 or GraphPad Prism five software packages .	Living things store their genetic material as molecules of DNA or a related chemical called RNA . Both DNA and RNA contain building blocks known as bases . There are several different types of bases and the specific order they appear in a DNA or RNA molecule encodes the genetic information . In RNA these bases are known as cytosine , guanine , adenine and uracil ( or C , G , A and U for short ) . The order that bases appear in DNA and RNA can be highly biased . For example , in RNAs from animals with backbones ( also known as vertebrates ) , cytosine followed by guanine and uracil followed by adenine occur less often than mathematics would predict . Viruses are particles that contain DNA or RNA surrounded by a coat made of proteins . They are unable to multiply by themselves and must therefore invade the cells of host organisms . Viruses that infect vertebrates mimic the base biases found in their host , a strategy that likely helps the virus’ genetic material to hide within host cells . Previous experiments have shown that viruses engineered to have more cytosines followed by guanines and uracils followed by adenines were easier to eliminate . However , it is not clear how this worked . Fros et al . investigated the ability of a virus called echovirus 7 to multiply inside the cells of humans and several other vertebrates . The experiments show that artificially increasing the number of cytosines followed by guanines and uracils followed by adenines in this virus reduced the ability of the virus to multiply immediately after the virus had entered the host cell . The location of the changes did not have any effect on how strongly the virus was inhibited . Furthermore , Fros et al . confirmed that these changes did not affect the ability of the virus’ genetic material to make the proteins it needs to multiply and make its coat . This suggests that the host specifically prevents the virus genetic material from being copied , solely based on the order of the bases in the viral genetic material . These findings provide evidence that human and other vertebrate cells contain factors that recognize and rapidly respond to foreign genetic material with biases in their genetic code that do not match their own . In the future , artificially increasing the frequency of specific orders of bases in viral genomes could be used to design more effective vaccines against diseases caused by viruses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2017	CpG and UpA dinucleotides in both coding and non-coding regions of echovirus 7 inhibit replication initiation post-entry
Indian Hedgehog ( IHH ) signaling , a key regulator of skeletal development , is highly activated in cartilage and bone tumors . Yet deletion of Ptch1 , encoding an inhibitor of IHH receptor Smoothened ( SMO ) , in chondrocyte or osteoblasts does not cause tumorigenesis . Here , we show that Ptch1 deletion in mice Prrx1+mesenchymal stem/stromal cells ( MSCs ) promotes MSC proliferation and osteogenic and chondrogenic differentiation but inhibits adipogenic differentiation . Moreover , Ptch1 deletion led to development of osteoarthritis-like phenotypes , exostoses , enchondroma , and osteosarcoma in Smo-Gli1/2-dependent manners . The cartilage and bone tumors are originated from Prrx1+ lineage cells and express low levels of osteoblast and chondrocyte markers , respectively . Mechanistically , Ptch1 deletion increases the expression of Wnt5a/6 and leads to enhanced β-Catenin activation . Inhibiting Wnt/β-Catenin pathway suppresses development of skeletal anomalies including enchondroma and osteosarcoma . These findings suggest that cartilage/bone tumors arise from their early progenitor cells and identify the Wnt/β-Catenin pathway as a pharmacological target for cartilage/bone neoplasms . The Hedgehog ( Hh ) signaling pathway controls embryonic pattern formation and organogenesis , adult stem cells homeostasis and tissue maintenance , and is involved in the etiology of various tumors ( Briscoe and Thérond , 2013 ) . Ligand ( Indian , Sonic , or Desert Hedgehog ) engagement to receptor Smoothened ( Smo ) relieves the inhibition of Patched 1 ( Ptch1 ) and upregulates Gli1/2 proteins , which increase the expression of proteins including Myc , Cyclin D , and Bcl2 and promote cell proliferation . Hedgehogs also activate the Rho/Rac pathway and increase the expression of Cyclin B in Smo-independent manners , which are regarded as the non-canonical pathway ( Briscoe and Thérond , 2013 ) . Human genetic studies have identified germline mutations in Ptch1 as the cause of Gorlin syndrome , which is characterized by basal cell carcinoma , medulloblastoma , cartilage tumors , and ectopic ossification during adolescence and early adulthood ( Hahn et al . , 1996 ) . Some of the patients also develop holoprosencephaly and autism ( Noor et al . , 2010 ) . Inhibitors for Smo or Gli1/2 are developed to treat the related tumors ( Amakye et al . , 2013 ) . IHH is mainly expressed in prehypertrophic chondrocytes and osteoblasts at puberty stages ( Kindblom et al . , 2002 ) . Genetic studies have shown that IHH signaling regulates proliferation and differentiation of osteoblasts and chondrocytes during skeletal development and repair ( Amano et al . , 2015; Lanske et al . , 1996; Maeda et al . , 2007; Ohba et al . , 2008; St-Jacques et al . , 1999 ) . IHH regulates chondrocyte proliferation and differentiation mainly via PTHrP ( Lanske et al . , 1996; Williams et al . , 2018 ) , while IHH regulates osteoblast differentiation by controlling Runx2 expression via the canonical and non-canonical pathways ( Shi et al . , 2015; Yuan et al . , 2016 ) . Interestingly , Wnt/β-Catenin signaling , a crucial regulator of skeletal development and remodeling , has been shown to mediate the effects of IHH signaling on osteoblast differentiation ( Canalis , 2013; Hill et al . , 2006; Hu et al . , 2005; Yoshida et al . , 2004 ) , but act upstream of and parallel to IHH signaling in chondrocyte survival and hypertrophy , respectively ( Mak et al . , 2006 ) . In addition , Hh signaling in mature osteoblasts upregulates RANKL expression and enhances osteoclastogenesis and bone resorption ( Mak et al . , 2008 ) . Thus , IHH signaling plays critical roles in skeletal development and remodeling . Enchondromas and osteosarcomas are among the most common skeleton tumors and they are generally resistant to conventional chemo- and radio-therapies ( Alman , 2015; Amakye et al . , 2013; Kansara et al . , 2014; Nazeri et al . , 2018 ) . There is an urgent need to identify druggable targets for treatment of these disorders , yet this is hampered by incomplete understanding of pathogenesis of these tumors and a lack of animal models that resemble the human disorders . Cartilage/bone tumors often show activated Hh signaling , resulted either from mutations in EXT1/2 , PTH1R , or SMO or from elevated expression of hedgehog ligands or Gli proteins ( Amary et al . , 2011; Hopyan et al . , 2002; Pansuriya et al . , 2011; Tarpey et al . , 2013; Tiet et al . , 2006 ) . However , activation of Hh signaling , for example by deletion of Ptch1 alone , in chondrocytes or osteoblasts does not cause tumorigenesis ( Bruce et al . , 2010; Chan et al . , 2014 ) . Here , we use Prrx1-CreERT; Ptch1f/f mice to study the functions of Hh signaling in mesenchymal stem/stromal cells ( MSCs ) during adolescence and show that activation of Hh signaling promotes MSC proliferation and osteogenic and chondrogenic differentiation but suppresses MSC adipogenic differentiation and leads to development of osteoarthritis-like phenotypes , enchondroma , and osteosarcoma . Ptch1 deletion executes these functions via the canonical Hh pathway and activation of Wnt/β-Catenin pathway . This study thus sheds light on the origin of enchondroma and osteosarcoma and identifies Wnt/β-Catenin as a drug target for cartilage/bone tumor treatment . To investigate the functions of Hh signaling in postnatal bone growth , we ablated Ptch1 in Prrx1+ MSCs using inducible Prrx1-CreERT mice ( Kawanami et al . , 2009 ) , due to embryonic lethality of Prrx1-Cre; Ptch1f/f mice ( Bruce et al . , 2010 ) . Prrx1 has been shown to mark osteoblasts and chondrocytes during development and growth and bone marrow ( BM ) cells marked by Prrx1 has been shown to have features of MSCs ( Miwa and Era , 2018 ) . Four doses of tamoxifen ( TAM ) was intraperitoneally injected into P14 Prrx1-CreERT; Ptch1f/f mice , which were euthanized 2 months later , when the peak bone mass was obtained ( Figure 1A ) . We found that Ptch1 deletion led to a decrease in Patched 1 but an increase in Hh target gene Gli1 in BM-MSCs ( Figure 1B ) . Prrx1-CreERT; Ptch1f/f mice were significantly smaller and showed decreases in body weight and length compared with age- and gender-matched control mice ( Figure 1C–1E ) . Similarly , femur and tibia lengths were decreased by 23 . 3% and 18 . 4% , respectively , in Prrx1-CreERT; Ptch1f/f mice ( Figure 1F ) . X-ray and micro-CT imaging revealed that the mutant mice had deformed knee joints and rough bone surfaces , which are indicative of exostoses ( Figure 1G ) . Joint deformation and exostoses were also observed in the phalanges of Prrx1-CreERT; Ptch1f/f mice , with the paws being swelled ( Figure 1H ) . These results suggest that ablation of Ptch1 in Prrx1+ MSCs disrupted bone and cartilage growth and led to joint deformation and ectopic bone growth . Histological analyses revealed that Prrx1-CreERT; Ptch1f/f mice exhibited signs of osteoarthritis with high OARSI scores at 2 . 5 months of age ( 2 months after TAM injection ) ( Figure 2A and B ) , and increased fibrotic cells lining the damaged cartilage surfaces , which were positive for fibrosis marker FSP1 ( Figure 2C and D ) . The articular cartilage showed reduced Col2-expressing cells but increased Col10-expressing cells ( Figure 2E ) . We isolated the articular cartilage from the mutant and control mice , carried out quantitative PCR analysis , and detected decreases in Timp3 ( encoding inhibitor of metalloproteinases-3 ) , Sox9 , and Acan but increases in Mmp13 ( encoding matrix degradation enzyme ) and Adamts5 in the mutant mice ( Figure 2F ) . Compared to control littermates , the thickness of calcified cartilage ( CC ) layer was greater whereas the thickness of hyaline cartilage ( HC ) layer was lesser in the mutant mice ( Figure 2—figure supplement 1A and B ) , accompanied by modest decreases in subchondral bone volume and bone mineral density ( BMD ) ( Figure 2—figure supplement 1C and D ) . These results suggest that Prrx1-CreERT; Ptch1f/f mice developed phenotypes resembling early osteoarthritis , consistent with the finding that expression of Smo in Col2+ chondrocytes leads to development of osteoarthritis ( Lin et al . , 2009 ) . Histological analyses of bone sections also revealed multiple enchondroma-like lesions ( referred to as EC hereafter ) at the growth plate , articular cartilage , and bone marrow ( likely originated from the growth plate , see late results ) in the mutant mice ( Figure 2G ) . In addition , chondrocytes in the growth plate of mutant mice was improperly aligned , which also displayed an increase in Ki67 staining in the tumor region and the unaffected region ( Figure 2H and I ) . Examination of mice at earlier time points post TAM administration detected EC-like overgrowth at both articular surfaces and growth plates at day 7 , which became larger and invaded the trabecular areas over time ( Figure 2—figure supplement 2A ) . Expression of Col10 but not Col2 was increased at the growth plate ( Figure 2J and K ) . Similar cartilage lesions were also observed in phalanges , humeri , and tibia but not in the vertebrae of mutant mice ( Figure 2—figure supplement 2B–2F ) . Overall , Ptch1 deletion in Prrx1+ MSCs promoted chondrocyte proliferation and enchondroma formation . Prrx1-CreERT;Ptch1f/f mice also developed tumors at the periosteal surfaces that have features of osteosarcoma: expansive osteoid lesions with mushroom-shaped appearance that were only located at cortical bones of the limbs ( Figure 3A–3D ) , which later transgressed the cortex ( Figure 3—figure supplement 1A ) . Note that exostoses , which were smaller and numerous , were observed in long bones and phalanges of Prrx1-CreERT; Ptch1f/f mice ( Figure 3—figure supplement 1B ) . Osteosarcoma-like lesions ( referred to as OS hereafter ) also showed increased angiogenesis , with the blood vessels mainly located outside of the tumors ( Figure 3—figure supplement 1C ) . All ( n = 9 ) Prrx1-CreERT; Ptch1f/f mice developed OS and they lived less than 7 months after TAM injection . Two of the mutant mice showed tumors in the lung ( Figure 3—figure supplement 1D ) , indicative of metastasis . Quantitative PCR and western blot analyses confirmed that Hh signaling was activated in the OS tumors , manifested by increases in the expression of Gli1 protein and other IHH target genes compared to cultured periosteal cells ( Figure 3E and F ) . We also isolated primary cells from osteosarcomas and cultured them for further analysis . Immunostaining results showed that there were more Ki67-poisitve cells in osteosarcoma cell cultures than normal periosteal cell cultures ( Figure 3G ) . A significant increase in cell proliferation rate was also observed in osteosarcoma cells ( Figure 3H ) . Wound healing assays revealed that osteosarcoma cells showed a significant increase in cell migration rate compared with periosteal cells ( Figure 3I ) . However , micro-CT analysis revealed no significant change in bone mass and the number or thickness of trabecular bones in Prrx1-CreERT; Ptch1f/f mice at 2 . 5 months of age ( Figure 3—figure supplement 2 ) . This could be due to the tight coupling between osteoblastogenesis and osteoclastogenesis caused by Hh activation ( Mak et al . , 2008 ) . The above results are in contrast to a previous study showing that Col2α1-Cre-mediated Ptch1 deletion only led to delayed chondrocyte hypertrophy ( Mak et al . , 2006 ) . We also deleted Ptch1 using Gli1-CreERT mouse , which was reported to label osteoprogenitors ( underneath the growth plate ) and chondrocytes ( Shi et al . , 2017 ) . We found that the mutant line did not develop tumors at all ( Figure 3—figure supplement 3 ) , yet , they showed deceases in trabecular bones ( Figure 3—figure supplement 3B ) . The lack of tumor formation in Gli1-CreERT; Ptch1f/f mice suggests that activation of Hh signaling in MSCs but not in chondrocytes or osteoprogenitors promotes tumorigenesis . Note that deletion of one allele of Gli1 , like in Gli1-CreERT mice , does not affect Ptch1 deficiency-induced tumorigenesis ( Kimura et al . , 2005 ) . To test the contribution of canonical Hh signaling to Ptch1 deficiency-induced skeletal defects and tumorigenesis , we treated Prrx1-CreERT; Ptch1f/f mice with Smo inhibitor cyclopamine or Gli1/2 inhibitor GANT61 for 2 months after TAM administration . X-ray and histological examination revealed that cyclopamine or GANT61 alleviated joint deformation and enchondroma formation , and restored the structures of articular cartilage and growth plate ( Figure 4A–4C ) . Cyclopamine or GANT61 also diminished the development of osteosarcoma in Prrx1-CreERT; Ptch1f/f mice ( Figure 4D and E ) . Overall , these findings indicate that Hh signaling regulates skeletal growth and promotes tumorigenesis via the Smo-Gli1/2 pathway . The above results indicate that deletion of Ptch1 alone in Prrx1+ MSCs resulted in development of both enchondroma and osteosarcoma in the same mouse . To trace the origin of these tumors , we generated Prrx1-CreERT;Ai14 and Prrx1-CreERT; Ptch1f/f; Ai14 mice . Lineage tracing revealed that immediately after TAM injection , a few Ai14+ cells were detected in the articular cartilage , growth plate , periosteum , and large numbers of Ai14+ cells in the trabecular bones , whereas Prrx1-CreERT; Ai14 mice without TAM administration showed no Ai14+ cells ( Figure 5A and Figure 5—figure supplement 1A ) . Over time , Ai14+ cells were expanded and replenished articular cartilage , growth plate , BM , and periosteum ( Figure 5A ) , suggesting that Prrx1 may label stem/progenitor cells at these locations . Ptch1 deletion led to further expansion of Prrx1 lineage cells ( Figure 5A ) . Furthermore , Ai14+ enchondroma ( at articular cartilage and growth plate ) and OS-like lesions ( at periosteal bone surfaces ) were detected in Prrx1-CreERT; Ptch1f/f; Ai14 mice at days 14 and 30 post TAM administration , respectively ( Figure 5A ) , suggesting that Ptch1 deletion quickly leads to overproliferation of MSCs or progenitors . More enchondromas and osteosarcomas were formed 2 months post TAM injection ( Figure 5B ) . The growth plate did not show much differences in the thickness of different zones ( Figure 5—figure supplement 1B ) . These results suggest that the tumors are originated from Prrx1 lineage cells located at articular cartilage , growth plate , and periosteal bones but not at trabecular bones . On the other hand , Prrx1 marked limited numbers of osteoblasts and no chondrocytes in the vertebrae ( Figure 5—figure supplement 1C ) , consistent with our observation that Prrx1-CreERT; Ptch1f/f mice did not develop tumors in vertebrae ( Figure 2—figure supplement 2E ) . Previous studies have shown that Prrx1-marked BM cells have osteoblast , chondrocyte , and adipocyte differentiation potentials ( Miwa and Era , 2018 ) . We found Prrx1-marked BM-MSCs accounted for more than 50% of the adherent MSCs ( Figure 5—figure supplement 2A ) . Moreover , Prrx1+ cells isolated from periosteal surfaces of Prrx1-CreERT;Ai14 mice ( right after four daily doses of TAM ) could form colony-forming units and had osteoblast , chondrocyte , and adipocyte differentiation potentials ( Figure 5—figure supplement 2B and C ) . Prrx1+ cells could also differentiate into chondrocytes and osteoblasts during bone fracture repair in vivo ( data not shown ) . Overall , these results suggest that Prrx1+ cells in bone marrow and periosteal bones are multipotent . Note that our tracing data indicate that enchondromas observed in BM cavities are originated from multipotent cells located in the growth plate but not BM MSCs . Immunostaining revealed that Ai14+ enchondroma expressed low levels of Col2 and Col10 , but not Col1 ( Figure 5C ) , whereas Ai14+ osteosarcoma cells , which were located at periosteal surfaces , expressed low levels of Col1 but not Col2 or Col10 ( Figure 5C and Figure 5—figure supplement 1D ) , suggesting that these tumor cells were in a low-degree of differentiation state and that enchondroma and osteosarcoma arise from early chondrocyte and osteoblast progenitors , respectively . It is predicted that tumors derived from multipotent MSCs would contain both osteoblasts and chondrocytes , yet no such tumor was detected in Prrx1-CreERT; Ptch1f/f mice . We found that BM-MSCs from 2 . 5 month-old Prrx1-CreERT; Ptch1f/f mice showed an increase in colony forming units ( Figure 6A ) , indicating that Ptch1 ablation led to an expansion of the BM-MSC pool in vivo and/or increased cell proliferation . BM-MSCs isolated from Prrx1-CreERT; Ptch1f/f; Ai14 mice indeed showed increased proliferation rates , manifested by an increase in Ki67-positive cells ( Figure 6B ) , enhanced osteogenic and chondrogenic differentiation , manifested by increased histological staining ( Figure 6C ) . We found that overexpression of Ptch1 in BM-MSCs suppressed osteoblast and chondrocyte differentiation ( Figure 6—figure supplement 1A–1C ) , demonstrating the negative roles for Ptch1 in osteoblast and chondrocyte differentiation . Ptch1-deficient BM-MSCs also showed suppressed adipogenic differentiation , manifested by reduced levels of Oil Red O staining ( Figure 6C ) , which is supported by the observation that Prrx1-CreERT; Ptch1f/f mice showed a decrease in Perilipin+ adipocytes in the bone marrow ( Figure 6—figure supplement 2 ) . Analyses of expression of osteoblast , chondrocyte , and adipocyte markers confirmed these histological staining results ( Figure 6D ) . We validated the effects of Ptch1 ablation on MSC differentiation by transplantation assays . Carrier particles of hydroxyapatite tricalcium phosphate ( HA/TCP ) were mixed with MSCs and implanted subcutaneously under the dorsal skin of nude mice . After 8 weeks , the implants were harvested and analyzed . It was found that mutant MSCs formed more bone and fewer adipocytes in vivo than control MSCs ( Figure 6E ) . We also used a cell pellet culture model to confirm the effects of Ptch1 ablation on chondrogenic differentiation . MSCs isolated from mutant and control mice were pelleted by centrifugation and maintained in chondrogenic culture medium for 3 weeks . We found that the mutant MSCs showed increases in chondrocyte pellet size , which may be attributable to increased hypertrophic growth and increased proliferation , which were manifested by H/E , toluidine blue , and Ki67 staining ( Figure 6F and G ) . Taken together , these results suggest that Hh signaling plays critical roles in MSC proliferation and differentiation . We next searched for possible downstream mediators of Ptch1 deficiency-induced tumorigenesis and other skeletal phenotypes . We compared activation of important signaling molecules that regulate chondrocyte and osteoblast proliferation and differentiation , in BM MSCs isolated from Prrx1-CreERT; Ptch1f/f and control mice by western blot . We found that Ptch1 deficiency increased the activation of β-Catenin and Akt1 but suppressed the activation of Smad1/5/8 without affecting activation of Smad2/3 or Erk1/2 ( Figure 7A ) . Immunostaining of bone sections also confirmed the changes in activation of these pathways ( Figure 7B ) . Ptch1 deficiency-induced increase in β-Catenin occurred in both the nucleus and cytoplasm ( Figure 7-figure supplement 1A ) . Furthermore , Hh signaling can activate β-Catenin in wildtype MSCs ( Figure 7—figure supplement 1B and C ) . Previous studies have reported that Hh signaling interacts with BMP and WNT pathways to regulate cell fate determination ( Zhao et al . , 2006 ) . However , suppressed Smad1/5/8 activation could not explain increased MSC differentiation into osteoblasts and chondrocytes . Since the Wnt/β-Catenin pathway promotes proliferation and differentiation of both osteoblasts and chondrocytes , we determined the expression of 19 Wnt molecules in Ptch1-/-BM MSCs . Quantitative PCR analysis showed that Wnt5a and Wnt6 were expressed at significantly higher levels in Ptch1-/-BM MSCs than control cells ( Figure 7C ) . To test whether Wnt5a and Wnt6 are target genes of Hh signaling , we performed chromatin-immunoprecipitation ( ChIP ) assays for Hh downstream transcription factor Gli1 . We found that both Wnt5a and Wnt6 contained two binding sites for Gli1 and moreover , binding of Gli1 to these sites were increased in Ptch1 deficient cells ( Figure 7D–7F ) . These results suggest that Hh signaling may activate transcription of Wnt5a and Wnt6 in MSCs via Gli-1 . The above mouse studies established a link between Hh signaling and Wnt/β–Catenin signaling in skeletal growth and enchondroma/osteosarcoma development . Wnt/β–Catenin activation is a common event in various human tumors , especially in colorectal cancer ( Anastas and Moon , 2013 ) . We then tested whether Wnt/β–Catenin pathway was activated in human cartilage/bone tumors . Immunohistochemical staining of 24 human cartilage/bone tumors revealed that β–Catenin levels were increased in most of the tumor samples compared with normal tissues ( Figure 7G and H and Figure 7—figure supplement 2 ) , which was correlated with increased levels of Gli1 ( Figure 7H ) . These results suggest that the link between Hh and Wnt/β–Catenin pathways also exists in human tumor samples . We then tested whether the Wnt/β-Catenin pathway mediated the effects of Ptch1 deficiency on MSC proliferation and chondrogenic/osteogenic differentiation . To this end , we used IWP2 to inhibit Wnt/β-Catenin activation in MSC cultures . IWP2 is a small molecule compound that specifically inhibits Wnt signaling and has been used in many studies ( Chen et al . , 2009; Jeong et al . , 2017 ) . We found that accelerated MSC proliferation caused by Ptch1 ablation was markedly suppressed by IWP2 but not by BMP2 ( Figure 8A ) . In addition , IWP2 blunted accelerated osteogenic and chondrogenic differentiation of Ptch1-/-MSCs while BMP2 enhanced the differentiation ( Figure 8B ) . In addition , FH535 , another Wnt/β-Catenin inhibitor , also suppressed accelerated osteogenic and chondrogenic differentiation of Ptch1-/-MSCs ( Figure 8—figure supplement 1 ) . Overall , these results indicate that Wnt/β-Catenin mediates accelerated MSC proliferation and differentiation caused by Ptch1 deletion . Encouraged by the in vitro results on the effects of IWP2 on MSC proliferation and differentiation , we tested whether IWP2 could rescue the skeletal phonotypes observed in Prrx1-CreERT; Ptch1f/f mice . We treated Prrx1-CreERT; Ptch1f/f mice with IWP2 for 2 months right after TAM administration . IWP2 caused a decrease in the β–Catenin levels on bone sections without affecting the levels of Gli1 ( Figure 8C ) . Importantly , IWP2 treatment rescued joint deformation and growth plate defects and diminished enchondroma and osteosarcoma formation of Prrx1-CreERT; Ptch1f/f mice ( Figure 8D–8H and Figure 8—figure supplement 2 ) . Overall , these results suggest that enhanced Wnt/β–Catenin signaling is responsible for cartilage/bone growth defects and tumor development caused by Ptch1 deficiency in MSCs . This study for the first time shows that deletion of Ptch1 alone in Prrx1+ MSCs results in development of enchondromas and osteosarcomas in the same mice . This is in stark contrast to deletion of Ptch1 in osteoblasts or chondrocytes and suggests that stemness or differentiation status of the cell plays a critical role in cartilage/bone tumorigenesis . We further show that Ptch1-/-enchondromas and osteosarcomas are derived from Prrx1+ lineage located at cartilage and periosteal bone , respectively and they express limited levels of markers for differentiated chondrocytes and osteoblasts , respectively . This is in line with the roles of Hh signaling in maintaining cancer stem cells . Although Prrx1+ MSCs located at different places including the periosteal bones have tri-lineage differentiation potentials , no tumor contains both chondrocytes and osteoblasts in Prrx1-CreERT; Ptch1f/f mice . These results , together with previous studies showing that multipotent stem cells must commit to unipotent progenitors in order for Ptch1 deficiency to promote growth of medulloblastoma ( Schüller et al . , 2008; Yang et al . , 2008b ) , suggest that enchondroma and osteosarcoma are originated from the early chondrocyte and osteoblast progenitors derived from the Prrx1+ MSCs . The underlying reason may be that progenitor cells have much greater proliferation activity than differentiated cells and multipotent stem cells . The Prrx1-CreERT; Ptch1f/f mouse line is one of the few models for cartilage/bone tumors . Other models include mice with chondrocyte-specific Gli-2 overexpression and mice with chondrocyte-specific Ext1 deletion ( Hirata et al . , 2015; Hopyan et al . , 2002 ) . While Gli proteins can be activated by Erks , Akt1 , and other pro-proliferating signals , Ext1/2 control synthesis of heparin sulphate and mutations of Ext genes may cause IHH diffusion and other effects ( Jones et al . , 2010 ) . In addition , deletion of Ptch1 in osteoblasts using human osteocalcin-Cre in p53+/-but not p53+/+ background could induce osteosarcoma formation ( Chan et al . , 2014 ) . Here , we show that Ptch1 deletion in Gli1-marked chondrocytes and osteoprogenitors does not cause tumorigenesis . Thus , Prrx1-CreERT; Ptch1f/f mouse represents a unique model useful for dissecting the initiation and progression of both enchondroma and osteosarcoma and for testing drug candidates to target these two tumors . In addition , we show that Hh signaling promotes MSC proliferation and MSC osteogenic and chondrogenic differentiation but inhibits MSC adipogenic differentiation . While MSC-specific Ptch1 deletion reproduces some of the phenotypes observed in chondrocyte- and/or osteoblast-specific Ptch1 knockout mice , for example decreased body weight , body length and/or joint deformation , differences are also evident . Prrx1-CreERT; Ptch1f/f mice show unaltered bone mass , whereas Ptch1 deletion in mature osteoblasts leads to a loss of bone mass due to increased bone resorption ( Mak et al . , 2008 ) . While deletion of Ptch1 in Col2 chondrocytes results in a delay in hypertrophic growth ( Mak et al . , 2006 ) , we show that deletion of Ptch1 in MSCs results in enhanced hypertrophic growth . In addition , Ptch1 deletion in MSCs causes much severe joint deformation and exostoses and only Ptch1 deletion in MSCs induces cartilage/bone tumor formation . These findings suggest that Hh activation-induced overproliferation and tumorigenesis occur in MSCs but not in differentiated daughter cells . We have previously reported that ectopic expression of HB-EGF or deletion of Tsc1 ( an mTOR inhibitor ) in MSCs produces much severe cartilage and bone defects than similar genetic manipulation in chondrocytes and/or osteoblasts ( Li et al . , 2019 ) . Taken together , these results indicate that in the process of osteoblast and chondrocyte production , important regulation may occur at stages of stem cell expansion and commitment . It is known that Hh signaling promotes cell proliferation , inhibits apoptosis , and promotes tumorigenesis via the canonical and non-canonical pathways . Our current study clearly shows that the canonical pathway plays a critical role in Ptch1 deficiency-induced enchondroma and osteosarcoma formation , as inhibitors for Smo or Gli1/2 , which are developed as candidate drugs to treat cartilage and bone tumors ( Amakye et al . , 2013 ) , suppressed tumorigenesis . In addition , we find that Wnt5a/6 are up-regulated in-Ptch1 deficient MSCs and that β-Catenin is highly activated in mouse enchondroma and osteosarcoma samples . Previous studies have shown that canonical Wnt/β-Catenin signaling is essential for skeletal lineage differentiation ( Hill et al . , 2005 ) , and Wnt/β-Catenin signaling is also activated in most of the human cartilage/bone tumor samples , which is correlated with activation of Hh signaling . Functionally , inhibition of Wnt/β-Catenin signaling impedes development of cartilage/bone tumors and other skeletal growth defects . These findings underscore the roles played by the Wnt/β-Catenin pathway in Hh signaling-induced cell proliferation and tumorigenesis and suggest that inhibitors for Wnt/β-Catenin pathway may be useful for treating cartilage/bone tumors , especially the ones that are resistant to Hh signaling inhibitors . Our findings thus uncover a functional link between Hh signaling and Wnt/β-Catenin signaling in MSC proliferation/differentiation , skeletal growth , and tumor formation . However , previous studies have revealed that in Col2+ chondrocytes , Wnt/β-Catenin may act upstream of or parallel to IHH pathway in controlling cell survival and joint development ( Mak et al . , 2006 ) . Although Hh pathway-driven development of basal cell carcinoma can be mediated by increased Wnt/β-Catenin signaling ( Yang et al . , 2008a ) , this is not a common theme as only a limited number of samples display this link ( Adolphe et al . , 2006 ) . Overall , these findings suggest that the functional link between Hh signaling and Wnt/β-Catenin may be specific to MSCs and progenitors of osteoblasts and chondrocytes . This is consistent with the pro-proliferation activity of Wnt/β-Catenin signaling in osteoblastogenesis and chondrogenesis . In summary , our study has uncovered unanticipated functions of Hh signaling in MSCs , many of which are not observed in the daughter cells of MSCs . Hh activation promotes MSC proliferation and osteoblastic and chondrogenic differentiation but inhibits MSC adipogenic differentiation and leads to development of osteoarthritis , exostoses , and cartilage/bone tumors . Mechanistically , Hh signaling executes its pleiotropic effects via increasing the expression of a couple of Wnt molecules and enhancing Wnt/β-Catenin activation . Moreover , our genetic evidence suggests that enchondroma and osteosarcoma are derived from early progenitors of chondrocytes and osteoblasts and that Wnt/β-Catenin can be targeted to treat cartilage/bone tumors . All mouse work was carried out following the recommendations from the National Research Council Guide for the Care and Use of Laboratory Animals , with the protocols approved by the Institutional Animal Care and Use Committee of Shanghai , China [SYXK ( SH ) 2011–0112] . ROSA-Ai14 ( stock #007914 ) , Gli1-CreERT ( stock #007913 ) , and Floxed Ptch1 ( stock #030494 ) mouse lines were purchased from The Jackson Laboratory . These mice were kept in the SPF facility of Shanghai Jiao Tong University . Ptch1f/f mice were crossed with Prrx1-CreERT mice to generate Prrx1-CreERT; Ptch1f/f mice . Prrx1-CreERT; Ptch1f/f mice were crossed with ROSA-Ai14 mice to generate Prrx1-CreERT; Ptch1f/f;Ai14 mice . Ptch1f/f mice were crossed with Gli1-CreERT mice to generate Gli1-CreERT; Ptch1f/f mice . In all mice , only one allele of CreERT is used . The whole-body and femur radiographs were taken using Cabinet X-Ray system ( LX-60 , Faxitron Bioptics ) with standardized settings ( 45Kv for 8 s ) . Quantitative analysis was performed in mouse femur on a SkyScan-1176 micro-CT Scanner ( Bruker micro-CT , Belgium ) , following the procedures provided by the manufacturer . Briefly , scanning was performed using 8 . 96 µm voxel size , 45KV , 500 µA and 0 . 6 degrees rotation step ( 180 degrees angular range ) through the whole-length of the femora and extended proximally for 1400 slices . We started morphometric analysis with the first slice , where the femoral condyles were fully merged and extended for 150 slices proximally . Using a contouring tool , we segmented the trabecular bone from the cortical shell manually on key slices , and morphed the contours automatically to segment the trabecular bone on all slices . The three-dimensional structure and morphometry were constructed and analyzed for BV/TV ( % ) , BMD ( mg HA/mm3 ) , Tb . N ( mm–1 ) , Tb . Th ( mm ) and Tb . Sp ( mm ) . Femur and tibiae bones were fixed in 4% PBS-buffered paraformaldehyde overnight at 4°C . The samples were then stored in 70% alcohol for further experiments . For paraffin sections , samples were decalcified in 15% EDTA for 2 weeks and then dehydrated in alcohol , cleared with xylene , and embedded in paraffin . Four-μm-thick sections were cut using microtome ( Lexica Microsystems Nussle GmbH ) . Hematoxylin and eosin ( H/E ) or Safranin O staining was carried out on bone sections . Stained slides were photographed under a light microscope ( Olympus Microsystems ) . Osteoarthritis severity was quantified by the Osteoarthritis Research Society International ( OARSI ) scoring system ( 0–6 for grade and 0–24 for score ) , which was assessed by a single observer who was blinded to the experimental groups . The average thickness of the articular cartilage of the femoral plateau was measured using Image-Pro Plus software . For BM MSC isolation , mice were euthanized and the femur and tibia were extracted and cleaned . The bone ends were cut off and the bone marrow was flushed out with α-MEM . Single cell suspension was filtered through a 70 μm mesh to remove the debris . BM MSCs were cultured in α-MEM containing 15% FBS , 100 μg/ml penicillin , and 100 μg/ml streptomycin , at 37°C for 5 days . The non-adherent cells were washed out and the BM MSCs were used for further experiments . For periosteal cell isolation , mice were euthanized and the femur and tibia were extracted and cleaned . The whole bone was digested by dispase and type II collagenase at 37°C for 3 hr without flushing bone marrow out . The digested periosteal tissues were filtered through 70 μm cell strainers and then centrifuged . The cell pellet was resuspended in PBS for FACS sorting . The whole cells or sorted Ai14+ cells were cultured for further studies . BM MSCs were plated at a density of 5 × 106 cells per well in 12-well plates . After 7 days of culture with medium replaced every 3 days , the cultures were fixed and stained with crystal violet staining solution in methanol for 20 min . The dishes were washed with water and allowed to dry . Colonies were counted macroscopically , and data were reported as colony numbers per well . Cell proliferation was assessed using the Cell Counting Kit-8 ( Sangon Biotech Co . , Ltd . , Shanghai , China ) . Brieﬂy , cells were seeded into duplicate wells of 96-well plates at a density of 1 × 103 cells/well . At days 1 , 2 , 3 , 4 , and 5 , 10 μl of cck-8 solution was added to each well . The samples were incubated for 4 hr at 37°C . The absorbance of each well was determined at 450 nm . Three independent experiments were performed . Cells were pretreated with mitomycin C ( 4 mg/ml , Sigma-Aldrich ) for 2 hr prior to analysis . Cell migration was monitored at 0 , 10 , 20 hr by introduction of a scratch in confluent cells . To induce BM MSC tri-lineage differentiation , sorted Prrx1+ lineage BM MSCs were seeded at 5 × 104/well in 12-well plates . The next day , the cells were switched into osteogenic medium with α-MEM medium containing 15% FBS , 10 mM β-glycerol phosphate , and 50 μg/ml ascorbic acid , for 7–10 days , with medium changed every 2 days . The cells were then fixed in 4% paraformaldehyde and stained for ALP using an Alkaline Phosphatase Kit ( Sigma-Aldrich ) or Alizarin red . For mineralization assay , the cells were cultured for 21 days , which were stained in 5% silver nitrate solution under ultraviolet light or 1% alizarin red S . The silver staining was terminated by adding sodium thiosulfate solution . For adipocyte differentiation , BM MSCs were plated at 1 × 105/well in a 12-well plate and cultured in α-MEM containing 15% FBS , 100 nM dexamethasone , and 5 μM insulin for 2 weeks . The cells were then fixed and stained with Oil red O solution . For chondrogenesis assays , BM MSCs were suspended at a concentration of 1 . 6 × 107 cells/ml . We generated micro-mass cultures by seeding 10 µl droplets of cell suspension at the center of 12-well plates . Cells were allowed to attach for 2 hr before adding α-MEM containing 15% FBS , 100 IU/ml penicillin , and 100 μg/ml streptomycin . After 1 day , the medium was replaced with chondrogenic medium in α-MEM containing 15% FBS , 100 nM dexamethasone , 10 ng/ml TGFβ1 , and 1 μM ascorbate-2-phosphate . Cultures were maintained for 21 days , with the medium changed every 3 days , and were lastly stained with Alcian Blue or Toluidine blue . A total of 2 × 106 isolated BM MSCs were collected and incubated with 40 mg hydroxyapatite/tricalcium phosphate carrier ( HA/TCP: 12 . 5:87 . 5 by weight ) ( Bioengineering Research Center of Sichuan University , China ) scaffolds for 6 hr at 37°C in humidifying incubator , and then implanted subcutaneously onto the back of 2-month-old BALB/C homozygous nude ( nu/nu ) mice ( four mice per group ) . Mice were euthanized 10 weeks later after transplantation . The implants were fixed in 4% paraformaldehyde and then decalcified for 10 days . The sections were stained with H/E . To quantify the bone-like tissues , 10 images of each sample were taken randomly to measure the area of new bone formation versus total area . BM MSCs were suspended in chondrogenic medium consisting of high-glucose DMEM supplemented with 10 ng/ml recombinant human transforming growth factor-β3 ( TGF-β3; R&D ) , 100 nM dexamethasone ( Sigma ) , 50 μg L-ascorbic acid/ml ( Sigma ) , 1 mM sodium pyruvate , 40 μg proline/ml and ITS + premix ( Sigma; final concentrations: 6 . 25 μg/ml bovine insulin , 6 . 25 μg/ml transferrin , 6 . 25 μg/ml selenous acid , 5 . 33 μg/ml linoleic acid and 1 . 25 mg/ml bovine serum albumin ) . Aliquots of 5 × 105 cells , suspended in 500 μl chondrogenic medium , were centrifuged at 300 g for 5 min in 15 ml polypropylene conical tubes . Pelleted cells were incubated at 37°C under 5% CO2 with loosened caps to permit gas exchange . Within 24 hr of incubation , the sedimented cells formed a spherical aggregate at the bottom of the tube . The medium was changed every 3 days and pellets were harvested after 6 weeks . Total RNA was extracted using Trizol regent ( Invitrogen ) , which was reverse transcribed using Transcriptor Universal cDNA Master ( Roche ) following the manufacturer’s instructions . Quantitative PCR was carried out using Fast Start Universal SYBR Green Master kit ( Roche ) on ABI Prism 7500 Sequence Detection System ( Applied Biosystems ) using primers listed in Supplementary file 1 Table S1 . The levels of different mRNA species were calculated with the delta-delta CT method and normalized to GAPDH . Significant difference was analyzed using two-tailed Student's t-test . Decalcified bone sections were deparaffinized in xylene , rehydrated in ethanol , and permeabilized with 0 . 1% Triton X-100 for 20 min at room temperature . Antigen retrieval was performed in a citrate buffer at 95°C for 20 min . For immunostaining , the bone slides or cells on cover glass were blocked with 10% goat serum for 60 min , incubated with primary antibodies overnight at 4°C , washed in PBS , incubated with secondary antibody for 1 hr at 37°C , and washed with PBS before mounted on ProLong Gold DAPI ( Life Technologies ) . The antibodies used in this study were: Col1α ( ab21286 , Abcam ) , Col2 ( ab34712 , Abcam ) , Col10 ( ab58632 , Abcam ) , CD31 ( 550274 , BD ) , Vimentin ( ab92547 , Abcam ) , Perilipin ( ab3526 , Abcam ) , Non-immune immunoglobulin G ( IgG ) ( of the same species as the primary antibodies ) used as negative control . The secondary antibodies were goat anti-rabbit Alexa Fluor 488 ( ThermoFisher Scientific ) . Slides were mounted with antifade mounting medium with DAPI ( ThermoFisher Scientific ) . Cell proliferation was determined by Ki67 immunofluorescence staining ( Abcam , ab15580 ) . Images were taken under Olympus DP72 microscope ( Olympus Microsystems ) . For immunohistochemical staining , endogenous peroxidase activity was quenched with 3% H2O2 in methanol for 20 mins followed by washing with PBS before primary antibody incubation . After incubation with secondary antibody , sections were developed with diaminobenzidine and counterstained with hematoxylin and then dehydrated and mounted in neutral resins . Primary antibodies were: Ki67 ( ab15580 , Abcam ) , p-Akt1 ( 4060S , CST ) , p-Smad1/5/8 ( 13820S , CST ) , FSP1 ( ab27957 , Abcam ) , Gli1 ( 43926 , SAB ) , β-Catenin ( ab32572 , Abcam ) , Non-immune immunoglobulin G ( IgG ) ( of the same species as the primary antibodies ) was used as negative controls . The human bone tumor tissue chips were provided by Alenabio ( Xi’an , China ) . Each tissue chip was included 25 samples of human bone tumor with 10 chondrosarcoma , 8 osteosarcoma and 6 Ewing’s sarcoma and one normal cartilage as control . The chips were stained with Gli1 ( ( 43926 , SAB ) or β-Catenin ( ( ab32572 , Abcam ) antibodies under standard IHC protocol . The stained slides were examined under Olympus DP72 microscope ( Olympus Microsystems ) , and images were acquired . Total proteins were extracted from cells or tissues with TNEN buffer containing phosphatase and proteinase inhibitors , quantitated by the Bradford method ( Bio-Rad assay ) , and subjected to SDS-PAGE gel electrophoresis , which were transferred onto nitrocellulose membranes . The proteins were detected with specific antibodies using standard western blot method . The following antibodies were used: β-Actin ( sc-47778 , Santa Cruz ) , p-Akt1 ( 9271S , CST ) , Akt1 ( 9272S , CST ) , p-Erk1/2 ( 4377 , CST ) , Erk1/2 ( 9102S , CST ) , p-Smad1/5/8 ( 9511L , CST ) , Smad1 ( 9743L , CST ) , Ptch1 ( 17520 , Proteintech ) , Gli1 ( 2553S , CST ) , p-Smad2 ( 3101S , CST ) , p-Smad3 ( 9520S , CST ) , Smad2/3 ( 3102 , CST ) , β-Catenin ( ab32572 , Abcam ) . Immunoreactivity was detected using a Western Chemiluminescent HRP Substrate Kit ( Millipore ) and imaged with FluorChem M system ( Protein Simple ) . For in vitro experiments , the IWP2 ( Selleck Chemicals , USA ) and FH535 ( Selleck Chemicals , USA ) were dissolved in dimethyl sulfoxide and applied at a final concentration of 10 μM and 20 μM in cell culture medium , respectively . Recombinant human sonic hedgehog ( R and D systems , USA ) was dissolved in PBS and applied at a final concentration of 5 μg/ml in cell culture medium . For in vivo experiments , Cyclopamine ( Selleck Chemicals , USA ) , GANT61 ( Selleck Chemicals , USA ) , and IWP2 were reconstituted in corn oil . Mice were treated with 20 mg/kg cyclopamine , 30 mg/kg GANT61 , or 10 mg/kg IWP2 through intragastric administration every other day after tamoxifen injection . Mice were treated for 2 months and then evaluated . The chromatin immunoprecipitation ( ChIP ) assay was carried out following the manufacturer’s protocol ( SimpleChIP Enzymatic Chromatin IP Kit , Agarose Beads , #9002 ) . Briefly , BMSCs were crosslinked with 1% formaldehyde and blocked with glycine , which were then washed and digested by micrococcal nuclease . The nuclear pellet was suspended in ChIP buffer and sheared by sonication . An aliquot of each sheared chromatin sample was set aside as input control . The remained chromatin was then incubated with anti-Gli1 antibody ( Santa Cruz ) . Rabbit immunoglobulin G ( IgG , #9002 , CST ) was used as a control . The immunoprecipitated chromatins were then eluted with ChIP elution buffer . The DNA fragments were then released by treatment with ribonuclease A and then with proteinase K at 65°C . The released DNA fragments were purified with columns and amplified by site-specific primers by Real-time Quantitative PCR assay . The data were analyzed by the following formula: percent ( % ) input recovery = ( 100/ ( input fold dilution/bound fold dilution ) ) ×2 ( input CT−bound CT ) . Pairs of primers designed to amplify the specific target sequences of the putative promoters were listed in Supplementary file 1 Table S2 . Cells were washed with ice-cold PBS and scraped from the wells with the plates on ice . A nuclear and cytoplasmic protein extraction kit ( Beyotime , china ) was applied to separate these two cellular components following the manufacturer's instructions . The immunoblotting procedure was performed as described before and the following antibodies were used: GAPDH ( T40004S , abmart ) , H3 ( ap50907 , abgent ) , and β-Catenin ( ab32572 , Abcam ) antibodies . To overexpress Ptch1 in BMSCs , PCR-amplified full-length human Ptch1 cDNA was tagged with Flag and subcloned into the pHBLV-CMV-Puro vector ( Hanbio Biotechnology , China ) . Lentiviral vector carrying GFP was constructed as negative control . Recombinant viruses were collected and purified and titer was determined . For lentivirus infections , BMSCs were plated at a density of 1 × 105 cells in 12 well-plate . LV-GFP , LV-hPtch1 , or LV-Cre viruses were added at a multiplicity of infection of 20 when the cells were 30–50% confluent . After 72 hr , the transfected cells were harvested and gene overexpression was verified by Western blotting . For transient Ptch1 deletion , primary Ptch1f/f MSCs were infected with Cre-expressing lentivirus . Numerical data and histograms were expressed as the mean ± SD . Comparisons between two groups were analyzed using two-tailed unpaired Student’s t test . p<0 . 05 was considered statistically significant . Analysis of mice was litter-based and at least three litters were analyzed for every parameter . All the experiments were repeated at least three times .	Bone and cartilage tumors are among the most common tumors in the skeleton , often affecting the limbs . Bone tumors , also called osteosarcomas , usually occur in growing children and teenagers , and they are often resistant to conventional chemo- and radio-therapies . Surgery is the only treatment option , but this can lead to long-lasting damage that impairs the quality of life of these patients . Thus , there is a need to find new drug targets for these diseases . Unfortunately , no good laboratory-based systems exist that mimic these human cancers , hindering research into these tumors . One way to create a laboratory-based model for cartilage tumors and osteosarcomas is to reproduce the signaling that is present in the human tumors in a mouse . A signaling pathway called Hedgehog signaling is overactive in human cartilage and bone tumors . The activity of this pathway can be increased by deleting a gene called Ptch1; but mice do not form tumors when this gene is deleted in their mature cartilage and bone cells . Now , Deng , Li et al . report that deleting Ptch1 in mesenchymal stem cells , early-stage cells that can give rise to cartilage and bone cells , generates a mouse model for osteosarcoma and cartilage tumors . The mice with these Ptch1 deficient cells developed tumors with overactive Hedgehog signaling in cartilage and bone . Deng , Li et al . also performed biochemical experiments to show that Hedgehog signaling turned on another signaling pathway called Wnt signaling . Treating the mice that had mesenchymal cells lacking Ptch1 with a drug that inhibits Wnt signaling reduced the growth of cartilage and bone tumors . These data suggest that deleting Ptch1 in mouse mesenchymal stem cells can mimic human cartilage tumors and osteosarcomas . More experiments will be needed to explain how the Hedgehog and Wnt signaling pathways interact in these tumors . Finally , further studies will need to investigate if inhibiting Wnt signaling might become a useful therapy for human patients with osteosarcoma in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2019	Activation of hedgehog signaling in mesenchymal stem cells induces cartilage and bone tumor formation via Wnt/β-Catenin

Subsets and Splits

No community queries yet

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